docs(groot): add note on inference.queue_threshold value for stable inference

Public-facing Hub references (--dataset.repo_id, --policy.repo_id) shown as
concrete IDs; local-only values ($OUTPUT_DIR, $JOB_NAME) stay as placeholders.

README.md

@@ -105,7 +105,7 @@ lerobot-train \
 | -------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | **Imitation Learning**     | [ACT](./docs/source/policy_act_README.md), [Diffusion](./docs/source/policy_diffusion_README.md), [VQ-BeT](./docs/source/policy_vqbet_README.md), [Multitask DiT Policy](./docs/source/policy_multi_task_dit_README.md)                                                                                                                                    |
 | **Reinforcement Learning** | [HIL-SERL](./docs/source/hilserl.mdx), [TDMPC](./docs/source/policy_tdmpc_README.md) & QC-FQL (coming soon)                                                                                                                                                                                                                                                |
-| **VLAs Models**            | [Pi0](./docs/source/pi0.mdx), [Pi0Fast](./docs/source/pi0fast.mdx), [Pi0.5](./docs/source/pi05.mdx), [GR00T N1.5](./docs/source/policy_groot_README.md), [SmolVLA](./docs/source/policy_smolvla_README.md), [XVLA](./docs/source/xvla.mdx), [EO-1](./docs/source/eo1.mdx), [MolmoAct2](./docs/source/molmoact2.mdx), [WALL-OSS](./docs/source/walloss.mdx) |
+| **VLAs Models**            | [Pi0](./docs/source/pi0.mdx), [Pi0Fast](./docs/source/pi0fast.mdx), [Pi0.5](./docs/source/pi05.mdx), [GR00T N1.7](./docs/source/policy_groot_README.md), [SmolVLA](./docs/source/policy_smolvla_README.md), [XVLA](./docs/source/xvla.mdx), [EO-1](./docs/source/eo1.mdx), [MolmoAct2](./docs/source/molmoact2.mdx), [WALL-OSS](./docs/source/walloss.mdx) |
 | **World Models**           | [VLA-JEPA](./docs/source/vla_jepa.mdx) (more coming soon)                                                                                                                                                                                                                                                                                                  |
 | **Reward Models**          | [SARM](./docs/source/sarm.mdx), [TOPReward](./docs/source/topreward.mdx), [Robometer](./docs/source/robometer.mdx)                                                                                                                                                                                                                                         |
 

docs/source/_toctree.yml

@@ -70,7 +70,7 @@
   - local: eo1
     title: EO-1
   - local: groot
-    title: NVIDIA GR00T N1.5
+    title: NVIDIA GR00T
   - local: xvla
     title: X-VLA
   - local: multi_task_dit

docs/source/cameras.mdx

@@ -157,6 +157,14 @@ finally:
 </hfoption>
 </hfoptions>
 
+### Working with depth
+
+The Intel RealSense and Reachy 2 cameras can capture both color and depth in lockstep. Calling `read()` returns the **color** frame as `(H, W, 3)` `uint8`. Calling `read_depth()` returns the **depth map** as `(H, W, 1)` `uint16`, where each pixel value is the distance from the sensor expressed in **millimetres**. A pixel value of `0` typically means "no measurement available" (out-of-range, occluded, or low-confidence).
+
+During recording, the control loop peeks the freshest buffered frames non-blockingly via `read_latest()` (color) and `read_latest_depth()` (depth), adding the depth map as a sibling feature (e.g. `front_depth` next to `front`).
+
+For how depth streams are stored and encoded when recording a dataset, see the [Depth streams](./video_encoding_parameters#depth-streams) section of the video encoding guide.
+
 ## Use your phone's camera
 
 <hfoptions id="use phone">

docs/source/cheat-sheet.mdx

@@ -89,6 +89,36 @@ Control the data recording flow using keyboard shortcuts:
 - Press **Left Arrow (`←`)**: Delete current episode and retry.
 - Press **Escape (`ESC`)**: Stop, encode videos, and upload.
 
+### Recording depth
+
+Intel RealSense cameras (`type: intelrealsense`) record a depth stream when you set `use_depth: true`. Depth is quantized to 12-bit codes and stored as its own video.
+
+```bash
+lerobot-record \
+    ... \
+    --robot.cameras="{ head: {type: intelrealsense, serial_number_or_name: \"0123456789\", width: 640, height: 480, fps: 30, use_depth: true} }" \
+    --dataset.repo_id=${HF_USER}/so101_depth_test \
+    --dataset.single_task="put the red brick in a bowl" \
+    --dataset.depth_encoder.depth_min=0.01 \
+    --dataset.depth_encoder.depth_max=10.0 \
+    --dataset.depth_encoder.shift=0.0 \
+    --dataset.depth_encoder.use_log=true
+```
+
+### Video encoding parameters
+
+RGB and depth streams are encoded independently via the `--dataset.rgb_encoder.*` and `--dataset.depth_encoder.*` keys.
+
+```bash
+lerobot-record \
+    ... \
+    --dataset.rgb_encoder.vcodec=h264 \
+    --dataset.rgb_encoder.pix_fmt=yuv420p \
+    --dataset.rgb_encoder.crf=23 \
+    --dataset.depth_encoder.vcodec=hevc \
+    --dataset.depth_encoder.extra_options='{"x265-params": "lossless=1"}'
+```
+
 ### Training
 
 Depending on your hardware training the policy might take a few hours. That's how you train simple `ACT` policy:

docs/source/earthrover_mini_plus.mdx

@@ -194,7 +194,7 @@ lerobot-record \
     --dataset.single_task="Navigate around obstacles" \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    # --dataset.camera_encoder.vcodec=auto \
+    # --dataset.rgb_encoder.vcodec=auto \
     --display_data=true
 ```
 

docs/source/envhub_isaaclab_arena.mdx

@@ -193,7 +193,7 @@ To learn more about training policies with LeRobot, please refer to the training
 
 - [SmolVLA](./smolvla)
 - [Pi0.5](./pi05)
-- [GR00T N1.5](./groot)
+- [GR00T N1.7](./groot)
 
 Sample IsaacLab Arena datasets are available on HuggingFace Hub for experimentation:
 

docs/source/groot.mdx

@@ -1,16 +1,19 @@
-# GR00T N1.5 Policy
+# GR00T Policy
 
-GR00T N1.5 is an open foundation model from NVIDIA designed for generalized humanoid robot reasoning and skills. It is a cross-embodiment model that accepts multimodal input, including language and images, to perform manipulation tasks in diverse environments.
+GR00T is an NVIDIA foundation model family for generalized humanoid robot reasoning and skills. It is a cross-embodiment policy that accepts multimodal input, including language, images, and proprioception, to perform manipulation tasks in diverse environments.
 
-This document outlines the specifics of its integration and usage within the LeRobot framework.
+LeRobot integrates GR00T N1.7 through the `groot` policy type.
+
+> [!WARNING]
+> **Breaking change:** GR00T N1.5 support was removed from LeRobot, and current releases support GR00T N1.7 only. N1.5 checkpoints and configs are rejected with a migration note. To keep using an N1.5 checkpoint, pin the last release that supports it: `pip install 'lerobot==0.5.1'`. To use the current release, migrate to GR00T N1.7 (base model [`nvidia/GR00T-N1.7-3B`](https://huggingface.co/nvidia/GR00T-N1.7-3B)).
 
 ## Model Overview
 
-NVIDIA Isaac GR00T N1.5 is an upgraded version of the GR00T N1 foundation model. It is built to improve generalization and language-following abilities for humanoid robots.
+GR00T N1.7 uses a Cosmos-Reason2/Qwen3-VL backbone and provides checkpoints for SimplerEnv, DROID, and LIBERO.
 
-Developers and researchers can post-train GR00T N1.5 with their own real or synthetic data to adapt it for specific humanoid robots or tasks.
+Developers and researchers can post-train GR00T with their own real or synthetic data to adapt it for specific humanoid robots or tasks.
 
-GR00T N1.5 (specifically the GR00T-N1.5-3B model) is built using pre-trained vision and language encoders. It utilizes a flow matching action transformer to model a chunk of actions, conditioned on vision, language, and proprioception.
+GR00T uses pre-trained vision and language encoders with a flow matching action transformer to model a chunk of actions conditioned on vision, language, and proprioception.
 
 <img
   src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/lerobot/lerobot-groot-paper1%20(1).png"
@@ -28,33 +31,24 @@ This approach allows the model to be highly adaptable through post-training for
 
 ## Installation Requirements
 
-As of today, GR00T N1.5 requires flash attention for it's internal working.
-
-We are working on making this optional, but in the meantime that means that we require an extra installation step and it can only be used in CUDA enabled devices.
-
-1. Following the Environment Setup of our [Installation Guide](./installation). **Attention** don't install `lerobot` in this step.
-2. Install [Flash Attention](https://github.com/Dao-AILab/flash-attention) by running:
+GR00T is intended for NVIDIA GPU-accelerated systems. Install LeRobot with the GR00T extra:
 
 ```bash
-# Check https://pytorch.org/get-started/locally/ for your system
-pip install "torch>=2.2.1,<2.8.0" "torchvision>=0.21.0,<0.23.0" # --index-url https://download.pytorch.org/whl/cu1XX
-pip install ninja "packaging>=24.2,<26.0" # flash attention dependencies
-pip install "flash-attn>=2.5.9,<3.0.0" --no-build-isolation
-python -c "import flash_attn; print(f'Flash Attention {flash_attn.__version__} imported successfully')"
+pip install "lerobot[groot]"
 ```
 
-3. Install LeRobot by running:
+For a source checkout:
 
 ```bash
-pip install lerobot[groot]
+pip install -e ".[groot]"
 ```
 
 ## Usage
 
-To use GR00T in your LeRobot configuration, specify the policy type as:
+To use GR00T N1.7:
 
-```python
-policy.type=groot
+```bash
+--policy.type=groot
 ```
 
 ## Training
@@ -63,72 +57,168 @@ policy.type=groot
 
 Here's a complete training command for finetuning the base GR00T model on your own dataset:
 
+This command is using the `new_embodiment` flag, which is used for the SO-101 robot, [read more about how GR00T handles different embodiments.](https://github.com/NVIDIA/Isaac-GR00T/blob/main/getting_started/policy.md#--embodiment-tag).
+
 ```bash
-# Using a multi-GPU setup
-accelerate launch \
-  --multi_gpu \
-  --num_processes=$NUM_GPUS \
-  $(which lerobot-train) \
-  --output_dir=$OUTPUT_DIR \
-  --save_checkpoint=true \
-  --batch_size=$BATCH_SIZE \
-  --steps=$NUM_STEPS \
-  --save_freq=$SAVE_FREQ \
-  --log_freq=$LOG_FREQ \
-  --policy.push_to_hub=true \
+# install extra deps for training
+pip install "lerobot[training]"
+
+hf auth login
+wandb login
+
+export DATASET_NAME=your_data_set
+export HF_USER=your_hf_username
+export DATASET=$HF_USER/$DATASET_NAME
+export REPO_ID="${DATASET}_GR00T17" #this is the model that will be uploaded to huggingface
+export OUTPUT_DIR=outputs/train/$REPO_ID
+
+lerobot-train \
+  --dataset.repo_id=$DATASET \
+  --dataset.image_transforms.enable=true \
   --policy.type=groot \
+  --policy.device=cuda \
+  --policy.base_model_path=nvidia/GR00T-N1.7-3B \
+  --policy.embodiment_tag=new_embodiment \
+  --policy.chunk_size=16 \
+  --policy.n_action_steps=16 \
+  --policy.use_relative_actions=true \
+  --policy.relative_exclude_joints='["gripper"]' \
+  --policy.use_bf16=true \
+  --policy.push_to_hub=true \
   --policy.repo_id=$REPO_ID \
-  --policy.tune_diffusion_model=false \
-  --dataset.repo_id=$DATASET_ID \
+  --seed=42 \
+  --batch_size=64 \
+  --steps=20000 \
+  --save_checkpoint=true \
+  --save_freq=5000 \
+  --use_policy_training_preset=true \
+  --env_eval_freq=0 \
+  --eval_steps=0 \
+  --log_freq=10 \
+  --output_dir=$OUTPUT_DIR \
+  --job_name=$DATASET \
   --wandb.enable=true \
-  --wandb.disable_artifact=true \
-  --job_name=$JOB_NAME
+  --wandb.disable_artifact=true
+
 ```
 
 ## Performance Results
 
-### Libero Benchmark Results
+### LIBERO Benchmark Results
 
 > [!NOTE]
-> Follow our instructions for Libero usage: [Libero](./libero)
+> Follow the [LIBERO](./libero) setup instructions before running `lerobot-eval`.
 
-GR00T has demonstrated strong performance on the Libero benchmark suite. To compare and test its LeRobot implementation, we finetuned the GR00T N1.5 model for 30k steps on the Libero dataset and compared the results to the GR00T reference results.
+GR00T N1.7 has demonstrated strong performance on the LIBERO benchmark suite. To reproduce LeRobot results, follow the instructions in the [LIBERO](./libero) section.
 
-| Benchmark          | LeRobot Implementation | GR00T Reference |
-| ------------------ | ---------------------- | --------------- |
-| **Libero Spatial** | 82.0%                  | 92.0%           |
-| **Libero Object**  | 99.0%                  | 92.0%           |
-| **Libero Long**    | 82.0%                  | 76.0%           |
-| **Average**        | 87.0%                  | 87.0%           |
+### Train on LIBERO
 
-These results demonstrate GR00T's strong generalization capabilities across diverse robotic manipulation tasks. To reproduce these results, you can follow the instructions in the [Libero](https://huggingface.co/docs/lerobot/libero) section.
+Example training command for a LIBERO suite (here `libero_spatial`):
+
+```bash
+IMAGE_TRANSFORMS='{
+  "brightness": {"weight": 1.0, "type": "ColorJitter", "kwargs": {"brightness": [0.7, 1.3]}},
+  "contrast":   {"weight": 1.0, "type": "ColorJitter", "kwargs": {"contrast":   [0.6, 1.4]}},
+  "saturation": {"weight": 1.0, "type": "ColorJitter", "kwargs": {"saturation": [0.5, 1.5]}},
+  "hue":        {"weight": 1.0, "type": "ColorJitter", "kwargs": {"hue":        [-0.08, 0.08]}}
+}'
+
+lerobot-train \
+  --dataset.repo_id=IPEC-COMMUNITY/libero_spatial_no_noops_1.0.0_lerobot \
+  --dataset.root=/datasets/libero_spatial \
+  --dataset.revision=main \
+  --dataset.video_backend=pyav \
+  --dataset.image_transforms.enable=true \
+  --dataset.image_transforms.max_num_transforms=4 \
+  --dataset.image_transforms.tfs="$IMAGE_TRANSFORMS" \
+  --policy.type=groot \
+  --policy.base_model_path=nvidia/GR00T-N1.7-3B \
+  --policy.embodiment_tag=libero_sim \
+  --policy.push_to_hub=false \
+  --policy.max_steps=20000 \
+  --batch_size=320 \
+  --steps=20000 \
+  --save_freq=2000 \
+  --env_eval_freq=0 \
+  --eval_steps=0 \
+  --log_freq=10 \
+  --wandb.enable=true \
+  --wandb.project=lerobot \
+  --wandb.mode=online \
+  --wandb.disable_artifact=true \
+  --num_workers=4 \
+  --prefetch_factor=2 \
+  --persistent_workers=true \
+  --output_dir=$OUTPUT_DIR \
+  --job_name=$JOB_NAME
+```
+
+### GR00T N1.7 LIBERO Results
+
+Preliminary LeRobot integration results (GR00T-LeRobot, `eval.n_episodes >= 50` per suite):
+
+| Suite                  | Success rate |
+| ---------------------- | -----------: |
+| LIBERO Spatial         |          94% |
+| LIBERO Object          |          98% |
+| LIBERO Goal            |          93% |
+| LIBERO 10 (Long)       |          90% |
+| **Average**            |   **93.75%** |
+
+```bash
+export MODEL_ID=your_trained_model_on_huggingface
+
+lerobot-eval \
+  --policy.type=groot \
+  --policy.base_model_path=$MODEL_ID \
+  --policy.embodiment_tag=libero_sim \
+  --env.type=libero \
+  --env.task=libero_spatial \
+  --eval.n_episodes=50
+```
+
+Use `eval.n_episodes >= 50` per suite when reporting success rates.
 
 ### Evaluate in your hardware setup
 
 Once you have trained your model using your parameters you can run inference in your downstream task. Follow the instructions in [Policy Deployment (lerobot-rollout)](./inference). For example:
 
 ```bash
-lerobot-rollout\
-  --strategy.type=sentry \
-  --strategy.upload_every_n_episodes=5 \
-  --robot.type=bi_so_follower \
-  --robot.left_arm_port=/dev/ttyACM1 \
-  --robot.right_arm_port=/dev/ttyACM0 \
-  --robot.id=bimanual_follower \
-  --robot.cameras='{ right: {"type": "opencv", "index_or_path": 0, "width": 640, "height": 480, "fps": 30},
-    left: {"type": "opencv", "index_or_path": 2, "width": 640, "height": 480, "fps": 30},
-    top: {"type": "opencv", "index_or_path": 4, "width": 640, "height": 480, "fps": 30},
-  }' \
+# install extra deps for roullout and real hardware
+pip install "lerobot[feetech,viz]"
+
+export MODEL_ID=your_trained_model_on_huggingface
+
+# make sure that camera index matches your setup! 
+# find index using `uv run lerobot-find-cameras opencv`
+WRIST_CAM='wrist: {type: opencv, index_or_path: 2, width: 640, height: 480, fps: 30, fourcc: "MJPG"}'
+FRONT_CAM='front: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30, fourcc: "MJPG"}'
+export ROBOT_CAMERAS="{ $WRIST_CAM, $FRONT_CAM }"
+export ROBOT_ID=follower_robot
+export ROBOT_PORT=/dev/ttyACM0
+
+uv run lerobot-rollout \
+  --strategy.type=base \
+  --policy.path=$MODEL_ID \
+  --policy.base_model_path=nvidia/GR00T-N1.7-3B \
+  --policy.n_action_steps=8 \
+  --robot.type=so101_follower \
+  --robot.port=$ROBOT_PORT \
+  --robot.id=$ROBOT_ID \
+  --robot.cameras="$ROBOT_CAMERAS" \
+  --task="place the vial in the rack" \
+  --duration=60 \
+  --device=cuda \
   --display_data=true \
-  --dataset.repo_id=<user>/eval_groot-bimanual  \
-  --dataset.single_task="Grab and handover the red cube to the other arm" \
-  --dataset.streaming_encoding=true \
-  --dataset.encoder_threads=2 \
-  # --dataset.camera_encoder.vcodec=auto \
-  --policy.path=<user>/groot-bimanual \ # your trained model
-  --duration=600
+  --inference.type=rtc \
+  --inference.rtc.enabled=false \
+  --inference.rtc.execution_horizon=8 \
+  --inference.queue_threshold=0
 ```
 
+> [!NOTE]
+> Value of `inference.queue_threshold` should not exeed 0.5 to ensure stable inference.
+
 ## License
 
-This model follows NVIDIA's proprietary license, consistent with the original [GR00T repository](https://github.com/NVIDIA/Isaac-GR00T). Future versions (starting from N1.7) will follow **Apache 2.0 License**.
+GR00T N1.7 is released under the [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/).

docs/source/hope_jr.mdx

@@ -232,7 +232,7 @@ lerobot-record \
     --dataset.private=true \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    # --dataset.camera_encoder.vcodec=auto \
+    # --dataset.rgb_encoder.vcodec=auto \
     --display_data=true
 ```
 
@@ -278,6 +278,6 @@ lerobot-record \
   --dataset.num_episodes=10 \
   --dataset.streaming_encoding=true \
   --dataset.encoder_threads=2 \
-  # --dataset.camera_encoder.vcodec=auto \
+  # --dataset.rgb_encoder.vcodec=auto \
   --policy.path=outputs/train/hopejr_hand/checkpoints/last/pretrained_model
 ```

docs/source/il_robots.mdx

@@ -207,7 +207,7 @@ lerobot-record \
     --dataset.num_episodes=5 \
     --dataset.single_task="Grab the black cube" \
     --dataset.streaming_encoding=true \
-    # --dataset.camera_encoder.vcodec=auto \
+    # --dataset.rgb_encoder.vcodec=auto \
     --dataset.encoder_threads=2
 ```
 </hfoption>

docs/source/lerobot-dataset-v3.mdx

@@ -44,7 +44,7 @@ lerobot-record \
   --dataset.num_episodes=5 \
   --dataset.single_task="Grab the black cube" \
   --dataset.streaming_encoding=true \
-  # --dataset.camera_encoder.vcodec=auto \
+  # --dataset.rgb_encoder.vcodec=auto \
   --dataset.encoder_threads=2
 ```
 

docs/source/policy_groot_README.md

@@ -1,6 +1,13 @@
 ## Research Paper
 
-Paper: https://research.nvidia.com/labs/gear/gr00t-n1_5/
+GR00T N1 technical report (covers the GR00T N1.x family, including N1.7): https://arxiv.org/abs/2503.14734
+
+GR00T N1.7 model card: https://huggingface.co/nvidia/GR00T-N1.7-3B
+
+GR00T N1.5 research page (earlier version): https://research.nvidia.com/labs/gear/gr00t-n1_5/
+
+> GR00T N1.5 support was removed from LeRobot; the last release supporting it is `lerobot==0.5.1`.
+> Current releases support GR00T N1.7 only.
 
 ## Repository
 
@@ -24,4 +31,103 @@ Code: https://github.com/NVIDIA/Isaac-GR00T
 
 Blog: https://developer.nvidia.com/isaac/gr00t
 
-Hugging Face Model: https://huggingface.co/nvidia/GR00T-N1.5-3B
+Hugging Face Models:
+
+- GR00T N1.7: https://huggingface.co/nvidia/GR00T-N1.7-3B
+- GR00T N1.7 LIBERO checkpoints: https://huggingface.co/nvidia/GR00T-N1.7-LIBERO
+
+## Original-vs-LeRobot parity test
+
+`tests/policies/groot/test_groot_vs_original.py` verifies this LeRobot
+reimplementation of GR00T N1.7 (Qwen3-VL backbone + flow-matching action head)
+against NVIDIA's original `gr00t` package with two comparisons, each parametrized
+over every embodiment tag present in the checkpoint:
+
+1. **Model parity** — given byte-identical pre-processed inputs and the same
+   flow-matching seed (recorded in each artifact), both implementations must produce
+   the **same raw model output** (`get_action(...)["action_pred"]`, the normalized
+   flow-matching prediction). Output shapes must match exactly; any action-horizon
+   or action-dim mismatch fails the test.
+2. **Preprocessor parity** — given the identical raw observations (per-camera
+   frames, state vectors, language instruction), LeRobot's own preprocessor pipeline
+   (real Qwen3-VL chat template / tokenizer / image packing + checkpoint-driven
+   state normalization, no mocks) must produce the **same collated model inputs**
+   (`input_ids`, `attention_mask`, `pixel_values`, `image_grid_thw`, `state`,
+   `embodiment_id`) as the original package's processor.
+
+### Why two environments
+
+The original `gr00t` package pins `transformers==4.57.3` (Python 3.10); this
+integration requires `transformers>=5.x` (Qwen3-VL). Under 5.x, `PretrainedConfig`
+is itself a defaulted dataclass, so the original config dataclasses fail to import
+(`non-default argument follows default argument`). The two implementations therefore
+**cannot be imported in the same Python process**.
+
+So the test uses a **producer / consumer** split across two venvs:
+
+1. **Producer** — `tests/policies/groot/utils/dump_original_n1_7.py`, run in the _original_
+   gr00t venv. For each embodiment it builds dummy inputs generically from the
+   checkpoint metadata (state dims from `statistics.json`; camera/language keys from
+   the processor modality configs), runs the original model, and saves to one `.npz`
+   per tag: the raw observations (`raw::` keys), the exact collated inputs
+   (`in::` keys), the seed, and the raw `action_pred`.
+2. **Consumer** — the pytest above, run in the _LeRobot_ venv. It discovers every
+   `.npz`; the model-parity case replays the byte-identical collated inputs through
+   the LeRobot model with the recorded seed and asserts the outputs match, and the
+   preprocessor-parity case replays the raw observations through LeRobot's full
+   preprocessor pipeline and asserts the collated tensors match.
+
+> Artifacts generated by older versions of the dump script contain no `raw::`
+> fields; the preprocessor-parity case then **skips** with a regeneration hint.
+> Re-run the producer to refresh them.
+
+### Fairness controls
+
+- **Same pre-processed inputs (model parity)** — the original processor's `input_ids`,
+  `pixel_values`, `image_grid_thw`, `attention_mask`, `state`, `embodiment_id` are
+  fed verbatim to the LeRobot model (no re-tokenization / re-normalization), so the
+  model comparison isolates the model. LeRobot's own tokenization / image packing is
+  covered separately by the preprocessor-parity case, which compares its output
+  against those same collated tensors from identical raw observations.
+- **Same precision + attention kernel** — both sides run **fp32 + SDPA**. The
+  original defaults to `use_flash_attention=True` (flash_attention_2 + bf16); the
+  producer forces SDPA + fp32. (With the defaults the gap is ~3e-2 — pure
+  kernel/rounding noise, not an implementation difference.)
+- **Same flow-matching seed** — fixed right before sampling on both sides; the
+  producer records it in each artifact (`--seed`, default 42) and the consumer
+  replays the recorded value.
+
+### How to run
+
+```bash
+# Resolve a local checkpoint (GR00T-N1.7-LIBERO / libero_10)
+CKPT=$(python - <<'PY'
+import os
+from huggingface_hub import snapshot_download
+print(os.path.join(snapshot_download("nvidia/GR00T-N1.7-LIBERO",
+      allow_patterns=["libero_10/*"]), "libero_10"))
+PY
+)
+
+# 1) Produce the original-side artifacts for all embodiments (original gr00t venv, CUDA)
+CUDA_VISIBLE_DEVICES=0 /path/to/Isaac-GR00T/.venv-original/bin/python \
+    tests/policies/groot/utils/dump_original_n1_7.py \
+    --ckpt "$CKPT" --out-dir tests/policies/groot/artifacts --device cuda --seed 42
+
+# 2) Run the parity test (LeRobot venv) — one parametrized case per embodiment
+CUDA_VISIBLE_DEVICES=0 GROOT_PARITY_DEVICE=cuda \
+    uv run pytest tests/policies/groot/test_groot_vs_original.py -v -s
+```
+
+The `.npz` artifacts are local-only (gitignored, ~6–10 MB each) and are regenerated by
+the producer; they are never committed. The tests **skip** (do not fail) on CI or
+when the checkpoint / artifacts are absent.
+
+#### Env knobs (all optional)
+
+| Var                                       | Default                          | Purpose                               |
+| ----------------------------------------- | -------------------------------- | ------------------------------------- |
+| `GROOT_N1_7_PARITY_DIR`                   | `tests/policies/groot/artifacts` | directory of per-tag `.npz` artifacts |
+| `GROOT_N1_7_LIBERO_CKPT`                  | auto (HF cache)                  | override checkpoint dir               |
+| `GROOT_PARITY_DEVICE`                     | `cuda` if available              | `cpu` or `cuda`                       |
+| `GROOT_PARITY_ATOL` / `GROOT_PARITY_RTOL` | `1e-3`                           | comparison tolerance                  |

docs/source/reachy2.mdx

@@ -161,7 +161,7 @@ lerobot-record \
     --dataset.private=true \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    # --dataset.camera_encoder.vcodec=auto \
+    # --dataset.rgb_encoder.vcodec=auto \
     --display_data=true
 ```
 
@@ -203,7 +203,7 @@ lerobot-record \
     --dataset.private=true \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    # --dataset.camera_encoder.vcodec=auto \
+    # --dataset.rgb_encoder.vcodec=auto \
     --display_data=true
 ```
 

docs/source/streaming_video_encoding.mdx

@@ -17,7 +17,7 @@ This makes `save_episode()` near-instant (the video is already encoded by the ti
 | Parameter               | CLI Flag                          | Type          | Default       | Description                                                       |
 | ----------------------- | --------------------------------- | ------------- | ------------- | ----------------------------------------------------------------- |
 | `streaming_encoding`    | `--dataset.streaming_encoding`    | `bool`        | `True`        | Enable real-time encoding during capture                          |
-| `vcodec`                | `--dataset.camera_encoder.vcodec` | `str`         | `"libsvtav1"` | Video codec. `"auto"` detects best HW encoder                     |
+| `vcodec`                | `--dataset.rgb_encoder.vcodec`    | `str`         | `"libsvtav1"` | Video codec. `"auto"` detects best HW encoder                     |
 | `encoder_threads`       | `--dataset.encoder_threads`       | `int \| None` | `None` (auto) | Threads per encoder instance. `None` will leave the vcoded decide |
 | `encoder_queue_maxsize` | `--dataset.encoder_queue_maxsize` | `int`         | `30`          | Max buffered frames per camera (~1s at 30fps). Consumes RAM       |
 
@@ -82,15 +82,15 @@ Use HW encoding when:
 
 ### Available HW Encoders
 
-| Encoder             | Platform      | Hardware                                                                                         | CLI Value                                           |
-| ------------------- | ------------- | ------------------------------------------------------------------------------------------------ | --------------------------------------------------- |
-| `h264_videotoolbox` | macOS         | Apple Silicon / Intel                                                                            | `--dataset.camera_encoder.vcodec=h264_videotoolbox` |
-| `hevc_videotoolbox` | macOS         | Apple Silicon / Intel                                                                            | `--dataset.camera_encoder.vcodec=hevc_videotoolbox` |
-| `h264_nvenc`        | Linux/Windows | NVIDIA GPU                                                                                       | `--dataset.camera_encoder.vcodec=h264_nvenc`        |
-| `hevc_nvenc`        | Linux/Windows | NVIDIA GPU                                                                                       | `--dataset.camera_encoder.vcodec=hevc_nvenc`        |
-| `h264_vaapi`        | Linux         | Intel/AMD GPU                                                                                    | `--dataset.camera_encoder.vcodec=h264_vaapi`        |
-| `h264_qsv`          | Linux/Windows | Intel Quick Sync                                                                                 | `--dataset.camera_encoder.vcodec=h264_qsv`          |
-| `auto`              | Any           | Probes the system for available HW encoders. Falls back to `libsvtav1` if no HW encoder is found | `--dataset.camera_encoder.vcodec=auto`              |
+| Encoder             | Platform      | Hardware                                                                                         | CLI Value                                        |
+| ------------------- | ------------- | ------------------------------------------------------------------------------------------------ | ------------------------------------------------ |
+| `h264_videotoolbox` | macOS         | Apple Silicon / Intel                                                                            | `--dataset.rgb_encoder.vcodec=h264_videotoolbox` |
+| `hevc_videotoolbox` | macOS         | Apple Silicon / Intel                                                                            | `--dataset.rgb_encoder.vcodec=hevc_videotoolbox` |
+| `h264_nvenc`        | Linux/Windows | NVIDIA GPU                                                                                       | `--dataset.rgb_encoder.vcodec=h264_nvenc`        |
+| `hevc_nvenc`        | Linux/Windows | NVIDIA GPU                                                                                       | `--dataset.rgb_encoder.vcodec=hevc_nvenc`        |
+| `h264_vaapi`        | Linux         | Intel/AMD GPU                                                                                    | `--dataset.rgb_encoder.vcodec=h264_vaapi`        |
+| `h264_qsv`          | Linux/Windows | Intel Quick Sync                                                                                 | `--dataset.rgb_encoder.vcodec=h264_qsv`          |
+| `auto`              | Any           | Probes the system for available HW encoders. Falls back to `libsvtav1` if no HW encoder is found | `--dataset.rgb_encoder.vcodec=auto`              |
 
 > [!NOTE]
 > In order to use the HW accelerated encoders you might need to upgrade your GPU drivers.
@@ -100,15 +100,15 @@ Use HW encoding when:
 
 ## 5. Troubleshooting
 
-| Symptom                                                            | Likely Cause                                 | Fix                                                                                                                                                                                                                                                                                                 |
-| ------------------------------------------------------------------ | -------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| System freezes or choppy robot movement or Rerun visualization lag | CPU starved (100% load usage)                | Close other apps, reduce encoding throughput, lower `encoder_threads`, use `h264`, use `display_data=False`. If the CPU continues to be at 100% then it might be insufficient for your setup, consider `--dataset.streaming_encoding=false` or HW encoding (`--dataset.camera_encoder.vcodec=auto`) |
-| "Encoder queue full" warnings or dropped frames in dataset         | Encoder can't keep up (Queue overflow)       | If CPU is not at 100%: Increase `encoder_threads`, increase `encoder_queue_maxsize` or use HW encoding (`--dataset.camera_encoder.vcodec=auto`).                                                                                                                                                    |
-| High RAM usage                                                     | Queue filling faster than encoding           | `encoder_threads` too low or CPU insufficient. Reduce `encoder_queue_maxsize` or use HW encoding                                                                                                                                                                                                    |
-| Large video files                                                  | Using HW encoder or H.264                    | Expected trade-off. Switch to `libsvtav1` if CPU allows                                                                                                                                                                                                                                             |
-| `save_episode()` still slow                                        | `streaming_encoding` is `False`              | Set `--dataset.streaming_encoding=true`                                                                                                                                                                                                                                                             |
-| Encoder thread crash                                               | Codec not available or invalid settings      | Check `vcodec` is installed, try `--dataset.camera_encoder.vcodec=auto`                                                                                                                                                                                                                             |
-| Recorded dataset is missing frames                                 | CPU/GPU starvation or occasional load spikes | If ~5% of frames are missing, your system is likely overloaded — follow the recommendations above. If fewer frames are missing (~2%), they are probably due to occasional transient load spikes (often at startup) and can be considered expected.                                                  |
+| Symptom                                                            | Likely Cause                                 | Fix                                                                                                                                                                                                                                                                                              |
+| ------------------------------------------------------------------ | -------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| System freezes or choppy robot movement or Rerun visualization lag | CPU starved (100% load usage)                | Close other apps, reduce encoding throughput, lower `encoder_threads`, use `h264`, use `display_data=False`. If the CPU continues to be at 100% then it might be insufficient for your setup, consider `--dataset.streaming_encoding=false` or HW encoding (`--dataset.rgb_encoder.vcodec=auto`) |
+| "Encoder queue full" warnings or dropped frames in dataset         | Encoder can't keep up (Queue overflow)       | If CPU is not at 100%: Increase `encoder_threads`, increase `encoder_queue_maxsize` or use HW encoding (`--dataset.rgb_encoder.vcodec=auto`).                                                                                                                                                    |
+| High RAM usage                                                     | Queue filling faster than encoding           | `encoder_threads` too low or CPU insufficient. Reduce `encoder_queue_maxsize` or use HW encoding                                                                                                                                                                                                 |
+| Large video files                                                  | Using HW encoder or H.264                    | Expected trade-off. Switch to `libsvtav1` if CPU allows                                                                                                                                                                                                                                          |
+| `save_episode()` still slow                                        | `streaming_encoding` is `False`              | Set `--dataset.streaming_encoding=true`                                                                                                                                                                                                                                                          |
+| Encoder thread crash                                               | Codec not available or invalid settings      | Check `vcodec` is installed, try `--dataset.rgb_encoder.vcodec=auto`                                                                                                                                                                                                                             |
+| Recorded dataset is missing frames                                 | CPU/GPU starvation or occasional load spikes | If ~5% of frames are missing, your system is likely overloaded — follow the recommendations above. If fewer frames are missing (~2%), they are probably due to occasional transient load spikes (often at startup) and can be considered expected.                                               |
 
 ## 6. Recommended Configurations
 
@@ -146,7 +146,7 @@ On very constrained systems, streaming encoding may compete too heavily with the
 # 2camsx 640x480x3 @30fps: Requires some tuning.
 
 # Use H.264, disable streaming, consider batching encoding
-lerobot-record --dataset.camera_encoder.vcodec=h264 --dataset.streaming_encoding=false ...
+lerobot-record --dataset.rgb_encoder.vcodec=h264 --dataset.streaming_encoding=false ...
 ```
 
 ## 7. Closing note

docs/source/using_dataset_tools.mdx

@@ -11,8 +11,9 @@ LeRobot provides several utilities for manipulating datasets:
 3. **Merge Datasets** - Combine multiple datasets into one. The datasets must have identical features, and episodes are concatenated in the order specified in `repo_ids`
 4. **Add Features** - Add new features to a dataset
 5. **Remove Features** - Remove features from a dataset
-6. **Convert to Video** - Convert image-based datasets to video format for efficient storage
-7. **Show the Info of Datasets** - Show the summary of datasets information such as number of episode etc.
+6. **Convert to Video** - Convert image-based datasets to video format for efficient storage (RGB and depth cameras are encoded with separate encoders)
+7. **Re-encode Videos** - Re-encode an existing video dataset's RGB and/or depth streams with new encoder settings
+8. **Show the Info of Datasets** - Show the summary of datasets information such as number of episode etc.
 
 The core implementation is in `lerobot.datasets.dataset_tools`.
 An example script detailing how to use the tools API is available in `examples/dataset/use_dataset_tools.py`.
@@ -117,10 +118,19 @@ lerobot-edit-dataset \
     --repo_id lerobot/pusht_image \
     --operation.type convert_image_to_video \
     --operation.output_dir outputs/pusht_video \
-    --operation.camera_encoder.vcodec libsvtav1 \
-    --operation.camera_encoder.pix_fmt yuv420p \
-    --operation.camera_encoder.g 2 \
-    --operation.camera_encoder.crf 30
+    --operation.rgb_encoder.vcodec libsvtav1 \
+    --operation.rgb_encoder.pix_fmt yuv420p \
+    --operation.rgb_encoder.g 2 \
+    --operation.rgb_encoder.crf 30
+
+# Convert a dataset that includes depth maps, customizing the depth encoder
+lerobot-edit-dataset \
+    --repo_id lerobot/pusht_image \
+    --operation.type convert_image_to_video \
+    --operation.output_dir outputs/pusht_video \
+    --operation.depth_encoder.depth_min 0.01 \
+    --operation.depth_encoder.depth_max 10.0 \
+    --operation.depth_encoder.use_log true
 
 # Convert only specific episodes
 lerobot-edit-dataset \
@@ -147,11 +157,42 @@ lerobot-edit-dataset \
 **Parameters:**
 
 - `output_dir`: Custom output directory (optional - by default uses `new_repo_id` or `{repo_id}_video`)
-- `camera_encoder`: Video encoder settings — all sub-fields accessible via `--operation.camera_encoder.<field>. See [Video Encoding Parameters](./video_encoding_parameters) for more details.
+- `rgb_encoder`: Video encoder settings applied to RGB cameras — all sub-fields accessible via `--operation.rgb_encoder.<field>`. See [Video Encoding Parameters](./video_encoding_parameters) for more details.
+- `depth_encoder`: Video encoder settings applied to depth-map cameras (e.g. from an Intel RealSense). In addition to the standard encoder fields it exposes the depth quantization knobs (`depth_min`, `depth_max`, `shift`, `use_log`), accessible via `--operation.depth_encoder.<field>`. These quantization settings are persisted to the dataset metadata so depth can be dequantized back to physical units on load. See the [Depth streams](./video_encoding_parameters#depth-streams) section for details.
 - `episode_indices`: List of specific episodes to convert (default: all episodes)
 - `num_workers`: Number of parallel workers for processing (default: 4)
 
-**Note:** The resulting dataset will be a proper LeRobotDataset with all cameras encoded as videos in the `videos/` directory, with parquet files containing only metadata (no raw image data). All episodes, stats, and tasks are preserved.
+**Note:** The resulting dataset will be a proper LeRobotDataset with all cameras encoded as videos in the `videos/` directory, with parquet files containing only metadata (no raw image data). Depth-map cameras are detected automatically and routed to the `depth_encoder`, while RGB cameras use the `rgb_encoder`. All episodes, stats, and tasks are preserved.
+
+#### Re-encode Videos
+
+Re-encode the videos of an existing video dataset with different encoder settings, without going back to raw frames. RGB videos use the `rgb_encoder` and depth videos use the `depth_encoder`. Provide only the encoder(s) you want to re-encode; the other stream type is left untouched.
+
+```bash
+# Re-encode all RGB videos with new settings (saves to lerobot/pusht_reencoded by default)
+lerobot-edit-dataset \
+    --repo_id lerobot/pusht \
+    --operation.type reencode_videos \
+    --operation.rgb_encoder.vcodec h264 \
+    --operation.rgb_encoder.pix_fmt yuv420p \
+    --operation.rgb_encoder.crf 23
+
+# Re-encode both RGB and depth videos in a dataset with depth maps
+lerobot-edit-dataset \
+    --repo_id lerobot/pusht_depth \
+    --operation.type reencode_videos \
+    --operation.rgb_encoder.vcodec h264 \
+    --operation.depth_encoder.crf 50
+```
+
+**Parameters:**
+
+- `rgb_encoder`: Encoder settings applied to every RGB video. Omit to skip re-encoding RGB videos.
+- `depth_encoder`: Encoder settings applied to every depth video. Omit to skip re-encoding depth videos.
+- `num_workers`: Number of parallel workers for processing.
+
+> [!NOTE]
+> When re-encoding depth videos, the existing depth quantization parameters (`depth_min`, `depth_max`, `shift`, `use_log`) and the `is_depth_map` flag are **preserved** — re-encoding only changes the codec/quality of the stored stream, not how depth is dequantized on load.
 
 ### Show the information of datasets
 

docs/source/video_encoding_parameters.mdx

@@ -2,15 +2,15 @@
 
 When video storage is enabled, LeRobot stores each camera stream as an **MP4** file instead of saving one image file per timestep. Video encoding compresses across time, which usually cuts dataset size and I/O compared to a pile of PNG, while keeping MP4 — a format every player and loader understands.
 
-Encoding frames into an MP4 is a full FFmpeg pipeline: choice of encoder, pixel format, GOP/keyframes, quality vs. speed, and optional extra encoder flags. Most of these knobs are user-tunable through `camera_encoder`, a nested `VideoEncoderConfig` (`lerobot.configs.video.VideoEncoderConfig`) passed through PyAV.
+Encoding frames into an MP4 is a full FFmpeg pipeline: choice of encoder, pixel format, GOP/keyframes, quality vs. speed, and optional extra encoder flags. Most of these knobs are user-tunable through `rgb_encoder`, a nested `RGBEncoderConfig` (`lerobot.configs.video.RGBEncoderConfig`) passed through PyAV.
 
-You can set these parameters from the CLI with `--dataset.camera_encoder.<field>` (e.g. with `lerobot-record` or `lerobot-rollout`). The same block applies to every camera video stream in that run.
+You can set these parameters from the CLI with `--dataset.rgb_encoder.<field>` (e.g. with `lerobot-record` or `lerobot-rollout`). The same block applies to every camera video stream in that run.
 
 <Tip>
-  Video storage must be on for `camera_encoder` to have any effect —
+  Video storage must be on for `rgb_encoder` to have any effect —
   `use_videos=True` in Python APIs, or `--dataset.video=true` on the CLI (the
-  recording default). With video off, inputs stay as images and `camera_encoder`
-  is ignored.
+  recording default). With video off, inputs stay as images and `rgb_encoder` is
+  ignored.
 </Tip>
 
 For details on **when** frames are written vs. encoded (streaming vs. post-episode), queues, and other top-level `--dataset.*` switches, see [Streaming Video Encoding](./streaming_video_encoding). For an encoding-parameter comparison and experiments, see the [video-benchmark Space](https://huggingface.co/spaces/lerobot/video-benchmark).
@@ -33,9 +33,9 @@ lerobot-record \
     --dataset.single_task="Grab the cube" \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    --dataset.camera_encoder.vcodec=h264 \
-    --dataset.camera_encoder.preset=fast \
-    --dataset.camera_encoder.extra_options={"tune": "film", "profile:v": "high", "bf": 2} \
+    --dataset.rgb_encoder.vcodec=h264 \
+    --dataset.rgb_encoder.preset=fast \
+    --dataset.rgb_encoder.extra_options={"tune": "film", "profile:v": "high", "bf": 2} \
     --display_data=true
 ```
 
@@ -50,7 +50,7 @@ Only override these parameters if you have a specific reason to, and measure the
 
 </Tip>
 
-All flags below are prefixed with `--dataset.camera_encoder.` on the CLI.
+All flags below are prefixed with `--dataset.rgb_encoder.` on the CLI.
 
 | Parameter       | Type             | Default       | Description                                                                                                                                                                            |
 | --------------- | ---------------- | ------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
@@ -65,6 +65,77 @@ All flags below are prefixed with `--dataset.camera_encoder.` on the CLI.
 
 ---
 
+## Depth streams
+
+Depth maps (Intel RealSense, Reachy 2) are stored as their **own video streams** alongside the RGB streams. Raw depth (`uint16` millimetres or `float32` metres) can't survive an 8-bit codec, so LeRobot **quantizes** each map to a 12-bit code (`[0, 4095]`) — logarithmically by default, to match the `1/depth` error profile of depth sensors — then packs it into a high-bit-depth pixel format (`gray12le`) and encodes it with a 12-bit codec.
+
+```mermaid
+flowchart LR
+    A["Raw depth (uint16 mm / float32 m)"] --> B["Clip to depth_min, depth_max"]
+    B --> C["Quantize to 12-bit code 0–4095 (log or linear)"]
+    C --> D["Pack into gray12le"]
+    D --> E["Encode video (hevc Main 12)"]
+    E --> F[("MP4 + metadata: depth_min/max, shift, use_log")]
+    F -. "load time (depth_output_unit)" .-> G["Dequantize to mm or m"]
+
+    classDef input fill:#e3f2fd,stroke:#1565c0,color:#0d47a1;
+    classDef encode fill:#ede7f6,stroke:#5e35b1,color:#311b92;
+    classDef store fill:#fff8e1,stroke:#f9a825,color:#e65100;
+    classDef load fill:#e8f5e9,stroke:#2e7d32,color:#1b5e20;
+
+    class A input;
+    class B,C,D,E encode;
+    class F store;
+    class G load;
+```
+
+Configure the depth pipeline through a parallel **`depth_encoder`** block (`DepthEncoderConfig`). It shares every `RGBEncoderConfig` field (`vcodec`, `pix_fmt`, `crf`, …) and adds four quantizer knobs, set via `--dataset.depth_encoder.<field>`:
+
+```bash
+lerobot-record \
+    ... \
+    --dataset.depth_encoder.vcodec=hevc \
+    --dataset.depth_encoder.depth_min=0.05 \
+    --dataset.depth_encoder.depth_max=5.0 \
+    --dataset.depth_encoder.use_log=true
+```
+
+| Parameter       | Type    | Default                         | Description                                                                                                                            |
+| --------------- | ------- | ------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------- |
+| `vcodec`        | `str`   | `"hevc"`                        | HEVC Main 12 (a 12-bit-capable codec, MP4-compatible).                                                                                 |
+| `extra_options` | `dict`  | `{"x265-params": "lossless=1"}` | **Depth defaults to lossless** (exact round-trip); `crf` is ignored. Pass `extra_options={}` and set `crf` for a smaller lossy stream. |
+| `pix_fmt`       | `str`   | `"gray12le"`                    | Single-channel 12-bit pixel format used to carry the quantized codes.                                                                  |
+| `depth_min`     | `float` | `0.01`                          | Depth in metres mapped to quantum `0`. Values below are clipped on decode.                                                             |
+| `depth_max`     | `float` | `10.0`                          | Depth in metres mapped to quantum `4095`. Values above are clipped on decode.                                                          |
+| `shift`         | `float` | `3.5`                           | Pre-log offset (metres) used in logarithmic quantization for numerical stability near zero. Must satisfy `depth_min + shift > 0`.      |
+| `use_log`       | `bool`  | `True`                          | If `true`, quantize in log-space (recommended for typical depth sensors). Set to `false` for uniform/linear quantization.              |
+
+> [!TIP]
+> `depth_min`, `depth_max`, and `shift` are always interpreted in **metres**, regardless of the input depth's unit. Inputs are auto-detected: integer arrays (e.g. `uint16` millimetres straight from a RealSense) are treated as millimetres, floating arrays as metres.
+> Pick `depth_min` / `depth_max` to bracket the actual working range of your sensor — quanta outside that range saturate, which can crush detail at the boundaries.
+
+Depth features are flagged with `"is_depth_map": true` in `meta/info.json`, and their quantizer settings (`video.depth_min`, `video.depth_max`, `video.shift`, `video.use_log`) are persisted — which is what lets depth be **dequantized back to physical units** on load.
+
+### Output unit at load time
+
+`depth_encoder` is a **record-time** concern. The unit that depth maps are dequantized to on _load_ (e.g. during training) is set separately by the read-time flag `--dataset.depth_output_unit`:
+
+```bash
+lerobot-train \
+    --dataset.repo_id=<my_username>/<my_dataset_name> \
+    --dataset.depth_output_unit=m \
+    --policy.type=act
+```
+
+| Parameter           | Type  | Default | Description                                                                                  |
+| ------------------- | ----- | ------- | -------------------------------------------------------------------------------------------- |
+| `depth_output_unit` | `str` | `"mm"`  | Physical unit depth maps are dequantized to on load: `"mm"` (millimetres) or `"m"` (metres). |
+
+> [!TIP]
+> This is purely a decode-time presentation choice — it does **not** alter the stored video or its metadata, so the same dataset can be read as `mm` or `m` without re-encoding. It has no effect on datasets without depth cameras.
+
+---
+
 ## Persistence in dataset metadata
 
 After the first episode of a video stream is encoded, the encoder configuration is **persisted into the dataset metadata** (`meta/info.json`) under each video feature, alongside the values probed from the file itself. For a video feature `observation.images.<camera>`, the layout in `info.json` is:
@@ -82,7 +153,7 @@ After the first episode of a video stream is encoded, the encoder configuration
         "video.pix_fmt": "yuv420p",
         "video.fps": 30,
         "video.channels": 3,
-        "video.is_depth_map": false,
+        "is_depth_map": false,
         "video.g": 2,
         "video.crf": 30,
         "video.preset": "fast",
@@ -97,12 +168,12 @@ After the first episode of a video stream is encoded, the encoder configuration
 
 Two sources contribute to the `info` block:
 
-- **Stream-derived** (read back from the encoded MP4 with PyAV): `video.height`, `video.width`, `video.codec`, `video.pix_fmt`, `video.fps`, `video.channels`, `video.is_depth_map`, plus `audio.*` if an audio stream is present.
-- **Encoder-derived** (taken from `VideoEncoderConfig`): `video.g`, `video.crf`, `video.preset`, `video.fast_decode`, `video.video_backend`, `video.extra_options`.
+- **Stream-derived** (read back from the encoded MP4 with PyAV): `video.height`, `video.width`, `video.codec`, `video.pix_fmt`, `video.fps`, `video.channels`, `is_depth_map`, plus `audio.*` if an audio stream is present.
+- **Encoder-derived** (taken from `RGBEncoderConfig` or `DepthEncoderConfig`): `video.g`, `video.crf`, `video.preset`, `video.fast_decode`, `video.video_backend`, `video.extra_options`.
 
 <Tip>
   This block is populated **once**, from the **first** episode. It assumes every
-  episode in the dataset was encoded with the same `camera_encoder`. Changing
+  episode in the dataset was encoded with the same `rgb_encoder`. Changing
   encoder settings partway through a recording is not supported — the
   `info.json` will only reflect the parameters used for the first episode.
 </Tip>

pyproject.toml

@@ -218,11 +218,10 @@ groot = [
     "lerobot[transformers-dep]",
     "lerobot[peft-dep]",
     "lerobot[diffusers-dep]",
+    "lerobot[dataset]", # NOTE: processor_groot builds a LeRobotDataset for relative-action training stats
     "dm-tree>=0.1.8,<1.0.0",
     "timm>=1.0.0,<1.1.0",
     "decord>=0.6.0,<1.0.0; (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "ninja>=1.11.1,<2.0.0",
-    "flash-attn>=2.5.9,<3.0.0 ; sys_platform != 'darwin'"
 ]
 sarm = ["lerobot[transformers-dep]", "pydantic>=2.0.0,<3.0.0", "faker>=33.0.0,<35.0.0", "lerobot[matplotlib-dep]", "lerobot[qwen-vl-utils-dep]"]
 robometer = ["lerobot[transformers-dep]", "lerobot[qwen-vl-utils-dep]", "lerobot[peft-dep]"]
@@ -308,7 +307,7 @@ all = [
     "lerobot[pi]",
     "lerobot[molmoact2]",
     "lerobot[smolvla]",
-    # "lerobot[groot]", TODO(Steven): Gr00t requires specific installation instructions for flash-attn
+    "lerobot[groot]",
     "lerobot[xvla]",
     "lerobot[hilserl]",
     "lerobot[vla_jepa]",

src/lerobot/annotations/steerable_pipeline/frames.py

@@ -36,7 +36,7 @@ from typing import Any, Protocol
 import PIL.Image
 import torch
 
-from lerobot.configs.video import VideoEncoderConfig
+from lerobot.configs import RGBEncoderConfig
 from lerobot.datasets.video_utils import decode_video_frames, reencode_video
 
 from .reader import EpisodeRecord, snap_to_frame
@@ -164,7 +164,9 @@ class VideoFrameProvider:
         # only for video-stored cameras. Image-stored cameras (also in
         # ``camera_keys``) would KeyError, so restrict the list — and the
         # default — to video keys.
-        keys = list(self._meta.video_keys)
+        # Depth cameras are excluded from the annotation pipeline for now.
+        depth_keys = set(self._meta.depth_keys)
+        keys = [key for key in self._meta.video_keys if key not in depth_keys]
         # Last-resort fallback: if metadata didn't surface any video keys but
         # the caller explicitly named a camera (``--vlm.camera_key=...``),
         # trust them — the key is by definition known to exist on the dataset.
@@ -276,12 +278,12 @@ class VideoFrameProvider:
         from_timestamp = float(ep[f"videos/{self.camera_key}/from_timestamp"])
         to_timestamp = float(ep[f"videos/{self.camera_key}/to_timestamp"])
         src = self.root / self._meta.get_video_file_path(record.episode_index, self.camera_key)
-        encoder = VideoEncoderConfig(vcodec="h264", pix_fmt="yuv420p", g=None, crf=23, preset="ultrafast")
+        encoder = RGBEncoderConfig(vcodec="h264", pix_fmt="yuv420p", g=None, crf=23, preset="ultrafast")
         try:
             reencode_video(
                 src,
                 out_path,
-                camera_encoder=encoder,
+                video_encoder=encoder,
                 overwrite=True,
                 start_time_s=from_timestamp,
                 end_time_s=to_timestamp,

src/lerobot/async_inference/helpers.py

@@ -105,8 +105,9 @@ def raw_observation_to_observation(
 
 
 def prepare_image(image: torch.Tensor) -> torch.Tensor:
-    """Minimal preprocessing to turn int8 images to float32 in [0, 1], and create a memory-contiguous tensor"""
-    image = image.type(torch.float32) / 255
+    """Minimal preprocessing to turn RGB uint8 images to float32 in [0, 1], and create a memory-contiguous tensor"""
+    if image.dtype == torch.uint8:
+        image = image.type(torch.float32) / 255
     image = image.contiguous()
 
     return image

src/lerobot/cameras/opencv/camera_opencv.py

@@ -436,7 +436,7 @@ class OpenCVCamera(Camera):
         Internal loop run by the background thread for asynchronous reading.
 
         On each iteration:
-        1. Reads a color frame
+        1. Reads a color frame (blocking call)
         2. Stores result in latest_frame and updates timestamp (thread-safe)
         3. Sets new_frame_event to notify listeners
 
@@ -485,6 +485,8 @@ class OpenCVCamera(Camera):
 
         if self.thread is not None and self.thread.is_alive():
             self.thread.join(timeout=2.0)
+            if self.thread.is_alive():
+                logger.warning(f"{self} read thread did not terminate within timeout.")
 
         self.thread = None
         self.stop_event = None

src/lerobot/cameras/realsense/camera_realsense.py

@@ -128,6 +128,7 @@ class RealSenseCamera(Camera):
 
         self.fps = config.fps
         self.color_mode = config.color_mode
+        self.use_rgb = config.use_rgb
         self.use_depth = config.use_depth
         self.warmup_s = config.warmup_s
 
@@ -195,12 +196,15 @@ class RealSenseCamera(Camera):
         # NOTE(Steven/Caroline): Enforcing at least one second of warmup as RS cameras need a bit of time before the first read. If we don't wait, the first read from the warmup will raise.
         self.warmup_s = max(self.warmup_s, 1)
 
+        warmup_read = self.async_read if self.use_rgb else self.async_read_depth
         start_time = time.time()
         while time.time() - start_time < self.warmup_s:
-            self.async_read(timeout_ms=self.warmup_s * 1000)
+            warmup_read(timeout_ms=self.warmup_s * 1000)
             time.sleep(0.1)
         with self.frame_lock:
-            if self.latest_color_frame is None or self.use_depth and self.latest_depth_frame is None:
+            if (self.use_rgb and self.latest_color_frame is None) or (
+                self.use_depth and self.latest_depth_frame is None
+            ):
                 raise ConnectionError(f"{self} failed to capture frames during warmup.")
 
         logger.info(f"{self} connected.")
@@ -268,13 +272,13 @@ class RealSenseCamera(Camera):
             )
 
         if len(found_devices) > 1:
-            serial_numbers = [dev["serial_number"] for dev in found_devices]
+            serial_numbers = [dev["id"] for dev in found_devices]
             raise ValueError(
                 f"Multiple RealSense cameras found with name '{name}'. "
                 f"Please use a unique serial number instead. Found SNs: {serial_numbers}"
             )
 
-        serial_number = str(found_devices[0]["serial_number"])
+        serial_number = str(found_devices[0]["id"])
         return serial_number
 
     def _configure_rs_pipeline_config(self, rs_config: Any) -> None:
@@ -282,15 +286,17 @@ class RealSenseCamera(Camera):
         rs.config.enable_device(rs_config, self.serial_number)
 
         if self.width and self.height and self.fps:
-            rs_config.enable_stream(
-                rs.stream.color, self.capture_width, self.capture_height, rs.format.rgb8, self.fps
-            )
+            if self.use_rgb:
+                rs_config.enable_stream(
+                    rs.stream.color, self.capture_width, self.capture_height, rs.format.rgb8, self.fps
+                )
             if self.use_depth:
                 rs_config.enable_stream(
                     rs.stream.depth, self.capture_width, self.capture_height, rs.format.z16, self.fps
                 )
         else:
-            rs_config.enable_stream(rs.stream.color)
+            if self.use_rgb:
+                rs_config.enable_stream(rs.stream.color)
             if self.use_depth:
                 rs_config.enable_stream(rs.stream.depth)
 
@@ -298,8 +304,9 @@ class RealSenseCamera(Camera):
     def _configure_capture_settings(self) -> None:
         """Sets fps, width, and height from device stream if not already configured.
 
-        Uses the color stream profile to update unset attributes. Handles rotation by
-        swapping width/height when needed. Original capture dimensions are always stored.
+        Uses the color stream profile (or the depth stream profile when the color
+        stream is disabled) to update unset attributes. Handles rotation by swapping
+        width/height when needed. Original capture dimensions are always stored.
 
         Raises:
             DeviceNotConnectedError: If device is not connected.
@@ -308,7 +315,8 @@ class RealSenseCamera(Camera):
         if self.rs_profile is None:
             raise RuntimeError(f"{self}: rs_profile must be initialized before use.")
 
-        stream = self.rs_profile.get_stream(rs.stream.color).as_video_stream_profile()
+        rs_stream = rs.stream.color if self.use_rgb else rs.stream.depth
+        stream = self.rs_profile.get_stream(rs_stream).as_video_stream_profile()
 
         if self.fps is None:
             self.fps = stream.fps()
@@ -323,6 +331,14 @@ class RealSenseCamera(Camera):
                 self.width, self.height = actual_width, actual_height
                 self.capture_width, self.capture_height = actual_width, actual_height
 
+    def _read(self, read_depth: bool = False) -> NDArray[Any]:
+        """Shared helper for :meth:`read`/:meth:`read_depth`: wait for a fresh color or depth frame."""
+        if self.thread is None or not self.thread.is_alive():
+            raise RuntimeError(f"{self} read thread is not running.")
+
+        self.new_frame_event.clear()
+        return self._async_read(timeout_ms=10000, read_depth=read_depth)
+
     @check_if_not_connected
     def read_depth(self, timeout_ms: int = 200) -> NDArray[Any]:
         """
@@ -332,8 +348,8 @@ class RealSenseCamera(Camera):
         from the camera hardware via the RealSense pipeline.
 
         Returns:
-            np.ndarray: The depth map as a NumPy array (height, width)
-                  of type `np.uint16` (raw depth values in millimeters) and rotation.
+            np.ndarray: The depth map as a NumPy array (height, width, 1)
+                  of type `np.uint16` (raw depth values in millimeters).
 
         Raises:
             DeviceNotConnectedError: If the camera is not connected.
@@ -349,20 +365,7 @@ class RealSenseCamera(Camera):
                 f"Failed to capture depth frame '.read_depth()'. Depth stream is not enabled for {self}."
             )
 
-        if self.thread is None or not self.thread.is_alive():
-            raise RuntimeError(f"{self} read thread is not running.")
-
-        self.new_frame_event.clear()
-
-        _ = self.async_read(timeout_ms=10000)
-
-        with self.frame_lock:
-            depth_map = self.latest_depth_frame
-
-        if depth_map is None:
-            raise RuntimeError("No depth frame available. Ensure camera is streaming.")
-
-        return depth_map
+        return self._read(read_depth=True)
 
     def _read_from_hardware(self):
         if self.rs_pipeline is None:
@@ -405,12 +408,10 @@ class RealSenseCamera(Camera):
                 f"{self} read() timeout_ms parameter is deprecated and will be removed in future versions."
             )
 
-        if self.thread is None or not self.thread.is_alive():
-            raise RuntimeError(f"{self} read thread is not running.")
+        if not self.use_rgb:
+            raise RuntimeError(f"{self}: cannot read color — camera was configured with use_rgb=False.")
 
-        self.new_frame_event.clear()
-
-        frame = self.async_read(timeout_ms=10000)
+        frame = self._read()
 
         read_duration_ms = (time.perf_counter() - start_time) * 1e3
         logger.debug(f"{self} read took: {read_duration_ms:.1f}ms")
@@ -465,8 +466,8 @@ class RealSenseCamera(Camera):
         Internal loop run by the background thread for asynchronous reading.
 
         On each iteration:
-        1. Reads a color frame with 500ms timeout
-        2. Stores result in latest_frame and updates timestamp (thread-safe)
+        1. Reads a color/depth frame (blocking call with 10s timeout)
+        2. Stores result in latest_color_frame/latest_depth_frame and updates timestamp (thread-safe)
         3. Sets new_frame_event to notify listeners
 
         Stops on DeviceNotConnectedError, logs other errors and continues.
@@ -479,19 +480,24 @@ class RealSenseCamera(Camera):
         while not stop_event.is_set():
             try:
                 frame = self._read_from_hardware()
-                color_frame_raw = frame.get_color_frame()
-                color_frame = np.asanyarray(color_frame_raw.get_data())
-                processed_color_frame = self._postprocess_image(color_frame)
+
+                if self.use_rgb:
+                    color_frame_raw = frame.get_color_frame()
+                    color_frame = np.asanyarray(color_frame_raw.get_data())
+                    processed_color_frame = self._postprocess_image(color_frame)
 
                 if self.use_depth:
                     depth_frame_raw = frame.get_depth_frame()
                     depth_frame = np.asanyarray(depth_frame_raw.get_data())
                     processed_depth_frame = self._postprocess_image(depth_frame, depth_frame=True)
+                    if processed_depth_frame.ndim == 2:  # (H, W) -> (H, W, 1)
+                        processed_depth_frame = processed_depth_frame[..., np.newaxis]
 
                 capture_time = time.perf_counter()
 
                 with self.frame_lock:
-                    self.latest_color_frame = processed_color_frame
+                    if self.use_rgb:
+                        self.latest_color_frame = processed_color_frame
                     if self.use_depth:
                         self.latest_depth_frame = processed_depth_frame
                     self.latest_timestamp = capture_time
@@ -523,6 +529,8 @@ class RealSenseCamera(Camera):
 
         if self.thread is not None and self.thread.is_alive():
             self.thread.join(timeout=2.0)
+            if self.thread.is_alive():  # pragma: no cover
+                logger.warning(f"{self} read thread did not terminate within timeout.")
 
         self.thread = None
         self.stop_event = None
@@ -533,7 +541,26 @@ class RealSenseCamera(Camera):
             self.latest_timestamp = None
             self.new_frame_event.clear()
 
-    # NOTE(Steven): Missing implementation for depth for now
+    def _async_read(self, timeout_ms: float, read_depth: bool = False) -> NDArray[Any]:
+        """Shared helper for :meth:`async_read`/:meth:`async_read_depth`: return the latest buffered frame."""
+        if self.thread is None or not self.thread.is_alive():
+            raise RuntimeError(f"{self} read thread is not running.")
+
+        if not self.new_frame_event.wait(timeout=timeout_ms / 1000.0):
+            raise TimeoutError(
+                f"Timed out waiting for frame from camera {self} after {timeout_ms} ms. "
+                f"Read thread alive: {self.thread.is_alive()}."
+            )
+
+        with self.frame_lock:
+            frame = self.latest_depth_frame if read_depth else self.latest_color_frame
+            self.new_frame_event.clear()
+
+        if frame is None:
+            raise RuntimeError(f"Internal error: Event set but no frame available for {self}.")
+
+        return frame
+
     @check_if_not_connected
     def async_read(self, timeout_ms: float = 200) -> NDArray[Any]:
         """
@@ -558,25 +585,31 @@ class RealSenseCamera(Camera):
             RuntimeError: If the background thread died unexpectedly or another error occurs.
         """
 
+        if not self.use_rgb:
+            raise RuntimeError(f"{self}: cannot read color — camera was configured with use_rgb=False.")
+
+        return self._async_read(timeout_ms=timeout_ms)
+
+    def _read_latest(self, max_age_ms: int, read_depth: bool = False) -> NDArray[Any]:
+        """Shared helper for :meth:`read_latest`/:meth:`read_latest_depth`: peek the latest buffered frame."""
         if self.thread is None or not self.thread.is_alive():
             raise RuntimeError(f"{self} read thread is not running.")
 
-        if not self.new_frame_event.wait(timeout=timeout_ms / 1000.0):
-            raise TimeoutError(
-                f"Timed out waiting for frame from camera {self} after {timeout_ms} ms. "
-                f"Read thread alive: {self.thread.is_alive()}."
-            )
-
         with self.frame_lock:
-            frame = self.latest_color_frame
-            self.new_frame_event.clear()
+            frame = self.latest_depth_frame if read_depth else self.latest_color_frame
+            timestamp = self.latest_timestamp
 
-        if frame is None:
-            raise RuntimeError(f"Internal error: Event set but no frame available for {self}.")
+        if frame is None or timestamp is None:
+            raise RuntimeError(f"{self} has not captured any frames yet.")
+
+        age_ms = (time.perf_counter() - timestamp) * 1e3
+        if age_ms > max_age_ms:
+            raise TimeoutError(
+                f"{self} latest frame is too old: {age_ms:.1f} ms (max allowed: {max_age_ms} ms)."
+            )
 
         return frame
 
-    # NOTE(Steven): Missing implementation for depth for now
     @check_if_not_connected
     def read_latest(self, max_age_ms: int = 500) -> NDArray[Any]:
         """Return the most recent (color) frame captured immediately (Peeking).
@@ -593,24 +626,48 @@ class RealSenseCamera(Camera):
             DeviceNotConnectedError: If the camera is not connected.
             RuntimeError: If the camera is connected but has not captured any frames yet.
         """
+        if not self.use_rgb:
+            raise RuntimeError(f"{self}: cannot read color — camera was configured with use_rgb=False.")
 
-        if self.thread is None or not self.thread.is_alive():
-            raise RuntimeError(f"{self} read thread is not running.")
+        return self._read_latest(max_age_ms=max_age_ms)
 
-        with self.frame_lock:
-            frame = self.latest_color_frame
-            timestamp = self.latest_timestamp
+    @check_if_not_connected
+    def async_read_depth(self, timeout_ms: float = 200) -> NDArray[np.uint16]:
+        """Read the latest depth frame asynchronously, in millimeters.
 
-        if frame is None or timestamp is None:
-            raise RuntimeError(f"{self} has not captured any frames yet.")
+        Mirrors :meth:`async_read` but returns the depth stream rather than the
+        color stream. Output is ``np.uint16`` of shape ``(H, W, 1)``, where each
+        pixel is the distance from the sensor in millimeters.
 
-        age_ms = (time.perf_counter() - timestamp) * 1e3
-        if age_ms > max_age_ms:
-            raise TimeoutError(
-                f"{self} latest frame is too old: {age_ms:.1f} ms (max allowed: {max_age_ms} ms)."
-            )
+        Raises:
+            DeviceNotConnectedError: If the camera is not connected.
+            RuntimeError: If ``use_depth`` is ``False`` for this camera, or if
+                the background read thread is not running.
+            TimeoutError: If no frame becomes available within ``timeout_ms``.
+        """
+        if not self.use_depth:
+            raise RuntimeError(f"{self}: cannot read depth — camera was configured with use_depth=False.")
 
-        return frame
+        return self._async_read(timeout_ms=timeout_ms, read_depth=True)
+
+    @check_if_not_connected
+    def read_latest_depth(self, max_age_ms: int = 500) -> NDArray[Any]:
+        """Return the most recent depth frame in millimeters (peeking).
+
+        Non-blocking counterpart of :meth:`read_latest` for the depth stream.
+        Output is ``np.uint16`` of shape ``(H, W, 1)``, where each pixel is the
+        distance from the sensor in millimeters.
+
+        Raises:
+            DeviceNotConnectedError: If the camera is not connected.
+            RuntimeError: If ``use_depth`` is ``False`` for this camera, or if
+                no depth frame has been captured yet.
+            TimeoutError: If the latest depth frame is older than ``max_age_ms``.
+        """
+        if not self.use_depth:
+            raise RuntimeError(f"{self}: cannot read depth — camera was configured with use_depth=False.")
+
+        return self._read_latest(max_age_ms=max_age_ms, read_depth=True)
 
     def disconnect(self) -> None:
         """

src/lerobot/cameras/realsense/configuration_realsense.py

@@ -42,12 +42,14 @@ class RealSenseCameraConfig(CameraConfig):
         height: Requested frame height in pixels for the color stream.
         serial_number_or_name: Unique serial number or human-readable name to identify the camera.
         color_mode: Color mode for image output (RGB or BGR). Defaults to RGB.
+        use_rgb: Whether to enable the color stream. Defaults to True.
         use_depth: Whether to enable depth stream. Defaults to False.
         rotation: Image rotation setting (0°, 90°, 180°, or 270°). Defaults to no rotation.
         warmup_s: Time reading frames before returning from connect (in seconds)
 
     Note:
         - Either name or serial_number must be specified.
+        - At least one of `use_rgb` or `use_depth` must be enabled.
         - Depth stream configuration (if enabled) will use the same FPS as the color stream.
         - The actual resolution and FPS may be adjusted by the camera to the nearest supported mode.
         - For `fps`, `width` and `height`, either all of them need to be set, or none of them.
@@ -55,6 +57,7 @@ class RealSenseCameraConfig(CameraConfig):
 
     serial_number_or_name: str
     color_mode: ColorMode = ColorMode.RGB
+    use_rgb: bool = True
     use_depth: bool = False
     rotation: Cv2Rotation = Cv2Rotation.NO_ROTATION
     warmup_s: int = 1
@@ -63,6 +66,9 @@ class RealSenseCameraConfig(CameraConfig):
         self.color_mode = ColorMode(self.color_mode)
         self.rotation = Cv2Rotation(self.rotation)
 
+        if not self.use_rgb and not self.use_depth:
+            raise ValueError("At least one of `use_rgb` or `use_depth` must be enabled.")
+
         values = (self.fps, self.width, self.height)
         if any(v is not None for v in values) and any(v is None for v in values):
             raise ValueError(

src/lerobot/cameras/zmq/camera_zmq.py

@@ -293,6 +293,8 @@ class ZMQCamera(Camera):
 
         if self.thread is not None and self.thread.is_alive():
             self.thread.join(timeout=2.0)
+            if self.thread.is_alive():
+                logger.warning(f"{self} read thread did not terminate within timeout.")
 
         self.thread = None
         self.stop_event = None

src/lerobot/configs/__init__.py

@@ -33,10 +33,15 @@ from .types import (
     RTCAttentionSchedule,
 )
 from .video import (
+    DEFAULT_DEPTH_UNIT,
     VALID_VIDEO_CODECS,
     VIDEO_ENCODER_INFO_KEYS,
+    DepthEncoderConfig,
+    RGBEncoderConfig,
     VideoEncoderConfig,
-    camera_encoder_defaults,
+    depth_encoder_defaults,
+    encoder_config_from_video_info,
+    rgb_encoder_defaults,
 )
 
 __all__ = [
@@ -57,9 +62,15 @@ __all__ = [
     "WandBConfig",
     "load_recipe",
     "VideoEncoderConfig",
+    "RGBEncoderConfig",
+    "DepthEncoderConfig",
     # Defaults
-    "camera_encoder_defaults",
+    "rgb_encoder_defaults",
+    "depth_encoder_defaults",
+    # Factories
+    "encoder_config_from_video_info",
     # Constants
+    "DEFAULT_DEPTH_UNIT",
     "VALID_VIDEO_CODECS",
     "VIDEO_ENCODER_INFO_KEYS",
 ]

src/lerobot/configs/dataset.py

@@ -18,7 +18,7 @@ from dataclasses import dataclass, field
 from datetime import datetime
 from pathlib import Path
 
-from .video import VideoEncoderConfig, camera_encoder_defaults
+from .video import DepthEncoderConfig, RGBEncoderConfig, depth_encoder_defaults, rgb_encoder_defaults
 
 
 @dataclass
@@ -58,8 +58,10 @@ class DatasetRecordConfig:
     # Set to 1 for immediate encoding (default behavior), or higher for batched encoding
     video_encoding_batch_size: int = 1
     # Video encoder settings for camera MP4s (codec, quality, GOP, etc.). Tuned via CLI nested keys,
-    # e.g. ``--dataset.camera_encoder.vcodec=h264`` (see ``VideoEncoderConfig``).
-    camera_encoder: VideoEncoderConfig = field(default_factory=camera_encoder_defaults)
+    # e.g. ``--dataset.rgb_encoder.vcodec=h264`` (see ``RGBEncoderConfig``).
+    rgb_encoder: RGBEncoderConfig = field(default_factory=rgb_encoder_defaults)
+    # Video encoder settings for depth-map MP4s (codec, quality, GOP, etc.). Tuned via CLI nested keys.
+    depth_encoder: DepthEncoderConfig = field(default_factory=depth_encoder_defaults)
     # Enable streaming video encoding: encode frames in real-time during capture instead
     # of writing PNG images first. Makes save_episode() near-instant. More info in the documentation: https://huggingface.co/docs/lerobot/streaming_video_encoding
     streaming_encoding: bool = False

src/lerobot/configs/default.py

@@ -19,6 +19,8 @@ from dataclasses import dataclass, field
 from lerobot.transforms import ImageTransformsConfig
 from lerobot.utils.import_utils import get_safe_default_video_backend
 
+from .video import DEFAULT_DEPTH_UNIT, DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT
+
 
 @dataclass
 class DatasetConfig:
@@ -35,14 +37,21 @@ class DatasetConfig:
     revision: str | None = None
     use_imagenet_stats: bool = True
     video_backend: str = field(default_factory=get_safe_default_video_backend)
-    # When True, video frames are returned as uint8 tensors (0-255) instead of float32 (0.0-1.0).
+    # When True, RGB video frames are returned as uint8 tensors (0-255) instead of float32 (0.0-1.0).
     # This reduces memory and speeds up DataLoader IPC. The training pipeline handles the conversion.
     return_uint8: bool = False
+    # Physical unit depth maps are dequantized to at load time: "mm" (millimeters) or "m" (metres).
+    # Has no effect on datasets without depth cameras.
+    depth_output_unit: str = DEFAULT_DEPTH_UNIT
     streaming: bool = False
     # Fraction of episodes held out per task for offline evaluation (0.0 = disabled).
     eval_split: float = 0.0
 
     def __post_init__(self) -> None:
+        if self.depth_output_unit not in (DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT):
+            raise ValueError(
+                f"depth_output_unit must be '{DEPTH_METER_UNIT}' or '{DEPTH_MILLIMETER_UNIT}', got {self.depth_output_unit!r}"
+            )
         if not (0.0 <= self.eval_split < 1.0):
             raise ValueError(f"eval_split must be in [0.0, 1.0), got {self.eval_split}")
         if self.episodes is not None:

src/lerobot/configs/video.py

@@ -20,7 +20,7 @@ from __future__ import annotations
 
 import logging
 from dataclasses import dataclass, field
-from typing import Any
+from typing import Any, ClassVar, Self
 
 from lerobot.utils.import_utils import require_package
 
@@ -40,7 +40,6 @@ VALID_VIDEO_CODECS: frozenset[str] = frozenset({"h264", "hevc", "libsvtav1", "au
 # Aliases for legacy video codec names.
 VIDEO_CODECS_ALIASES: dict[str, str] = {"av1": "libsvtav1"}
 
-
 LIBSVTAV1_DEFAULT_PRESET: int = 12
 
 # Keys persisted under ``features[*]["info"]`` as ``video.<name>`` (from :class:`VideoEncoderConfig`).
@@ -52,40 +51,45 @@ VIDEO_ENCODER_INFO_KEYS: frozenset[str] = frozenset(
     f"video.{name}" for name in VIDEO_ENCODER_INFO_FIELD_NAMES
 )
 
+# Default depth quantization and encoding parameters.
+DEPTH_QUANT_BITS: int = 12
+DEPTH_QMAX: int = (1 << DEPTH_QUANT_BITS) - 1  # 4095
+
+DEFAULT_DEPTH_MIN: float = 0.01
+DEFAULT_DEPTH_MAX: float = 10.0
+DEFAULT_DEPTH_SHIFT: float = 3.5
+DEFAULT_DEPTH_USE_LOG: bool = True
+DEFAULT_DEPTH_PIX_FMT: str = "gray12le"
+
+DEPTH_METER_UNIT: str = "m"
+DEPTH_MILLIMETER_UNIT: str = "mm"
+DEFAULT_DEPTH_UNIT: str = DEPTH_MILLIMETER_UNIT
+
+# Depth-specific tuning fields persisted under ``features[*]["info"]`` as ``video.<name>``.
+DEPTH_ENCODER_INFO_FIELD_NAMES: frozenset[str] = frozenset({"depth_min", "depth_max", "shift", "use_log"})
+
 
 @dataclass
 class VideoEncoderConfig:
-    """Video encoder configuration.
+    """Video encoder configuration."""
 
-    Attributes:
-        vcodec: Video encoder name. ``"auto"`` is resolved during
-            construction (HW encoder if available, else ``libsvtav1``).
-        pix_fmt: Pixel format (e.g. ``"yuv420p"``).
-        g: GOP size (keyframe interval).
-        crf: Quality level — mapped to the native quality parameter of the
-            codec (``crf`` for software, ``qp`` for NVENC/VAAPI,
-            ``q:v`` for VideoToolbox, ``global_quality`` for QSV).
-        preset: Speed/quality preset. Accepted type is per-codec.
-        fast_decode: Fast-decode tuning. For ``libsvtav1`` this is a level (0-2)
-            embedded in ``svtav1-params``. For ``h264`` and ``hevc`` non-zero values
-            set ``tune=fastdecode``. Ignored for other codecs.
-        video_backend: Python to be used for encoding. Only ``"pyav"``
-            is currently supported.
-        extra_options: Free-form dictionary of additional video encoder options
-            (e.g. ``{"tune": "film", "profile:v": "high", "bf": 2}``).
-    """
-
-    vcodec: str = "libsvtav1"  # TODO(CarolinePascal): rename to codec ?
-    pix_fmt: str = "yuv420p"
-    g: int | None = 2
-    crf: int | float | None = 30
-    preset: int | str | None = None
-    fast_decode: int = 0
+    vcodec: str = "libsvtav1"  # Video codec name. "auto" picks a hardware codec if available, else libsvtav1.
+    pix_fmt: str = "yuv420p"  # Pixel format (e.g. yuv420p).
+    g: int | None = 2  # GOP size (keyframe interval).
+    crf: int | float | None = 30  # Quality level. Lower means better quality and larger files.
+    preset: int | str | None = None  # Speed/quality preset. Accepted values are codec-specific.
+    fast_decode: int = 0  # Fast-decode tuning. Accepted values are codec-specific, 0 disables it.
     # TODO(CarolinePascal): add torchcodec support + find a way to unify the
     # two backends (encoding and decoding).
-    video_backend: str = "pyav"
+    video_backend: str = "pyav"  # Encoding backend. Only "pyav" is currently supported.
+    # Extra codec options merged last, e.g. {"tune": "film"}.
     extra_options: dict[str, Any] = field(default_factory=dict)
 
+    # Source-data channel count this encoder is expected to handle. ``None``
+    # disables the pix_fmt channel-count check; concrete subclasses set it
+    # (3 for RGB, 1 for depth, etc.).
+    _DEFAULT_CHANNELS: ClassVar[int | None] = None
+
     def __post_init__(self) -> None:
         self.resolve_vcodec()
         # Empty-constructor ergonomics: ``VideoEncoderConfig()`` must "just work".
@@ -94,9 +98,9 @@ class VideoEncoderConfig:
         self.validate()
 
     @classmethod
-    def from_video_info(cls, video_info: dict | None) -> VideoEncoderConfig:
-        """Reconstruct a :class:`VideoEncoderConfig` from a video feature's ``info`` block.
-        Missing or ``None`` values fall back to the class defaults.
+    def _kwargs_from_video_info(cls, video_info: dict | None) -> dict[str, Any]:
+        """Parse the ``video.*`` keys of a feature ``info`` block into
+        constructor kwargs.
         """
         video_info = video_info or {}
         kwargs: dict[str, Any] = {}
@@ -115,7 +119,15 @@ class VideoEncoderConfig:
                 continue
             kwargs[field_name] = value
 
-        return cls(**kwargs)
+        return kwargs
+
+    @classmethod
+    def from_video_info(cls, video_info: dict | None) -> Self:
+        """Reconstruct an encoder config from a video feature's ``info`` block.
+
+        Missing or ``None`` values fall back to the class defaults.
+        """
+        return cls(**cls._kwargs_from_video_info(video_info))
 
     def detect_available_encoders(self, encoders: list[str] | str) -> list[str]:
         """Return the subset of available encoders based on the specified video backend.
@@ -138,7 +150,9 @@ class VideoEncoderConfig:
             require_package("av", extra="dataset")
             from lerobot.datasets import check_video_encoder_parameters_pyav
 
-            check_video_encoder_parameters_pyav(self.vcodec, self.pix_fmt, self.get_codec_options())
+            check_video_encoder_parameters_pyav(
+                self.vcodec, self.pix_fmt, self.get_codec_options(), channels=self._DEFAULT_CHANNELS
+            )
 
     def resolve_vcodec(self) -> None:
         """Check ``vcodec`` and, when it is ``"auto"``, pick a concrete encoder.
@@ -230,6 +244,79 @@ class VideoEncoderConfig:
         return opts
 
 
-def camera_encoder_defaults() -> VideoEncoderConfig:
-    """Return a :class:`VideoEncoderConfig` with RGB-camera defaults."""
-    return VideoEncoderConfig()
+@dataclass
+class RGBEncoderConfig(VideoEncoderConfig):
+    """Encoder configuration for RGB camera streams.
+
+    Identical to :class:`VideoEncoderConfig` but declares the 3-channel
+    source-data layout so ``pix_fmt`` is validated against RGB inputs.
+    """
+
+    _DEFAULT_CHANNELS: ClassVar[int] = 3
+
+
+def rgb_encoder_defaults() -> RGBEncoderConfig:
+    """Return a :class:`RGBEncoderConfig` with RGB-camera defaults."""
+    return RGBEncoderConfig()
+
+
+@dataclass
+class DepthEncoderConfig(VideoEncoderConfig):
+    """Encoder configuration for depth-map streams.
+
+    Inherits the full :class:`VideoEncoderConfig` surface (codec, GOP, CRF,
+    preset, ``extra_options``…) and adds the parameters of the depth quantizer.
+    Defaults flip ``vcodec`` to ``"hevc"`` (Main 12 profile) and ``pix_fmt`` to
+    ``"gray12le"``.
+    """
+
+    vcodec: str = "hevc"  # Video codec name. Defaults to HEVC Main 12 (a 12-bit-capable codec).
+    pix_fmt: str = "gray12le"  # Pixel format. Defaults to 12-bit grayscale.
+    extra_options: dict[str, Any] = field(default_factory=lambda: {"x265-params": "lossless=1"})
+
+    depth_min: float = DEFAULT_DEPTH_MIN  # Minimum depth in meters, mapped to the lowest quantum.
+    depth_max: float = DEFAULT_DEPTH_MAX  # Maximum depth in meters, mapped to the highest quantum.
+    shift: float = DEFAULT_DEPTH_SHIFT  # Pre-log offset in meters for numerical stability near zero.
+    use_log: bool = DEFAULT_DEPTH_USE_LOG  # Use logarithmic quantization (True) or linear (False).
+
+    _DEFAULT_CHANNELS: ClassVar[int] = 1
+
+    @classmethod
+    def _kwargs_from_video_info(cls, video_info: dict | None) -> dict[str, Any]:
+        """Layer the depth-specific tuning (``depth_min`` / ``depth_max`` /
+        ``shift`` / ``use_log``) on top of the base parser. Missing keys
+        fall back to the class defaults.
+        """
+        kwargs = super()._kwargs_from_video_info(video_info)
+        video_info = video_info or {}
+        for name in DEPTH_ENCODER_INFO_FIELD_NAMES:
+            value = video_info.get(f"video.{name}")
+            if value is not None:
+                kwargs[name] = value
+        return kwargs
+
+
+def depth_encoder_defaults() -> DepthEncoderConfig:
+    """Return a :class:`DepthEncoderConfig` with depth-camera defaults."""
+    return DepthEncoderConfig()
+
+
+def encoder_config_from_video_info(video_info: dict | None) -> VideoEncoderConfig:
+    """Build the appropriate encoder config from a feature's ``info`` block.
+
+    Dispatches to :class:`DepthEncoderConfig` when the dict marks the feature
+    as a depth map and to :class:`RGBEncoderConfig`
+    otherwise.
+
+    Args:
+        video_info: A feature's ``info`` dict as persisted in ``info.json``,
+            or ``None`` (treated as an empty dict).
+
+    Returns:
+        A :class:`DepthEncoderConfig` for depth features, otherwise a
+        :class:`RGBEncoderConfig`.
+    """
+    video_info = video_info or {}
+    is_depth = bool(video_info.get("is_depth_map") or video_info.get("video.is_depth_map"))
+    cls: type[VideoEncoderConfig] = DepthEncoderConfig if is_depth else RGBEncoderConfig
+    return cls.from_video_info(video_info)

src/lerobot/datasets/compute_stats.py

@@ -242,12 +242,12 @@ def sample_images(image_paths: list[str]) -> np.ndarray:
     images = None
     for i, idx in enumerate(sampled_indices):
         path = image_paths[idx]
-        # we load as uint8 to reduce memory usage
+        # we load RGB images as uint8 to reduce memory usage; depth keeps its native dtype
         img = load_image_as_numpy(path, dtype=np.uint8, channel_first=True)
         img = auto_downsample_height_width(img)
 
         if images is None:
-            images = np.empty((len(sampled_indices), *img.shape), dtype=np.uint8)
+            images = np.empty((len(sampled_indices), *img.shape), dtype=img.dtype)
 
         images[i] = img
 
@@ -506,8 +506,10 @@ def compute_episode_stats(
         Each statistics dictionary contains min, max, mean, std, count, and quantiles.
 
     Note:
-        Image statistics are normalized to [0,1] range and have shape (3,1,1) for
-        per-channel values when dtype is 'image' or 'video'.
+        For 'image'/'video' features, stats are computed per channel and kept with a
+        leading channel axis (e.g. shape (3, 1, 1) for RGB). RGB stats are divided by
+        255 to land in [0, 1]; depth maps (features flagged with ``is_depth_map``) skip
+        this rescaling and remain in their stored units.
     """
     if quantile_list is None:
         quantile_list = DEFAULT_QUANTILES
@@ -531,8 +533,12 @@ def compute_episode_stats(
         )
 
         if features[key]["dtype"] in ["image", "video"]:
+            normalization_factor = (
+                255.0 if not (features[key].get("info") or {}).get("is_depth_map", False) else 1.0
+            )
             ep_stats[key] = {
-                k: v if k == "count" else np.squeeze(v / 255.0, axis=0) for k, v in ep_stats[key].items()
+                k: v if k == "count" else np.squeeze(v / normalization_factor, axis=0)
+                for k, v in ep_stats[key].items()
             }
 
     return ep_stats
@@ -552,8 +558,10 @@ def _validate_stat_value(value: np.ndarray, key: str, feature_key: str) -> None:
     if key == "count" and value.shape != (1,):
         raise ValueError(f"Shape of 'count' must be (1), but is {value.shape} instead.")
 
-    if "image" in feature_key and key != "count" and value.shape != (3, 1, 1):
-        raise ValueError(f"Shape of quantile '{key}' must be (3,1,1), but is {value.shape} instead.")
+    if "image" in feature_key and key != "count" and value.shape not in ((3, 1, 1), (1, 1, 1)):
+        raise ValueError(
+            f"Shape of quantile '{key}' must be (3,1,1) or (1,1,1) but is {value.shape} instead."
+        )
 
 
 def _assert_type_and_shape(stats_list: list[dict[str, dict]]):

src/lerobot/datasets/dataset_metadata.py

@@ -14,7 +14,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import contextlib
-from collections.abc import Callable
+import logging
+from collections.abc import Callable, Iterable
 from copy import deepcopy
 from pathlib import Path
 
@@ -338,6 +339,25 @@ class LeRobotDatasetMetadata:
         """Keys to access visual modalities stored as videos."""
         return [key for key, ft in self.features.items() if ft["dtype"] == "video"]
 
+    @property
+    def depth_keys(self) -> list[str]:
+        """Keys to access depth-map modalities stored as videos or images.
+
+        A depth key is a feature whose ``info`` dict carries ``"is_depth_map": True``
+        (or the legacy ``"video.is_depth_map"`` inside ``info`` or ``video_info``).
+        """
+
+        def _is_depth(ft: dict) -> bool:
+            info = ft.get("info") or {}
+            video_info = ft.get("video_info") or {}
+            return (
+                info.get("is_depth_map", False)
+                or info.get("video.is_depth_map", False)
+                or video_info.get("video.is_depth_map", False)
+            )
+
+        return [key for key, ft in self.features.items() if _is_depth(ft)]
+
     @property
     def camera_keys(self) -> list[str]:
         """Keys to access visual modalities (regardless of their storage method)."""
@@ -581,29 +601,48 @@ class LeRobotDatasetMetadata:
     def update_video_info(
         self,
         video_key: str | None = None,
-        camera_encoder: VideoEncoderConfig | None = None,
+        video_encoder: VideoEncoderConfig | None = None,
+        preserve_keys: Iterable[str] | None = None,
     ) -> None:
-        """Populate per-feature video info in ``info.json``.
+        """Populate or refresh per-feature video info in ``info.json``.
 
         Warning: this function writes info from first episode videos, implicitly assuming that all videos have
         been encoded the same way. Also, this means it assumes the first episode exists.
 
+        Always re-probes the videos and overwrites existing info for every recomputed
+        key. ``preserve_keys`` lists keys whose existing values must be kept (e.g.
+        data-intrinsic entries like ``is_depth_map`` and depth quantization params)
+        instead of being recomputed.
+
         Args:
             video_key: If provided, only update this video key. Otherwise update
                 all video keys in the dataset.
-            camera_encoder: Encoder configuration used to produce the
+            video_encoder: Encoder configuration used to produce the
                 videos. When provided, its fields are recorded as
                 ``video.<field>`` entries alongside the stream-derived
                 ``video.*`` entries (see :func:`get_video_info`).
+            preserve_keys: Keys whose existing values are kept instead of being
+                recomputed. ``None`` (default) recomputes every key.
         """
         if video_key is not None and video_key not in self.video_keys:
             raise ValueError(f"Video key {video_key} not found in dataset")
 
         video_keys = [video_key] if video_key is not None else self.video_keys
+        preserve_set = set(preserve_keys or ())
         for key in video_keys:
-            if not self.features[key].get("info", None):
-                video_path = self.root / self.video_path.format(video_key=key, chunk_index=0, file_index=0)
-                self.info.features[key]["info"] = get_video_info(video_path, camera_encoder=camera_encoder)
+            existing = self.features[key].get("info") or {}
+            video_path = self.root / self.video_path.format(video_key=key, chunk_index=0, file_index=0)
+            new_info = get_video_info(video_path, video_encoder=video_encoder)
+            # Drop preserved keys so the existing values win on merge.
+            new_info = {k: v for k, v in new_info.items() if k not in preserve_set}
+            merged = {**existing, **new_info}
+            # Migrate the legacy depth marker to the canonical key.
+            if "video.is_depth_map" in merged:
+                logging.warning(
+                    f"Migrating legacy 'video.is_depth_map' to 'is_depth_map' for feature {key!r}."
+                )
+                merged.setdefault("is_depth_map", merged.pop("video.is_depth_map"))
+            self.info.features[key]["info"] = merged
 
     def update_chunk_settings(
         self,

src/lerobot/datasets/dataset_reader.py

@@ -22,7 +22,10 @@ from pathlib import Path
 import datasets
 import torch
 
+from lerobot.configs import DEFAULT_DEPTH_UNIT, DepthEncoderConfig
+
 from .dataset_metadata import LeRobotDatasetMetadata
+from .depth_utils import dequantize_depth
 from .feature_utils import (
     check_delta_timestamps,
     get_delta_indices,
@@ -51,6 +54,7 @@ class DatasetReader:
         delta_timestamps: dict[str, list[float]] | None,
         image_transforms: Callable | None,
         return_uint8: bool = False,
+        depth_output_unit: str = DEFAULT_DEPTH_UNIT,
     ):
         """Initialize the reader with metadata, filtering, and transform config.
 
@@ -68,6 +72,10 @@ class DatasetReader:
                 relative timestamp offsets for temporal context windows.
             image_transforms: Optional torchvision v2 transform applied to
                 visual features.
+            return_uint8: If True, return RGB video frames as raw uint8 tensors
+                instead of normalized float32.
+            depth_output_unit: Physical unit depth maps are dequantized to
+                (``"m"`` or ``"mm"``). Defaults to ``"mm"``.
         """
         self._meta = meta
         self.root = root
@@ -78,6 +86,7 @@ class DatasetReader:
             raise TypeError("image_transforms must be callable or None.")
         self._image_transforms = image_transforms
         self._return_uint8 = return_uint8
+        self._depth_output_unit = depth_output_unit
 
         self.hf_dataset: datasets.Dataset | None = None
         self._absolute_to_relative_idx: dict[int, int] | None = None
@@ -88,6 +97,11 @@ class DatasetReader:
             check_delta_timestamps(delta_timestamps, meta.fps, tolerance_s)
             self.delta_indices = get_delta_indices(delta_timestamps, meta.fps)
 
+        self._depth_encoder_configs: dict[str, DepthEncoderConfig] = {
+            vid_key: DepthEncoderConfig.from_video_info(self._meta.features[vid_key].get("info"))
+            for vid_key in self._meta.depth_keys
+        }
+
     def set_image_transforms(self, image_transforms: Callable | None) -> None:
         """Replace the transform applied to visual observations."""
         if image_transforms is not None and not callable(image_transforms):
@@ -259,7 +273,18 @@ class DatasetReader:
                 self._tolerance_s,
                 self._video_backend,
                 return_uint8=self._return_uint8,
+                is_depth=vid_key in self._meta.depth_keys,
             )
+            if vid_key in self._meta.depth_keys:
+                depth_encoder = self._depth_encoder_configs[vid_key]
+                frames = dequantize_depth(
+                    frames,
+                    depth_min=depth_encoder.depth_min,
+                    depth_max=depth_encoder.depth_max,
+                    shift=depth_encoder.shift,
+                    use_log=depth_encoder.use_log,
+                    output_unit=self._depth_output_unit,
+                )
             return vid_key, frames.squeeze(0)
 
         items = list(query_timestamps.items())
@@ -299,8 +324,9 @@ class DatasetReader:
             item = {**video_frames, **item}
 
         if self._image_transforms is not None:
-            image_keys = self._meta.camera_keys
-            for cam in image_keys:
+            for cam in self._meta.camera_keys:
+                if cam in self._meta.depth_keys:
+                    continue
                 item[cam] = self._image_transforms(item[cam])
 
         # Add task as a string

src/lerobot/datasets/dataset_tools.py

@@ -37,7 +37,15 @@ import pyarrow.parquet as pq
 import torch
 from tqdm import tqdm
 
-from lerobot.configs import VideoEncoderConfig, camera_encoder_defaults
+from lerobot.configs import (
+    DepthEncoderConfig,
+    RGBEncoderConfig,
+    VideoEncoderConfig,
+    depth_encoder_defaults,
+    encoder_config_from_video_info,
+    rgb_encoder_defaults,
+)
+from lerobot.configs.video import DEPTH_ENCODER_INFO_FIELD_NAMES
 from lerobot.utils.constants import ACTION, HF_LEROBOT_HOME, OBS_IMAGE, OBS_STATE
 from lerobot.utils.utils import flatten_dict
 
@@ -48,6 +56,7 @@ from .compute_stats import (
     compute_relative_action_stats,
 )
 from .dataset_metadata import LeRobotDatasetMetadata
+from .image_writer import write_image
 from .io_utils import (
     get_parquet_file_size_in_mb,
     load_episodes,
@@ -62,12 +71,13 @@ from .utils import (
     DEFAULT_DATA_FILE_SIZE_IN_MB,
     DEFAULT_DATA_PATH,
     DEFAULT_EPISODES_PATH,
+    DEPTH_FILE_PATTERN,
+    IMAGE_FILE_PATTERN,
     VIDEO_DIR,
     update_chunk_file_indices,
 )
 from .video_utils import (
     encode_video_frames,
-    get_video_info,
     reencode_video,
 )
 
@@ -601,7 +611,7 @@ def _keep_episodes_from_video_with_av(
     output_path: Path,
     episodes_to_keep: list[tuple[int, int]],
     fps: float,
-    camera_encoder: VideoEncoderConfig,
+    video_encoder: VideoEncoderConfig,
 ) -> None:
     """Keep only specified episodes from a video file using PyAV.
 
@@ -615,7 +625,7 @@ def _keep_episodes_from_video_with_av(
             Ranges are half-open intervals: [start_frame, end_frame), where start_frame
             is inclusive and end_frame is exclusive.
         fps: Frame rate of the video.
-        camera_encoder: Video encoder settings used to re-encode the kept frames.
+        video_encoder: Video encoder settings used to re-encode the kept frames.
     """
     from fractions import Fraction
 
@@ -640,13 +650,13 @@ def _keep_episodes_from_video_with_av(
 
     # Convert fps to Fraction for PyAV compatibility.
     fps_fraction = Fraction(fps).limit_denominator(1000)
-    codec_options = camera_encoder.get_codec_options(as_strings=True)
-    v_out = out.add_stream(camera_encoder.vcodec, rate=fps_fraction, options=codec_options)
+    codec_options = video_encoder.get_codec_options(as_strings=True)
+    v_out = out.add_stream(video_encoder.vcodec, rate=fps_fraction, options=codec_options)
 
     # PyAV type stubs don't distinguish video streams from audio/subtitle streams.
     v_out.width = v_in.codec_context.width
     v_out.height = v_in.codec_context.height
-    v_out.pix_fmt = camera_encoder.pix_fmt
+    v_out.pix_fmt = video_encoder.pix_fmt
 
     # Set time_base to match the frame rate for proper timestamp handling.
     v_out.time_base = Fraction(1, int(fps))
@@ -733,7 +743,7 @@ def _copy_and_reindex_videos(
 
     for video_key in src_dataset.meta.video_keys:
         logging.info(f"Processing videos for {video_key}")
-        camera_encoder = VideoEncoderConfig.from_video_info(
+        video_encoder = encoder_config_from_video_info(
             src_dataset.meta.info.features.get(video_key, {}).get("info")
         )
 
@@ -817,7 +827,7 @@ def _copy_and_reindex_videos(
                     dst_video_path,
                     episodes_to_keep_ranges,
                     src_dataset.meta.fps,
-                    camera_encoder,
+                    video_encoder,
                 )
 
                 cumulative_ts = 0.0
@@ -874,11 +884,11 @@ def _copy_and_reindex_episodes_metadata(
             episode_meta.update(video_metadata[new_idx])
 
         # Extract episode statistics from parquet metadata.
-        # Note (maractingi): When pandas/pyarrow serializes numpy arrays with shape (3, 1, 1) to parquet,
+        # When pandas/pyarrow serializes numpy arrays with shape (C, 1, 1) to parquet,
         # they are being deserialized as nested object arrays like:
         #   array([array([array([0.])]), array([array([0.])]), array([array([0.])])])
         # This happens particularly with image/video statistics. We need to detect and flatten
-        # these nested structures back to proper (3, 1, 1) arrays so aggregate_stats can process them.
+        # these nested structures back to proper (C, 1, 1) arrays so aggregate_stats can process them.
         episode_stats = {}
         for key in src_episode_full:
             if key.startswith("stats/"):
@@ -894,15 +904,16 @@ def _copy_and_reindex_episodes_metadata(
                     if feature_name in src_dataset.meta.features:
                         feature_dtype = src_dataset.meta.features[feature_name]["dtype"]
                         if feature_dtype in ["image", "video"] and stat_name != "count":
+                            # Stats are channel-first (C, 1, 1)
                             if isinstance(value, np.ndarray) and value.dtype == object:
                                 flat_values = []
                                 for item in value:
                                     while isinstance(item, np.ndarray):
                                         item = item.flatten()[0]
                                     flat_values.append(item)
-                                value = np.array(flat_values, dtype=np.float64).reshape(3, 1, 1)
-                            elif isinstance(value, np.ndarray) and value.shape == (3,):
-                                value = value.reshape(3, 1, 1)
+                                value = np.array(flat_values, dtype=np.float64).reshape(-1, 1, 1)
+                            elif isinstance(value, np.ndarray) and value.ndim == 1:
+                                value = value.reshape(-1, 1, 1)
 
                     episode_stats[feature_name][stat_name] = value
 
@@ -1153,15 +1164,15 @@ def _save_episode_images_for_video(
     # Get all items for this episode
     episode_dataset = imgs_dataset.select(range(from_idx, to_idx))
 
+    is_depth = img_key in dataset.meta.depth_keys
+    frame_pattern = DEPTH_FILE_PATTERN if is_depth else IMAGE_FILE_PATTERN
+
     # Define function to save a single image
     def save_single_image(i_item_tuple):
         i, item = i_item_tuple
-        img = item[img_key]
-        # Use frame-XXXXXX.png format to match encode_video_frames expectations
-        img.save(str(imgs_dir / f"frame-{i:06d}.png"), quality=100)
+        write_image(item[img_key], imgs_dir / frame_pattern.format(frame_index=i))
         return i
 
-    # Save images with proper naming convention for encode_video_frames (frame-XXXXXX.png)
     items = list(enumerate(episode_dataset))
 
     with ThreadPoolExecutor(max_workers=num_workers) as executor:
@@ -1193,13 +1204,14 @@ def _save_batch_episodes_images(
     hf_dataset = dataset.hf_dataset.with_format(None)
     imgs_dataset = hf_dataset.select_columns(img_key)
 
+    is_depth = img_key in dataset.meta.depth_keys
+    frame_pattern = DEPTH_FILE_PATTERN if is_depth else IMAGE_FILE_PATTERN
+
     # Define function to save a single image with global frame index
     # Defined once outside the loop to avoid repeated closure creation
     def save_single_image(i_item_tuple, base_frame_idx, img_key_param):
         i, item = i_item_tuple
-        img = item[img_key_param]
-        # Use global frame index for naming
-        img.save(str(imgs_dir / f"frame-{base_frame_idx + i:06d}.png"), quality=100)
+        write_image(item[img_key_param], imgs_dir / frame_pattern.format(frame_index=base_frame_idx + i))
         return i
 
     episode_durations = []
@@ -1290,7 +1302,7 @@ def _estimate_frame_size_via_calibration(
     episode_indices: list[int],
     temp_dir: Path,
     fps: int,
-    camera_encoder: VideoEncoderConfig,
+    video_encoder: VideoEncoderConfig,
     num_calibration_frames: int = 30,
 ) -> float:
     """Estimate MB per frame by encoding a small calibration sample.
@@ -1304,7 +1316,7 @@ def _estimate_frame_size_via_calibration(
         episode_indices: List of episode indices being processed.
         temp_dir: Temporary directory for calibration files.
         fps: Frames per second for video encoding.
-        camera_encoder: Video encoder settings used for calibration encoding.
+        video_encoder: Video encoder settings used for calibration encoding.
         num_calibration_frames: Number of frames to use for calibration (default: 30).
 
     Returns:
@@ -1329,10 +1341,11 @@ def _estimate_frame_size_via_calibration(
         hf_dataset = dataset.hf_dataset.with_format(None)
         sample_indices = range(from_idx, from_idx + num_frames)
 
-        # Save calibration frames
+        # Save calibration frames using the suffix/format the encoder expects.
+        is_depth = img_key in dataset.meta.depth_keys
+        frame_pattern = DEPTH_FILE_PATTERN if is_depth else IMAGE_FILE_PATTERN
         for i, idx in enumerate(sample_indices):
-            img = hf_dataset[idx][img_key]
-            img.save(str(calibration_dir / f"frame-{i:06d}.png"), quality=100)
+            write_image(hf_dataset[idx][img_key], calibration_dir / frame_pattern.format(frame_index=i))
 
         # Encode calibration video
         calibration_video_path = calibration_dir / "calibration.mp4"
@@ -1340,7 +1353,7 @@ def _estimate_frame_size_via_calibration(
             imgs_dir=calibration_dir,
             video_path=calibration_video_path,
             fps=fps,
-            camera_encoder=camera_encoder,
+            video_encoder=video_encoder,
             overwrite=True,
         )
 
@@ -1613,6 +1626,7 @@ def recompute_stats(
         raise ValueError(f"No parquet files found in {data_dir}")
 
     all_episode_stats = []
+    # TODO: enable image and video stats re-computation
     numeric_keys = [k for k, v in features_to_compute.items() if v["dtype"] not in ["image", "video"]]
 
     for parquet_path in tqdm(parquet_files, desc="Computing stats from data files"):
@@ -1658,7 +1672,8 @@ def convert_image_to_video_dataset(
     dataset: LeRobotDataset,
     output_dir: Path | None = None,
     repo_id: str | None = None,
-    camera_encoder: VideoEncoderConfig | None = None,
+    rgb_encoder: RGBEncoderConfig | None = None,
+    depth_encoder: DepthEncoderConfig | None = None,
     episode_indices: list[int] | None = None,
     num_workers: int = 4,
     max_episodes_per_batch: int | None = None,
@@ -1670,21 +1685,32 @@ def convert_image_to_video_dataset(
     LeRobot dataset structure with videos stored in chunked MP4 files.
 
     Args:
-        dataset: The source LeRobot dataset with images
-        output_dir: Root directory where the edited dataset will be stored. If not specified, defaults to $HF_LEROBOT_HOME/repo_id. Equivalent to new_root in EditDatasetConfig.
-        repo_id: Edited dataset identifier. Equivalent to new_repo_id in EditDatasetConfig.
-        camera_encoder: Video encoder settings
-            (``None`` uses :func:`~lerobot.configs.camera_encoder_defaults`).
-        episode_indices: List of episode indices to convert (None = all episodes)
-        num_workers: Number of threads for parallel processing (default: 4)
-        max_episodes_per_batch: Maximum episodes per video batch to avoid memory issues (None = no limit)
-        max_frames_per_batch: Maximum frames per video batch to avoid memory issues (None = no limit)
+        dataset: The source LeRobot dataset with images.
+        output_dir: Root directory where the converted dataset will be stored. When
+            ``None``, defaults to ``$HF_LEROBOT_HOME/repo_id``. Equivalent to
+            ``new_root`` in ``EditDatasetConfig``.
+        repo_id: Converted dataset identifier. Equivalent to ``new_repo_id`` in
+            ``EditDatasetConfig``.
+        rgb_encoder: Video encoder settings applied to RGB cameras. When ``None``,
+            :func:`~lerobot.configs.video.rgb_encoder_defaults` is used.
+        depth_encoder: Video encoder settings applied to depth-map cameras, including
+            the quantization parameters persisted to the dataset metadata. When
+            ``None``, :func:`~lerobot.configs.video.depth_encoder_defaults` is used.
+        episode_indices: Episode indices to convert. When ``None``, all episodes are
+            converted.
+        num_workers: Number of threads for parallel processing.
+        max_episodes_per_batch: Maximum episodes per video batch, to bound memory use.
+            ``None`` means no limit.
+        max_frames_per_batch: Maximum frames per video batch, to bound memory use.
+            ``None`` means no limit.
 
     Returns:
-        New LeRobotDataset with images encoded as videos
+        A new :class:`LeRobotDataset` with images encoded as videos.
     """
-    if camera_encoder is None:
-        camera_encoder = camera_encoder_defaults()
+    if rgb_encoder is None:
+        rgb_encoder = rgb_encoder_defaults()
+    if depth_encoder is None:
+        depth_encoder = depth_encoder_defaults()
 
     # Check that it's an image dataset
     if len(dataset.meta.video_keys) > 0:
@@ -1709,10 +1735,7 @@ def convert_image_to_video_dataset(
     logging.info(
         f"Converting {len(episode_indices)} episodes with {len(img_keys)} cameras from {dataset.repo_id}"
     )
-    logging.info(
-        f"Video codec: {camera_encoder.vcodec}, pixel format: {camera_encoder.pix_fmt}, "
-        f"GOP: {camera_encoder.g}, CRF: {camera_encoder.crf}"
-    )
+    logging.info(f"RGB video encoder: {rgb_encoder}, depth video encoder: {depth_encoder}")
 
     # Create new features dict, converting image features to video features
     new_features = {}
@@ -1774,6 +1797,8 @@ def convert_image_to_video_dataset(
         episode_lengths = {ep_idx: dataset.meta.episodes["length"][ep_idx] for ep_idx in episode_indices}
 
         for img_key in tqdm(img_keys, desc="Processing cameras"):
+            target_encoder = depth_encoder if img_key in dataset.meta.depth_keys else rgb_encoder
+
             # Estimate size per frame by encoding a small calibration sample
             # This provides accurate compression ratio for the specific codec parameters
             size_per_frame_mb = _estimate_frame_size_via_calibration(
@@ -1782,7 +1807,7 @@ def convert_image_to_video_dataset(
                 episode_indices=episode_indices,
                 temp_dir=temp_dir,
                 fps=fps,
-                camera_encoder=camera_encoder,
+                video_encoder=target_encoder,
             )
 
             logging.info(f"Processing camera: {img_key}")
@@ -1824,7 +1849,7 @@ def convert_image_to_video_dataset(
                     imgs_dir=imgs_dir,
                     video_path=video_path,
                     fps=fps,
-                    camera_encoder=camera_encoder,
+                    video_encoder=target_encoder,
                     overwrite=True,
                 )
 
@@ -1863,16 +1888,11 @@ def convert_image_to_video_dataset(
         new_meta.info.total_tasks = dataset.meta.total_tasks
         new_meta.info.splits = {"train": f"0:{len(episode_indices)}"}
 
-        # Update video info for all image keys (now videos)
-        # We need to manually set video info since update_video_info() checks video_keys first
+        # Update video info for all image keys (now videos). They are registered as
+        # video features above, so update_video_info populates their (still-empty) info.
         for img_key in img_keys:
-            if not new_meta.features[img_key].get("info", None):
-                video_path = new_meta.root / new_meta.video_path.format(
-                    video_key=img_key, chunk_index=0, file_index=0
-                )
-                new_meta.info.features[img_key]["info"] = get_video_info(
-                    video_path, camera_encoder=camera_encoder
-                )
+            target_encoder = depth_encoder if img_key in dataset.meta.depth_keys else rgb_encoder
+            new_meta.update_video_info(video_key=img_key, video_encoder=target_encoder)
 
         write_info(new_meta.info, new_meta.root)
 
@@ -1899,11 +1919,11 @@ def convert_image_to_video_dataset(
 
 def _reencode_video_worker(args: tuple) -> Path:
     """Picklable worker for :func:`reencode_dataset`'s process pool."""
-    video_path, camera_encoder, encoder_threads = args
+    video_path, video_encoder, encoder_threads = args
     reencode_video(
         input_video_path=video_path,
         output_video_path=video_path,
-        camera_encoder=camera_encoder,
+        video_encoder=video_encoder,
         encoder_threads=encoder_threads,
         overwrite=True,
     )
@@ -1912,7 +1932,8 @@ def _reencode_video_worker(args: tuple) -> Path:
 
 def reencode_dataset(
     dataset: LeRobotDataset,
-    camera_encoder: VideoEncoderConfig,
+    rgb_encoder: RGBEncoderConfig | None = None,
+    depth_encoder: DepthEncoderConfig | None = None,
     encoder_threads: int | None = None,
     num_workers: int | None = None,
 ) -> LeRobotDataset:
@@ -1923,8 +1944,11 @@ def reencode_dataset(
     Args:
         dataset: An existing :class:`LeRobotDataset` whose videos will be
             re-encoded.
-        camera_encoder: Target encoder configuration applied to every video
-            file.
+        rgb_encoder: Target encoder configuration applied to every RGB video
+            file. If ``None``, re-encoding is skipped for RGB videos.
+        depth_encoder: Target encoder configuration applied to every depth video
+            file. If ``None``, re-encoding is skipped for depth videos.
+            Quantization parameters will not override the ones in the current dataset.
         encoder_threads: Per-encoder thread count forwarded to
             :func:`reencode_video`. ``None`` lets the codec decide.
         num_workers: Number of parallel processes. ``None`` or ``0`` means
@@ -1936,23 +1960,35 @@ def reencode_dataset(
         on disk.
     """
     meta = dataset.meta
-    video_paths_list = []
+    video_keys_encoders_dict = {}
+    video_keys_paths_dict = {}
+
+    if rgb_encoder is None and depth_encoder is None:
+        raise ValueError("Either rgb_encoder or depth_encoder must be provided")
 
     # Only re-encode if the videos are not already encoded with the given video encoding parameters
     for video_key in meta.video_keys:
         current_info = meta.info.features[video_key].get("info", {})
-        current_encoder = VideoEncoderConfig.from_video_info(current_info)
-        if current_encoder != camera_encoder:
-            video_paths_list.extend((meta.root / VIDEO_DIR / video_key).rglob("*.mp4"))
+        current_encoder = encoder_config_from_video_info(current_info)
+        target_encoder = depth_encoder if video_key in meta.depth_keys else rgb_encoder
+        if target_encoder is None:
+            logging.info(f"No encoder provided for {video_key} video. Skipping re-encoding.")
+        elif current_encoder != target_encoder:
+            video_keys_paths_dict[video_key] = list((meta.root / VIDEO_DIR / video_key).rglob("*.mp4"))
+            video_keys_encoders_dict[video_key] = target_encoder
         else:
-            logging.info(f"{video_key} videos are already encoded with {camera_encoder}. Nothing to do.")
+            logging.info(f"{video_key} videos are already encoded with {target_encoder}. Nothing to do.")
 
-    if len(video_paths_list) == 0:
+    if len(video_keys_paths_dict) == 0:
         logging.warning("Dataset has no videos to re-encode.")
         return dataset
-    logging.info(f"Re-encoding {len(video_paths_list)} video file(s) with {camera_encoder}")
+    logging.info(f"Re-encoding {sum(len(paths) for paths in video_keys_paths_dict.values())} video file(s).")
 
-    worker_args = [(vp, camera_encoder, encoder_threads) for vp in video_paths_list]
+    worker_args = [
+        (path, encoder, encoder_threads)
+        for video_key, encoder in video_keys_encoders_dict.items()
+        for path in video_keys_paths_dict[video_key]
+    ]
     if num_workers and num_workers > 1:
         with ProcessPoolExecutor(max_workers=num_workers) as pool:
             futures = [pool.submit(_reencode_video_worker, args) for args in worker_args]
@@ -1966,10 +2002,15 @@ def reencode_dataset(
         for args in tqdm(worker_args, desc="Re-encoding videos"):
             _reencode_video_worker(args)
 
-    # Refresh video info in metadata for every video key.
-    for vid_key in meta.video_keys:
-        video_path = meta.root / meta.get_video_file_path(0, vid_key)
-        meta.info.features[vid_key]["info"] = get_video_info(video_path, camera_encoder=camera_encoder)
+    # Refresh video info in metadata for every re-encoded key. Re-encoding only
+    # changes codec/container params, so for depth videos we preserve ``is_depth_map``
+    # and the depth quantization params (``video.depth_min`` / ``video.depth_max`` /
+    # ...), which describe the data rather than the codec and must survive a transcode.
+    # RGB videos pass an empty set: still a refresh, but nothing to preserve.
+    depth_preserve_keys = {"is_depth_map", *(f"video.{n}" for n in DEPTH_ENCODER_INFO_FIELD_NAMES)}
+    for video_key, encoder in video_keys_encoders_dict.items():
+        preserve_keys = depth_preserve_keys if video_key in meta.depth_keys else set()
+        meta.update_video_info(video_key=video_key, video_encoder=encoder, preserve_keys=preserve_keys)
 
     write_info(meta.info, meta.root)
     logging.info("Dataset metadata updated.")

src/lerobot/datasets/dataset_writer.py

@@ -31,7 +31,13 @@ import PIL.Image
 import pyarrow.parquet as pq
 import torch
 
-from lerobot.configs import VideoEncoderConfig, camera_encoder_defaults
+from lerobot.configs import (
+    DepthEncoderConfig,
+    RGBEncoderConfig,
+    VideoEncoderConfig,
+    depth_encoder_defaults,
+    rgb_encoder_defaults,
+)
 
 from .compute_stats import compute_episode_stats
 from .dataset_metadata import LeRobotDatasetMetadata
@@ -48,6 +54,7 @@ from .io_utils import (
     write_info,
 )
 from .utils import (
+    DEFAULT_DEPTH_PATH,
     DEFAULT_EPISODES_PATH,
     DEFAULT_IMAGE_PATH,
     update_chunk_file_indices,
@@ -67,17 +74,22 @@ def _encode_video_worker(
     episode_index: int,
     root: Path,
     fps: int,
-    camera_encoder: VideoEncoderConfig | None = None,
+    video_encoder: VideoEncoderConfig | None = None,
     encoder_threads: int | None = None,
 ) -> Path:
     temp_path = Path(tempfile.mkdtemp(dir=root)) / f"{video_key}_{episode_index:03d}.mp4"
-    fpath = DEFAULT_IMAGE_PATH.format(image_key=video_key, episode_index=episode_index, frame_index=0)
+    path_template = (
+        DEFAULT_DEPTH_PATH
+        if video_encoder is not None and isinstance(video_encoder, DepthEncoderConfig)
+        else DEFAULT_IMAGE_PATH
+    )
+    fpath = path_template.format(image_key=video_key, episode_index=episode_index, frame_index=0)
     img_dir = (root / fpath).parent
     encode_video_frames(
         img_dir,
         temp_path,
         fps,
-        camera_encoder=camera_encoder,
+        video_encoder=video_encoder,
         encoder_threads=encoder_threads,
         overwrite=True,
     )
@@ -96,7 +108,8 @@ class DatasetWriter:
         self,
         meta: LeRobotDatasetMetadata,
         root: Path,
-        camera_encoder: VideoEncoderConfig | None,
+        rgb_encoder: RGBEncoderConfig | None,
+        depth_encoder: DepthEncoderConfig | None,
         encoder_threads: int | None,
         batch_encoding_size: int,
         streaming_encoder: StreamingVideoEncoder | None = None,
@@ -108,8 +121,11 @@ class DatasetWriter:
             meta: Dataset metadata instance (used for feature schema, chunk
                 settings, and episode persistence).
             root: Local dataset root directory.
-            camera_encoder: Video encoder settings applied to all cameras.
-                ``None`` uses :func:`~lerobot.configs.camera_encoder_defaults`.
+            rgb_encoder: Video encoder settings applied to RGB cameras. When
+                ``None``, :func:`~lerobot.configs.video.rgb_encoder_defaults` is used.
+            depth_encoder: Video encoder settings applied to depth cameras, including
+                the quantization parameters. When ``None``,
+                :func:`~lerobot.configs.video.depth_encoder_defaults` is used.
             encoder_threads: Number of encoder threads (global). ``None``
                 lets the codec decide.
             batch_encoding_size: Number of episodes to accumulate before
@@ -120,7 +136,8 @@ class DatasetWriter:
         """
         self._meta = meta
         self._root = root
-        self._camera_encoder = camera_encoder or camera_encoder_defaults()
+        self._rgb_encoder = rgb_encoder or rgb_encoder_defaults()
+        self._depth_encoder = depth_encoder or depth_encoder_defaults()
         self._encoder_threads = encoder_threads
         self._batch_encoding_size = batch_encoding_size
         self._streaming_encoder = streaming_encoder
@@ -145,7 +162,8 @@ class DatasetWriter:
         return ep_buffer
 
     def _get_image_file_path(self, episode_index: int, image_key: str, frame_index: int) -> Path:
-        fpath = DEFAULT_IMAGE_PATH.format(
+        path_template = DEFAULT_DEPTH_PATH if image_key in self._meta.depth_keys else DEFAULT_IMAGE_PATH
+        fpath = path_template.format(
             image_key=image_key, episode_index=episode_index, frame_index=frame_index
         )
         return self._root / fpath
@@ -195,6 +213,7 @@ class DatasetWriter:
         if frame_index == 0 and self._streaming_encoder is not None:
             self._streaming_encoder.start_episode(
                 video_keys=list(self._meta.video_keys),
+                depth_video_keys=list(self._meta.depth_keys),
                 temp_dir=self._root,
             )
 
@@ -282,10 +301,13 @@ class DatasetWriter:
         if use_streaming:
             streaming_results = self._streaming_encoder.finish_episode()
             for video_key in self._meta.video_keys:
+                normalization_factor = 255.0 if video_key not in self._meta.depth_keys else 1.0
                 temp_path, video_stats = streaming_results[video_key]
                 if video_stats is not None:
                     ep_stats[video_key] = {
-                        k: v if k == "count" else np.squeeze(v.reshape(1, -1, 1, 1) / 255.0, axis=0)
+                        k: v
+                        if k == "count"
+                        else np.squeeze(v.reshape(1, -1, 1, 1) / normalization_factor, axis=0)
                         for k, v in video_stats.items()
                     }
                 ep_metadata.update(self._save_episode_video(video_key, episode_index, temp_path=temp_path))
@@ -300,7 +322,7 @@ class DatasetWriter:
                             episode_index,
                             self._root,
                             self._meta.fps,
-                            self._camera_encoder,
+                            self._depth_encoder if video_key in self._meta.depth_keys else self._rgb_encoder,
                             self._encoder_threads,
                         ): video_key
                         for video_key in self._meta.video_keys
@@ -511,7 +533,12 @@ class DatasetWriter:
 
         # Update video info (only needed when first episode is encoded)
         if episode_index == 0:
-            self._meta.update_video_info(video_key, camera_encoder=self._camera_encoder)
+            self._meta.update_video_info(
+                video_key,
+                video_encoder=self._depth_encoder
+                if video_key in self._meta.depth_keys
+                else self._rgb_encoder,
+            )
             write_info(self._meta.info, self._meta.root)
 
         metadata = {
@@ -578,13 +605,14 @@ class DatasetWriter:
             self.image_writer.wait_until_done()
 
     def _encode_temporary_episode_video(self, video_key: str, episode_index: int) -> Path:
-        """Use ffmpeg to convert frames stored as png into mp4 videos."""
+        """Use ffmpeg to convert frames stored as png/tiff into mp4 videos."""
+        is_depth = video_key in self._meta.depth_keys
         return _encode_video_worker(
             video_key,
             episode_index,
             self._root,
             self._meta.fps,
-            self._camera_encoder,
+            self._depth_encoder if is_depth else self._rgb_encoder,
             self._encoder_threads,
         )
 

src/lerobot/datasets/depth_utils.py

@@ -0,0 +1,268 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Depth encoding/decoding helpers for :class:`DepthEncoderConfig`.
+"""
+
+import math
+from typing import Literal
+
+import av
+import numpy as np
+import torch
+from numpy.typing import NDArray
+
+from lerobot.configs.video import (
+    DEFAULT_DEPTH_MAX,
+    DEFAULT_DEPTH_MIN,
+    DEFAULT_DEPTH_PIX_FMT,
+    DEFAULT_DEPTH_SHIFT,
+    DEFAULT_DEPTH_USE_LOG,
+    DEPTH_METER_UNIT,
+    DEPTH_MILLIMETER_UNIT,
+    DEPTH_QMAX,
+)
+
+from .image_writer import squeeze_single_channel
+from .pyav_utils import write_u16_plane
+
+_MM_PER_METRE = 1000.0
+_UINT16_MAX = 65535
+
+
+def _validate_log_quant_params(depth_min: float, shift: float) -> None:
+    """Ensure ``log(depth_min + shift)`` is finite."""
+    if depth_min + shift <= 0:
+        raise ValueError(
+            f"depth_min + shift must be positive for logarithmic quantization, "
+            f"got depth_min={depth_min} + shift={shift} = {depth_min + shift}"
+        )
+
+
+def _depth_input_to_float32_and_unit(
+    depth: NDArray[np.integer] | NDArray[np.floating],
+    input_unit: Literal["auto", DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT],
+) -> tuple[NDArray[np.float32], Literal[DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT]]:
+    """Convert depth to float32 in the chosen unit, and return the resolved unit."""
+    resolved_unit = (
+        (DEPTH_METER_UNIT if np.issubdtype(depth.dtype, np.floating) else DEPTH_MILLIMETER_UNIT)
+        if input_unit == "auto"
+        else input_unit
+    )
+    return depth.astype(np.float32, order="K"), resolved_unit
+
+
+def quantize_depth(
+    depth: NDArray[np.uint16] | NDArray[np.float32] | torch.Tensor,
+    depth_min: float = DEFAULT_DEPTH_MIN,
+    depth_max: float = DEFAULT_DEPTH_MAX,
+    shift: float = DEFAULT_DEPTH_SHIFT,
+    use_log: bool = DEFAULT_DEPTH_USE_LOG,
+    pix_fmt: str = DEFAULT_DEPTH_PIX_FMT,
+    video_backend: str | None = "pyav",
+    input_unit: Literal["auto", DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT] = "auto",
+) -> NDArray[np.uint16] | av.VideoFrame:
+    """Quantize depth to 12-bit codes (``uint16``, values ``0…DEPTH_QMAX``).
+
+    Depth maps are packed into 12-bit integer frames so they fit in standard
+    high-bit-depth pixel formats (e.g. ``yuv420p12le`` / ``gray12le``)
+    and can be encoded by widely supported video codecs (e.g. HEVC Main 12).
+    Logarithmic quantization is the default because it allocates more quanta
+    to near-range depth, which matches the (1/depth) error profile of typical
+    depth sensors. Math is ported from BEHAVIOR-1K's ``obs_utils.py``.
+
+    **Input units**:
+
+    - ``input_unit="auto"`` (default): infer from dtype (floating = m, non-floating = mm).
+    - ``input_unit="mm"``: interpret input values as millimetres.
+    - ``input_unit="m"``: interpret input values as metres.
+
+    Quantization math runs in the **resolved input unit**.
+
+    ``depth_min``, ``depth_max``, and ``shift`` are always in **metres**.
+
+    Args:
+        depth: Depth map; ``torch.Tensor`` is moved to CPU for conversion.
+        depth_min: Depth (metres) at quantum ``0``.
+        depth_max: Depth (metres) at quantum :data:`DEPTH_QMAX`.
+        shift: Depth shift (metres); used in log mode. Must satisfy ``depth_min + shift > 0``.
+        use_log: If ``True`` (default), quantize in log space.
+        video_backend: Video backend to use for encoding. Defaults to "pyav".
+        input_unit: Input unit policy (``"auto"``, ``"mm"``, ``"m"``).
+
+    Returns:
+        ``numpy.ndarray``, ``dtype=uint16``, same shape as ``depth``, values in
+        ``[0, DEPTH_QMAX]``.
+
+    Raises:
+        ValueError: If ``input_unit`` is not ``"auto"``, ``"mm"``, or ``"m"``.
+        ValueError: If ``use_log=True`` and ``depth_min + shift <= 0``.
+    """
+    if input_unit not in ("auto", DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT):
+        raise ValueError(
+            f"input_unit must be 'auto', '{DEPTH_METER_UNIT}', or '{DEPTH_MILLIMETER_UNIT}', got {input_unit!r}"
+        )
+
+    if isinstance(depth, torch.Tensor):
+        depth = depth.detach().cpu().numpy()
+
+    # Squeeze single-channel dim: (H, W, 1) or (1, H, W) → (H, W)
+    depth = squeeze_single_channel(depth)
+
+    depth_f, resolved_unit = _depth_input_to_float32_and_unit(depth, input_unit=input_unit)
+
+    # Convert depth_min, depth_max, and shift to the resolved input unit.
+    depth_min_u = (
+        np.float32(depth_min) if resolved_unit == DEPTH_METER_UNIT else np.float32(depth_min * _MM_PER_METRE)
+    )
+    depth_max_u = (
+        np.float32(depth_max) if resolved_unit == DEPTH_METER_UNIT else np.float32(depth_max * _MM_PER_METRE)
+    )
+    shift_u = np.float32(shift) if resolved_unit == DEPTH_METER_UNIT else np.float32(shift * _MM_PER_METRE)
+
+    # Normalization and quantization is performed in the resolved input unit.
+    if use_log:
+        _validate_log_quant_params(depth_min, shift)
+        log_min = math.log(float(depth_min_u + shift_u))
+        log_max = math.log(float(depth_max_u + shift_u))
+        norm = (np.log(depth_f + shift_u) - log_min) / (log_max - log_min)
+    else:
+        norm = (depth_f - depth_min_u) / (depth_max_u - depth_min_u)
+
+    quantized = np.rint(norm * DEPTH_QMAX).clip(0, DEPTH_QMAX).astype(np.uint16, copy=False)
+
+    if video_backend == "pyav":
+        frame = av.VideoFrame.from_ndarray(quantized, format=pix_fmt)
+        write_u16_plane(frame.planes[0], quantized)
+        return frame
+    else:
+        return quantized
+
+
+def dequantize_depth(
+    quantized: NDArray[np.uint16] | av.VideoFrame | torch.Tensor,
+    depth_min: float = DEFAULT_DEPTH_MIN,
+    depth_max: float = DEFAULT_DEPTH_MAX,
+    shift: float = DEFAULT_DEPTH_SHIFT,
+    use_log: bool = DEFAULT_DEPTH_USE_LOG,
+    pix_fmt: str = DEFAULT_DEPTH_PIX_FMT,
+    output_unit: Literal[DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT] = DEPTH_MILLIMETER_UNIT,
+    output_tensor: bool = True,
+    output_channel_last: bool = False,
+) -> NDArray[np.uint16] | NDArray[np.float32] | torch.Tensor:
+    """Inverse of :func:`quantize_depth`.
+
+    Decoding inverts the same normalized code mapping as :func:`quantize_depth`
+    using ``depth_min`` / ``depth_max`` / ``shift`` (in metres), then returns
+    the requested output unit. Tuning arguments **must match** :func:`quantize_depth`.
+
+    Accepted input layouts :
+
+    - ``(H, W, 1)`` or ``(H, W)`` — single frame with channel-last.
+    - ``(..., 1, H, W)`` — batched frames with channel-first.
+    - ``(..., H, W, 1)`` — batched frames with channel-last.
+    Output layout is determined by ``output_channel_last``.
+
+    Args:
+        quantized: 12-bit codes in ``[0, DEPTH_QMAX]``. ``np.ndarray``,
+            ``av.VideoFrame``, or ``torch.Tensor`` (any integer or float dtype).
+        depth_min, depth_max, shift, use_log: Same as :func:`quantize_depth` (metres).
+        pix_fmt: Pixel format used to extract the plane from an ``av.VideoFrame``.
+        output_unit: ``"mm"`` returns ``uint16`` millimetres (rint, clip
+            ``[0, 65535]``) when returning a numpy array, or ``float32`` mm when
+            ``output_tensor=True``. ``"m"`` returns ``float32`` metres in
+            ``[depth_min, depth_max]``.
+        output_tensor: If True, return a ``torch.Tensor`` instead of a numpy array.
+
+    Returns:
+        Depth map in the requested unit and dtype.
+
+    Raises:
+        ValueError: If ``output_unit`` is not ``"m"`` or ``"mm"``.
+        ValueError: If ``use_log=True`` and ``depth_min + shift <= 0``.
+    """
+    if output_unit not in (DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT):
+        raise ValueError(
+            f"output_unit must be '{DEPTH_METER_UNIT}' or '{DEPTH_MILLIMETER_UNIT}', got {output_unit!r}"
+        )
+    if use_log:
+        _validate_log_quant_params(depth_min, shift)
+
+    if isinstance(quantized, av.VideoFrame):
+        quantized = quantized.to_ndarray(format=pix_fmt)
+
+    # Compute the scale and offset first.
+    depth_min_m = float(depth_min)
+    depth_max_m = float(depth_max)
+    shift_m = float(shift)
+    if use_log:
+        log_min = math.log(depth_min_m + shift_m)
+        log_max = math.log(depth_max_m + shift_m)
+        scale = (log_max - log_min) / DEPTH_QMAX
+        offset = log_min
+    else:
+        scale = (depth_max_m - depth_min_m) / DEPTH_QMAX
+        offset = depth_min_m
+
+    # ── Torch path: stay on the input device, single fp32 allocation. ────────
+    if isinstance(quantized, torch.Tensor):
+        if quantized.ndim >= 3:
+            # Drop the single-channel dimension so the math runs on (..., H, W).
+            quantized = quantized.squeeze(-3) if quantized.shape[-3] == 1 else quantized.squeeze(-1)
+
+        # Single allocation we own; everything else is in-place.
+        buf = quantized.to(dtype=torch.float32, copy=True)
+        buf.mul_(scale).add_(offset)
+        if use_log:
+            buf.exp_().sub_(shift_m)
+        buf.clamp_(depth_min_m, depth_max_m)
+        buf.unsqueeze_(-1) if output_channel_last else buf.unsqueeze_(-3)
+
+        if output_unit == DEPTH_METER_UNIT:
+            return buf if output_tensor else buf.cpu().numpy()
+
+        # mm path: round + clamp in float32, skipping the uint16 round-trip
+        # when returning a tensor (torch.uint16 is poorly supported).
+        buf.mul_(_MM_PER_METRE).round_().clamp_(0.0, _UINT16_MAX)
+        if output_tensor:
+            return buf
+        return buf.cpu().numpy().astype(np.uint16, copy=False)
+
+    # ── NumPy path: single fp32 allocation, ``out=`` for in-place math. ─────
+    arr = np.asarray(quantized)
+    if arr.ndim >= 3:
+        # Drop the single-channel dimension so the math runs on (..., H, W).
+        arr = np.squeeze(arr, axis=-3) if arr.shape[-3] == 1 else np.squeeze(arr, axis=-1)
+
+    buf = np.empty(arr.shape, dtype=np.float32)
+    np.multiply(arr, scale, out=buf)
+    np.add(buf, offset, out=buf)
+    if use_log:
+        np.exp(buf, out=buf)
+        np.subtract(buf, shift_m, out=buf)
+    np.clip(buf, depth_min_m, depth_max_m, out=buf)
+    buf = np.expand_dims(buf, axis=-1) if output_channel_last else np.expand_dims(buf, axis=-3)
+
+    if output_unit == DEPTH_METER_UNIT:
+        return torch.from_numpy(buf) if output_tensor else buf
+
+    np.multiply(buf, _MM_PER_METRE, out=buf)
+    np.rint(buf, out=buf)
+    np.clip(buf, 0.0, _UINT16_MAX, out=buf)
+    if output_tensor:
+        # torch.uint16 support is very limited; return float32 millimetres.
+        return torch.from_numpy(buf)
+    return buf.astype(np.uint16, copy=False)

src/lerobot/datasets/factory.py

@@ -97,6 +97,7 @@ def make_dataset(cfg: TrainPipelineConfig) -> LeRobotDataset | MultiLeRobotDatas
                 revision=cfg.dataset.revision,
                 video_backend=cfg.dataset.video_backend,
                 return_uint8=True,
+                depth_output_unit=cfg.dataset.depth_output_unit,
                 tolerance_s=cfg.tolerance_s,
             )
         else:
@@ -127,6 +128,8 @@ def make_dataset(cfg: TrainPipelineConfig) -> LeRobotDataset | MultiLeRobotDatas
 
     if cfg.dataset.use_imagenet_stats:
         for key in dataset.meta.camera_keys:
+            if key in dataset.meta.depth_keys:
+                continue  # Exclude depth keys from ImageNet stats
             for stats_type, stats in IMAGENET_STATS.items():
                 dataset.meta.stats[key][stats_type] = torch.tensor(stats, dtype=torch.float32)
 

src/lerobot/datasets/feature_utils.py

@@ -336,7 +336,7 @@ def validate_feature_image_or_video(
 
     Args:
         name (str): The name of the feature.
-        expected_shape (list[str]): The expected shape (C, H, W).
+        expected_shape (list[str]): The expected shape, e.g. (C, H, W) or (H, W, C).
         value: The image data to validate.
 
     Returns:

src/lerobot/datasets/image_writer.py

@@ -41,11 +41,51 @@ def safe_stop_image_writer(func):
     return wrapper
 
 
-def image_array_to_pil_image(image_array: np.ndarray, range_check: bool = True) -> PIL.Image.Image:
-    # TODO(aliberts): handle 1 channel and 4 for depth images
-    if image_array.ndim != 3:
-        raise ValueError(f"The array has {image_array.ndim} dimensions, but 3 is expected for an image.")
+def squeeze_single_channel(array: np.ndarray) -> np.ndarray:
+    """Drop a leading or trailing singleton channel dim: ``(1, H, W)`` / ``(H, W, 1)`` -> ``(H, W)``.
 
+    Unlike ``array.squeeze()``, this only removes the channel axis, never an ``H`` or ``W`` of size 1.
+    """
+    if array.ndim == 3:
+        if array.shape[0] == 1:
+            return array[0]
+        if array.shape[-1] == 1:
+            return array[..., 0]
+    return array
+
+
+def image_array_to_pil_image(image_array: np.ndarray, range_check: bool = True) -> PIL.Image.Image:
+    """Convert a NumPy array to a PIL Image, preserving precision for grayscale.
+
+    Behaviour by shape:
+
+    - ``(H, W)`` or ``(1, H, W)`` / ``(H, W, 1)``: single-channel grayscale.
+      The native dtype is preserved using the matching PIL mode
+      (``I;16`` / ``F``). This is the path used for raw depth maps (no rescaling, clamping, or downcasting)
+    - ``(3, H, W)`` / ``(H, W, 3)``: RGB. Channels-first inputs are transposed
+      to channels-last. Float inputs in ``[0, 1]`` are scaled to ``uint8``
+      (existing behaviour, gated by ``range_check``).
+
+    Other shapes / channel counts raise ``NotImplementedError`` or
+    ``ValueError``.
+    """
+    # TODO(CarolinePascal): 4 dimensions RGB-D images
+    if image_array.ndim not in (2, 3):
+        raise ValueError(f"The array has {image_array.ndim} dimensions, but 2 or 3 is expected for an image.")
+
+    # Squeeze 3D single-channel inputs to 2D so depth maps work whether the
+    # caller emits (H, W), (1, H, W), or (H, W, 1).
+    image_array = squeeze_single_channel(image_array)
+
+    if image_array.ndim == 2:
+        if image_array.dtype not in [np.uint16, np.float32]:
+            raise ValueError(
+                f"Unsupported single-channel image dtype: {image_array.dtype}. "
+                f"Supported dtypes: {sorted(str(d) for d in [np.uint16, np.float32])}."
+            )
+        return PIL.Image.fromarray(np.ascontiguousarray(image_array))
+
+    # 3D path: must be RGB (3 channels), channels-first or channels-last.
     if image_array.shape[0] == 3:
         # Transpose from pytorch convention (C, H, W) to (H, W, C)
         image_array = image_array.transpose(1, 2, 0)
@@ -71,13 +111,29 @@ def image_array_to_pil_image(image_array: np.ndarray, range_check: bool = True)
     return PIL.Image.fromarray(image_array)
 
 
+def save_kwargs_for_path(fpath: Path, compress_level: int) -> dict:
+    """Pick the right format-specific kwargs for :meth:`PIL.Image.Image.save`.
+
+    PNG uses ``compress_level`` (0-9, zlib). TIFF uses ``compression`` (raw) for lossless raw depth maps.
+    """
+    suffix = Path(fpath).suffix.lower()
+    if suffix == ".png":
+        return {"compress_level": compress_level}
+    if suffix in (".tif", ".tiff"):
+        return {"compression": "raw"}
+    else:
+        raise ValueError(f"Unsupported image file extension: {suffix}")
+
+
 def write_image(image: np.ndarray | PIL.Image.Image, fpath: Path, compress_level: int = 1):
     """
     Saves a NumPy array or PIL Image to a file.
 
     This function handles both NumPy arrays and PIL Image objects, converting
     the former to a PIL Image before saving. It includes error handling for
-    the save operation.
+    the save operation. The output format is inferred from the *fpath*
+    extension: ``.png`` → PNG with ``compress_level``, ``.tiff`` / ``.tif``
+    → lossless raw depth maps (TIFF).
 
     Args:
         image (np.ndarray | PIL.Image.Image): The image data to save.
@@ -101,7 +157,7 @@ def write_image(image: np.ndarray | PIL.Image.Image, fpath: Path, compress_level
             img = image
         else:
             raise TypeError(f"Unsupported image type: {type(image)}")
-        img.save(fpath, compress_level=compress_level)
+        img.save(fpath, **save_kwargs_for_path(fpath, compress_level))
     except Exception as e:
         logger.error("Error writing image %s: %s", fpath, e)
 

src/lerobot/datasets/io_utils.py

@@ -226,28 +226,50 @@ def load_image_as_numpy(
     Args:
         fpath (str | Path): Path to the image file.
         dtype (np.dtype): The desired data type of the output array. If floating,
-            pixels are scaled to [0, 1].
+            pixels are scaled to [0, 1]. Only used for RGB images.
         channel_first (bool): If True, converts the image to (C, H, W) format.
             Otherwise, it remains in (H, W, C) format.
 
     Returns:
         np.ndarray: The image as a numpy array.
     """
-    img = PILImage.open(fpath).convert("RGB")
-    img_array = np.array(img, dtype=dtype)
+    is_depth = fpath.endswith(".tiff") or fpath.endswith(".tif")
+    if is_depth:
+        # Preserve the native depth dtype (uint16 -> "I;16", float32 -> "F").
+        img = PILImage.open(fpath)
+        img_array = np.array(img)
+    else:
+        img = PILImage.open(fpath).convert("RGB")
+        img_array = np.array(img, dtype=dtype)
+        if np.issubdtype(dtype, np.floating):
+            img_array /= 255.0
     if channel_first:  # (H, W, C) -> (C, H, W)
-        img_array = np.transpose(img_array, (2, 0, 1))
-    if np.issubdtype(dtype, np.floating):
-        img_array /= 255.0
+        img_array = img_array[np.newaxis, ...] if img_array.ndim == 2 else np.transpose(img_array, (2, 0, 1))
     return img_array
 
 
+# PIL modes for 16-bit unsigned depth maps.
+UINT16_PIL_MODES = {"I;16", "I;16B", "I;16L"}
+
+
+def pil_to_chw_tensor(img: PILImage.Image) -> torch.Tensor:
+    """Convert a PIL image to a channel-first tensor.
+
+    ``uint16`` depth maps become ``float32 (1, H, W)`` in native units (``ToTensor``
+    would overflow them to ``int16``); all other modes use the standard ``ToTensor`` path.
+    """
+    if img.mode in UINT16_PIL_MODES:
+        return torch.from_numpy(np.array(img, dtype=np.float32))[None, ...]
+    return transforms.ToTensor()(img)
+
+
 def hf_transform_to_torch(items_dict: dict[str, list[Any]]) -> dict[str, list[torch.Tensor | str]]:
     """Convert a batch from a Hugging Face dataset to torch tensors.
 
     This transform function converts items from Hugging Face dataset format (pyarrow)
-    to torch tensors. Importantly, images are converted from PIL objects (H, W, C, uint8)
-    to a torch image representation (C, H, W, float32) in the range [0, 1]. Other
+    to torch tensors. RGB images are converted from PIL objects (H, W, C, uint8)
+    to a torch image representation (C, H, W, float32) in the range [0, 1]. Depth
+    maps are returned as float32 (1, H, W) in their native units. Other
     types are converted to torch.tensor.
 
     Args:
@@ -262,8 +284,7 @@ def hf_transform_to_torch(items_dict: dict[str, list[Any]]) -> dict[str, list[to
             continue
         first_item = items_dict[key][0]
         if isinstance(first_item, PILImage.Image):
-            to_tensor = transforms.ToTensor()
-            items_dict[key] = [to_tensor(img) for img in items_dict[key]]
+            items_dict[key] = [pil_to_chw_tensor(img) for img in items_dict[key]]
         elif first_item is None or isinstance(first_item, dict):
             pass
         else:
@@ -329,7 +350,11 @@ def item_to_torch(item: dict) -> dict:
     """
     skip_keys = {"task", *LANGUAGE_COLUMNS}
     for key, val in item.items():
-        if isinstance(val, (np.ndarray | list)) and key not in skip_keys:
+        if key in skip_keys:
+            continue
+        if isinstance(val, PILImage.Image):
+            item[key] = pil_to_chw_tensor(val)
+        elif isinstance(val, (np.ndarray | list)):
             # Convert numpy arrays and lists to torch tensors
             item[key] = torch.tensor(val)
     return item

src/lerobot/datasets/lerobot_dataset.py

@@ -24,7 +24,7 @@ import torch.utils
 from huggingface_hub import HfApi, snapshot_download
 from huggingface_hub.errors import RevisionNotFoundError
 
-from lerobot.configs import VideoEncoderConfig
+from lerobot.configs import DEFAULT_DEPTH_UNIT, DepthEncoderConfig, RGBEncoderConfig
 from lerobot.utils.constants import HF_LEROBOT_HUB_CACHE
 
 from .dataset_metadata import CODEBASE_VERSION, LeRobotDatasetMetadata
@@ -58,8 +58,10 @@ class LeRobotDataset(torch.utils.data.Dataset):
         download_videos: bool = True,
         video_backend: str | None = None,
         return_uint8: bool = False,
+        depth_output_unit: str = DEFAULT_DEPTH_UNIT,
         batch_encoding_size: int = 1,
-        camera_encoder: VideoEncoderConfig | None = None,
+        rgb_encoder: RGBEncoderConfig | None = None,
+        depth_encoder: DepthEncoderConfig | None = None,
         encoder_threads: int | None = None,
         streaming_encoding: bool = False,
         encoder_queue_maxsize: int = 30,
@@ -183,8 +185,11 @@ class LeRobotDataset(torch.utils.data.Dataset):
                 You can also use the 'pyav' decoder used by Torchvision, which used to be the default option, or 'video_reader' which is another decoder of Torchvision.
             batch_encoding_size (int, optional): Number of episodes to accumulate before batch encoding videos.
                 Set to 1 for immediate encoding (default), or higher for batched encoding. Defaults to 1.
-            camera_encoder (VideoEncoderConfig | None, optional): Video encoder settings for cameras
-                (codec, quality, etc.). When ``None``, :func:`~lerobot.configs.video.camera_encoder_defaults`
+            rgb_encoder (RGBEncoderConfig | None, optional): Video encoder settings for cameras
+                (codec, quality, etc.). When ``None``, :func:`~lerobot.configs.video.rgb_encoder_defaults`
+                is used by the writer.
+            depth_encoder (DepthEncoderConfig | None, optional): Video encoder settings for depth cameras
+                (codec, quality, etc.). When ``None``, :func:`~lerobot.configs.video.depth_encoder_defaults`
                 is used by the writer.
             encoder_threads (int | None, optional): Number of encoder threads (global). ``None`` lets the
                 codec decide.
@@ -206,6 +211,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         self.revision = revision if revision else CODEBASE_VERSION
         self._video_backend = video_backend if video_backend else get_safe_default_video_backend()
         self._return_uint8 = return_uint8
+        self._depth_output_unit = depth_output_unit
         self._batch_encoding_size = batch_encoding_size
         self._encoder_threads = encoder_threads
 
@@ -246,6 +252,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
             delta_timestamps=delta_timestamps,
             image_transforms=image_transforms,
             return_uint8=self._return_uint8,
+            depth_output_unit=self._depth_output_unit,
         )
         self.image_transforms = image_transforms
 
@@ -271,14 +278,16 @@ class LeRobotDataset(torch.utils.data.Dataset):
             if streaming_encoding and len(self.meta.video_keys) > 0:
                 streaming_enc = self._build_streaming_encoder(
                     self.meta.fps,
-                    camera_encoder,
+                    rgb_encoder,
+                    depth_encoder,
                     encoder_queue_maxsize,
                     encoder_threads,
                 )
             self.writer = DatasetWriter(
                 meta=self.meta,
                 root=self.root,
-                camera_encoder=camera_encoder,
+                rgb_encoder=rgb_encoder,
+                depth_encoder=depth_encoder,
                 encoder_threads=encoder_threads,
                 batch_encoding_size=batch_encoding_size,
                 streaming_encoder=streaming_enc,
@@ -314,19 +323,22 @@ class LeRobotDataset(torch.utils.data.Dataset):
                 delta_timestamps=self.delta_timestamps,
                 image_transforms=self.image_transforms,
                 return_uint8=self._return_uint8,
+                depth_output_unit=self._depth_output_unit,
             )
         return self.reader
 
     @staticmethod
     def _build_streaming_encoder(
         fps: int,
-        camera_encoder: VideoEncoderConfig | None,
+        rgb_encoder: RGBEncoderConfig | None,
+        depth_encoder: DepthEncoderConfig | None,
         encoder_queue_maxsize: int,
         encoder_threads: int | None,
     ) -> StreamingVideoEncoder:
         return StreamingVideoEncoder(
             fps=fps,
-            camera_encoder=camera_encoder,
+            rgb_encoder=rgb_encoder,
+            depth_encoder=depth_encoder,
             queue_maxsize=encoder_queue_maxsize,
             encoder_threads=encoder_threads,
         )
@@ -655,7 +667,8 @@ class LeRobotDataset(torch.utils.data.Dataset):
         image_writer_threads: int = 0,
         video_backend: str | None = None,
         batch_encoding_size: int = 1,
-        camera_encoder: VideoEncoderConfig | None = None,
+        rgb_encoder: RGBEncoderConfig | None = None,
+        depth_encoder: DepthEncoderConfig | None = None,
         metadata_buffer_size: int = 10,
         streaming_encoding: bool = False,
         encoder_queue_maxsize: int = 30,
@@ -686,8 +699,10 @@ class LeRobotDataset(torch.utils.data.Dataset):
             video_backend: Video decoding backend (used when reading back).
             batch_encoding_size: Number of episodes to accumulate before
                 batch-encoding videos. ``1`` means encode immediately.
-            camera_encoder: Video encoder settings for cameras (codec, quality, etc.).
-                When ``None``, :func:`~lerobot.configs.video.camera_encoder_defaults` is used.
+            rgb_encoder: Video encoder settings for cameras (codec, quality, etc.).
+                When ``None``, :func:`~lerobot.configs.video.rgb_encoder_defaults` is used.
+            depth_encoder: Video encoder settings for depth cameras (codec, quality, etc.).
+                When ``None``, :func:`~lerobot.configs.video.depth_encoder_defaults` is used.
             encoder_threads: Number of encoder threads (global). ``None``
                 lets the codec decide.
             metadata_buffer_size: Number of episode metadata records to buffer
@@ -722,6 +737,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         obj.episodes = None
         obj._video_backend = video_backend if video_backend is not None else get_safe_default_video_backend()
         obj._return_uint8 = False
+        obj._depth_output_unit = DEFAULT_DEPTH_UNIT
         obj._batch_encoding_size = batch_encoding_size
         obj._encoder_threads = encoder_threads
 
@@ -731,12 +747,13 @@ class LeRobotDataset(torch.utils.data.Dataset):
         streaming_enc = None
         if streaming_encoding and len(obj.meta.video_keys) > 0:
             streaming_enc = cls._build_streaming_encoder(
-                fps, camera_encoder, encoder_queue_maxsize, encoder_threads
+                fps, rgb_encoder, depth_encoder, encoder_queue_maxsize, encoder_threads
             )
         obj.writer = DatasetWriter(
             meta=obj.meta,
             root=obj.root,
-            camera_encoder=camera_encoder,
+            rgb_encoder=rgb_encoder,
+            depth_encoder=depth_encoder,
             encoder_threads=encoder_threads,
             batch_encoding_size=batch_encoding_size,
             streaming_encoder=streaming_enc,
@@ -759,7 +776,8 @@ class LeRobotDataset(torch.utils.data.Dataset):
         force_cache_sync: bool = False,
         video_backend: str | None = None,
         batch_encoding_size: int = 1,
-        camera_encoder: VideoEncoderConfig | None = None,
+        rgb_encoder: RGBEncoderConfig | None = None,
+        depth_encoder: DepthEncoderConfig | None = None,
         encoder_threads: int | None = None,
         image_writer_processes: int = 0,
         image_writer_threads: int = 0,
@@ -787,8 +805,10 @@ class LeRobotDataset(torch.utils.data.Dataset):
             video_backend: Video decoding backend for reading back data.
             batch_encoding_size: Number of episodes to accumulate before
                 batch-encoding videos.
-            camera_encoder: Video encoder settings for cameras (codec, quality, etc.).
-                When ``None``, :func:`~lerobot.configs.video.camera_encoder_defaults` is used.
+            rgb_encoder: Video encoder settings for cameras (codec, quality, etc.).
+                When ``None``, :func:`~lerobot.configs.video.rgb_encoder_defaults` is used.
+            depth_encoder: Video encoder settings for depth cameras (codec, quality, etc.).
+                When ``None``, :func:`~lerobot.configs.video.depth_encoder_defaults` is used.
             encoder_threads: Number of encoder threads (global). ``None``
                 lets the codec decide.
             image_writer_processes: Subprocesses for async image writing.
@@ -816,6 +836,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         obj.episodes = None
         obj._video_backend = video_backend if video_backend else get_safe_default_video_backend()
         obj._return_uint8 = False
+        obj._depth_output_unit = DEFAULT_DEPTH_UNIT
         obj._batch_encoding_size = batch_encoding_size
 
         if obj._requested_root is not None:
@@ -835,12 +856,13 @@ class LeRobotDataset(torch.utils.data.Dataset):
         streaming_enc = None
         if streaming_encoding and len(obj.meta.video_keys) > 0:
             streaming_enc = cls._build_streaming_encoder(
-                obj.meta.fps, camera_encoder, encoder_queue_maxsize, encoder_threads
+                obj.meta.fps, rgb_encoder, depth_encoder, encoder_queue_maxsize, encoder_threads
             )
         obj.writer = DatasetWriter(
             meta=obj.meta,
             root=obj.root,
-            camera_encoder=camera_encoder,
+            rgb_encoder=rgb_encoder,
+            depth_encoder=depth_encoder,
             encoder_threads=encoder_threads,
             batch_encoding_size=batch_encoding_size,
             streaming_encoder=streaming_enc,

src/lerobot/datasets/pyav_utils.py

@@ -24,6 +24,7 @@ import logging
 from typing import Any
 
 import av
+import numpy as np
 
 logger = logging.getLogger(__name__)
 
@@ -31,6 +32,34 @@ FFMPEG_NUMERIC_OPTION_TYPES = ("INT", "INT64", "UINT64", "FLOAT", "DOUBLE")
 FFMPEG_INTEGER_OPTION_TYPES = ("INT", "INT64", "UINT64")
 
 
+def write_u16_plane(plane: av.video.plane.VideoPlane, src: np.ndarray, fill_value: int | None = None) -> None:
+    """Copy a 2D ``uint16`` image into the plane's memory buffer, row by row.
+
+    For speed, each row is padded to a wider size than ``width``, so the true row width in
+    memory is ``plane.line_size`` (bytes), not ``width``. Copying as one straight stream
+    would skew the image, so we write only the first ``width`` columns of each row and
+    leave the padding untouched.
+
+    Args:
+        plane: Destination 16-bit plane.
+        src: Source image, shape ``(height, width)``, dtype ``uint16``.
+        fill_value: If given, every pixel (padding included) is set to this first, so the
+            padding holds clean data instead of garbage.
+    """
+    height, width = src.shape
+    stride_u16 = plane.line_size // np.dtype(np.uint16).itemsize
+    dst = np.frombuffer(plane, dtype=np.uint16).reshape(height, stride_u16)
+    if fill_value is not None:
+        dst.fill(fill_value)
+    dst[:, :width] = src
+
+
+@functools.cache
+def get_pix_fmt_channels(pix_fmt: str) -> int:
+    """Return the number of components (channels) for *pix_fmt*."""
+    return len(av.VideoFormat(pix_fmt).components)
+
+
 @functools.cache
 def get_codec(vcodec: str) -> av.codec.Codec | None:
     """PyAV write-mode ``Codec`` for *vcodec*, or ``None`` if unavailable."""
@@ -92,7 +121,7 @@ def _check_option_value(vcodec: str, label: str, value: Any, opt: av.option.Opti
                     f"{label}={value!r} is not numeric; codec {vcodec!r} expects a number for this option."
                 ) from e
         elif isinstance(value, (float, int)):
-            num_val = value
+            num_val = float(value)
         else:
             raise ValueError(
                 f"{label}={value!r} is not numeric; codec {vcodec!r} expects a number for this option."
@@ -142,6 +171,16 @@ def _check_pixel_format(vcodec: str, pix_fmt: str) -> None:
         )
 
 
+def _check_pix_fmt_channels(pix_fmt: str, channels: int) -> None:
+    """Ensure *pix_fmt* can carry at least *channels* components."""
+    pix_fmt_channels = get_pix_fmt_channels(pix_fmt)
+    if pix_fmt_channels < channels:
+        raise ValueError(
+            f"pix_fmt={pix_fmt!r} carries only {pix_fmt_channels} component(s) "
+            f"but the source data has {channels} channel(s)."
+        )
+
+
 def _check_codec_options(vcodec: str, codec_options: dict[str, Any]) -> None:
     """Validate merged encoder options (typed) against the codec's published AVOptions."""
     supported_options = _get_codec_options_by_name(vcodec)
@@ -156,12 +195,18 @@ def _check_codec_options(vcodec: str, codec_options: dict[str, Any]) -> None:
         _check_option_value(vcodec, key, value, supported_options[key])
 
 
-def check_video_encoder_parameters_pyav(vcodec: str, pix_fmt: str, codec_options: dict[str, Any]) -> None:
+def check_video_encoder_parameters_pyav(
+    vcodec: str,
+    pix_fmt: str,
+    codec_options: dict[str, Any],
+    channels: int | None = None,
+) -> None:
     """Verify *config* is compatible with the bundled FFmpeg build.
 
     Checks pixel format, abstract tuning-field compatibility, and each merged
     encoder option from :meth:`~lerobot.configs.video.VideoEncoderConfig.get_codec_options`
     against PyAV (including numeric ``extra_options`` present in that dict).
+    When given, additionally verify that *pix_fmt* carries as many components as the source data channels.
     No-op when ``config.vcodec`` isn't in the local FFmpeg build.
 
     Raises:
@@ -171,4 +216,6 @@ def check_video_encoder_parameters_pyav(vcodec: str, pix_fmt: str, codec_options
     if not options:
         raise ValueError(f"Codec {vcodec!r} is not available in the bundled FFmpeg build")
     _check_pixel_format(vcodec, pix_fmt)
+    if channels is not None:
+        _check_pix_fmt_channels(pix_fmt, channels)
     _check_codec_options(vcodec, codec_options)

src/lerobot/datasets/streaming_dataset.py

@@ -22,9 +22,11 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
+from lerobot.configs import DEFAULT_DEPTH_UNIT, DepthEncoderConfig
 from lerobot.utils.constants import HF_LEROBOT_HOME, LOOKAHEAD_BACKTRACKTABLE, LOOKBACK_BACKTRACKTABLE
 
 from .dataset_metadata import CODEBASE_VERSION, LeRobotDatasetMetadata
+from .depth_utils import dequantize_depth
 from .feature_utils import get_delta_indices
 from .io_utils import item_to_torch
 from .utils import (
@@ -35,6 +37,7 @@ from .utils import (
 )
 from .video_utils import (
     VideoDecoderCache,
+    decode_video_frames,
     decode_video_frames_torchcodec,
 )
 
@@ -252,6 +255,7 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
         rng: np.random.Generator | None = None,
         shuffle: bool = True,
         return_uint8: bool = False,
+        depth_output_unit: str = DEFAULT_DEPTH_UNIT,
     ):
         """Initialize a StreamingLeRobotDataset.
 
@@ -272,6 +276,8 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
             seed (int, optional): Reproducibility random seed.
             rng (np.random.Generator | None, optional): Random number generator.
             shuffle (bool, optional): Whether to shuffle the dataset across exhaustions. Defaults to True.
+            depth_output_unit (str, optional): Physical unit depth maps are dequantized to ("m" or "mm").
+                Defaults to "mm".
         """
         super().__init__()
         self.repo_id = repo_id
@@ -290,6 +296,7 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
         self.streaming = streaming
         self.buffer_size = buffer_size
         self._return_uint8 = return_uint8
+        self._depth_output_unit = depth_output_unit
 
         # We cache the video decoders to avoid re-initializing them at each frame (avoiding a ~10x slowdown)
         self.video_decoder_cache = None
@@ -306,6 +313,11 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
         # Check version
         check_version_compatibility(self.repo_id, self.meta._version, CODEBASE_VERSION)
 
+        self._depth_encoder_configs: dict[str, DepthEncoderConfig] = {
+            vid_key: DepthEncoderConfig.from_video_info(self.meta.features[vid_key].get("info"))
+            for vid_key in self.meta.depth_keys
+        }
+
         self.delta_timestamps = None
         self.delta_indices = None
 
@@ -554,13 +566,34 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
         for video_key, query_ts in query_timestamps.items():
             root = self.meta.url_root if self.streaming and not self.streaming_from_local else self.root
             video_path = f"{root}/{self.meta.get_video_file_path(ep_idx, video_key)}"
-            frames = decode_video_frames_torchcodec(
-                video_path,
-                query_ts,
-                self.tolerance_s,
-                decoder_cache=self.video_decoder_cache,
-                return_uint8=self._return_uint8,
-            )
+            if video_key in self.meta.depth_keys:
+                # Depth maps are 12-bit quantized and only decodable via pyav; dequantize back
+                # to physical units to match the non-streaming reader.
+                frames = decode_video_frames(
+                    video_path,
+                    query_ts,
+                    self.tolerance_s,
+                    backend="pyav",
+                    return_uint8=False,
+                    is_depth=True,
+                )
+                depth_encoder = self._depth_encoder_configs[video_key]
+                frames = dequantize_depth(
+                    frames,
+                    depth_min=depth_encoder.depth_min,
+                    depth_max=depth_encoder.depth_max,
+                    shift=depth_encoder.shift,
+                    use_log=depth_encoder.use_log,
+                    output_unit=self._depth_output_unit,
+                )
+            else:
+                frames = decode_video_frames_torchcodec(
+                    video_path,
+                    query_ts,
+                    self.tolerance_s,
+                    decoder_cache=self.video_decoder_cache,
+                    return_uint8=self._return_uint8,
+                )
 
             item[video_key] = frames.squeeze(0) if len(query_ts) == 1 else frames
 

src/lerobot/datasets/utils.py

@@ -87,11 +87,14 @@ DATA_DIR = "data"
 VIDEO_DIR = "videos"
 
 CHUNK_FILE_PATTERN = "chunk-{chunk_index:03d}/file-{file_index:03d}"
+IMAGE_FILE_PATTERN = "frame-{frame_index:06d}.png"
+DEPTH_FILE_PATTERN = "frame-{frame_index:06d}.tiff"
 DEFAULT_TASKS_PATH = "meta/tasks.parquet"
 DEFAULT_EPISODES_PATH = EPISODES_DIR + "/" + CHUNK_FILE_PATTERN + ".parquet"
 DEFAULT_DATA_PATH = DATA_DIR + "/" + CHUNK_FILE_PATTERN + ".parquet"
 DEFAULT_VIDEO_PATH = VIDEO_DIR + "/{video_key}/" + CHUNK_FILE_PATTERN + ".mp4"
-DEFAULT_IMAGE_PATH = "images/{image_key}/episode-{episode_index:06d}/frame-{frame_index:06d}.png"
+DEFAULT_IMAGE_PATH = "images/{image_key}/episode-{episode_index:06d}/" + IMAGE_FILE_PATTERN
+DEFAULT_DEPTH_PATH = "images/{image_key}/episode-{episode_index:06d}/" + DEPTH_FILE_PATTERN
 
 LEGACY_EPISODES_PATH = "meta/episodes.jsonl"
 LEGACY_EPISODES_STATS_PATH = "meta/episodes_stats.jsonl"

src/lerobot/datasets/video_utils.py

@@ -39,11 +39,17 @@ from datasets.features.features import register_feature
 from PIL import Image
 
 from lerobot.configs import (
+    DepthEncoderConfig,
+    RGBEncoderConfig,
     VideoEncoderConfig,
-    camera_encoder_defaults,
+    depth_encoder_defaults,
+    rgb_encoder_defaults,
 )
 from lerobot.utils.import_utils import get_safe_default_video_backend
 
+from .depth_utils import quantize_depth
+from .pyav_utils import get_pix_fmt_channels
+
 logger = logging.getLogger(__name__)
 
 
@@ -53,6 +59,7 @@ def decode_video_frames(
     tolerance_s: float,
     backend: str | None = None,
     return_uint8: bool = False,
+    is_depth: bool = False,
 ) -> torch.Tensor:
     """
     Decodes video frames using the specified backend.
@@ -64,23 +71,35 @@ def decode_video_frames(
         backend (str, optional): Backend to use for decoding. Defaults to "torchcodec" when available
             in the platform; otherwise, defaults to "pyav". The legacy value "video_reader" is
             accepted for one release as an alias for "pyav" and will be removed in a future version.
-        return_uint8 (bool): If True, return raw uint8 frames without float32 normalization.
+        return_uint8 (bool): For RGB videos, if True return raw uint8 frames without float32 normalization.
             This reduces memory for DataLoader IPC; normalization can be done on GPU afterward.
+        is_depth (bool): Set to True if the video is a depth map (1 channel, uint12).
 
     Returns:
-        torch.Tensor: Decoded frames (float32 in [0,1] by default, or uint8 if return_uint8=True).
+        torch.Tensor: Decoded frames (RGB: float32 in [0,1] by default, or uint8 if return_uint8=True, Depth: uint12).
 
     Currently supports torchcodec on cpu and pyav.
     """
+    if backend != "pyav" and is_depth:
+        logger.debug("Decoding depth maps is only supported with the 'pyav' backend, falling back to pyav.")
+        # We do not actually return uint8 here, but we avoid the 255 normalization step.
+        return decode_video_frames_pyav(
+            video_path, timestamps, tolerance_s, return_uint8=False, is_depth=True
+        )
+
     if backend is None:
         backend = get_safe_default_video_backend()
     if backend == "torchcodec":
         return decode_video_frames_torchcodec(video_path, timestamps, tolerance_s, return_uint8=return_uint8)
     elif backend == "pyav":
-        return decode_video_frames_pyav(video_path, timestamps, tolerance_s, return_uint8=return_uint8)
+        return decode_video_frames_pyav(
+            video_path, timestamps, tolerance_s, return_uint8=return_uint8, is_depth=is_depth
+        )
     elif backend == "video_reader":
         logger.warning("backend='video_reader' is deprecated and now aliases to 'pyav'.")
-        return decode_video_frames_pyav(video_path, timestamps, tolerance_s, return_uint8=return_uint8)
+        return decode_video_frames_pyav(
+            video_path, timestamps, tolerance_s, return_uint8=return_uint8, is_depth=is_depth
+        )
     else:
         raise ValueError(f"Unsupported video backend: {backend}")
 
@@ -91,6 +110,7 @@ def decode_video_frames_pyav(
     tolerance_s: float,
     log_loaded_timestamps: bool = False,
     return_uint8: bool = False,
+    is_depth: bool = False,
 ) -> torch.Tensor:
     """Loads frames associated to the requested timestamps of a video using PyAV.
 
@@ -109,8 +129,9 @@ def decode_video_frames_pyav(
         tolerance_s: Allowed deviation in seconds between a queried timestamp and the closest
             decoded frame.
         log_loaded_timestamps: When True, log every decoded frame's timestamp at INFO level.
-        return_uint8: When True, return raw uint8 frames (C, H, W). Otherwise, return float32 in
-            [0, 1] range.
+        return_uint8: For RGB videos, if True return raw uint8 frames (C, H, W).
+            Otherwise, return float32 in [0, 1] range.
+        is_depth: Set to True if the video is a depth map (1 channel, uint12).
 
     Returns:
         torch.Tensor of shape (len(timestamps), C, H, W).
@@ -132,7 +153,13 @@ def decode_video_frames_pyav(
     # https://pyav.basswood-io.com/docs/stable/api/container.html#av.container.InputContainer.seek
     with av.open(video_path) as container:
         stream = container.streams.video[0]
-        container.seek(int(first_ts * av.time_base), backward=True)
+        # Seek to the nearest keyframe at or before `first_ts` with a 1 frame margin
+        container.seek(
+            round(first_ts / stream.time_base) - 1,
+            backward=True,
+            any_frame=False,
+            stream=stream,
+        )
 
         for frame in container.decode(stream):
             if frame.pts is None:
@@ -140,9 +167,13 @@ def decode_video_frames_pyav(
             current_ts = float(frame.pts * stream.time_base)
             if log_loaded_timestamps:
                 logger.info(f"frame loaded at timestamp={current_ts:.4f}")
-            # Convert to CHW uint8 to match torchcodec's output layout.
-            arr = frame.to_ndarray(format="rgb24")  # H, W, 3
-            loaded_frames.append(torch.from_numpy(arr).permute(2, 0, 1).contiguous())
+            if is_depth:
+                arr = frame.to_ndarray(format="gray12le")  # (H, W) uint12
+                loaded_frames.append(torch.from_numpy(arr).unsqueeze(0).contiguous())
+            else:
+                arr = frame.to_ndarray(format="rgb24")  # (H, W, 3)
+                # Convert to CHW uint8 to match torchcodec's output layout.
+                loaded_frames.append(torch.from_numpy(arr).permute(2, 0, 1).contiguous())
             loaded_ts.append(current_ts)
             if current_ts >= last_ts:
                 break
@@ -185,7 +216,7 @@ def decode_video_frames_pyav(
             f"number of queried timestamps ({len(timestamps)})"
         )
 
-    if return_uint8:
+    if return_uint8 or is_depth:
         return closest_frames
 
     # convert to the pytorch format which is float32 in [0,1] range (and channel first)
@@ -406,17 +437,38 @@ def encode_video_frames(
     imgs_dir: Path | str,
     video_path: Path | str,
     fps: int,
-    camera_encoder: VideoEncoderConfig | None = None,
+    video_encoder: VideoEncoderConfig | None = None,
     encoder_threads: int | None = None,
     *,
     log_level: int | None = av.logging.WARNING,
     overwrite: bool = False,
 ) -> None:
-    """More info on ffmpeg arguments tuning on `benchmark/video/README.md`"""
-    if camera_encoder is None:
-        camera_encoder = camera_encoder_defaults()
-    vcodec = camera_encoder.vcodec
-    pix_fmt = camera_encoder.pix_fmt
+    """Encode a directory of image frames into an MP4 video.
+
+    When ``video_encoder`` is a :class:`~lerobot.configs.video.DepthEncoderConfig`,
+    frames are read from ``.tiff`` files and quantized to 12-bit depth codes using the
+    encoder's ``depth_min`` / ``depth_max`` / ``shift`` / ``use_log``; otherwise ``.png``
+    RGB frames are encoded directly.
+
+    Args:
+        imgs_dir: Directory containing the frames to encode, named ``frame-000000``
+            onwards (``.png`` for RGB, ``.tiff`` for depth).
+        video_path: Output path for the encoded ``.mp4`` file.
+        fps: Frame rate of the output video.
+        video_encoder: Encoder settings (codec, pixel format, quality, ...). When
+            ``None``, :func:`rgb_encoder_defaults` is used. Pass a
+            :class:`~lerobot.configs.video.DepthEncoderConfig` to encode depth frames.
+        encoder_threads: Per-encoder thread count forwarded to the codec. ``None``
+            lets the codec decide.
+        log_level: libav log level to set while encoding, or ``None`` to leave the
+            current logging configuration unchanged.
+        overwrite: When ``False`` and ``video_path`` already exists, skip encoding and
+            log a warning. When ``True``, re-encode and replace the existing file.
+    """
+    if video_encoder is None:
+        video_encoder = rgb_encoder_defaults()
+    vcodec = video_encoder.vcodec
+    pix_fmt = video_encoder.pix_fmt
 
     video_path = Path(video_path)
     imgs_dir = Path(imgs_dir)
@@ -428,17 +480,19 @@ def encode_video_frames(
     video_path.parent.mkdir(parents=True, exist_ok=True)
 
     # Get input frames
-    template = "frame-" + ("[0-9]" * 6) + ".png"
+    is_depth = isinstance(video_encoder, DepthEncoderConfig)
+    suffix = ".png" if not is_depth else ".tiff"
+    template = "frame-" + ("[0-9]" * 6) + suffix
     input_list = sorted(
         glob.glob(str(imgs_dir / template)), key=lambda x: int(x.split("-")[-1].split(".")[0])
     )
 
     if len(input_list) == 0:
-        raise FileNotFoundError(f"No images found in {imgs_dir}.")
+        raise FileNotFoundError(f"No images with suffix {suffix} found in {imgs_dir}.")
     with Image.open(input_list[0]) as dummy_image:
         width, height = dummy_image.size
 
-    video_options = camera_encoder.get_codec_options(encoder_threads, as_strings=True)
+    video_options = video_encoder.get_codec_options(encoder_threads, as_strings=True)
 
     # Set logging level
     if log_level is not None:
@@ -455,8 +509,19 @@ def encode_video_frames(
         # Loop through input frames and encode them
         for input_data in input_list:
             with Image.open(input_data) as input_image:
-                input_image = input_image.convert("RGB")
-                input_frame = av.VideoFrame.from_image(input_image)
+                if is_depth:
+                    input_frame = quantize_depth(
+                        np.array(input_image),
+                        depth_min=video_encoder.depth_min,
+                        depth_max=video_encoder.depth_max,
+                        shift=video_encoder.shift,
+                        use_log=video_encoder.use_log,
+                        pix_fmt=video_encoder.pix_fmt,
+                        video_backend="pyav",
+                    )
+                else:
+                    input_image = input_image.convert("RGB")
+                    input_frame = av.VideoFrame.from_image(input_image)
                 packet = output_stream.encode(input_frame)
                 if packet:
                     output.mux(packet)
@@ -477,7 +542,7 @@ def encode_video_frames(
 def reencode_video(
     input_video_path: Path | str,
     output_video_path: Path | str,
-    camera_encoder: VideoEncoderConfig | None = None,
+    video_encoder: VideoEncoderConfig | None = None,
     encoder_threads: int | None = None,
     log_level: int | None = av.logging.WARNING,
     overwrite: bool = False,
@@ -489,7 +554,7 @@ def reencode_video(
     Args:
         input_video_path: Existing video file to read.
         output_video_path: Path for the re-encoded file.
-        camera_encoder: Encoder configuration. Defaults to :func:`camera_encoder_defaults`.
+        video_encoder: Encoder configuration. Defaults to :func:`rgb_encoder_defaults`.
         encoder_threads: Optional thread count forwarded to :meth:`VideoEncoderConfig.get_codec_options`.
         log_level: libav log level while encoding, or ``None`` to leave logging unchanged. Defaults to WARNING.
         overwrite: When ``False`` and ``output_video_path`` already exists, skip and log a warning.
@@ -497,7 +562,7 @@ def reencode_video(
         end_time_s: When set, trim the output to end at this timestamp (seconds, exclusive).
     """
 
-    camera_encoder = camera_encoder or camera_encoder_defaults()
+    video_encoder = video_encoder or rgb_encoder_defaults()
 
     if (start_time_s is not None and start_time_s < 0) or (end_time_s is not None and end_time_s < 0):
         raise ValueError(f"Trim times must be non-negative, got start={start_time_s}, end={end_time_s}.")
@@ -512,9 +577,9 @@ def reencode_video(
 
     output_video_path.parent.mkdir(parents=True, exist_ok=True)
 
-    video_options = camera_encoder.get_codec_options(encoder_threads, as_strings=True)
-    vcodec = camera_encoder.vcodec
-    pix_fmt = camera_encoder.pix_fmt
+    video_options = video_encoder.get_codec_options(encoder_threads, as_strings=True)
+    vcodec = video_encoder.vcodec
+    pix_fmt = video_encoder.pix_fmt
 
     with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp_named_file:
         tmp_output_video_path = tmp_named_file.name
@@ -696,22 +761,21 @@ class _CameraEncoderThread(threading.Thread):
         self,
         video_path: Path,
         fps: int,
-        vcodec: str,
-        pix_fmt: str,
-        codec_options: dict[str, str],
+        video_encoder: VideoEncoderConfig,
         frame_queue: queue.Queue,
         result_queue: queue.Queue,
         stop_event: threading.Event,
+        encoder_threads: int | None = None,
     ):
         super().__init__(daemon=True)
         self.video_path = video_path
         self.fps = fps
-        self.vcodec = vcodec
-        self.pix_fmt = pix_fmt
-        self.codec_options = codec_options
+        self.video_encoder = video_encoder
+        self.is_depth = isinstance(video_encoder, DepthEncoderConfig)
         self.frame_queue = frame_queue
         self.result_queue = result_queue
         self.stop_event = stop_event
+        self.encoder_threads = encoder_threads
 
     def run(self) -> None:
         from .compute_stats import RunningQuantileStats, auto_downsample_height_width
@@ -736,12 +800,12 @@ class _CameraEncoderThread(threading.Thread):
                     # Sentinel: flush and close
                     break
 
-                # Ensure HWC uint8 numpy array
+                # Ensure HWC (RGB or depth) uint8 (RGB only) numpy array
                 if isinstance(frame_data, np.ndarray):
-                    if frame_data.ndim == 3 and frame_data.shape[0] == 3:
+                    if frame_data.ndim == 3 and frame_data.shape[0] in (1, 3):
                         # CHW -> HWC
                         frame_data = frame_data.transpose(1, 2, 0)
-                    if frame_data.dtype != np.uint8:
+                    if not self.is_depth and frame_data.dtype != np.uint8:
                         frame_data = (frame_data * 255).astype(np.uint8)
 
                 # Open container on first frame (to get width/height)
@@ -749,15 +813,29 @@ class _CameraEncoderThread(threading.Thread):
                     height, width = frame_data.shape[:2]
                     Path(self.video_path).parent.mkdir(parents=True, exist_ok=True)
                     container = av.open(str(self.video_path), "w")
-                    output_stream = container.add_stream(self.vcodec, self.fps, options=self.codec_options)
-                    output_stream.pix_fmt = self.pix_fmt
+                    output_stream = container.add_stream(
+                        self.video_encoder.vcodec,
+                        self.fps,
+                        options=self.video_encoder.get_codec_options(self.encoder_threads, as_strings=True),
+                    )
+                    output_stream.pix_fmt = self.video_encoder.pix_fmt
                     output_stream.width = width
                     output_stream.height = height
                     output_stream.time_base = Fraction(1, self.fps)
 
                 # Encode frame with explicit timestamps
-                pil_img = Image.fromarray(frame_data)
-                video_frame = av.VideoFrame.from_image(pil_img)
+                if not self.is_depth:
+                    pil_img = Image.fromarray(frame_data)
+                    video_frame = av.VideoFrame.from_image(pil_img)
+                else:
+                    video_frame = quantize_depth(
+                        frame_data,
+                        depth_min=self.video_encoder.depth_min,
+                        depth_max=self.video_encoder.depth_max,
+                        shift=self.video_encoder.shift,
+                        use_log=self.video_encoder.use_log,
+                        video_backend=self.video_encoder.video_backend,
+                    )
                 video_frame.pts = frame_count
                 video_frame.time_base = Fraction(1, self.fps)
                 packet = output_stream.encode(video_frame)
@@ -815,22 +893,27 @@ class StreamingVideoEncoder:
     def __init__(
         self,
         fps: int,
-        camera_encoder: VideoEncoderConfig | None = None,
+        rgb_encoder: RGBEncoderConfig | None = None,
+        depth_encoder: DepthEncoderConfig | None = None,
         queue_maxsize: int = 30,
         encoder_threads: int | None = None,
     ):
         """
         Args:
             fps: Frames per second for the output videos.
-            camera_encoder: Video encoder settings applied to all cameras.
-                When ``None``, :func:`camera_encoder_defaults` is used.
-            encoder_threads: Number of encoder threads (global setting).
-                ``None`` lets the codec decide.
+            rgb_encoder: Video encoder settings applied to all RGB cameras.
+                When ``None``, :func:`rgb_encoder_defaults` is used.
+            depth_encoder: Video encoder settings applied to all depth cameras,
+                including the depth quantization parameters. When ``None``,
+                :func:`depth_encoder_defaults` is used.
             queue_maxsize: Max frames to buffer per camera before
                 back-pressure drops frames.
+            encoder_threads: Number of encoder threads (global setting).
+                ``None`` lets the codec decide.
         """
         self.fps = fps
-        self._camera_encoder = camera_encoder or camera_encoder_defaults()
+        self._rgb_encoder = rgb_encoder or rgb_encoder_defaults()
+        self._depth_encoder = depth_encoder or depth_encoder_defaults()
         self._encoder_threads = encoder_threads
         self.queue_maxsize = queue_maxsize
 
@@ -843,18 +926,25 @@ class StreamingVideoEncoder:
         self._episode_active = False
         self._closed = False
 
-    def start_episode(self, video_keys: list[str], temp_dir: Path) -> None:
+    def start_episode(
+        self, video_keys: list[str], temp_dir: Path, depth_video_keys: list[str] | None = None
+    ) -> None:
         """Start encoder threads for a new episode.
 
         Args:
             video_keys: List of video feature keys (e.g. ["observation.images.laptop"])
             temp_dir: Base directory for temporary MP4 files
+            depth_video_keys: List of video or image feature keys that carry depth maps (e.g.
+                ["observation.images.laptop_depth"]).  Defaults to ``[]`` (no depth keys).
         """
         if self._episode_active:
             self.cancel_episode()
 
         self._dropped_frames.clear()
 
+        if depth_video_keys is None:
+            depth_video_keys = []
+
         for video_key in video_keys:
             frame_queue: queue.Queue = queue.Queue(maxsize=self.queue_maxsize)
             result_queue: queue.Queue = queue.Queue(maxsize=1)
@@ -863,17 +953,15 @@ class StreamingVideoEncoder:
             temp_video_dir = Path(tempfile.mkdtemp(dir=temp_dir))
             video_path = temp_video_dir / f"{video_key.replace('/', '_')}_streaming.mp4"
 
-            vcodec = self._camera_encoder.vcodec
-            codec_options = self._camera_encoder.get_codec_options(self._encoder_threads, as_strings=True)
+            encoder = self._depth_encoder if video_key in depth_video_keys else self._rgb_encoder
             encoder_thread = _CameraEncoderThread(
                 video_path=video_path,
                 fps=self.fps,
-                vcodec=vcodec,
-                pix_fmt=self._camera_encoder.pix_fmt,
-                codec_options=codec_options,
+                video_encoder=encoder,
                 frame_queue=frame_queue,
                 result_queue=result_queue,
                 stop_event=stop_event,
+                encoder_threads=self._encoder_threads,
             )
             encoder_thread.start()
 
@@ -1080,15 +1168,23 @@ def get_audio_info(video_path: Path | str) -> dict:
 
 def get_video_info(
     video_path: Path | str,
-    camera_encoder: VideoEncoderConfig | None = None,
+    video_encoder: VideoEncoderConfig | None = None,
 ) -> dict:
     """Build the ``video.*`` / ``audio.*`` info dict persisted in ``info.json``.
 
     Args:
         video_path: Path to the encoded video file to probe.
-        camera_encoder: If provided, record the exact encoder settings used to encode this
+        video_encoder: If provided, record the exact encoder settings used to encode this
             video. Stream-derived values take precedence — encoder fields are only written for keys
-            not already populated from the video file itself.
+            not already populated from the video file itself. When a
+            :class:`~lerobot.configs.video.DepthEncoderConfig` is passed, the depth
+            quantization parameters (``depth_min`` / ``depth_max`` / ``shift`` /
+            ``use_log``) are recorded so frames can be dequantized on read.
+
+    Returns:
+        The ``video.*`` / ``audio.*`` info dict, including ``is_depth_map`` which is
+        ``True`` only when ``video_encoder`` is a
+        :class:`~lerobot.configs.video.DepthEncoderConfig`.
     """
     logging.getLogger("libav").setLevel(av.logging.WARNING)
 
@@ -1106,13 +1202,10 @@ def get_video_info(
         video_info["video.width"] = video_stream.width
         video_info["video.codec"] = video_stream.codec.canonical_name
         video_info["video.pix_fmt"] = video_stream.pix_fmt
-        video_info["video.is_depth_map"] = False
 
         # Calculate fps from r_frame_rate
         video_info["video.fps"] = int(video_stream.base_rate)
-
-        pixel_channels = get_video_pixel_channels(video_stream.pix_fmt)
-        video_info["video.channels"] = pixel_channels
+        video_info["video.channels"] = get_pix_fmt_channels(video_stream.pix_fmt)
 
     # Reset logging level
     av.logging.restore_default_callback()
@@ -1121,27 +1214,18 @@ def get_video_info(
     video_info.update(**get_audio_info(video_path))
 
     # Add additional encoder configuration if provided
-    if camera_encoder is not None:
-        for field_name, field_value in asdict(camera_encoder).items():
+    if video_encoder is not None:
+        for field_name, field_value in asdict(video_encoder).items():
             # vcodec is already populated from the video stream
             if field_name == "vcodec":
                 continue
             video_info.setdefault(f"video.{field_name}", field_value)
 
+    video_info["is_depth_map"] = isinstance(video_encoder, DepthEncoderConfig)
+
     return video_info
 
 
-def get_video_pixel_channels(pix_fmt: str) -> int:
-    if "gray" in pix_fmt or "depth" in pix_fmt or "monochrome" in pix_fmt:
-        return 1
-    elif "rgba" in pix_fmt or "yuva" in pix_fmt:
-        return 4
-    elif "rgb" in pix_fmt or "yuv" in pix_fmt:
-        return 3
-    else:
-        raise ValueError("Unknown format")
-
-
 def get_video_duration_in_s(video_path: Path | str) -> float:
     """
     Get the duration of a video file in seconds using PyAV.
@@ -1202,10 +1286,13 @@ class VideoEncodingManager:
         img_dir = self.dataset.root / "images"
         if img_dir.exists():
             png_files = list(img_dir.rglob("*.png"))
-            if len(png_files) == 0:
+            tiff_files = list(img_dir.rglob("*.tiff"))
+            if len(png_files) == 0 and len(tiff_files) == 0:
                 shutil.rmtree(img_dir)
                 logger.debug("Cleaned up empty images directory")
             else:
-                logger.debug(f"Images directory is not empty, containing {len(png_files)} PNG files")
+                logger.debug(
+                    f"Images directory is not empty, containing {len(png_files)} PNG and {len(tiff_files)} TIFF files"
+                )
 
         return False  # Don't suppress the original exception

src/lerobot/policies/factory.py

@@ -281,26 +281,23 @@ def make_pre_post_processors(
             policy configuration type.
     """
     if pretrained_path:
-        # TODO(Steven): Temporary patch, implement correctly the processors for Gr00t
         if isinstance(policy_cfg, GrootConfig):
-            # GROOT handles normalization in groot_pack_inputs_v3 step
-            # Need to override both stats AND normalize_min_max since saved config might be empty
-            preprocessor_overrides = {}
-            postprocessor_overrides = {}
-            preprocessor_overrides["groot_pack_inputs_v3"] = {
-                "stats": kwargs.get("dataset_stats"),
-                "normalize_min_max": True,
-            }
+            from .groot.processor_groot import make_groot_pre_post_processors_from_pretrained
 
-            # Also ensure postprocessing slices to env action dim and unnormalizes with dataset stats
-            env_action_dim = policy_cfg.output_features[ACTION].shape[0]
-            postprocessor_overrides["groot_action_unpack_unnormalize_v1"] = {
-                "stats": kwargs.get("dataset_stats"),
-                "normalize_min_max": True,
-                "env_action_dim": env_action_dim,
-            }
-            kwargs["preprocessor_overrides"] = preprocessor_overrides
-            kwargs["postprocessor_overrides"] = postprocessor_overrides
+            return make_groot_pre_post_processors_from_pretrained(
+                config=policy_cfg,
+                pretrained_path=pretrained_path,
+                dataset_stats=kwargs.get("dataset_stats"),
+                dataset_meta=kwargs.get("dataset_meta"),
+                preprocessor_overrides=kwargs.get("preprocessor_overrides"),
+                postprocessor_overrides=kwargs.get("postprocessor_overrides"),
+                preprocessor_config_filename=kwargs.get(
+                    "preprocessor_config_filename", f"{POLICY_PREPROCESSOR_DEFAULT_NAME}.json"
+                ),
+                postprocessor_config_filename=kwargs.get(
+                    "postprocessor_config_filename", f"{POLICY_POSTPROCESSOR_DEFAULT_NAME}.json"
+                ),
+            )
 
         preprocessor = PolicyProcessorPipeline.from_pretrained(
             pretrained_model_name_or_path=pretrained_path,
@@ -406,6 +403,7 @@ def make_pre_post_processors(
         processors = make_groot_pre_post_processors(
             config=policy_cfg,
             dataset_stats=kwargs.get("dataset_stats"),
+            dataset_meta=kwargs.get("dataset_meta"),
         )
 
     elif isinstance(policy_cfg, XVLAConfig):
@@ -540,6 +538,7 @@ def make_policy(
         set_dataset_feature_metadata = getattr(cfg, "set_dataset_feature_metadata", None)
         if callable(set_dataset_feature_metadata):
             set_dataset_feature_metadata(ds_meta.features)
+        cfg._runtime_dataset_meta = ds_meta
 
     kwargs["config"] = cfg
 

src/lerobot/policies/groot/__init__.py

@@ -18,4 +18,12 @@ from .configuration_groot import GrootConfig
 from .modeling_groot import GrootPolicy
 from .processor_groot import make_groot_pre_post_processors
 
-__all__ = ["GrootConfig", "GrootPolicy", "make_groot_pre_post_processors"]
+__all__ = ["GR00TN17", "GR00TN17Config", "GrootConfig", "GrootPolicy", "make_groot_pre_post_processors"]
+
+
+def __getattr__(name: str):
+    if name in {"GR00TN17", "GR00TN17Config"}:
+        from .groot_n1_7 import GR00TN17, GR00TN17Config
+
+        return {"GR00TN17": GR00TN17, "GR00TN17Config": GR00TN17Config}[name]
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

src/lerobot/policies/groot/action_head/action_encoder.py

@@ -1,54 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import torch
-import torch.nn as nn
-
-
-def swish(x):
-    return x * torch.sigmoid(x)
-
-
-class SinusoidalPositionalEncoding(nn.Module):
-    """
-    Produces a sinusoidal encoding of shape (B, T, w)
-    given timesteps of shape (B, T).
-    """
-
-    def __init__(self, embedding_dim):
-        super().__init__()
-        self.embedding_dim = embedding_dim
-
-    def forward(self, timesteps):
-        # timesteps: shape (B, T)
-        # We'll compute sin/cos frequencies across dim T
-        timesteps = timesteps.float()  # ensure float
-
-        b, t = timesteps.shape
-        device = timesteps.device
-
-        half_dim = self.embedding_dim // 2
-        # typical log space frequencies for sinusoidal encoding
-        exponent = -torch.arange(half_dim, dtype=torch.float, device=device) * (
-            torch.log(torch.tensor(10000.0)) / half_dim
-        )
-        # Expand timesteps to (B, T, 1) then multiply
-        freqs = timesteps.unsqueeze(-1) * exponent.exp()  # (B, T, half_dim)
-
-        sin = torch.sin(freqs)
-        cos = torch.cos(freqs)
-        enc = torch.cat([sin, cos], dim=-1)  # (B, T, w)
-
-        return enc

src/lerobot/policies/groot/action_head/cross_attention_dit.py

@@ -14,6 +14,7 @@
 # limitations under the License.
 
 
+import logging
 from typing import TYPE_CHECKING
 
 import torch
@@ -42,6 +43,9 @@ else:
     Timesteps = None
 
 
+logger = logging.getLogger(__name__)
+
+
 class TimestepEncoder(nn.Module):
     def __init__(self, embedding_dim, compute_dtype=torch.float32):
         require_package("diffusers", extra="groot")
@@ -181,8 +185,7 @@ class BasicTransformerBlock(nn.Module):
         attn_output = self.attn1(
             norm_hidden_states,
             encoder_hidden_states=encoder_hidden_states,
-            attention_mask=attention_mask,
-            # encoder_attention_mask=encoder_attention_mask,
+            attention_mask=encoder_attention_mask if encoder_hidden_states is not None else attention_mask,
         )
         if self.final_dropout:
             attn_output = self.final_dropout(attn_output)
@@ -266,8 +269,8 @@ class DiT(ModelMixin, ConfigMixin):
         self.norm_out = nn.LayerNorm(self.inner_dim, elementwise_affine=False, eps=1e-6)
         self.proj_out_1 = nn.Linear(self.inner_dim, 2 * self.inner_dim)
         self.proj_out_2 = nn.Linear(self.inner_dim, self.config.output_dim)
-        print(
-            "Total number of DiT parameters: ",
+        logger.debug(
+            "Total number of DiT parameters: %d",
             sum(p.numel() for p in self.parameters() if p.requires_grad),
         )
 
@@ -318,6 +321,71 @@ class DiT(ModelMixin, ConfigMixin):
             return self.proj_out_2(hidden_states)
 
 
+class AlternateVLDiT(DiT):
+    """N1.7 DiT variant that alternates cross-attention over image and text tokens."""
+
+    def __init__(self, *args, attend_text_every_n_blocks: int = 2, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.attend_text_every_n_blocks = attend_text_every_n_blocks
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: torch.Tensor,
+        timestep: torch.LongTensor | None = None,
+        encoder_attention_mask: torch.Tensor | None = None,
+        return_all_hidden_states: bool = False,
+        image_mask: torch.Tensor | None = None,
+        backbone_attention_mask: torch.Tensor | None = None,
+    ):
+        if image_mask is None:
+            raise ValueError("image_mask is required for AlternateVLDiT.")
+        if backbone_attention_mask is None:
+            raise ValueError("backbone_attention_mask is required for AlternateVLDiT.")
+
+        temb = self.timestep_encoder(timestep)
+        hidden_states = hidden_states.contiguous()
+        encoder_hidden_states = encoder_hidden_states.contiguous()
+
+        image_attention_mask = image_mask & backbone_attention_mask
+        non_image_attention_mask = (~image_mask) & backbone_attention_mask
+
+        all_hidden_states = [hidden_states]
+        if not self.config.interleave_self_attention:
+            raise ValueError("AlternateVLDiT requires interleave_self_attention=True.")
+
+        for idx, block in enumerate(self.transformer_blocks):
+            if idx % 2 == 1:
+                hidden_states = block(
+                    hidden_states,
+                    attention_mask=None,
+                    encoder_hidden_states=None,
+                    encoder_attention_mask=None,
+                    temb=temb,
+                )
+            else:
+                curr_encoder_attention_mask = (
+                    non_image_attention_mask
+                    if idx % (2 * self.attend_text_every_n_blocks) == 0
+                    else image_attention_mask
+                )
+                hidden_states = block(
+                    hidden_states,
+                    attention_mask=None,
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_attention_mask=curr_encoder_attention_mask,
+                    temb=temb,
+                )
+            all_hidden_states.append(hidden_states)
+
+        conditioning = temb
+        shift, scale = self.proj_out_1(F.silu(conditioning)).chunk(2, dim=1)
+        hidden_states = self.norm_out(hidden_states) * (1 + scale[:, None]) + shift[:, None]
+        if return_all_hidden_states:
+            return self.proj_out_2(hidden_states), all_hidden_states
+        return self.proj_out_2(hidden_states)
+
+
 class SelfAttentionTransformer(ModelMixin, ConfigMixin):
     _supports_gradient_checkpointing = True
 
@@ -362,8 +430,8 @@ class SelfAttentionTransformer(ModelMixin, ConfigMixin):
                 for _ in range(self.config.num_layers)
             ]
         )
-        print(
-            "Total number of SelfAttentionTransformer parameters: ",
+        logger.debug(
+            "Total number of SelfAttentionTransformer parameters: %d",
             sum(p.numel() for p in self.parameters() if p.requires_grad),
         )
 

src/lerobot/policies/groot/action_head/flow_matching_action_head.py

@@ -1,408 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from dataclasses import field
-from typing import TYPE_CHECKING
-
-import torch
-import torch.nn.functional as F  # noqa: N812
-from torch import nn
-from torch.distributions import Beta
-
-from lerobot.utils.import_utils import _transformers_available
-
-# Conditional import for type checking and lazy loading
-if TYPE_CHECKING or _transformers_available:
-    from transformers import PretrainedConfig
-    from transformers.feature_extraction_utils import BatchFeature
-else:
-    PretrainedConfig = object
-    BatchFeature = None
-
-from .action_encoder import (
-    SinusoidalPositionalEncoding,
-    swish,
-)
-from .cross_attention_dit import DiT, SelfAttentionTransformer
-
-
-class CategorySpecificLinear(nn.Module):
-    def __init__(self, num_categories, input_dim, hidden_dim):
-        super().__init__()
-        self.num_categories = num_categories
-        # For each category, we have separate weights and biases.
-        self.W = nn.Parameter(0.02 * torch.randn(num_categories, input_dim, hidden_dim))
-        self.b = nn.Parameter(torch.zeros(num_categories, hidden_dim))
-
-    def forward(self, x, cat_ids):
-        selected_w = self.W[cat_ids]
-        selected_b = self.b[cat_ids]
-        return torch.bmm(x, selected_w) + selected_b.unsqueeze(1)
-
-
-class CategorySpecificMLP(nn.Module):
-    def __init__(self, num_categories, input_dim, hidden_dim, output_dim):
-        super().__init__()
-        self.num_categories = num_categories
-        self.layer1 = CategorySpecificLinear(num_categories, input_dim, hidden_dim)
-        self.layer2 = CategorySpecificLinear(num_categories, hidden_dim, output_dim)
-
-    def forward(self, x, cat_ids):
-        hidden = F.relu(self.layer1(x, cat_ids))
-        return self.layer2(hidden, cat_ids)
-
-
-class MultiEmbodimentActionEncoder(nn.Module):
-    def __init__(self, action_dim, hidden_size, num_embodiments):
-        super().__init__()
-        self.hidden_size = hidden_size
-        self.num_embodiments = num_embodiments
-
-        # W1: R^{w x d}, W2: R^{w x 2w}, W3: R^{w x w}
-        self.W1 = CategorySpecificLinear(num_embodiments, action_dim, hidden_size)  # (d -> w)
-        self.W2 = CategorySpecificLinear(num_embodiments, 2 * hidden_size, hidden_size)  # (2w -> w)
-        self.W3 = CategorySpecificLinear(num_embodiments, hidden_size, hidden_size)  # (w -> w)
-        self.pos_encoding = SinusoidalPositionalEncoding(hidden_size)
-
-    def forward(self, actions, timesteps, cat_ids):
-        """
-        actions:   shape (B, T, action_dim)
-        timesteps: shape (B,)  -- a single scalar per batch item
-        cat_ids:   shape (B,)
-        returns:   shape (B, T, hidden_size)
-        """
-        b, t, _ = actions.shape
-
-        # 1) Expand each batch's single scalar time 'tau' across all T steps
-        #    so that shape => (B, T)
-        #    e.g. if timesteps is (B,), replicate across T
-        if timesteps.dim() == 1 and timesteps.shape[0] == b:
-            # shape (B,) => (B,T)
-            timesteps = timesteps.unsqueeze(1).expand(-1, t)
-        else:
-            raise ValueError("Expected `timesteps` to have shape (B,) so we can replicate across T.")
-
-        # 2) Standard action MLP step for shape => (B, T, w)
-        a_emb = self.W1(actions, cat_ids)
-
-        # 3) Get the sinusoidal encoding (B, T, w)
-        tau_emb = self.pos_encoding(timesteps).to(dtype=a_emb.dtype)
-
-        # 4) Concat along last dim => (B, T, 2w), then W2 => (B, T, w), swish
-        x = torch.cat([a_emb, tau_emb], dim=-1)
-        x = swish(self.W2(x, cat_ids))
-
-        # 5) Finally W3 => (B, T, w)
-        x = self.W3(x, cat_ids)
-        return x
-
-
-class FlowmatchingActionHeadConfig(PretrainedConfig):
-    """NOTE: N1.5 uses XEmbFlowmatchingPolicyHeadConfig as action head"""
-
-    add_pos_embed: bool = field(default=True, metadata={"help": "Whether to add positional embedding"})
-    model_dtype: str = field(default="float32", metadata={"help": "Model data type."})
-    diffusion_model_cfg: dict = field(default=None, metadata={"help": "Diffusion model configuration."})
-    input_embedding_dim: int = field(default=1536, metadata={"help": "Input embedding channel dimension."})
-    backbone_embedding_dim: int = field(
-        default=1536, metadata={"help": "Backbone embedding channel dimension."}
-    )
-
-    hidden_size: int = field(default=1024, metadata={"help": "Input embedding dimension."})
-    max_seq_len: int = field(default=1024, metadata={"help": "Maximum Sequence Length"})
-    action_dim: int = field(default=None, metadata={"help": "Action dimension."})
-    action_horizon: int = field(default=None, metadata={"help": "Action horizon."})
-    noise_beta_alpha: float = field(default=1.5, metadata={"help": ""})
-    noise_beta_beta: float = field(default=1.0, metadata={"help": ""})
-    noise_s: float = field(default=0.999, metadata={"help": "Flow matching noise Beta distribution s."})
-    num_timestep_buckets: int = field(
-        default=1000, metadata={"help": "Number of timestep discretization buckets."}
-    )
-    num_inference_timesteps: int = field(
-        default=None,
-        metadata={"help": "Number of inference steps for noise diffusion."},
-    )
-    max_num_embodiments: int = field(default=32, metadata={"help": "Number of embodiments."})
-    tune_projector: bool = field(default=True, metadata={"help": "Whether to tune the projector."})
-    tune_diffusion_model: bool = field(
-        default=True, metadata={"help": "Whether to tune the diffusion model."}
-    )
-    load_pretrained_det_decode_layer_path: str = field(
-        default=None, metadata={"help": "Path to pretrained detection model."}
-    )
-    detection_coeff: float = field(default=1.0, metadata={"help": "Detection coefficient."})
-
-    freeze_decode_layer: bool = field(default=False)
-    expand_batch: int = field(default=None)
-    use_vlln: bool = field(default=True)
-
-    vl_self_attention_cfg: dict = field(default=None)
-    num_target_vision_tokens: int = field(default=32, metadata={"help": "Number of target vision tokens."})
-
-    def __init__(self, **kwargs):
-        super().__init__(**kwargs)
-        for key, value in kwargs.items():
-            setattr(self, key, value)
-
-
-class FlowmatchingActionHead(nn.Module):
-    config_class = FlowmatchingActionHeadConfig
-    supports_gradient_checkpointing = True
-
-    def __init__(
-        self,
-        config: FlowmatchingActionHeadConfig,
-    ):
-        super().__init__()
-        self.hidden_size = config.hidden_size
-        self.input_embedding_dim = config.input_embedding_dim
-
-        self.model = DiT(**config.diffusion_model_cfg)
-        self.action_dim = config.action_dim
-        self.action_horizon = config.action_horizon
-        self.num_inference_timesteps = config.num_inference_timesteps
-
-        self.state_encoder = CategorySpecificMLP(
-            num_categories=config.max_num_embodiments,
-            input_dim=config.max_state_dim,
-            hidden_dim=self.hidden_size,
-            output_dim=self.input_embedding_dim,
-        )
-        self.action_encoder = MultiEmbodimentActionEncoder(
-            action_dim=config.action_dim,
-            hidden_size=self.input_embedding_dim,
-            num_embodiments=config.max_num_embodiments,
-        )
-        self.action_decoder = CategorySpecificMLP(
-            num_categories=config.max_num_embodiments,
-            input_dim=self.hidden_size,
-            hidden_dim=self.hidden_size,
-            output_dim=self.action_dim,
-        )
-        self.future_tokens = nn.Embedding(config.num_target_vision_tokens, self.input_embedding_dim)
-        nn.init.normal_(self.future_tokens.weight, mean=0.0, std=0.02)
-
-        self.vlln = nn.LayerNorm(config.backbone_embedding_dim) if config.use_vlln else nn.Identity()
-        self.vl_self_attention = (
-            SelfAttentionTransformer(**config.vl_self_attention_cfg) if config.use_vlln else nn.Identity()
-        )
-
-        if config.add_pos_embed:
-            self.position_embedding = nn.Embedding(config.max_seq_len, self.input_embedding_dim)
-            nn.init.normal_(self.position_embedding.weight, mean=0.0, std=0.02)
-
-        self._noise_beta_alpha = config.noise_beta_alpha
-        self._noise_beta_beta = config.noise_beta_beta
-        self._beta_dist = None
-        self.num_timestep_buckets = config.num_timestep_buckets
-        self.config = config
-        self.set_trainable_parameters(config.tune_projector, config.tune_diffusion_model)
-
-    def set_trainable_parameters(self, tune_projector: bool, tune_diffusion_model: bool):
-        self.tune_projector = tune_projector
-        self.tune_diffusion_model = tune_diffusion_model
-        for p in self.parameters():
-            p.requires_grad = True
-        if not tune_projector:
-            self.state_encoder.requires_grad_(False)
-            self.action_encoder.requires_grad_(False)
-            self.action_decoder.requires_grad_(False)
-            if self.config.add_pos_embed:
-                self.position_embedding.requires_grad_(False)
-        if not tune_diffusion_model:
-            self.model.requires_grad_(False)
-        print(f"Tune action head projector: {self.tune_projector}")
-        print(f"Tune action head diffusion model: {self.tune_diffusion_model}")
-        # Check if any parameters are still trainable. If not, print a warning.
-        if not tune_projector and not tune_diffusion_model:
-            for name, p in self.named_parameters():
-                if p.requires_grad:
-                    print(f"Action head trainable parameter: {name}")
-        if not any(p.requires_grad for p in self.parameters()):
-            print("Warning: No action head trainable parameters found.")
-
-    def set_frozen_modules_to_eval_mode(self):
-        """
-        Huggingface will call model.train() at each training_step. To ensure
-        the expected behaviors for modules like dropout, batchnorm, etc., we
-        need to call model.eval() for the frozen modules.
-        """
-        if self.training:
-            if not self.tune_projector:
-                self.state_encoder.eval()
-                self.action_encoder.eval()
-                self.action_decoder.eval()
-                if self.config.add_pos_embed:
-                    self.position_embedding.eval()
-            if not self.tune_diffusion_model:
-                self.model.eval()
-
-    def sample_time(self, batch_size, device, dtype):
-        if self._beta_dist is None:
-            self._beta_dist = Beta(self._noise_beta_alpha, self._noise_beta_beta, validate_args=False)
-        sample = self._beta_dist.sample([batch_size]).to(device, dtype=dtype)
-        return (self.config.noise_s - sample) / self.config.noise_s
-
-    def prepare_input(self, batch: dict) -> BatchFeature:
-        return BatchFeature(data=batch)
-
-    def process_backbone_output(self, backbone_output: BatchFeature) -> BatchFeature:
-        backbone_features = backbone_output["backbone_features"]
-        backbone_features = self.vlln(backbone_features)
-        backbone_features = self.vl_self_attention(backbone_features)
-        backbone_output["backbone_features"] = backbone_features
-        return backbone_output
-
-    def forward(self, backbone_output: BatchFeature, action_input: BatchFeature) -> BatchFeature:
-        # Set frozen modules to eval
-        self.set_frozen_modules_to_eval_mode()
-
-        backbone_output = self.process_backbone_output(backbone_output)
-
-        if self.config.expand_batch is not None:
-            for k, v in backbone_output.items():
-                ndim = len(v.shape)
-                factors = [self.config.expand_batch]
-                while len(factors) < ndim:
-                    factors.append(1)
-                factors = tuple(factors)
-                expanded = v.repeat(*factors)
-                backbone_output[k] = expanded
-
-            for k, v in action_input.items():
-                ndim = len(v.shape)
-                factors = [self.config.expand_batch]
-                while len(factors) < ndim:
-                    factors.append(1)
-                factors = tuple(factors)
-                expanded = v.repeat(*factors)
-                action_input[k] = expanded
-
-        # Get vision and language embeddings.
-        vl_embs = backbone_output.backbone_features
-        device = vl_embs.device
-
-        # Get embodiment ID.
-        embodiment_id = action_input.embodiment_id
-
-        # Embed state.
-        state_features = self.state_encoder(action_input.state, embodiment_id)
-
-        # Embed noised action trajectory.
-        actions = action_input.action
-        noise = torch.randn(actions.shape, device=actions.device, dtype=actions.dtype)
-        t = self.sample_time(actions.shape[0], device=actions.device, dtype=actions.dtype)
-        t = t[:, None, None]  # shape (B,1,1) for broadcast
-
-        noisy_trajectory = (1 - t) * noise + t * actions
-        velocity = actions - noise
-
-        # Convert (continuous) t -> discrete if needed
-        t_discretized = (t[:, 0, 0] * self.num_timestep_buckets).long()
-        action_features = self.action_encoder(noisy_trajectory, t_discretized, embodiment_id)
-
-        # Maybe add position embedding.
-        if self.config.add_pos_embed:
-            pos_ids = torch.arange(action_features.shape[1], dtype=torch.long, device=device)
-            pos_embs = self.position_embedding(pos_ids).unsqueeze(0)
-            action_features = action_features + pos_embs
-
-        # Join vision, language, state and action embedding along sequence dimension.
-        future_tokens = self.future_tokens.weight.unsqueeze(0).expand(vl_embs.shape[0], -1, -1)
-        sa_embs = torch.cat((state_features, future_tokens, action_features), dim=1)
-
-        vl_attn_mask = backbone_output.backbone_attention_mask
-
-        model_output = self.model(
-            hidden_states=sa_embs,
-            encoder_hidden_states=vl_embs,
-            encoder_attention_mask=vl_attn_mask,
-            timestep=t_discretized,
-            return_all_hidden_states=False,  # NOTE (YL): not using flare now
-        )
-        pred = self.action_decoder(model_output, embodiment_id)
-        pred_actions = pred[:, -actions.shape[1] :]
-
-        # Slice out only the action portion of pred and target.
-        action_mask = action_input.action_mask
-        loss = F.mse_loss(pred_actions, velocity, reduction="none") * action_mask
-        loss = loss.sum() / action_mask.sum()
-        output_dict = {
-            "loss": loss,
-        }
-        return BatchFeature(data=output_dict)
-
-    @torch.no_grad()
-    def get_action(self, backbone_output: BatchFeature, action_input: BatchFeature) -> BatchFeature:
-        backbone_output = self.process_backbone_output(backbone_output)
-
-        # Get vision and language embeddings.
-        vl_embs = backbone_output.backbone_features
-        embodiment_id = action_input.embodiment_id
-
-        # Embed state.
-        state_features = self.state_encoder(action_input.state, embodiment_id)
-
-        # Set initial actions as the sampled noise.
-        batch_size = vl_embs.shape[0]
-        device = vl_embs.device
-        actions = torch.randn(
-            size=(batch_size, self.config.action_horizon, self.config.action_dim),
-            dtype=vl_embs.dtype,
-            device=device,
-        )
-
-        num_steps = self.num_inference_timesteps
-        dt = 1.0 / num_steps
-
-        # Run denoising steps.
-        for t in range(num_steps):
-            t_cont = t / float(num_steps)  # e.g. goes 0, 1/N, 2/N, ...
-            t_discretized = int(t_cont * self.num_timestep_buckets)
-
-            # Embed noised action trajectory.
-            timesteps_tensor = torch.full(size=(batch_size,), fill_value=t_discretized, device=device)
-            action_features = self.action_encoder(actions, timesteps_tensor, embodiment_id)
-            # Maybe add position embedding.
-            if self.config.add_pos_embed:
-                pos_ids = torch.arange(action_features.shape[1], dtype=torch.long, device=device)
-                pos_embs = self.position_embedding(pos_ids).unsqueeze(0)
-                action_features = action_features + pos_embs
-
-            # Join vision, language, state and action embedding along sequence dimension.
-            future_tokens = self.future_tokens.weight.unsqueeze(0).expand(vl_embs.shape[0], -1, -1)
-            sa_embs = torch.cat((state_features, future_tokens, action_features), dim=1)
-
-            # Run model forward.
-            model_output = self.model(
-                hidden_states=sa_embs,
-                encoder_hidden_states=vl_embs,
-                timestep=timesteps_tensor,
-            )
-            pred = self.action_decoder(model_output, embodiment_id)
-
-            pred_velocity = pred[:, -self.action_horizon :]
-
-            # Update actions using euler integration.
-            actions = actions + dt * pred_velocity
-        return BatchFeature(data={"action_pred": actions})
-
-    @property
-    def device(self):
-        return next(iter(self.parameters())).device
-
-    @property
-    def dtype(self):
-        return next(iter(self.parameters())).dtype

src/lerobot/policies/groot/configuration_groot.py

@@ -14,12 +14,229 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
+import math
 from dataclasses import dataclass, field
+from pathlib import Path
 
 from lerobot.configs import FeatureType, NormalizationMode, PolicyFeature, PreTrainedConfig
-from lerobot.optim import AdamWConfig, CosineDecayWithWarmupSchedulerConfig
+from lerobot.optim import AdamWConfig, DiffuserSchedulerConfig
 from lerobot.utils.constants import ACTION, OBS_STATE
 
+from .utils import read_json
+
+logger = logging.getLogger(__name__)
+
+GROOT_N1_7 = "n1.7"
+# Legacy GR00T N1.5 identifier. N1.5 is NOT a supported model_version (it is
+# intentionally absent from _GROOT_MODEL_VERSION_ALIASES so normalize_groot_model_version
+# still rejects it). It is retained only so that infer_groot_model_version can recognise
+# an N1.5 base path/checkpoint and the N1.7 config/loader can reject the mismatch.
+GROOT_N1_5 = "n1.5"
+# Canonical guidance appended to every error raised when an N1.5 checkpoint, config,
+# or processor pipeline is detected. Keep this message in sync with docs/source/groot.mdx.
+GROOT_N1_5_REMOVAL_GUIDANCE = (
+    "GR00T N1.5 support was removed from LeRobot. "
+    "To keep using an N1.5 checkpoint, pin the last release that supports it: "
+    "`pip install 'lerobot==0.5.1'`. To use the current release, migrate to GR00T N1.7 "
+    "(model_version='n1.7', base model nvidia/GR00T-N1.7-3B)."
+)
+GROOT_N1_7_BASE_MODEL = "nvidia/GR00T-N1.7-3B"
+GROOT_N1_7_BACKBONE_MODEL = "nvidia/Cosmos-Reason2-2B"
+# Default GR00T N1.7 training resolution. Fallback if processor_config lacks sizing. Prevents mismatched
+# full-res patchification by forcing a resize. Mirrored by GR00T_N1_7_DEFAULTS in groot_n1_7.py.
+N1_7_DEFAULT_IMAGE_TARGET_SIZE = (256, 256)
+N1_7_DEFAULT_IMAGE_CROP_SIZE = (230, 230)
+GROOT_ACTION_DECODE_TRANSFORM_LIBERO = "libero"
+# Sentinel meaning "the user did not pick an action decode transform": __post_init__ resolves it
+# to the embodiment default ('libero' for 'libero_sim', otherwise None). It is distinct from an
+# explicit 'none' (resolved to None) so an opt-out survives a draccus save/load round-trip.
+GROOT_ACTION_DECODE_TRANSFORM_AUTO = "auto"
+
+_GROOT_MODEL_VERSION_ALIASES = {
+    "n1.7": GROOT_N1_7,
+    "n1_7": GROOT_N1_7,
+    "n1d7": GROOT_N1_7,
+    "n17": GROOT_N1_7,
+    "1.7": GROOT_N1_7,
+}
+
+# Legacy N1.5 spellings, kept ONLY so they can be detected and rejected with
+# GROOT_N1_5_REMOVAL_GUIDANCE (see GROOT_N1_5 above). Never map these to a supported version.
+_GROOT_N1_5_VERSION_ALIASES = {"n1.5", "n1_5", "n1d5", "n15", "1.5"}
+
+_GROOT_ACTION_DECODE_TRANSFORM_ALIASES = {
+    GROOT_ACTION_DECODE_TRANSFORM_AUTO: GROOT_ACTION_DECODE_TRANSFORM_AUTO,
+    "none": None,
+    "": None,
+    GROOT_ACTION_DECODE_TRANSFORM_LIBERO: GROOT_ACTION_DECODE_TRANSFORM_LIBERO,
+}
+
+
+def normalize_groot_model_version(model_version: str) -> str:
+    normalized = _GROOT_MODEL_VERSION_ALIASES.get(model_version.lower())
+    if normalized is None:
+        supported = GROOT_N1_7
+        message = f"Unsupported GR00T model_version '{model_version}'. Supported versions: {supported}."
+        if model_version.lower() in _GROOT_N1_5_VERSION_ALIASES:
+            message = f"{message} {GROOT_N1_5_REMOVAL_GUIDANCE}"
+        raise ValueError(message)
+    return normalized
+
+
+def normalize_groot_action_decode_transform(transform: str | None) -> str | None:
+    if transform is None:
+        return None
+    normalized = _GROOT_ACTION_DECODE_TRANSFORM_ALIASES.get(transform.lower())
+    if normalized is None and transform.lower() not in _GROOT_ACTION_DECODE_TRANSFORM_ALIASES:
+        supported = ", ".join(
+            sorted(key for key, value in _GROOT_ACTION_DECODE_TRANSFORM_ALIASES.items() if value is not None)
+        )
+        raise ValueError(
+            f"Unsupported GR00T N1.7 action decode transform '{transform}'. "
+            f"Supported transforms: none, {supported}."
+        )
+    return normalized
+
+
+def infer_groot_model_version(model_path: str | None) -> str | None:
+    if not model_path:
+        return None
+    model_path_lower = model_path.lower()
+    if "gr00t-n1.7" in model_path_lower or "gr00t_n1.7" in model_path_lower:
+        return GROOT_N1_7
+    # Detect legacy N1.5 paths so the N1.7 config/loader can reject the mismatch.
+    # N1.5 is unsupported, but it must still be recognised here to fail loudly
+    # rather than silently treating an N1.5 checkpoint as N1.7.
+    if "gr00t-n1.5" in model_path_lower or "gr00t_n1.5" in model_path_lower:
+        return GROOT_N1_5
+    config_version = _infer_groot_model_version_from_local_config(model_path)
+    if config_version is not None:
+        return config_version
+    return None
+
+
+def is_raw_groot_n1_7_checkpoint(model_path: str | Path | None) -> bool:
+    if model_path is None:
+        return False
+
+    path = Path(model_path).expanduser()
+    if path.is_dir():
+        config_path = path / "config.json"
+    elif path.name == "config.json":
+        config_path = path
+    else:
+        return False
+
+    config = read_json(config_path)
+    return "type" not in config and _infer_groot_model_version_from_config(config) == GROOT_N1_7
+
+
+def infer_groot_n1_7_embodiment_tag(model_path: str | Path | None) -> str | None:
+    if model_path is None:
+        return None
+
+    processor_config_path = Path(model_path).expanduser() / "processor_config.json"
+    processor_config = read_json(processor_config_path)
+
+    modality_configs = processor_config.get("processor_kwargs", {}).get("modality_configs", {})
+    if not isinstance(modality_configs, dict):
+        return None
+    if "libero_sim" in modality_configs:
+        return "libero_sim"
+    if len(modality_configs) == 1:
+        return next(iter(modality_configs))
+    return None
+
+
+def infer_groot_n1_7_action_horizon(
+    model_path: str | Path | None, embodiment_tag: str | None = None
+) -> int | None:
+    if model_path is None:
+        return None
+
+    processor_config_path = Path(model_path).expanduser() / "processor_config.json"
+    processor_config = read_json(processor_config_path)
+
+    processor_kwargs = processor_config.get("processor_kwargs", {})
+    if not isinstance(processor_kwargs, dict):
+        return None
+    modality_configs = processor_kwargs.get("modality_configs", {})
+    if not isinstance(modality_configs, dict):
+        return None
+
+    if embodiment_tag is None:
+        embodiment_tag = infer_groot_n1_7_embodiment_tag(model_path)
+    if embodiment_tag is None:
+        return None
+
+    embodiment_config = modality_configs.get(embodiment_tag, {})
+    if not isinstance(embodiment_config, dict):
+        return None
+    action_config = embodiment_config.get("action", {})
+    if not isinstance(action_config, dict):
+        return None
+    delta_indices = action_config.get("delta_indices", [])
+    if not isinstance(delta_indices, list):
+        return None
+    return len(delta_indices) or None
+
+
+def infer_groot_n1_7_action_execution_horizon(
+    model_path: str | Path | None, embodiment_tag: str | None = None
+) -> int | None:
+    action_horizon = infer_groot_n1_7_action_horizon(model_path, embodiment_tag)
+    if action_horizon is None:
+        return None
+
+    if embodiment_tag is None:
+        embodiment_tag = infer_groot_n1_7_embodiment_tag(model_path)
+    if embodiment_tag == "libero_sim":
+        # NVIDIA's N1.7 LIBERO rollout wrapper replans after 8 of the 16 decoded
+        # actions. Keeping that execution cadence avoids stale open-loop chunks.
+        return min(action_horizon, 8)
+    return action_horizon
+
+
+def _infer_groot_model_version_from_local_config(model_path: str) -> str | None:
+    path = Path(model_path).expanduser()
+    if path.is_dir():
+        config_path = path / "config.json"
+    elif path.name == "config.json":
+        config_path = path
+    else:
+        return None
+
+    return _infer_groot_model_version_from_config(read_json(config_path))
+
+
+def _infer_groot_model_version_from_config(config: dict) -> str | None:
+    model_version = config.get("model_version")
+    if isinstance(model_version, str):
+        if model_version.lower() in _GROOT_N1_5_VERSION_ALIASES:
+            return GROOT_N1_5
+        try:
+            return normalize_groot_model_version(model_version)
+        except ValueError:
+            return None
+
+    candidates = [config.get("model_type"), *(config.get("architectures") or [])]
+    for candidate in candidates:
+        if not isinstance(candidate, str):
+            continue
+        normalized = candidate.lower().replace("-", "_")
+        if normalized in {"gr00tn1d7", "gr00t_n1d7", "gr00t_n1_7"}:
+            return GROOT_N1_7
+        if normalized in {"gr00t_n1_5", "gr00tn1_5", "gr00t_n15", "gr00t_n1d5", "gr00tn1d5"}:
+            return GROOT_N1_5
+    if config.get("model_name") == GROOT_N1_7_BACKBONE_MODEL:
+        return GROOT_N1_7
+    # The Eagle VLM backbone is specific to pre-N1.7 GR00T checkpoints (N1.7 uses Cosmos/Qwen3-VL).
+    backbone_cfg = config.get("backbone_cfg")
+    if isinstance(backbone_cfg, dict) and "eagle_path" in backbone_cfg:
+        return GROOT_N1_5
+    return None
+
 
 @PreTrainedConfig.register_subclass("groot")
 @dataclass
@@ -28,35 +245,44 @@ class GrootConfig(PreTrainedConfig):
 
     # Basic policy settings
     n_obs_steps: int = 1
-    chunk_size: int = 50
-    n_action_steps: int = 50
+    chunk_size: int = 40
+    n_action_steps: int = 40
 
     # Dimension settings (must match pretrained GR00T model expectations)
     # Maximum state dimension. Shorter states will be zero-padded.
-    max_state_dim: int = 64
+    max_state_dim: int = 132
 
     # Maximum action dimension. Shorter actions will be zero-padded.
-    max_action_dim: int = 32
+    max_action_dim: int = 132
 
-    # Normalization (start with identity, adjust as needed)
+    # GR00T normalizes state/action internally in its processor steps (min/max with
+    # q01/q99 percentiles, per embodiment), and the Qwen3-VL backbone's image processor
+    # handles image normalization. The policy therefore does NOT use LeRobot's
+    # NormalizerProcessorStep/UnnormalizerProcessorStep, so this mapping is intentionally
+    # IDENTITY for every feature and is not consulted by make_groot_pre_post_processors.
     normalization_mapping: dict[str, NormalizationMode] = field(
         default_factory=lambda: {
             "VISUAL": NormalizationMode.IDENTITY,
-            "STATE": NormalizationMode.MEAN_STD,
-            "ACTION": NormalizationMode.MEAN_STD,
+            "STATE": NormalizationMode.IDENTITY,
+            "ACTION": NormalizationMode.IDENTITY,
         }
     )
 
-    # Image preprocessing (adjust to match Groot's expected input)
-    image_size: tuple[int, int] = (224, 224)
+    # Groot-specific model parameters
 
-    # Groot-specific model parameters (from groot_finetune_script.py)
+    # Path or HuggingFace model ID for the base GR00T N1.7 model whose backbone weights and
+    # checkpoint sidecars (statistics.json, processor_config.json, ...) are loaded. This is the
+    # model *source*, and is intentionally distinct from the inherited `pretrained_path`:
+    # `pretrained_path` (`--policy.path`) points at a saved LeRobot checkpoint directory whose
+    # `config.json` carries a `type` field, whereas a raw NVIDIA GR00T checkpoint has no such
+    # field and so can only be loaded through `base_model_path` (`--policy.base_model_path`).
+    # Defaults to GROOT_N1_7_BASE_MODEL when unset (resolved in __post_init__).
+    base_model_path: str | None = None
 
-    # Path or HuggingFace model ID for the base Groot model
-    base_model_path: str = "nvidia/GR00T-N1.5-3B"
-
-    # HF repo ID (or local path) that hosts vocab.json and merges.txt for Eagle tokenizer.
-    tokenizer_assets_repo: str = "lerobot/eagle2hg-processor-groot-n1p5"
+    # Optional named action transform applied after raw N1.7 checkpoint decoding and before env.step().
+    # 'auto' (default) resolves to the embodiment default ('libero' for 'libero_sim', otherwise no
+    # transform). Pass 'none' to explicitly disable the transform, including for 'libero_sim'.
+    action_decode_transform: str | None = GROOT_ACTION_DECODE_TRANSFORM_AUTO
 
     # Embodiment tag to use for training (e.g. 'new_embodiment', 'gr1')
     embodiment_tag: str = "new_embodiment"
@@ -75,38 +301,67 @@ class GrootConfig(PreTrainedConfig):
     # Whether to fine-tune the diffusion model
     tune_diffusion_model: bool = True
 
-    # LoRA parameters (from groot_finetune_script.py)
-    # Rank for the LORA model. If 0, no LORA will be used.
-    lora_rank: int = 0
+    # Whether to fine-tune the VL LayerNorm + VL self-attention projector in the action head.
+    tune_vlln: bool = True
 
-    # Alpha value for the LORA model
-    lora_alpha: int = 16
+    # Number of top LLM backbone layers to fine-tune (0 = none). Lets you adapt just the final
+    # language layers without unfreezing the whole backbone; independent of `tune_llm`, which tunes
+    # the entire LLM.
+    tune_top_llm_layers: int = 0
 
-    # Dropout rate for the LORA model
-    lora_dropout: float = 0.1
+    # Inference-time knob: Number of flow-matching denoising steps used to decode an action chunk.
+    # Trades inference latency for action quality.
+    # None keeps the checkpoint value (GR00T N1.7 default: 4).
+    num_inference_timesteps: int | None = None
 
-    # Whether to use the full model for LORA
-    lora_full_model: bool = False
+    # Inference-time knob: Real-Time Chunking (RTC) overlap-blend ramp rate, used when the RTC engine
+    # supplies a previous-chunk prefix. Higher values blend the overlapping prefix more aggressively.
+    # None keeps the checkpoint value (GR00T N1.7 default: 6.0).
+    rtc_ramp_rate: float | None = None
 
-    # Training parameters (matching groot_finetune_script.py)
+    # Inference-time knob: Whether to request the flash-attention-2 kernel for the Qwen3-VL backbone.
+    # flash-attn is an optional, user-managed optimization; when it is absent (the default),
+    # the backbone transparently falls back to SDPA, which is numerically equivalent.
+    # Set to True only after installing a flash-attn build matching your torch/CUDA env.
+    use_flash_attention: bool = False
+
+    # Enable GR00T-style state-relative action chunks (action chunk expressed relative to the current
+    # observation state).
+    use_relative_actions: bool = False
+
+    # relative_exclude_joints names the action dimensions that stay absolute; the
+    # match is substring/case-insensitive against the dataset action feature names. With the empty
+    # default every dimension is treated as relative, including the gripper -- set e.g. ["gripper"] to
+    # keep the gripper absolute, matching the Isaac-GR00T single-arm + absolute-gripper convention.
+    relative_exclude_joints: list[str] = field(default_factory=list)
+
+    # Training parameters
     optimizer_lr: float = 1e-4
-    optimizer_betas: tuple[float, float] = (0.95, 0.999)
+    # Isaac-GR00T N1.7 fine-tunes with AdamW betas (0.9, 0.999).
+    optimizer_betas: tuple[float, float] = (0.9, 0.999)
     optimizer_eps: float = 1e-8
     optimizer_weight_decay: float = 1e-5
     warmup_ratio: float = 0.05
     use_bf16: bool = True
+    # The native N1.7 fine-tuning recipe keeps model parameters in FP32 and computes under BF16 autocast.
+    model_params_fp32: bool = True
 
-    # Dataset parameters
-    # Video backend to use for training ('decord' or 'torchvision_av')
+    # TODO(Steven): Remove these deprecated fields in a future release.
+    # Deprecated Isaac-GR00T runner / GR00T N1.5 fields, plus the (never-wired) LoRA fields — all
+    # unused by the LeRobot N1.7 implementation except the `tokenizer_assets_repo` N1.5 tripwire and
+    # the `image_size` legacy remap in __post_init__. They are kept ONLY so a config.json saved by an
+    # earlier lerobot release (notably a GR00T N1.5 checkpoint) still parses under draccus — which
+    # rejects unknown fields — and is then rejected with a clear N1.5 removal message rather than an
+    # opaque draccus decoding error.
+    image_size: tuple[int, int] = (256, 256)  # image sizing is handled by the backbone's image processor.
+    tokenizer_assets_repo: str | None = None
+    lora_rank: int = 0
+    lora_alpha: int = 16
+    lora_dropout: float = 0.1
+    lora_full_model: bool = False
     video_backend: str = "decord"
-
-    # Whether to balance dataset weights in mixture datasets
     balance_dataset_weights: bool = True
-
-    # Whether to sample trajectories weighted by their length
     balance_trajectory_weights: bool = True
-
-    # Optional dataset paths for delegating training to Isaac-GR00T runner
     dataset_paths: list[str] | None = None
     output_dir: str = "./tmp/gr00t"
     save_steps: int = 1000
@@ -117,6 +372,65 @@ class GrootConfig(PreTrainedConfig):
     resume: bool = False
 
     def __post_init__(self):
+        if self.tokenizer_assets_repo is not None:
+            raise ValueError(
+                "Config sets 'tokenizer_assets_repo', which only existed for GR00T N1.5; this looks "
+                f"like a legacy GR00T N1.5 checkpoint or config. {GROOT_N1_5_REMOVAL_GUIDANCE}"
+            )
+
+        self.action_decode_transform = normalize_groot_action_decode_transform(self.action_decode_transform)
+        if self.base_model_path is None:
+            self.base_model_path = GROOT_N1_7_BASE_MODEL
+
+        # The N1.7 LIBERO checkpoints emit a [0, 1] gripper action, but the LIBERO
+        # simulator expects the OpenVLA/[-1, 1] sign convention. NVIDIA's rollout
+        # wrapper applies this conversion; mirror it here so eval on the
+        # 'libero_sim' embodiment grasps correctly instead of scoring 0% success.
+        # This matches the embodiment-specific handling already done for the
+        # action execution horizon (see infer_groot_n1_7_action_execution_horizon).
+        # Only the 'auto' sentinel resolves to the embodiment default; an explicit
+        # 'none' (normalized to None above) keeps the transform disabled.
+        if self.action_decode_transform == GROOT_ACTION_DECODE_TRANSFORM_AUTO:
+            self.action_decode_transform = (
+                GROOT_ACTION_DECODE_TRANSFORM_LIBERO if self.embodiment_tag == "libero_sim" else None
+            )
+
+        # GR00T N1.5-era default values (e.g. --policy.chunk_size=50 from old commands or
+        # stale configs) are migrated to the values the N1.7 checkpoints expect, with a
+        # warning. The dataclass defaults are already the N1.7 values, so a plain
+        # GrootConfig() never triggers this.
+        legacy_default_remaps = (
+            ("max_state_dim", 64, 132),
+            ("max_action_dim", 32, 132),
+            ("chunk_size", 50, 40),
+            ("n_action_steps", 50, 40),
+            ("image_size", (224, 224), (256, 256)),
+        )
+        for field_name, legacy_value, n1_7_value in legacy_default_remaps:
+            current_value = getattr(self, field_name)
+            if isinstance(legacy_value, tuple):
+                current_value = tuple(current_value)
+            if current_value == legacy_value:
+                logger.warning(
+                    "GrootConfig.%s=%s matches a legacy GR00T N1.5-era default; remapping it to %s, "
+                    "the value expected by GR00T N1.7 checkpoints. Set a different value explicitly "
+                    "if this is not what you want.",
+                    field_name,
+                    legacy_value,
+                    n1_7_value,
+                )
+                setattr(self, field_name, n1_7_value)
+
+        inferred_version = infer_groot_model_version(self.base_model_path)
+        if inferred_version is not None and inferred_version != GROOT_N1_7:
+            message = (
+                f"GR00T model_version '{GROOT_N1_7}' does not match base_model_path "
+                f"'{self.base_model_path}', which looks like '{inferred_version}'."
+            )
+            if inferred_version == GROOT_N1_5:
+                message = f"{message} {GROOT_N1_5_REMOVAL_GUIDANCE}"
+            raise ValueError(message)
+
         super().__post_init__()
 
         if self.n_action_steps > self.chunk_size:
@@ -124,9 +438,6 @@ class GrootConfig(PreTrainedConfig):
                 f"n_action_steps ({self.n_action_steps}) cannot exceed chunk_size ({self.chunk_size})"
             )
 
-        # groot_repo_path is now optional since we ported the components
-        # No validation needed
-
     def validate_features(self) -> None:
         """Validate and set up input/output features for Groot."""
         image_features = [key for key, feat in self.input_features.items() if feat.type == FeatureType.VISUAL]
@@ -173,15 +484,20 @@ class GrootConfig(PreTrainedConfig):
             betas=self.optimizer_betas,
             eps=self.optimizer_eps,
             weight_decay=self.optimizer_weight_decay,
+            grad_clip_norm=1.0,
         )
 
-    def get_scheduler_preset(self) -> CosineDecayWithWarmupSchedulerConfig:
-        """Return scheduler configuration."""
-        return CosineDecayWithWarmupSchedulerConfig(
-            num_warmup_steps=int(10000 * self.warmup_ratio),  # 5% warmup by default
-            num_decay_steps=10000,  # Adjust based on training steps
-            peak_lr=self.optimizer_lr,
-            decay_lr=self.optimizer_lr * 0.1,
+    def get_scheduler_preset(self) -> DiffuserSchedulerConfig:
+        """Return scheduler configuration.
+
+        Isaac-GR00T uses the HF Trainer cosine schedule with ~5% warmup over the
+        actual training update count; DiffuserSchedulerConfig wraps the same
+        diffusers/transformers `get_scheduler("cosine")` implementation and
+        derives num_training_steps from the outer --steps value at runtime.
+        """
+        return DiffuserSchedulerConfig(
+            name="cosine",
+            num_warmup_steps=math.ceil(self.max_steps * self.warmup_ratio),
         )
 
     @property
@@ -192,7 +508,15 @@ class GrootConfig(PreTrainedConfig):
     @property
     def action_delta_indices(self) -> list[int]:
         """Return indices for delta actions."""
-        return list(range(min(self.chunk_size, 16)))
+        model_action_horizon = (
+            infer_groot_n1_7_action_horizon(self.base_model_path, self.embodiment_tag) or 40
+        )
+        return list(range(min(self.chunk_size, model_action_horizon)))
+
+    @property
+    def drop_n_last_frames(self) -> int:
+        """Exclude episode tails that cannot supply a complete N1.7 action chunk."""
+        return max(0, len(self.action_delta_indices) - 1)
 
     @property
     def reward_delta_indices(self) -> None:

src/lerobot/policies/groot/eagle2_hg_model/configuration_eagle2_5_vl.py

@@ -1,135 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-
-import copy
-
-from transformers.configuration_utils import PretrainedConfig
-from transformers.models.llama.configuration_llama import LlamaConfig
-from transformers.models.qwen2.configuration_qwen2 import Qwen2Config
-from transformers.models.qwen3.configuration_qwen3 import Qwen3Config
-from transformers.models.siglip.configuration_siglip import SiglipVisionConfig
-from transformers.utils import logging
-
-logger = logging.get_logger(__name__)
-
-
-class Eagle25VLConfig(PretrainedConfig):
-    model_type = "eagle_2_5_vl"
-    is_composition = True
-    sub_configs = {"vision_config": SiglipVisionConfig, "text_config": Qwen2Config}
-
-    def __init__(
-        self,
-        vision_config=None,
-        text_config=None,
-        use_backbone_lora=0,
-        use_llm_lora=0,
-        pad2square=False,
-        select_layer=-4,
-        force_image_size=None,
-        downsample_ratio=0.5,
-        template=None,
-        dynamic_image_size=False,
-        use_thumbnail=False,
-        loss_version="v1",
-        min_dynamic_tiles=1,
-        max_dynamic_tiles=6,
-        mlp_checkpoint=False,
-        initializer_range=0.02,
-        _attn_implementation="flash_attention_2",
-        _attn_implementation_autoset=False,
-        llm_config=None,
-        image_token_index=None,
-        use_pixel_shuffle=True,
-        mlp_connector_layers=2,
-        **kwargs,
-    ):
-        super().__init__(**kwargs)
-
-        if vision_config is None:
-            vision_config = {"model_type": "siglip_vision_model"}
-            logger.info("vision_config is None. Initializing the InternVisionConfig with default values.")
-
-        if text_config is None:
-            text_config = {"architectures": ["Qwen2ForCausalLM"]}
-            logger.info(
-                "text_config is None. Initializing the LlamaConfig config with default values (`LlamaConfig`)."
-            )
-
-        if vision_config["model_type"] == "siglip_vision_model":
-            self.vision_config = SiglipVisionConfig(**vision_config)
-        else:
-            raise ValueError("Unsupported model_type: {}".format(vision_config["model_type"]))
-
-        if text_config["architectures"][0] == "LlamaForCausalLM":
-            self.text_config = LlamaConfig(**text_config)
-        elif text_config["architectures"][0] == "Qwen2ForCausalLM":
-            self.text_config = Qwen2Config(**text_config)
-        elif text_config["architectures"][0] == "Qwen3ForCausalLM":
-            self.text_config = Qwen3Config(**text_config)
-        else:
-            raise ValueError("Unsupported architecture: {}".format(text_config["architectures"][0]))
-        self.use_backbone_lora = use_backbone_lora
-        self.use_llm_lora = use_llm_lora
-        self.mlp_checkpoint = mlp_checkpoint
-        self.pad2square = pad2square
-        self.select_layer = select_layer
-        self.force_image_size = force_image_size
-        self.downsample_ratio = downsample_ratio
-        self.template = template
-        self.dynamic_image_size = dynamic_image_size
-        self.use_thumbnail = use_thumbnail
-        self.loss_version = loss_version
-        self.initializer_range = initializer_range
-        self.min_dynamic_tiles = min_dynamic_tiles
-        self.max_dynamic_tiles = max_dynamic_tiles
-        self.tie_word_embeddings = self.text_config.tie_word_embeddings
-        self._attn_implementation = _attn_implementation
-        self._attn_implementation_autoset = _attn_implementation_autoset
-        self.image_token_index = image_token_index
-        self.use_pixel_shuffle = use_pixel_shuffle
-        self.mlp_connector_layers = mlp_connector_layers
-        logger.info(f"min_dynamic_tiles: {self.min_dynamic_tiles}")
-        logger.info(f"max_dynamic_tiles: {self.max_dynamic_tiles}")
-
-    def to_dict(self):
-        """
-        Serializes this instance to a Python dictionary. Override the default [`~PretrainedConfig.to_dict`].
-
-        Returns:
-            `Dict[str, any]`: Dictionary of all the attributes that make up this configuration instance,
-        """
-        output = copy.deepcopy(self.__dict__)
-        output["vision_config"] = self.vision_config.to_dict()
-        output["text_config"] = self.text_config.to_dict()
-        output["model_type"] = self.__class__.model_type
-        output["use_backbone_lora"] = self.use_backbone_lora
-        output["use_llm_lora"] = self.use_llm_lora
-        output["pad2square"] = self.pad2square
-        output["select_layer"] = self.select_layer
-        output["force_image_size"] = self.force_image_size
-        output["downsample_ratio"] = self.downsample_ratio
-        output["template"] = self.template
-        output["dynamic_image_size"] = self.dynamic_image_size
-        output["use_thumbnail"] = self.use_thumbnail
-        output["min_dynamic_tiles"] = self.min_dynamic_tiles
-        output["max_dynamic_tiles"] = self.max_dynamic_tiles
-        output["tie_word_embeddings"] = self.tie_word_embeddings
-        output["_attn_implementation"] = self._attn_implementation
-        output["_attn_implementation_autoset"] = self._attn_implementation_autoset
-        output["use_pixel_shuffle"] = self.use_pixel_shuffle
-        output["mlp_connector_layers"] = self.mlp_connector_layers
-        return output

src/lerobot/policies/groot/eagle2_hg_model/image_processing_eagle2_5_vl_fast.py

@@ -1,503 +0,0 @@
-# --------------------------------------------------------
-# NVIDIA
-# Copyright (c) 2025 NVIDIA
-# Licensed under The MIT License [see LICENSE for details]
-# --------------------------------------------------------
-
-from __future__ import annotations
-
-# copy from https://github.com/huggingface/transformers/blob/main/src/transformers/models/llava_onevision/image_processing_llava_onevision_fast.py
-from transformers.image_processing_utils import (
-    BatchFeature,
-    get_patch_output_size,
-)
-from transformers.image_processing_utils_fast import (
-    BaseImageProcessorFast,
-    ImagesKwargs,
-    group_images_by_shape,
-    reorder_images,
-)
-from transformers.image_utils import (
-    IMAGENET_STANDARD_MEAN,  # 0.5, 0.5, 0.5
-    IMAGENET_STANDARD_STD,  # 0.5, 0.5, 0.5
-    ChannelDimension,
-    ImageInput,
-    PILImageResampling,
-    SizeDict,
-    get_image_size,
-    make_flat_list_of_images,
-    validate_kwargs,
-)
-from transformers.processing_utils import Unpack
-from transformers.utils import (
-    TensorType,
-    add_start_docstrings,
-    is_torch_available,
-    is_torchvision_v2_available,
-)
-from transformers.video_utils import VideoInput
-
-if is_torch_available():
-    import torch
-if is_torchvision_v2_available():
-    from torchvision.transforms.v2 import functional as F  # noqa: N812
-    from transformers.image_utils import pil_torch_interpolation_mapping
-else:
-    from torchvision.transforms import functional as F  # noqa: N812
-
-
-def crop(img: torch.Tensor, left: int, top: int, right: int, bottom: int) -> torch.Tensor:
-    """Crop the given numpy array.
-
-    Args:
-        img (torch.Tensor): Image to be cropped. Format should be (C, H, W).
-        left (int): The left coordinate of the crop box.
-        top (int): The top coordinate of the crop box.
-        right (int): The right coordinate of the crop box.
-        bottom (int): The bottom coordinate of the crop box.
-
-    Returns:
-        torch.Tensor: Cropped image.
-    """
-    if not isinstance(img, torch.Tensor):
-        raise TypeError(f"img should be torch.Tensor. Got {type(img)}")
-
-    if img.ndim not in [2, 3]:
-        raise ValueError(f"Image should have 2 or 3 dimensions. Got {img.ndim}")
-
-    img_height = img.shape[1]
-    img_width = img.shape[2]
-    if top < 0 or left < 0 or bottom > img_height or right > img_width:
-        raise ValueError("Crop coordinates out of bounds")
-
-    if top >= bottom or left >= right:
-        raise ValueError("Invalid crop coordinates")
-
-    return img[:, top:bottom, left:right]
-
-
-class Eagle25VLFastImageProcessorKwargs(ImagesKwargs):
-    max_dynamic_tiles: int | None
-    min_dynamic_tiles: int | None
-    use_thumbnail: bool | None
-    pad_during_tiling: bool | None
-    do_pad: bool | None
-
-
-@add_start_docstrings(
-    "Constructs a fast ConvNeXT image processor. Based on [`SiglipImageProcessor`] with incorporation of processing each video frame.",
-    # BASE_IMAGE_PROCESSOR_FAST_DOCSTRING, TODO: this was depreciated from transformers remove!
-    """
-        image_grid_pinpoints (`List[List[int]]`, *optional*):
-            A list of possible resolutions to use for processing high resolution images. The best resolution is selected
-            based on the original size of the image. Can be overridden by `image_grid_pinpoints` in the `preprocess`
-            method. Not used for processing videos.
-        do_pad (`bool`, *optional*):
-            Whether to pad the image. If `True`, will pad the patch dimension of the images in the batch to the largest
-            number of patches in the batch. Padding will be applied to the bottom and right with zeros.
-    """,
-)
-class Eagle25VLImageProcessorFast(BaseImageProcessorFast):
-    resample = PILImageResampling.BICUBIC
-    image_mean = IMAGENET_STANDARD_MEAN
-    image_std = IMAGENET_STANDARD_STD
-    size = {"height": 448, "width": 448}
-    default_to_square = False
-    crop_size = None
-    do_resize = True
-    do_center_crop = None
-    do_rescale = True
-    do_normalize = True
-    do_convert_rgb = True
-    do_pad = True
-    max_dynamic_tiles = 12
-    min_dynamic_tiles = 1
-    use_thumbnail = True
-    pad_during_tiling = False
-    valid_kwargs = Eagle25VLFastImageProcessorKwargs
-    model_input_names = ["pixel_values_videos"]
-
-    def __init__(self, **kwargs: Unpack[Eagle25VLFastImageProcessorKwargs]):
-        super().__init__(**kwargs)
-
-    @add_start_docstrings(
-        # BASE_IMAGE_PROCESSOR_FAST_DOCSTRING_PREPROCESS, TODO: this was depreciated from transformers remove!
-        """
-            max_dynamic_tiles (`int`, *optional*):
-                The maximum number of dynamic tiles to use for processing high resolution images.
-            min_dynamic_tiles (`int`, *optional*):
-                The minimum number of dynamic tiles to use for processing high resolution images.
-            use_thumbnail (`bool`, *optional*):
-                Whether to use a thumbnail for processing high resolution images.
-            pad_during_tiling (`bool`, *optional*):
-                Whether to pad the image during tiling.
-            do_pad (`bool`, *optional*):
-                    Whether to pad the image. If `True`, will pad the patch dimension of the images in the batch to the largest
-                    number of patches in the batch. Padding will be applied to the bottom and right with zeros.
-        """,
-    )
-
-    # NOTE(YL): we will overload the preprocess method to add the image_flags
-    # def preprocess(
-    #     self, images: ImageInput, **kwargs: Unpack[Eagle25VLFastImageProcessorKwargs]
-    # ) -> BatchFeature:
-    #     return super().preprocess(images, **kwargs)
-
-    def _prepare_images_structure(
-        self,
-        images: ImageInput,
-        expected_ndims: int = 3,
-    ) -> ImageInput:
-        """
-        Prepare the images structure for processing.
-
-        Args:
-            images (`ImageInput`):
-                The input images to process.
-            expected_ndims (`int`, *optional*, defaults to 3):
-                Expected number of dimensions for the images (added for transformers >=4.53.0 compatibility).
-
-        Returns:
-            `ImageInput`: The images with a valid nesting.
-        """
-        return make_flat_list_of_images(images)
-
-    def _resize_for_patching(
-        self,
-        image: torch.Tensor,
-        target_resolution: tuple,
-        interpolation: F.InterpolationMode,
-        input_data_format: ChannelDimension,
-    ) -> torch.Tensor:
-        """
-        Resizes an image to a target resolution while maintaining aspect ratio.
-
-        Args:
-            image ("torch.Tensor"):
-                The input image.
-            target_resolution (tuple):
-                The target resolution (height, width) of the image.
-            interpolation (`InterpolationMode`):
-                Resampling filter to use if resizing the image.
-            input_data_format (`ChannelDimension` or `str`):
-                The channel dimension format of the input image.
-
-        Returns:
-            "torch.Tensor": The resized and padded image.
-        """
-        new_height, new_width = get_patch_output_size(image, target_resolution, input_data_format)
-
-        # Resize the image
-        resized_image = F.resize(image, (new_height, new_width), interpolation=interpolation)
-
-        return resized_image
-
-    def find_closest_aspect_ratio(self, aspect_ratio, target_ratios, width, height, image_size):
-        """
-        previous version mainly focus on ratio.
-        We also consider area ratio here.
-        """
-        best_factor = float("-inf")
-        best_ratio = (1, 1)
-        area = width * height
-        for ratio in target_ratios:
-            target_aspect_ratio = ratio[0] / ratio[1]
-            # ratio_diff = abs(aspect_ratio - target_aspect_ratio)
-            # area_ratio = (ratio[0] * ratio[1] * image_size * image_size) / area
-            """
-            new area > 60% of original image area is enough.
-            """
-            factor_based_on_area_n_ratio = min(
-                (ratio[0] * ratio[1] * image_size * image_size) / area, 0.6
-            ) * min(target_aspect_ratio / aspect_ratio, aspect_ratio / target_aspect_ratio)
-
-            if factor_based_on_area_n_ratio > best_factor:
-                best_factor = factor_based_on_area_n_ratio
-                best_ratio = ratio
-
-        return best_ratio
-
-    def _pad_for_patching(
-        self, image: torch.Tensor, target_resolution: tuple, input_data_format: ChannelDimension
-    ) -> torch.Tensor:
-        """
-        Pad an image to a target resolution while maintaining aspect ratio.
-        """
-        target_height, target_width = target_resolution
-        new_height, new_width = get_patch_output_size(image, target_resolution, input_data_format)
-
-        paste_x = (target_width - new_width) // 2
-        paste_y = (target_height - new_height) // 2
-
-        padded_image = F.pad(image, padding=[paste_x, paste_y, paste_x, paste_y])
-
-        return padded_image
-
-    def _get_image_patches(
-        self,
-        image: torch.Tensor,
-        min_num: int,
-        max_num: int,
-        size: tuple,
-        tile_size: int,
-        use_thumbnail: bool,
-        interpolation: F.InterpolationMode,
-        pad_during_tiling: bool,
-    ) -> list[torch.Tensor]:
-        image_size = get_image_size(image, channel_dim=ChannelDimension.FIRST)
-        orig_height, orig_width = image_size
-        aspect_ratio = orig_width / orig_height
-
-        # calculate the existing image aspect ratio
-        target_ratios = {
-            (i, j)
-            for n in range(min_num, max_num + 1)
-            for i in range(1, n + 1)
-            for j in range(1, n + 1)
-            if i * j <= max_num and i * j >= min_num
-        }
-        target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])
-
-        # find the closest aspect ratio to the target
-        target_aspect_ratio = self.find_closest_aspect_ratio(
-            aspect_ratio, target_ratios, orig_width, orig_height, tile_size
-        )
-
-        # calculate the target width and height
-        target_width = tile_size * target_aspect_ratio[0]
-        target_height = tile_size * target_aspect_ratio[1]
-        blocks = target_aspect_ratio[0] * target_aspect_ratio[1]
-        if pad_during_tiling:
-            resized_image = self._resize_for_patching(
-                image,
-                (target_height, target_width),
-                interpolation=interpolation,
-                input_data_format=ChannelDimension.FIRST,
-            )
-            padded_image = self._pad_for_patching(
-                resized_image,
-                (target_height, target_width),
-                input_data_format=ChannelDimension.FIRST,
-            )
-            image_used_to_split = padded_image
-        else:
-            image_used_to_split = F.resize(image, (target_height, target_width), interpolation=interpolation)
-
-        processed_tiles = []
-        for i in range(blocks):
-            box = (
-                (i % (target_width // tile_size)) * tile_size,
-                (i // (target_width // tile_size)) * tile_size,
-                ((i % (target_width // tile_size)) + 1) * tile_size,
-                ((i // (target_width // tile_size)) + 1) * tile_size,
-            )
-            # split the image
-            split_img = crop(image_used_to_split, box[0], box[1], box[2], box[3])
-            processed_tiles.append(split_img)
-        assert len(processed_tiles) == blocks
-
-        if use_thumbnail and len(processed_tiles) != 1:
-            thumbnail_img = F.resize(image, (tile_size, tile_size), interpolation=interpolation)
-            processed_tiles.append(thumbnail_img)
-
-        return processed_tiles
-
-    def _pad_for_batching(
-        self,
-        pixel_values: list[torch.Tensor],
-    ) -> list[torch.Tensor]:
-        """
-        Pads images on the `num_of_patches` dimension with zeros to form a batch of same number of patches.
-
-        Args:
-            pixel_values (`List[torch.Tensor]`):
-                An array of pixel values of each images of shape (`batch_size`, `num_patches`, `image_in_3D`)
-
-        Returns:
-            List[`torch.Tensor`]: The padded images.
-        """
-        max_patch = max(len(x) for x in pixel_values)
-        pixel_values = [
-            torch.nn.functional.pad(image, pad=[0, 0, 0, 0, 0, 0, 0, max_patch - image.shape[0]])
-            for image in pixel_values
-        ]
-
-        return pixel_values
-
-    def _preprocess(
-        self,
-        images: list[torch.Tensor],
-        do_resize: bool,
-        size: SizeDict,
-        max_dynamic_tiles: int,
-        min_dynamic_tiles: int,
-        use_thumbnail: bool,
-        pad_during_tiling: bool,
-        interpolation: F.InterpolationMode | None,
-        do_center_crop: bool,
-        crop_size: SizeDict,
-        do_rescale: bool,
-        rescale_factor: float,
-        do_normalize: bool,
-        image_mean: float | list[float] | None,
-        image_std: float | list[float] | None,
-        do_pad: bool,
-        return_tensors: str | TensorType | None,
-        pad_size: SizeDict | None = None,  # Added for transformers >=4.53.0 compatibility
-        disable_grouping: bool | None = None,  # Added for transformers >=4.53.0 compatibility
-    ) -> BatchFeature:
-        processed_images = []
-        image_sizes = []
-        # Determine the size tuple
-        if size and size.height and size.width:
-            size_tuple = (size.height, size.width)
-        else:
-            size_tuple = (size.shortest_edge, size.shortest_edge)
-
-        # Determine the patch size
-        if crop_size and crop_size.height:
-            tile_size = crop_size.height
-        elif size and size.height:
-            tile_size = size.height
-        else:
-            tile_size = size.shortest_edge
-
-        for image in images:
-            image_patches = self._get_image_patches(
-                image,
-                min_num=min_dynamic_tiles,
-                max_num=max_dynamic_tiles,
-                size=size_tuple,
-                tile_size=tile_size,
-                use_thumbnail=use_thumbnail,
-                interpolation=interpolation,
-                pad_during_tiling=pad_during_tiling,
-            )
-
-            # Group images by size for batched processing
-            processed_image_patches_grouped = {}
-            # Added for transformers >=4.53.0 compatibility
-            grouped_image_patches, grouped_image_patches_index = group_images_by_shape(
-                image_patches,
-                disable_grouping=disable_grouping,
-            )
-
-            for shape, stacked_image_patches in grouped_image_patches.items():
-                if do_resize:
-                    stacked_image_patches = self.resize(
-                        image=stacked_image_patches,
-                        size=size,
-                        interpolation=interpolation,
-                    )
-                if do_center_crop:
-                    stacked_image_patches = self.center_crop(stacked_image_patches, crop_size)
-                # Fused rescale and normalize
-                stacked_image_patches = self.rescale_and_normalize(
-                    stacked_image_patches,
-                    do_rescale,
-                    rescale_factor,
-                    do_normalize,
-                    image_mean,
-                    image_std,
-                )
-                processed_image_patches_grouped[shape] = stacked_image_patches
-            processed_image_patches = reorder_images(
-                processed_image_patches_grouped, grouped_image_patches_index
-            )
-            processed_image_patches = (
-                torch.stack(processed_image_patches, dim=0) if return_tensors else processed_image_patches
-            )
-            processed_images.append(processed_image_patches)
-            image_sizes.append(get_image_size(image, ChannelDimension.FIRST))
-
-        if do_pad:
-            processed_images = self._pad_for_batching(processed_images)
-
-        # processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
-        processed_images = torch.cat(processed_images, dim=0) if return_tensors else processed_images
-        return BatchFeature(
-            data={"pixel_values": processed_images, "image_sizes": image_sizes},
-            tensor_type=return_tensors,
-        )
-
-    def preprocess(
-        self,
-        images: ImageInput,
-        videos: VideoInput = None,
-        **kwargs: Unpack[Eagle25VLFastImageProcessorKwargs],
-    ) -> BatchFeature:
-        validate_kwargs(
-            captured_kwargs=kwargs.keys(),
-            valid_processor_keys=self.valid_kwargs.__annotations__.keys(),
-        )
-        # Set default kwargs from self. This ensures that if a kwarg is not provided
-        # by the user, it gets its default value from the instance, or is set to None.
-        for kwarg_name in self.valid_kwargs.__annotations__:
-            kwargs.setdefault(kwarg_name, getattr(self, kwarg_name, None))
-
-        # Extract parameters that are only used for preparing the input images
-        do_convert_rgb = kwargs.pop("do_convert_rgb")
-        input_data_format = kwargs.pop("input_data_format")
-        device = kwargs.pop("device")
-        # Prepare input images
-        # transformers >= 4.53.0: uses _prepare_image_like_inputs instead of _prepare_input_images
-        if images is not None:
-            images = self._prepare_image_like_inputs(
-                images=images,
-                do_convert_rgb=do_convert_rgb,
-                input_data_format=input_data_format,
-                device=device,
-            )
-
-        if videos is not None:
-            videos = self._prepare_image_like_inputs(
-                images=videos,
-                do_convert_rgb=do_convert_rgb,
-                input_data_format=input_data_format,
-                device=device,
-            )
-
-        # Update kwargs that need further processing before being validated
-        kwargs = self._further_process_kwargs(**kwargs)
-
-        # Validate kwargs
-        self._validate_preprocess_kwargs(**kwargs)
-
-        # torch resize uses interpolation instead of resample
-        # Added for transformers >=4.53.0 compatibility
-        resample = kwargs.pop("resample", self.resample)
-        kwargs["interpolation"] = (
-            pil_torch_interpolation_mapping[resample]
-            if isinstance(resample, PILImageResampling | int)
-            else resample
-        )
-
-        # Filter kwargs to only include those accepted by _preprocess
-        valid_preprocess_kwargs = {
-            "do_resize",
-            "size",
-            "max_dynamic_tiles",
-            "min_dynamic_tiles",
-            "use_thumbnail",
-            "pad_during_tiling",
-            "interpolation",
-            "do_center_crop",
-            "crop_size",
-            "do_rescale",
-            "rescale_factor",
-            "do_normalize",
-            "image_mean",
-            "image_std",
-            "do_pad",
-            "return_tensors",
-            "pad_size",
-            "disable_grouping",
-        }
-        filtered_kwargs = {k: v for k, v in kwargs.items() if k in valid_preprocess_kwargs}
-        if images is not None:
-            return self._preprocess(images, **filtered_kwargs)
-        elif videos is not None:
-            return self._preprocess(videos, **filtered_kwargs)
-
-
-__all__ = ["Eagle25VLImageProcessorFast"]

src/lerobot/policies/groot/eagle2_hg_model/modeling_eagle2_5_vl.py

@@ -1,396 +0,0 @@
-# --------------------------------------------------------
-# NVIDIA
-# Copyright (c) 2025 NVIDIA
-# Licensed under The MIT License [see LICENSE for details]
-# --------------------------------------------------------
-
-import inspect
-
-import torch
-import torch.utils.checkpoint as cp
-from peft import LoraConfig, get_peft_model
-from torch import nn
-from torch.nn import CrossEntropyLoss
-from transformers import GenerationConfig
-from transformers.generation import GenerationMixin
-from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.modeling_utils import PreTrainedModel
-from transformers.models.llama.modeling_llama import LlamaForCausalLM
-from transformers.models.qwen2.modeling_qwen2 import Qwen2ForCausalLM
-from transformers.models.qwen3.modeling_qwen3 import Qwen3ForCausalLM
-from transformers.models.siglip.modeling_siglip import SiglipVisionModel
-from transformers.utils import add_start_docstrings, logging
-
-from .configuration_eagle2_5_vl import Eagle25VLConfig
-
-logger = logging.get_logger(__name__)
-
-
-# copy from https://github.com/huggingface/transformers/blob/main/src/transformers/models/llava_onevision/modeling_llava_onevision.py#L241C1-L280C1
-EAGLE2_5_VL_START_DOCSTRING = r"""
-    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
-    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
-    etc.)
-
-    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
-    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
-    and behavior.
-
-    Parameters:
-        config ([`Eagle25VLConfig`]):
-            Model configuration class with all the parameters of the model. Initializing with a config file does not
-            load the weights associated with the model, only the configuration. Check out the
-            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
-"""
-
-
-@add_start_docstrings(
-    "The bare Eagle2_5_VL Model outputting raw hidden-states without any specific head on top.",
-    EAGLE2_5_VL_START_DOCSTRING,
-)
-class Eagle25VLPreTrainedModel(PreTrainedModel):
-    config_class = Eagle25VLConfig
-    base_model_prefix = "model"
-    main_input_name = "input_ids"
-    supports_gradient_checkpointing = True
-    _no_split_modules = [
-        "Qwen2DecoderLayer",
-        "LlamaDecoderLayer",
-        "Siglip2EncoderLayer",
-        "SiglipEncoderLayer",
-    ]
-    _skip_keys_device_placement = "past_key_values"
-    _supports_flash_attn = True
-    _supports_flash_attn_2 = True
-    _supports_cache_class = True
-    _supports_static_cache = True
-    _supports_quantized_cache = True
-    _supports_sdpa = True
-
-    def _init_weights(self, module):
-        std = self.config.initializer_range
-        if isinstance(module, nn.Linear | nn.Conv2d):
-            module.weight.data.normal_(mean=0.0, std=std)
-            if module.bias is not None:
-                module.bias.data.zero_()
-        elif isinstance(module, nn.Embedding):
-            module.weight.data.normal_(mean=0.0, std=std)
-            if module.padding_idx is not None:
-                module.weight.data[module.padding_idx].zero_()
-
-
-class Eagle25VLForConditionalGeneration(Eagle25VLPreTrainedModel, GenerationMixin):
-    config_class = Eagle25VLConfig
-
-    def __init__(self, config: Eagle25VLConfig, vision_model=None, language_model=None):
-        super().__init__(config)
-
-        image_size = config.force_image_size or config.vision_config.image_size
-        patch_size = config.vision_config.patch_size
-        self.patch_size = patch_size
-        if config.use_pixel_shuffle:
-            self.num_image_token = int((image_size // patch_size) ** 2 * (config.downsample_ratio**2))
-        else:
-            self.num_image_token = int((image_size // patch_size) ** 2)
-
-        self.select_layer = config.select_layer
-        self.downsample_ratio = config.downsample_ratio
-        self.loss_version = config.loss_version
-        self.mlp_checkpoint = config.mlp_checkpoint
-        self.use_pixel_shuffle = config.use_pixel_shuffle
-        self.mlp_connector_layers = config.mlp_connector_layers
-        logger.info(f"num_image_token: {self.num_image_token}")
-        logger.info(f"mlp_checkpoint: {self.mlp_checkpoint}")
-        if vision_model is not None:
-            self.vision_model = vision_model
-        else:
-            if config.vision_config.model_type == "siglip_vision_model":
-                config.vision_config._attn_implementation = "flash_attention_2"
-                self.vision_model = SiglipVisionModel(config.vision_config)
-            else:
-                raise NotImplementedError(f"{config.vision_config.model_type} is not implemented.")
-
-        if language_model is not None:
-            self.language_model = language_model
-        else:
-            if config.text_config.architectures[0] == "LlamaForCausalLM":
-                self.language_model = LlamaForCausalLM(config.text_config)
-            elif config.text_config.architectures[0] == "Phi3ForCausalLM":
-                raise NotImplementedError("Phi3 is not implemented.")
-                # self.language_model = Phi3ForCausalLM(config.text_config)
-            elif config.text_config.architectures[0] == "Qwen2ForCausalLM":
-                assert config.text_config._attn_implementation == "flash_attention_2", (
-                    f"Qwen2 must use flash_attention_2 but got {config.text_config._attn_implementation}"
-                )
-                self.language_model = Qwen2ForCausalLM(config.text_config)
-            elif config.text_config.architectures[0] == "Qwen3ForCausalLM":
-                self.language_model = Qwen3ForCausalLM(config.text_config)
-            else:
-                raise NotImplementedError(f"{config.text_config.architectures[0]} is not implemented.")
-
-        vit_hidden_size = config.vision_config.hidden_size
-        llm_hidden_size = config.text_config.hidden_size
-
-        if config.mlp_connector_layers == 2:
-            self.mlp1 = nn.Sequential(
-                nn.LayerNorm(vit_hidden_size * int(1 / self.downsample_ratio) ** 2),
-                nn.Linear(vit_hidden_size * int(1 / self.downsample_ratio) ** 2, llm_hidden_size),
-                nn.GELU(),
-                nn.Linear(llm_hidden_size, llm_hidden_size),
-            )
-        elif config.mlp_connector_layers == 1 and config.use_pixel_shuffle:
-            self.mlp1 = nn.Sequential(
-                nn.Linear(vit_hidden_size * int(1 / self.downsample_ratio) ** 2, llm_hidden_size),
-            )
-        elif config.mlp_connector_layers == 1 and not config.use_pixel_shuffle:
-            self.mlp1 = nn.Sequential(
-                nn.Linear(vit_hidden_size, llm_hidden_size),
-            )
-        else:
-            raise NotImplementedError(f"{config.mlp_connector_layers} is not implemented.")
-
-        self.image_token_index = config.image_token_index
-        self.neftune_alpha = None
-
-        if config.use_backbone_lora:
-            self.wrap_backbone_lora(r=config.use_backbone_lora, lora_alpha=2 * config.use_backbone_lora)
-
-        self.use_llm_lora = config.use_llm_lora
-        if config.use_llm_lora:
-            self.wrap_llm_lora(r=config.use_llm_lora, lora_alpha=2 * config.use_llm_lora)
-
-        self.check_forward_kwargs()
-
-    def check_forward_kwargs(self):
-        # We intentionally avoid using **kwargs in forward because Hugging Face Transformers
-        # has special handling for functions with **kwargs parameters that would affect
-        # how our model is processed during training and inference.
-        forward_params = inspect.signature(self.forward).parameters
-        assert not any(k.kind == inspect.Parameter.VAR_KEYWORD for k in forward_params.values())
-
-    def wrap_backbone_lora(self, r=128, lora_alpha=256, lora_dropout=0.05):
-        lora_config = LoraConfig(
-            r=r,
-            target_modules=[
-                "self_attn.q_proj",
-                "self_attn.k_proj",
-                "self_attn.v_proj",
-                "self_attn.out_proj",
-                "mlp.fc1",
-                "mlp.fc2",
-            ],
-            lora_alpha=lora_alpha,
-            lora_dropout=lora_dropout,
-        )
-        self.vision_model = get_peft_model(self.vision_model, lora_config)
-        self.vision_model.print_trainable_parameters()
-
-    def wrap_llm_lora(self, r=128, lora_alpha=256, lora_dropout=0.05):
-        lora_config = LoraConfig(
-            r=r,
-            target_modules=[
-                "self_attn.q_proj",
-                "self_attn.k_proj",
-                "self_attn.v_proj",
-                "self_attn.o_proj",
-                "mlp.gate_proj",
-                "mlp.down_proj",
-                "mlp.up_proj",
-            ],
-            lora_alpha=lora_alpha,
-            lora_dropout=lora_dropout,
-            task_type="CAUSAL_LM",
-        )
-        self.language_model = get_peft_model(self.language_model, lora_config)
-        self.language_model.enable_input_require_grads()
-        self.language_model.print_trainable_parameters()
-        self.use_llm_lora = True
-
-    def forward(
-        self,
-        pixel_values: torch.FloatTensor,
-        input_ids: torch.LongTensor = None,
-        attention_mask: torch.Tensor | None = None,
-        position_ids: torch.LongTensor | None = None,
-        image_flags: torch.LongTensor | None = None,
-        past_key_values: list[torch.FloatTensor] | None = None,
-        labels: torch.LongTensor | None = None,
-        use_cache: bool | None = None,
-        output_attentions: bool | None = None,
-        output_hidden_states: bool | None = None,
-        return_dict: bool | None = None,
-        num_tiles_list: list[torch.Tensor] | None = None,
-    ) -> tuple | CausalLMOutputWithPast:
-        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
-        input_embeds = self.language_model.get_input_embeddings()(input_ids)
-
-        vit_embeds = self.extract_feature(pixel_values)
-
-        if image_flags is not None:
-            image_flags = image_flags.view(-1)
-            vit_embeds = vit_embeds[image_flags == 1]
-
-        b, n, c = input_embeds.shape
-        input_embeds = input_embeds.reshape(b * n, c)
-
-        input_ids = input_ids.reshape(b * n)
-        selected = input_ids == self.image_token_index
-        try:
-            input_embeds[selected] = input_embeds[selected] * 0.0 + vit_embeds.reshape(-1, c)
-        except Exception as e:
-            vit_embeds = vit_embeds.reshape(-1, c)
-            print(
-                f"warning: {e}, input_embeds[selected].shape={input_embeds[selected].shape}, "
-                f"vit_embeds.shape={vit_embeds.shape}"
-            )
-            n_token = selected.sum()
-            input_embeds[selected] = input_embeds[selected] * 0.0 + vit_embeds[:n_token]
-
-        input_embeds = input_embeds.reshape(b, n, c)
-
-        outputs = self.language_model(
-            inputs_embeds=input_embeds,
-            attention_mask=attention_mask,
-            position_ids=position_ids,
-            past_key_values=past_key_values,
-            use_cache=use_cache,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-        )
-        logits = outputs.logits
-
-        loss = None
-        if labels is not None:
-            # Shift so that tokens < n predict n
-            shift_logits = logits[..., :-1, :].contiguous()
-            shift_labels = labels[..., 1:].contiguous()
-            # Flatten the tokens
-            loss_fct = CrossEntropyLoss()
-            shift_logits = shift_logits.view(-1, self.language_model.config.vocab_size)
-            shift_labels = shift_labels.view(-1)
-            # Enable model parallelism
-            shift_labels = shift_labels.to(shift_logits.device)
-            loss = loss_fct(shift_logits, shift_labels)
-
-        if not return_dict:
-            output = (logits,) + outputs[1:]
-            return (loss,) + output if loss is not None else output
-
-        return CausalLMOutputWithPast(
-            loss=loss,
-            logits=logits,
-            past_key_values=outputs.past_key_values,
-            hidden_states=outputs.hidden_states,
-            attentions=outputs.attentions,
-        )
-
-    def pixel_shuffle(self, x, scale_factor=0.5):
-        n, w, h, c = x.size()
-        # N, W, H, C --> N, W, H * scale, C // scale
-        x = x.view(n, w, int(h * scale_factor), int(c / scale_factor))
-        # N, W, H * scale, C // scale --> N, H * scale, W, C // scale
-        x = x.permute(0, 2, 1, 3).contiguous()
-        # N, H * scale, W, C // scale --> N, H * scale, W * scale, C // (scale ** 2)
-        x = x.view(n, int(h * scale_factor), int(w * scale_factor), int(c / (scale_factor * scale_factor)))
-
-        x = x.permute(0, 2, 1, 3).contiguous()
-        return x
-
-    def extract_feature(self, pixel_values):
-        if self.select_layer == -1:
-            vit_embeds = self.vision_model(
-                pixel_values=pixel_values, output_hidden_states=False, return_dict=True
-            )
-            if hasattr(vit_embeds, "last_hidden_state"):
-                vit_embeds = vit_embeds.last_hidden_state
-
-        else:
-            vit_embeds = self.vision_model(
-                pixel_values=pixel_values, output_hidden_states=True, return_dict=True
-            ).hidden_states[self.select_layer]
-
-        if self.use_pixel_shuffle:
-            h = w = int(vit_embeds.shape[1] ** 0.5)
-            vit_embeds = vit_embeds.reshape(vit_embeds.shape[0], h, w, -1)
-            vit_embeds = self.pixel_shuffle(
-                vit_embeds, scale_factor=self.downsample_ratio
-            )  # torch.Size([B, 1024, 1024]) -> torch.Size([B, 16, 16, 4096])
-            vit_embeds = vit_embeds.reshape(
-                vit_embeds.shape[0], -1, vit_embeds.shape[-1]
-            )  # torch.Size([B, 16, 16, 4096]) -> torch.Size([B, 256, 4096])
-
-        if self.mlp_checkpoint and vit_embeds.requires_grad:
-            vit_embeds = cp.checkpoint(self.mlp1, vit_embeds)
-        else:
-            vit_embeds = self.mlp1(vit_embeds)
-
-        return vit_embeds
-
-    @torch.no_grad()
-    def generate(
-        self,
-        pixel_values: torch.FloatTensor | None = None,
-        input_ids: torch.FloatTensor | None = None,
-        attention_mask: torch.LongTensor | None = None,
-        visual_features: torch.FloatTensor | None = None,
-        generation_config: GenerationConfig | None = None,
-        output_hidden_states: bool | None = None,
-        image_sizes: list[tuple[int, int]] | None = None,
-        **generate_kwargs,
-    ) -> torch.LongTensor:
-        if pixel_values is not None:
-            if visual_features is not None:
-                vit_embeds = visual_features
-            else:
-                vit_embeds = self.extract_feature(pixel_values)
-
-            input_embeds = self.language_model.get_input_embeddings()(input_ids)
-            b, n, c = input_embeds.shape
-            input_embeds = input_embeds.reshape(b * n, c)
-
-            input_ids = input_ids.reshape(b * n)
-            selected = input_ids == self.config.image_token_index
-            assert selected.sum() != 0
-            input_embeds[selected] = vit_embeds.reshape(-1, c).to(input_embeds.device)
-
-            input_embeds = input_embeds.reshape(b, n, c)
-        else:
-            input_embeds = self.language_model.get_input_embeddings()(input_ids)
-
-        if "use_cache" not in generate_kwargs:
-            generate_kwargs["use_cache"] = True
-
-        outputs = self.language_model.generate(
-            inputs_embeds=input_embeds,
-            attention_mask=attention_mask,
-            generation_config=generation_config,
-            output_hidden_states=output_hidden_states,
-            **generate_kwargs,
-        )
-
-        return outputs
-
-    # Copied from transformers.models.llava_next.modeling_llava_next.LlavaNextForConditionalGeneration.get_input_embeddings
-    def get_input_embeddings(self):
-        return self.language_model.get_input_embeddings()
-
-    # Copied from transformers.models.llava_next.modeling_llava_next.LlavaNextForConditionalGeneration.set_input_embeddings
-    def set_input_embeddings(self, value):
-        self.language_model.set_input_embeddings(value)
-
-    # Copied from transformers.models.llava_next.modeling_llava_next.LlavaNextForConditionalGeneration.get_output_embeddings
-    def get_output_embeddings(self):
-        return self.language_model.get_output_embeddings()
-
-    # Copied from transformers.models.llava_next.modeling_llava_next.LlavaNextForConditionalGeneration.set_output_embeddings
-    def set_output_embeddings(self, new_embeddings):
-        self.language_model.set_output_embeddings(new_embeddings)
-
-    # Copied from transformers.models.llava_next.modeling_llava_next.LlavaNextForConditionalGeneration.set_decoder
-    def set_decoder(self, decoder):
-        self.language_model.set_decoder(decoder)
-
-    # Copied from transformers.models.llava_next.modeling_llava_next.LlavaNextForConditionalGeneration.get_decoder
-    def get_decoder(self):
-        return self.language_model.get_decoder()

src/lerobot/policies/groot/eagle2_hg_model/processing_eagle2_5_vl.py

@@ -1,541 +0,0 @@
-# Copyright 2024 The HuggingFace Inc. team.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-Processor class for Eagle25VL.
-copy from https://github.com/huggingface/transformers/blob/main/src/transformers/models/llava_onevision/processing_llava_onevision.py
-"""
-
-import base64
-import os
-import re
-from io import BytesIO
-
-import requests
-import torch
-from PIL import Image
-from transformers.feature_extraction_utils import BatchFeature
-from transformers.image_utils import ImageInput
-from transformers.processing_utils import ProcessingKwargs, ProcessorMixin, Unpack
-from transformers.tokenization_utils_base import PreTokenizedInput, TextInput
-from transformers.utils import logging
-from transformers.video_utils import VideoInput
-
-logger = logging.get_logger(__name__)
-
-
-FRAME_FACTOR = 2
-FPS = 2.0
-FPS_MIN_FRAMES = 4
-FPS_MAX_FRAMES = 256
-
-
-def to_rgb(pil_image: Image.Image) -> Image.Image:
-    if pil_image.mode == "RGBA":
-        white_background = Image.new("RGB", pil_image.size, (255, 255, 255))
-        white_background.paste(pil_image, mask=pil_image.split()[3])  # Use alpha channel as mask
-        return white_background
-    else:
-        return pil_image.convert("RGB")
-
-
-def fetch_image(ele: dict[str, str | Image.Image]) -> Image.Image:
-    image = ele["image"] if "image" in ele else ele["image_url"]
-    image_obj = None
-    if isinstance(image, Image.Image):
-        image_obj = image
-    elif image.startswith("http://") or image.startswith("https://"):
-        response = requests.get(image, stream=True, timeout=10)
-        image_obj = Image.open(BytesIO(response.content))
-    elif image.startswith("file://"):
-        image_obj = Image.open(image[7:])
-    elif image.startswith("data:image"):
-        if "base64," in image:
-            _, base64_data = image.split("base64,", 1)
-            data = base64.b64decode(base64_data)
-            image_obj = Image.open(BytesIO(data))
-    else:
-        image_obj = Image.open(image)
-    if image_obj is None:
-        raise ValueError(
-            f"Unrecognized image input, support local path, http url, base64 and PIL.Image, got {image}"
-        )
-    image = to_rgb(image_obj)
-    if "scale_factor" in ele:
-        scale_factor = ele["scale_factor"]
-        image = image.resize((image.width * scale_factor, image.height * scale_factor), Image.BILINEAR)
-    return image
-
-
-class Eagle25VLProcessorKwargs(ProcessingKwargs, total=False):
-    # see processing_utils.ProcessingKwargs documentation for usage.
-    _defaults = {
-        "text_kwargs": {
-            "padding": False,
-        },
-        "images_kwargs": {},
-        "videos_kwargs": {"max_dynamic_tiles": 1},
-    }
-
-
-class Eagle25VLProcessor(ProcessorMixin):
-    r"""
-    Constructs a Eagle25VL processor which wraps a Eagle25VL video processor, Eagle25VL image processor and a Eagle25VL tokenizer into a single processor.
-
-    [`Eagle25VLProcessor`] offers all the functionalities of [`Eagle25VLVideoProcessor`], [`Eagle25VLImageProcessor`] and [`Eagle25VLTokenizer`]. See the
-    [`~Eagle25VLVideoProcessor.__call__`], [`~Eagle25VLProcessor.__call__`] and [`~Eagle25VLProcessor.decode`] for more information.
-
-    Args:
-        image_processor ([`LlavaOnevisionImageProcessor`], *optional*):
-            The image processor is a required input.
-        tokenizer ([`LlamaTokenizerFast`], *optional*):
-            The tokenizer is a required input.
-        num_image_tokens (`int`, *optional*):
-            Number of image tokens for one imagethat will be returned by vision tower.
-        vision_feature_select_strategy (`str`, *optional*):
-            The feature selection strategy used to select the vision feature from the vision backbone.
-            Should be same as in model's config
-        chat_template (`str`, *optional*): A Jinja template which will be used to convert lists of messages
-            in a chat into a tokenizable string.
-        image_token (`str`, *optional*, defaults to `"<image>"`):
-            Special token used to denote image location.
-        video_token (`str`, *optional*, defaults to `"<video>"`):
-            Special token used to denote video location.
-    """
-
-    attributes = ["image_processor", "tokenizer"]
-    valid_kwargs = [
-        "chat_template",
-        "num_image_tokens",
-        "vision_feature_select_strategy",
-        "image_token",
-        "video_token",
-        "images_kwargs",
-        "videos_kwargs",
-        "text_kwargs",
-    ]
-    tokenizer_class = "AutoTokenizer"
-
-    def __init__(
-        self,
-        image_processor=None,
-        tokenizer=None,
-        vision_feature_select_strategy=None,
-        chat_template=None,
-        image_token="<IMG_CONTEXT>",  # nosec: B107
-        video_token="<IMG_CONTEXT>",  # nosec: B107
-        tokens_per_tile=256,
-        image_placeholder="image",
-        video_placeholder="video",
-        image_start_token="<img>",
-        image_end_token="</img>",
-        **kwargs,
-    ):
-        self.vision_feature_select_strategy = vision_feature_select_strategy
-        self.image_token = tokenizer.image_token if hasattr(tokenizer, "image_token") else image_token
-        self.video_token = tokenizer.video_token if hasattr(tokenizer, "video_token") else video_token
-        self.image_token_id = (
-            tokenizer.image_token_id
-            if getattr(tokenizer, "image_token_id", None)
-            else tokenizer.convert_tokens_to_ids(self.image_token)
-        )
-        self.video_token_id = (
-            tokenizer.video_token_id
-            if getattr(tokenizer, "video_token_id", None)
-            else tokenizer.convert_tokens_to_ids(self.video_token)
-        )
-        self.image_placeholder = image_placeholder
-        self.video_placeholder = video_placeholder
-        self.tokens_per_tile = tokens_per_tile
-        self.image_start_token = image_start_token
-        self.image_end_token = image_end_token
-        if "auto_map" in kwargs:
-            self.auto_map = kwargs["auto_map"]
-        super().__init__(image_processor, tokenizer, chat_template=chat_template)
-
-    def replace_media_placeholder(
-        self, text, image_list, video_list, timestamps_list, fps_list, **output_kwargs
-    ):
-        num_of_images_in_this_sample = 0
-        num_of_videos_in_this_sample = 0
-        # Regular expression pattern to match formats like <image-1> or <video-2>
-        pattern = re.compile(rf"<({self.image_placeholder}|{self.video_placeholder})-(\d+)>")
-        unified_frame_list = []
-
-        # image_min_dynamic_tiles = output_kwargs["images_kwargs"].get(
-        #     "min_dynamic_tiles", self.image_processor.min_dynamic_tiles
-        # )
-        # image_max_dynamic_tiles = output_kwargs["images_kwargs"].get(
-        #     "max_dynamic_tiles", self.image_processor.max_dynamic_tiles
-        # )
-        # image_use_thumbnail = output_kwargs["images_kwargs"].get(
-        #     "use_thumbnail", self.image_processor.use_thumbnail
-        # )
-        video_min_dynamic_tiles = output_kwargs["videos_kwargs"].get(
-            "min_dynamic_tiles", self.image_processor.min_dynamic_tiles
-        )
-        video_max_dynamic_tiles = output_kwargs["videos_kwargs"].get(
-            "max_dynamic_tiles", self.image_processor.max_dynamic_tiles
-        )
-        video_use_thumbnail = output_kwargs["videos_kwargs"].get(
-            "use_thumbnail", self.image_processor.use_thumbnail
-        )
-
-        tile_size = self.image_processor.size.get("height", 448)
-
-        # Function to replace tags in a single text
-        def replace_in_text(text):
-            # repl callback function for each match replacement operation
-            def repl(match):
-                nonlocal unified_frame_list
-                nonlocal num_of_images_in_this_sample
-                nonlocal num_of_videos_in_this_sample
-                media_type = match.group(1)  # 'image' or 'video'
-                idx_in_list = int(match.group(2)) - 1  # Convert to list index (0-based)
-                # Select the corresponding path based on media type
-                idx_mapper = {
-                    0: "first",
-                    1: "second",
-                    2: "third",
-                    3: "fourth",
-                    4: "fifth",
-                    5: "sixth",
-                    6: "seventh",
-                    7: "eighth",
-                    8: "ninth",
-                    9: "tenth",
-                }
-                if media_type == "image":
-                    image_inputs = self.image_processor(
-                        images=[image_list[idx_in_list]],
-                        videos=None,
-                        **output_kwargs["images_kwargs"],
-                    )
-                    if isinstance(image_inputs["pixel_values"], list):
-                        _pv = image_inputs["pixel_values"]
-                        if _pv and isinstance(_pv[0], list):
-                            _pv = [t for sub in _pv for t in sub]
-                        image_inputs["pixel_values"] = torch.stack(
-                            [t if isinstance(t, torch.Tensor) else torch.as_tensor(t) for t in _pv]
-                        )
-                    num_all_tiles = image_inputs["pixel_values"].shape[0]
-                    special_placeholder = f"<image {idx_in_list + 1}>{self.image_start_token}{self.image_token * num_all_tiles * self.tokens_per_tile}{self.image_end_token}"
-                    unified_frame_list.append(image_inputs)
-                    num_of_images_in_this_sample += 1
-
-                elif media_type == "video":
-                    video_inputs = self.image_processor(
-                        images=None,
-                        videos=[video_list[idx_in_list]],
-                        **output_kwargs["videos_kwargs"],
-                    )
-                    if isinstance(video_inputs["pixel_values"], list):
-                        _pv = video_inputs["pixel_values"]
-                        if _pv and isinstance(_pv[0], list):
-                            _pv = [t for sub in _pv for t in sub]
-                        video_inputs["pixel_values"] = torch.stack(
-                            [t if isinstance(t, torch.Tensor) else torch.as_tensor(t) for t in _pv]
-                        )
-                    num_all_tiles = video_inputs["pixel_values"].shape[0]
-                    image_sizes = video_inputs["image_sizes"]
-                    if timestamps_list is not None and -1 not in timestamps_list:
-                        frame_timestamps = timestamps_list[idx_in_list]
-                    else:
-                        frame_timestamps = None
-                    sampled_fps = fps_list[idx_in_list] if fps_list is not None else None
-
-                    num_of_tiles_each_frame = [
-                        self.get_number_tiles_based_on_image_size(
-                            image_size,
-                            video_min_dynamic_tiles,
-                            video_max_dynamic_tiles,
-                            video_use_thumbnail,
-                            tile_size,
-                        )
-                        for image_size in image_sizes
-                    ]
-                    assert sum(num_of_tiles_each_frame) == num_all_tiles, (
-                        f"The number of tiles in each frame is not equal to the total number of tiles: {sum(num_of_tiles_each_frame)} != {num_all_tiles}"
-                    )
-
-                    if frame_timestamps is not None:
-                        assert len(frame_timestamps) == len(num_of_tiles_each_frame), (
-                            f"The number of timestamps is not equal to the number of frames: {len(frame_timestamps)} != {len(num_of_tiles_each_frame)}"
-                        )
-                        special_placeholder = [
-                            f"Frame {i + 1} sample at {frame_timestamps[i]:.2f}s: {self.image_start_token}{self.image_token * num_of_tiles * self.tokens_per_tile}{self.image_end_token}"
-                            for i, num_of_tiles in enumerate(num_of_tiles_each_frame)
-                        ]
-                    else:
-                        special_placeholder = [
-                            f"Frame {i + 1}: {self.image_start_token}{self.image_token * num_of_tiles * self.tokens_per_tile}{self.image_end_token}"
-                            for i, num_of_tiles in enumerate(num_of_tiles_each_frame)
-                        ]
-
-                    if sampled_fps is not None:
-                        special_placeholder = (
-                            f"The {idx_mapper[idx_in_list]} video sampled with {sampled_fps:.2f} fps: "
-                            + "".join(special_placeholder)
-                        )
-                    else:
-                        special_placeholder = f"The {idx_mapper[idx_in_list]} video: " + "".join(
-                            special_placeholder
-                        )
-                    unified_frame_list.append(video_inputs)
-                    num_of_videos_in_this_sample += 1
-                else:
-                    raise ValueError(f"Unknown media type: {media_type}")
-                return special_placeholder
-
-            return pattern.sub(repl, text)
-
-        text = replace_in_text(text)
-        if len(unified_frame_list) > 0:
-
-            def _to_tensor(v):
-                if isinstance(v, torch.Tensor):
-                    return v
-                if isinstance(v, list):
-                    if v and isinstance(v[0], list):
-                        v = [t for sub in v for t in sub]
-                    return torch.stack([t if isinstance(t, torch.Tensor) else torch.as_tensor(t) for t in v])
-                return torch.as_tensor(v)
-
-            pixel_values = torch.cat([_to_tensor(frame["pixel_values"]) for frame in unified_frame_list])
-            image_sizes = torch.cat([_to_tensor(frame["image_sizes"]) for frame in unified_frame_list])
-        else:
-            pixel_values = None
-            image_sizes = None
-        return (
-            text,
-            pixel_values,
-            image_sizes,
-            num_of_images_in_this_sample,
-            num_of_videos_in_this_sample,
-        )
-
-    def __call__(
-        self,
-        images: ImageInput = None,
-        text: TextInput | PreTokenizedInput | list[TextInput] | list[PreTokenizedInput] = None,
-        audio=None,
-        videos: VideoInput = None,
-        **kwargs: Unpack[Eagle25VLProcessorKwargs],
-    ) -> BatchFeature:
-        """
-        Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text`
-        and `kwargs` arguments to LlamaTokenizerFast's [`~LlamaTokenizerFast.__call__`] if `text` is not `None` to encode
-        the text. To prepare the image(s), this method forwards the `images` and `kwrags` arguments to
-        LlavaNextImageProcessor's [`~LlavaNextImageProcessor.__call__`] if `images` is not `None`. Please refer to the docstring
-        of the above two methods for more information.
-
-        Args:
-            images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`, `List[torch.Tensor]`):
-                The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
-                tensor. Both channels-first and channels-last formats are supported.
-            text (`str`, `List[str]`, `List[List[str]]`):
-                The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
-                (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
-                `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
-            videos (`np.ndarray`, `torch.Tensor`, `List[np.ndarray]`, `List[torch.Tensor]`):
-                The image or batch of videos to be prepared. Each video can be a 4D NumPy array or PyTorch
-
-        Returns:
-            [`BatchFeature`]: A [`BatchFeature`] with the following fields:
-
-            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
-            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
-              `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
-              `None`).
-            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
-            - **pixel_values_videos** -- Pixel values of a video input to be fed to a model. Returned when `videos` is not `None`.
-            - **image_sizes** -- Size of each image that will be used to unpad an image. Returned when `images` is not `None`.
-        """
-
-        output_kwargs = self._merge_kwargs(
-            Eagle25VLProcessorKwargs,
-            tokenizer_init_kwargs=self.tokenizer.init_kwargs,
-            **kwargs,
-        )
-
-        if isinstance(text, str):
-            text_list = [text]
-        elif not isinstance(text, list) and not isinstance(text[0], str):
-            raise ValueError("Invalid input text. Please provide a string, or a list of strings")
-        elif isinstance(text, list) and isinstance(text[0], str):
-            text_list = text
-
-        if images is None:
-            images = []
-        if videos is None:
-            videos = []
-
-        pixel_values_list = []
-        image_sizes_list = []
-        new_sample_list = []
-        image_start_idx = 0
-        video_start_idx = 0
-        timestamps_batch = output_kwargs["videos_kwargs"].pop("timestamps", None)
-        fps_batch = output_kwargs["videos_kwargs"].pop("fps", None)
-        for sample in text_list:
-            timestamps_list = timestamps_batch[video_start_idx:] if timestamps_batch is not None else None
-            fps_list = fps_batch[video_start_idx:] if fps_batch is not None else None
-            (
-                sample,
-                pixel_values,
-                image_sizes,
-                num_of_images_in_this_sample,
-                num_of_videos_in_this_sample,
-            ) = self.replace_media_placeholder(
-                sample,
-                images[image_start_idx:],
-                videos[video_start_idx:],
-                timestamps_list,
-                fps_list,
-                **output_kwargs,
-            )
-            new_sample_list.append(sample)
-            if pixel_values is not None:
-                pixel_values_list.append(pixel_values)
-                image_sizes_list.append(image_sizes)
-            image_start_idx += num_of_images_in_this_sample
-            video_start_idx += num_of_videos_in_this_sample
-
-        if len(pixel_values_list) > 0:
-            image_inputs = {
-                "pixel_values": torch.cat(pixel_values_list),
-                "image_sizes": torch.cat(image_sizes_list),
-            }
-        else:
-            image_inputs = {}
-        video_inputs = {}
-        text_inputs = self.tokenizer(new_sample_list, **output_kwargs["text_kwargs"])
-        return BatchFeature(data={**text_inputs, **image_inputs, **video_inputs})
-
-    def get_number_tiles_based_on_image_size(
-        self, image_size: tuple, min_num: int, max_num: int, use_thumbnail: bool, tile_size: int
-    ) -> int:
-        """
-        Get the number of tiles based on the image size.
-        """
-        orig_height, orig_width = image_size
-        aspect_ratio = orig_width / orig_height
-        # calculate the existing image aspect ratio
-        target_ratios = {
-            (i, j)
-            for n in range(min_num, max_num + 1)
-            for i in range(1, n + 1)
-            for j in range(1, n + 1)
-            if i * j <= max_num and i * j >= min_num
-        }
-        target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])
-
-        # find the closest aspect ratio to the target
-        target_aspect_ratio = self.image_processor.find_closest_aspect_ratio(
-            aspect_ratio, target_ratios, orig_width, orig_height, tile_size
-        )
-        tiles_num = target_aspect_ratio[0] * target_aspect_ratio[1]
-        if use_thumbnail and tiles_num > 1:
-            tiles_num += 1
-        return tiles_num
-
-    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.batch_decode with CLIP->Llama
-    def batch_decode(self, *args, **kwargs):
-        """
-        This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
-        refer to the docstring of this method for more information.
-        """
-        return self.tokenizer.batch_decode(*args, **kwargs)
-
-    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.decode with CLIP->Llama
-    def decode(self, *args, **kwargs):
-        """
-        This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
-        the docstring of this method for more information.
-        """
-        return self.tokenizer.decode(*args, **kwargs)
-
-    @property
-    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.model_input_names
-    def model_input_names(self):
-        tokenizer_input_names = self.tokenizer.model_input_names
-        image_processor_input_names = self.image_processor.model_input_names
-        return list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))
-
-    # override to save video-config in a separate config file
-    def save_pretrained(self, save_directory, **kwargs):
-        if os.path.isfile(save_directory):
-            raise ValueError(f"Provided path ({save_directory}) should be a directory, not a file")
-        os.makedirs(save_directory, exist_ok=True)
-
-        outputs = super().save_pretrained(save_directory, **kwargs)
-        return outputs
-
-    # override to load video-config from a separate config file
-    @classmethod
-    def from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
-        processor = super().from_pretrained(pretrained_model_name_or_path, **kwargs)
-
-        # if return_unused_kwargs a tuple is returned where the second element is 'unused_kwargs'
-        if isinstance(processor, tuple):
-            processor = processor[0]
-        return processor
-
-    # Copy from https://github.com/QwenLM/Qwen2.5-VL/blob/main/qwen-vl-utils/src/qwen_vl_utils/vision_process.py
-    def process_vision_info(
-        self,
-        conversations: list[dict] | list[list[dict]],
-        return_video_kwargs: bool = False,
-    ) -> tuple[list[Image.Image] | None, list[torch.Tensor | list[Image.Image]] | None, dict | None]:
-        vision_infos = self.extract_vision_info(conversations)
-        ## Read images or videos
-        image_inputs = []
-        video_inputs = []
-        video_sample_fps_list = []
-        video_timestamps_list = []
-        for vision_info in vision_infos:
-            if "image" in vision_info or "image_url" in vision_info:
-                image_inputs.append(fetch_image(vision_info))
-            else:
-                raise ValueError("image, image_url or video should in content.")
-        if len(image_inputs) == 0:
-            image_inputs = None
-        if len(video_inputs) == 0:
-            video_inputs = None
-        if return_video_kwargs:
-            return (
-                image_inputs,
-                video_inputs,
-                {"fps": video_sample_fps_list, "timestamps": video_timestamps_list},
-            )
-        return image_inputs, video_inputs
-
-    def extract_vision_info(self, conversations: list[dict] | list[list[dict]]) -> list[dict]:
-        vision_infos = []
-        if isinstance(conversations[0], dict):
-            conversations = [conversations]
-        for conversation in conversations:
-            for message in conversation:
-                if isinstance(message["content"], list):
-                    for ele in message["content"]:
-                        if (
-                            "image" in ele
-                            or "image_url" in ele
-                            or "video" in ele
-                            or ele["type"] in ("image", "image_url", "video")
-                        ):
-                            vision_infos.append(ele)
-        return vision_infos
-
-
-__all__ = ["Eagle25VLProcessor"]

src/lerobot/policies/groot/groot_n1.py

@@ -1,380 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from pathlib import Path
-from typing import TYPE_CHECKING, Any
-
-import numpy as np
-import torch
-import torch.nn as nn
-from huggingface_hub import snapshot_download
-from huggingface_hub.errors import HFValidationError, RepositoryNotFoundError
-
-from lerobot.utils.import_utils import _transformers_available
-
-# Conditional import for type checking and lazy loading
-if TYPE_CHECKING or _transformers_available:
-    from huggingface_hub.dataclasses import strict
-    from transformers import AutoConfig, AutoModel, PretrainedConfig, PreTrainedModel
-    from transformers.feature_extraction_utils import BatchFeature
-else:
-
-    def strict(cls):
-        return cls
-
-    AutoConfig = None
-    AutoModel = None
-    PretrainedConfig = object
-    PreTrainedModel = object
-    BatchFeature = None
-
-try:
-    import tree
-except ImportError:
-    tree = None
-
-from lerobot.utils.constants import ACTION, HF_LEROBOT_HOME
-
-from .action_head.flow_matching_action_head import (
-    FlowmatchingActionHead,
-    FlowmatchingActionHeadConfig,
-)
-from .utils import ensure_eagle_cache_ready
-
-DEFAULT_VENDOR_EAGLE_PATH = str((Path(__file__).resolve().parent / "eagle2_hg_model").resolve())
-DEFAULT_TOKENIZER_ASSETS_REPO = "lerobot/eagle2hg-processor-groot-n1p5"
-
-
-class EagleBackbone(nn.Module):
-    def __init__(
-        self,
-        tune_llm: bool = False,
-        tune_visual: bool = False,
-        select_layer: int = -1,
-        reproject_vision: bool = False,
-        use_flash_attention: bool = False,
-        load_bf16: bool = False,
-        eagle_path: str = DEFAULT_VENDOR_EAGLE_PATH,
-        tokenizer_assets_repo: str = DEFAULT_TOKENIZER_ASSETS_REPO,
-        project_to_dim: int = 1536,
-    ):
-        """
-        Args:
-            tune_llm: whether to tune the LLM model (default: True)
-            tune_visual: whether to tune the visual model (default: False)
-        """
-        super().__init__()
-        assert not reproject_vision, "Reproject vision is not implemented here, set to False"
-
-        # Prefer loading Eagle model config from the cache directory where vendor files were copied.
-        vendor_dir = DEFAULT_VENDOR_EAGLE_PATH
-        cache_dir = HF_LEROBOT_HOME / tokenizer_assets_repo
-        try:
-            ensure_eagle_cache_ready(vendor_dir, cache_dir, tokenizer_assets_repo)
-        except Exception as exc:  # nosec: B110
-            print(f"[GROOT] Warning: failed to prepare Eagle cache for backbone: {exc}")
-
-        config = AutoConfig.from_pretrained(str(cache_dir), trust_remote_code=True)
-        self.eagle_model = AutoModel.from_config(config, trust_remote_code=True)
-
-        if project_to_dim is not None:
-            self.eagle_linear = torch.nn.Linear(2048, project_to_dim)
-        else:
-            self.eagle_linear = torch.nn.Identity()
-
-        # needed since we don't use these layers. Also saves compute
-        while len(self.eagle_model.language_model.model.layers) > select_layer:
-            self.eagle_model.language_model.model.layers.pop(-1)
-
-        self.select_layer = select_layer
-        self.set_trainable_parameters(tune_llm, tune_visual)
-
-    def set_trainable_parameters(self, tune_llm: bool, tune_visual: bool):
-        self.tune_llm = tune_llm
-        self.tune_visual = tune_visual
-        for p in self.parameters():
-            p.requires_grad = True
-        if not tune_llm:
-            self.eagle_model.language_model.requires_grad_(False)
-        if not tune_visual:
-            self.eagle_model.vision_model.requires_grad_(False)
-            self.eagle_model.mlp1.requires_grad_(False)
-        print(f"Tune backbone llm: {self.tune_llm}")
-        print(f"Tune backbone visual: {self.tune_visual}")
-        # Check if any parameters are still trainable. If not, print a warning.
-        if not tune_llm and not tune_visual:
-            for name, p in self.named_parameters():
-                if p.requires_grad:
-                    print(f"Backbone trainable parameter: {name}")
-        if not any(p.requires_grad for p in self.parameters()):
-            print("Warning: No backbone trainable parameters found.")
-
-    def set_frozen_modules_to_eval_mode(self):
-        """
-        Huggingface will call model.train() at each training_step. To ensure
-        the expected behaviors for modules like dropout, batchnorm, etc., we
-        need to call model.eval() for the frozen modules.
-        """
-        if self.training:
-            if self.eagle_model.language_model and not self.tune_llm:
-                self.eagle_model.language_model.eval()
-            if self.eagle_model.vision_model and not self.tune_visual:
-                self.eagle_model.vision_model.eval()
-
-    def prepare_input(self, batch: dict) -> BatchFeature:
-        return BatchFeature(data=batch)
-
-    def forward_eagle(self, vl_input: BatchFeature) -> BatchFeature:
-        eagle_prefix = "eagle_"
-        eagle_input = {
-            k.removeprefix(eagle_prefix): v for k, v in vl_input.items() if k.startswith(eagle_prefix)
-        }
-        del eagle_input["image_sizes"]
-
-        eagle_output = self.eagle_model(**eagle_input, output_hidden_states=True, return_dict=True)
-        eagle_features = eagle_output.hidden_states[self.select_layer]
-
-        eagle_features = self.eagle_linear(eagle_features)
-        return eagle_features, eagle_input["attention_mask"]
-
-    def forward(self, vl_input: BatchFeature) -> BatchFeature:
-        self.set_frozen_modules_to_eval_mode()
-
-        eagle_embeds, eagle_mask = self.forward_eagle(vl_input)
-
-        # YL (TODO HACK): to resolve DDP issue when tune_visual=True
-        # Ensure all trainable parameters in vision_model are used in the forward pass for DDP compatibility
-        if self.training and self.tune_visual:
-            dummy_term = torch.tensor(
-                0.0, device=eagle_embeds.device, dtype=eagle_embeds.dtype, requires_grad=True
-            )
-            for param in self.eagle_model.vision_model.parameters():
-                if param.requires_grad:
-                    dummy_term = dummy_term + 0.0 * param.sum()
-            eagle_embeds = eagle_embeds + dummy_term
-
-        return BatchFeature(
-            data={"backbone_features": eagle_embeds, "backbone_attention_mask": eagle_mask}
-        )  # [B, T2, hidden_size]
-
-
-BACKBONE_FEATURE_KEY = "backbone_features"
-ACTION_KEY = "action_pred"
-LOSS_KEY = "loss"
-ERROR_MSG = "Error: unexpected input/output"
-N_COLOR_CHANNELS = 3
-
-
-# config
-@strict
-class GR00TN15Config(PretrainedConfig):
-    model_type = "gr00t_n1_5"
-
-    backbone_cfg: dict[str, Any] | None = None
-    action_head_cfg: dict[str, Any] | None = None
-    action_horizon: int = 0
-    action_dim: int = 0
-    compute_dtype: str = "float32"
-
-    def __post_init__(self, **kwargs):
-        self.backbone_cfg = {} if self.backbone_cfg is None else self.backbone_cfg
-        self.action_head_cfg = {} if self.action_head_cfg is None else self.action_head_cfg
-        super().__post_init__(**kwargs)
-
-
-# real model
-class GR00TN15(PreTrainedModel):
-    supports_gradient_checkpointing = True
-    config_class = GR00TN15Config
-    """
-    we expect the backbone output to have a key 'backbone_features' with shape (batch_size, n, hidden_size)
-    here n is variable and can be e.g. time, 1 or user specified
-    we expect the action head output to have a key 'action_pred' with shape (batch_size, time, action_dim) during inference time
-    we expect these to have type BatchFeature, and they can of course have many other user specified keys too
-    """
-
-    def __init__(
-        self,
-        config: GR00TN15Config,
-        local_model_path: str,
-    ):
-        assert isinstance(config.backbone_cfg, dict)
-        assert isinstance(config.action_head_cfg, dict)
-
-        super().__init__(config)
-        self.local_model_path = local_model_path
-
-        self.backbone = EagleBackbone(**config.backbone_cfg)
-        action_head_cfg = FlowmatchingActionHeadConfig(**config.action_head_cfg)
-        self.action_head = FlowmatchingActionHead(action_head_cfg)
-
-        self.action_horizon = config.action_horizon
-        self.action_dim = config.action_dim
-        self.compute_dtype = config.compute_dtype
-        self.post_init()
-
-    def validate_inputs(self, inputs):
-        # NOTE -- this should be handled internally by the model
-        # however, doing that will likely be breaking changes -- so we'll need to do it after the deadline
-
-        detected_error = False
-        error_msg = ERROR_MSG
-        if ACTION in inputs:
-            action = inputs[ACTION]
-            # In inference, action may be omitted or None; validate only when it's a tensor.
-            if action is None:
-                pass  # allow None during inference
-            elif isinstance(action, torch.Tensor):
-                shape_ok = (
-                    len(action.shape) == 3
-                    and action.shape[1] == self.action_horizon
-                    and action.shape[2] == self.action_dim
-                )
-                if not shape_ok:
-                    error_msg += f"\n{action.shape=}"
-                    detected_error = True
-            else:
-                # Unexpected non-tensor type provided for action
-                error_msg += f"\nInvalid type for action: {type(action)}"
-                detected_error = True
-
-        if "video" in inputs:
-            video = inputs["video"]
-            type_ok = isinstance(video, np.ndarray)
-            dtype_ok = video.dtype == np.uint8
-            shape_ok = len(video.shape) == 6 and video.shape[3] == N_COLOR_CHANNELS
-            if not type_ok:
-                error_msg += f"\n{type(video)=}"
-                detected_error = True
-            if not dtype_ok:
-                error_msg += f"\n{video.dtype=}"
-                detected_error = True
-            if not shape_ok:
-                error_msg += f"\n{video.shape=}"
-                detected_error = True
-
-        if detected_error:
-            raise ValueError(error_msg)
-
-    def validate_data(self, action_head_outputs, backbone_outputs, is_training):
-        fail_backbone = (
-            not isinstance(backbone_outputs, BatchFeature) or BACKBONE_FEATURE_KEY not in backbone_outputs
-        )
-
-        if fail_backbone:
-            error_msg = ERROR_MSG
-            error_msg += f"\n{isinstance(backbone_outputs, BatchFeature)=}"
-            error_msg += f"\n{BACKBONE_FEATURE_KEY in backbone_outputs=}"
-            error_msg += f"\n{backbone_outputs[BACKBONE_FEATURE_KEY].shape=}"
-            raise ValueError(error_msg)
-
-        fail_action_head = (not isinstance(action_head_outputs, BatchFeature)) or not (
-            (
-                LOSS_KEY in action_head_outputs and is_training
-            )  # there might not be an action prediction during training
-            or (
-                ACTION_KEY in action_head_outputs
-                and action_head_outputs[ACTION_KEY].shape[1] == self.action_horizon
-                and action_head_outputs[ACTION_KEY].shape[2] == self.action_dim
-            )
-        )
-
-        if fail_action_head:
-            error_msg = ERROR_MSG
-            error_msg += f"\n{isinstance(action_head_outputs, BatchFeature)=}"
-            error_msg += f"\n{LOSS_KEY in action_head_outputs=}"
-            error_msg += f"\n{action_head_outputs[ACTION_KEY].shape=}"
-            error_msg += f"\n{self.action_horizon=}"
-            error_msg += f"\n{self.action_dim=}"
-            raise ValueError(error_msg)
-
-    def forward(
-        self,
-        inputs: dict,
-    ) -> BatchFeature:
-        backbone_inputs, action_inputs = self.prepare_input(inputs)
-        backbone_outputs = self.backbone(backbone_inputs)
-        action_head_outputs = self.action_head(backbone_outputs, action_inputs)
-        self.validate_data(action_head_outputs, backbone_outputs, is_training=True)
-        return action_head_outputs
-
-    def get_action(
-        self,
-        inputs: dict,
-    ) -> BatchFeature:
-        backbone_inputs, action_inputs = self.prepare_input(inputs)
-        # Because the behavior of backbones remains the same for training and inference, we can use `forward` for backbones.
-        backbone_outputs = self.backbone(backbone_inputs)
-        action_head_outputs = self.action_head.get_action(backbone_outputs, action_inputs)
-        self.validate_data(action_head_outputs, backbone_outputs, is_training=False)
-        return action_head_outputs
-
-    def prepare_input(self, inputs) -> tuple[BatchFeature, BatchFeature]:
-        self.validate_inputs(inputs)
-        backbone_inputs = self.backbone.prepare_input(inputs)
-        action_inputs = self.action_head.prepare_input(inputs)
-
-        def to_device_with_maybe_dtype(x):
-            # Cast floating tensors to a memory-efficient compute dtype when requested.
-            # Rationale: Upcasting backbone activations to fp32 significantly increases VRAM.
-            # When compute_dtype is bfloat16, prefer bf16 for activations to match AMP behavior.
-            if not isinstance(x, torch.Tensor):
-                return x
-            if torch.is_floating_point(x):
-                if getattr(self, "compute_dtype", None) == "bfloat16":
-                    return x.to(self.device, dtype=torch.bfloat16)
-                # Fallback: preserve previous behavior if not using bf16 compute
-                return x.to(self.device, dtype=self.action_head.dtype)
-            # Non-floating tensors: move device only
-            return x.to(self.device)
-
-        backbone_inputs = tree.map_structure(to_device_with_maybe_dtype, backbone_inputs)
-        action_inputs = tree.map_structure(to_device_with_maybe_dtype, action_inputs)
-        return backbone_inputs, action_inputs
-
-    @classmethod
-    def from_pretrained(cls, pretrained_model_name_or_path: str, **kwargs):
-        tune_visual = kwargs.pop("tune_visual", True)
-        tune_llm = kwargs.pop("tune_llm", False)
-        tune_projector = kwargs.pop("tune_projector", True)
-        tune_diffusion_model = kwargs.pop("tune_diffusion_model", True)
-
-        print(f"Loading pretrained dual brain from {pretrained_model_name_or_path}")
-        print(f"Tune backbone vision tower: {tune_visual}")
-        print(f"Tune backbone LLM: {tune_llm}")
-        print(f"Tune action head projector: {tune_projector}")
-        print(f"Tune action head DiT: {tune_diffusion_model}")
-
-        # get the current model path being downloaded
-        try:
-            # NOTE(YL) This downloads the model to the local cache and returns the local path to the model
-            # saved in ~/.cache/huggingface/hub/
-            local_model_path = snapshot_download(pretrained_model_name_or_path, repo_type="model")
-            # HFValidationError, RepositoryNotFoundError
-        except (HFValidationError, RepositoryNotFoundError):
-            print(
-                f"Model not found or avail in the huggingface hub. Loading from local path: {pretrained_model_name_or_path}"
-            )
-            local_model_path = pretrained_model_name_or_path
-
-        pretrained_model = super().from_pretrained(
-            local_model_path, local_model_path=local_model_path, **kwargs
-        )
-
-        pretrained_model.backbone.set_trainable_parameters(tune_visual=tune_visual, tune_llm=tune_llm)
-        pretrained_model.action_head.set_trainable_parameters(
-            tune_projector=tune_projector, tune_diffusion_model=tune_diffusion_model
-        )
-        return pretrained_model

src/lerobot/policies/groot/groot_n1_7.py

@@ -0,0 +1,952 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import importlib
+import logging
+from contextlib import suppress
+from copy import deepcopy
+from typing import TYPE_CHECKING, Any
+
+import torch
+import torch.nn.functional as F  # noqa: N812
+from huggingface_hub import snapshot_download
+from huggingface_hub.errors import HFValidationError, RepositoryNotFoundError
+from torch import nn
+from torch.distributions import Beta
+
+from lerobot.utils.import_utils import _transformers_available, require_package
+
+from .action_head.cross_attention_dit import AlternateVLDiT, DiT, SelfAttentionTransformer
+from .configuration_groot import N1_7_DEFAULT_IMAGE_CROP_SIZE, N1_7_DEFAULT_IMAGE_TARGET_SIZE
+
+if TYPE_CHECKING or _transformers_available:
+    from transformers import (
+        AutoConfig,
+        AutoModel,
+        PretrainedConfig,
+        PreTrainedModel,
+        Qwen3VLConfig,
+        Qwen3VLForConditionalGeneration,
+    )
+    from transformers.feature_extraction_utils import BatchFeature
+else:
+    AutoConfig = None
+    AutoModel = None
+    PretrainedConfig = object
+    PreTrainedModel = object
+    BatchFeature = None
+    Qwen3VLConfig = None
+    Qwen3VLForConditionalGeneration = None
+
+try:
+    import tree
+except ImportError:
+    tree = None
+
+logger = logging.getLogger(__name__)
+
+
+def _tie_unused_qwen_lm_head(model: nn.Module) -> None:
+    """Restore the TF4 weight tie so the unused LM head stays frozen and is omitted on save."""
+    lm_head = getattr(model, "lm_head", None)
+    get_input_embeddings = getattr(model, "get_input_embeddings", None)
+    if lm_head is None or not callable(get_input_embeddings):
+        return
+    input_embeddings = get_input_embeddings()
+    embedding_weight = getattr(input_embeddings, "weight", None)
+    if embedding_weight is None:
+        return
+    lm_head.weight = embedding_weight
+
+
+GR00T_N1_7_DEFAULTS: dict[str, Any] = {
+    "model_dtype": "bfloat16",
+    "dtype": "bfloat16",
+    "model_name": "nvidia/Cosmos-Reason2-2B",
+    "backbone_model_type": "qwen",
+    "model_revision": None,
+    "tune_top_llm_layers": 0,
+    "backbone_embedding_dim": 2048,
+    "tune_llm": False,
+    "tune_visual": False,
+    "select_layer": 16,
+    "reproject_vision": False,
+    "use_flash_attention": False,
+    "load_bf16": False,
+    "backbone_trainable_params_fp32": True,
+    "image_crop_size": N1_7_DEFAULT_IMAGE_CROP_SIZE,
+    "image_target_size": N1_7_DEFAULT_IMAGE_TARGET_SIZE,
+    "shortest_image_edge": None,
+    "crop_fraction": None,
+    "random_rotation_angle": None,
+    "color_jitter_params": None,
+    "use_albumentations_transforms": True,
+    "extra_augmentation_config": None,
+    "formalize_language": True,
+    "apply_sincos_state_encoding": False,
+    "use_percentiles": True,
+    "use_relative_action": False,
+    "max_state_dim": 132,
+    "max_action_dim": 132,
+    "action_horizon": 40,
+    "hidden_size": 1024,
+    "input_embedding_dim": 1536,
+    "state_history_length": 1,
+    "add_pos_embed": True,
+    "attn_dropout": 0.2,
+    "use_vlln": True,
+    "max_seq_len": 1024,
+    "use_alternate_vl_dit": True,
+    "attend_text_every_n_blocks": 2,
+    "diffusion_model_cfg": {
+        "positional_embeddings": None,
+        "num_layers": 32,
+        "num_attention_heads": 32,
+        "attention_head_dim": 48,
+        "norm_type": "ada_norm",
+        "dropout": 0.2,
+        "final_dropout": True,
+        "output_dim": 1024,
+        "interleave_self_attention": True,
+    },
+    "vl_self_attention_cfg": {
+        "positional_embeddings": None,
+        "num_layers": 4,
+        "num_attention_heads": 32,
+        "attention_head_dim": 64,
+        "dropout": 0.2,
+        "final_dropout": True,
+    },
+    "num_inference_timesteps": 4,
+    "noise_beta_alpha": 1.5,
+    "noise_beta_beta": 1.0,
+    "noise_s": 0.999,
+    "num_timestep_buckets": 1000,
+    "tune_projector": True,
+    "tune_diffusion_model": True,
+    "tune_vlln": True,
+    "state_dropout_prob": 0.2,
+    "exclude_state": False,
+    "use_mean_std": False,
+    "max_num_embodiments": 32,
+    "rtc_ramp_rate": 6.0,
+}
+
+
+class GR00TN17Config(PretrainedConfig):
+    """Configuration for NVIDIA GR00T N1.7.
+
+    N1.7 uses the Cosmos-Reason2-2B / Qwen3-VL backbone and a multi-embodiment
+    flow-matching action head. This mirrors the public N1.7 checkpoint config
+    while keeping it local to LeRobot and independent from the external
+    Isaac-GR00T ``gr00t`` Python package.
+    """
+
+    model_type = "Gr00tN1d7"
+
+    _defaults = GR00T_N1_7_DEFAULTS
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        for key, value in GR00T_N1_7_DEFAULTS.items():
+            setattr(self, key, deepcopy(kwargs.pop(key, value)))
+        for key, value in kwargs.items():
+            setattr(self, key, value)
+
+
+class CategorySpecificLinear(nn.Module):
+    """Linear layer with category-specific weights for multi-embodiment support."""
+
+    def __init__(self, num_categories: int, input_dim: int, hidden_dim: int):
+        super().__init__()
+        self.num_categories = num_categories
+        self.W = nn.Parameter(0.02 * torch.randn(num_categories, input_dim, hidden_dim))
+        self.b = nn.Parameter(torch.zeros(num_categories, hidden_dim))
+
+    def forward(self, x: torch.Tensor, cat_ids: torch.Tensor) -> torch.Tensor:
+        selected_w = self.W[cat_ids]
+        selected_b = self.b[cat_ids]
+        return torch.bmm(x, selected_w) + selected_b.unsqueeze(1)
+
+
+class CategorySpecificMLP(nn.Module):
+    """Two-layer MLP with category-specific weights."""
+
+    def __init__(self, num_categories: int, input_dim: int, hidden_dim: int, output_dim: int):
+        super().__init__()
+        self.layer1 = CategorySpecificLinear(num_categories, input_dim, hidden_dim)
+        self.layer2 = CategorySpecificLinear(num_categories, hidden_dim, output_dim)
+
+    def forward(self, x: torch.Tensor, cat_ids: torch.Tensor) -> torch.Tensor:
+        hidden = F.relu(self.layer1(x, cat_ids))
+        return self.layer2(hidden, cat_ids)
+
+
+class SinusoidalPositionalEncoding(nn.Module):
+    """Sinusoidal encoding of shape ``(B, T, D)`` for timestep tensors ``(B, T)``.
+
+    The frequency scalar is intentionally created on CPU and then broadcast with
+    the device-local arange result. That mirrors Isaac-GR00T's N1.7 timestep
+    embedding and avoids tiny dtype/device construction differences in parity
+    tests.
+    """
+
+    def __init__(self, embedding_dim: int):
+        super().__init__()
+        self.embedding_dim = embedding_dim
+
+    def forward(self, timesteps: torch.Tensor) -> torch.Tensor:
+        timesteps = timesteps.float()
+        half_dim = self.embedding_dim // 2
+        exponent = -torch.arange(half_dim, dtype=torch.float, device=timesteps.device) * (
+            torch.log(torch.tensor(10000.0)) / half_dim
+        )
+        freqs = timesteps.unsqueeze(-1) * exponent.exp()
+        return torch.cat([torch.sin(freqs), torch.cos(freqs)], dim=-1)
+
+
+def swish(x: torch.Tensor) -> torch.Tensor:
+    return x * torch.sigmoid(x)
+
+
+class MultiEmbodimentActionEncoder(nn.Module):
+    """Action encoder with category-specific projections and sinusoidal time encoding."""
+
+    def __init__(self, action_dim: int, hidden_size: int, num_embodiments: int):
+        super().__init__()
+        self.W1 = CategorySpecificLinear(num_embodiments, action_dim, hidden_size)
+        self.W2 = CategorySpecificLinear(num_embodiments, 2 * hidden_size, hidden_size)
+        self.W3 = CategorySpecificLinear(num_embodiments, hidden_size, hidden_size)
+        self.pos_encoding = SinusoidalPositionalEncoding(hidden_size)
+
+    def forward(self, actions: torch.Tensor, timesteps: torch.Tensor, cat_ids: torch.Tensor) -> torch.Tensor:
+        batch_size, horizon, _ = actions.shape
+        if timesteps.dim() != 1 or timesteps.shape[0] != batch_size:
+            raise ValueError("Expected `timesteps` to have shape (B,).")
+        timesteps = timesteps.unsqueeze(1).expand(-1, horizon)
+        action_emb = self.W1(actions, cat_ids)
+        time_emb = self.pos_encoding(timesteps).to(dtype=action_emb.dtype)
+        x = swish(self.W2(torch.cat([action_emb, time_emb], dim=-1), cat_ids))
+        return self.W3(x, cat_ids)
+
+
+class Qwen3Backbone(nn.Module):
+    """Cosmos-Reason2/Qwen3-VL backbone used by GR00T N1.7.
+
+    The public checkpoint stores the action head in the GR00T checkpoint but
+    uses a Hugging Face Qwen3-VL-compatible backbone interface. This wrapper
+    keeps the nested HF module layout compatible across transformer versions
+    and exposes the hidden states consumed by the action head.
+    """
+
+    def __init__(
+        self,
+        model_name: str = "nvidia/Cosmos-Reason2-2B",
+        tune_llm: bool = False,
+        tune_visual: bool = False,
+        select_layer: int = -1,
+        reproject_vision: bool = False,
+        use_flash_attention: bool = False,
+        load_bf16: bool = False,
+        tune_top_llm_layers: int = 0,
+        trainable_params_fp32: bool = False,
+        transformers_loading_kwargs: dict[str, Any] | None = None,
+        load_pretrained_weights: bool = True,
+    ):
+        require_package("transformers", extra="groot")
+        if Qwen3VLForConditionalGeneration is None:
+            raise ImportError(
+                "Qwen3VLForConditionalGeneration is required for GR00T N1.7. "
+                "Install a transformers version with Qwen3-VL support."
+            )
+        super().__init__()
+        transformers_loading_kwargs = transformers_loading_kwargs or {"trust_remote_code": True}
+
+        extra_kwargs: dict[str, Any] = {}
+        if use_flash_attention:
+            try:
+                import flash_attn  # noqa: F401
+
+                extra_kwargs["attn_implementation"] = "flash_attention_2"
+            except ImportError:
+                logger.warning("flash_attn is not installed. Falling back to SDPA attention.")
+                extra_kwargs["attn_implementation"] = "sdpa"
+        if load_bf16:
+            extra_kwargs["torch_dtype"] = torch.bfloat16
+
+        if load_pretrained_weights:
+            self.model = Qwen3VLForConditionalGeneration.from_pretrained(
+                model_name,
+                **extra_kwargs,
+                **transformers_loading_kwargs,
+            ).eval()
+        else:
+            self.model = self._from_backbone_config(
+                model_name=model_name,
+                model_kwargs=extra_kwargs,
+                config_kwargs=transformers_loading_kwargs,
+            ).eval()
+
+        _tie_unused_qwen_lm_head(self.model)
+        while len(self.language_model.layers) > select_layer:
+            self.language_model.layers.pop(-1)
+
+        self.select_layer = select_layer
+        self.set_trainable_parameters(tune_llm, tune_visual, tune_top_llm_layers)
+        if load_bf16 and trainable_params_fp32:
+            for parameter in self.parameters():
+                if parameter.requires_grad:
+                    parameter.data = parameter.data.to(torch.float32)
+
+    def set_trainable_parameters(
+        self, tune_llm: bool, tune_visual: bool, tune_top_llm_layers: int = 0
+    ) -> None:
+        self.tune_llm = tune_llm
+        self.tune_visual = tune_visual
+        for parameter in self.parameters():
+            parameter.requires_grad = True
+        if not tune_llm:
+            self.language_model.requires_grad_(False)
+        if not tune_visual:
+            self.visual.requires_grad_(False)
+        if tune_top_llm_layers > 0:
+            for layer in self.language_model.layers[-tune_top_llm_layers:]:
+                for parameter in layer.parameters():
+                    parameter.requires_grad = True
+
+    def set_frozen_modules_to_eval_mode(self) -> None:
+        if self.training:
+            if self.language_model and not self.tune_llm:
+                self.language_model.eval()
+            if self.visual and not self.tune_visual:
+                self.visual.eval()
+
+    @property
+    def language_model(self) -> nn.Module:
+        return getattr(self.model, "model", self.model).language_model
+
+    @property
+    def visual(self) -> nn.Module:
+        return getattr(self.model, "model", self.model).visual
+
+    def _from_backbone_config(
+        self,
+        *,
+        model_name: str,
+        model_kwargs: dict[str, Any],
+        config_kwargs: dict[str, Any],
+    ) -> nn.Module:
+        if _is_cosmos_reason2_backbone(model_name):
+            backbone_config = _cosmos_reason2_qwen3_vl_config()
+        else:
+            backbone_config = AutoConfig.from_pretrained(model_name, **config_kwargs)
+        return Qwen3VLForConditionalGeneration._from_config(backbone_config, **model_kwargs)
+
+    def prepare_input(self, batch: dict[str, Any]) -> BatchFeature:
+        return BatchFeature(data=batch)
+
+    def _ensure_mm_token_type_ids(self, model_input: dict[str, torch.Tensor]) -> None:
+        if "mm_token_type_ids" in model_input:
+            return
+        if "image_grid_thw" not in model_input and "video_grid_thw" not in model_input:
+            return
+
+        input_ids = model_input.get("input_ids")
+        if input_ids is None:
+            return
+
+        mm_token_type_ids = torch.zeros(input_ids.shape, dtype=torch.int32, device=input_ids.device)
+        image_token_id = getattr(self.model.config, "image_token_id", None)
+        video_token_id = getattr(self.model.config, "video_token_id", None)
+        if image_token_id is not None:
+            mm_token_type_ids[input_ids == image_token_id] = 1
+        if video_token_id is not None:
+            mm_token_type_ids[input_ids == video_token_id] = 2
+
+        model_input["mm_token_type_ids"] = mm_token_type_ids
+
+    def _ensure_legacy_qwen3_position_ids(self, model_input: dict[str, torch.Tensor]) -> None:
+        """Restore the Qwen3-VL text position ids used by older Transformers releases.
+
+        Transformers 5.x computes 3-row multimodal RoPE ids for Qwen3-VL and then
+        drops text position ids before calling text-layer flash attention. GR00T
+        N1.7 was aligned against the older Transformers path, where a fourth text
+        position row is forwarded alongside the temporal/height/width rows. Adding
+        the row here preserves the newer multimodal position computation while
+        keeping flash attention on the legacy code path.
+        """
+
+        if "position_ids" in model_input:
+            return
+
+        qwen3_model = getattr(self.model, "model", self.model)
+        compute_3d_position_ids = getattr(qwen3_model, "compute_3d_position_ids", None)
+        if compute_3d_position_ids is None:
+            return
+
+        position_ids = compute_3d_position_ids(
+            input_ids=model_input.get("input_ids"),
+            image_grid_thw=model_input.get("image_grid_thw"),
+            video_grid_thw=model_input.get("video_grid_thw"),
+            inputs_embeds=None,
+            attention_mask=model_input.get("attention_mask"),
+            past_key_values=None,
+            mm_token_type_ids=model_input.get("mm_token_type_ids"),
+        )
+        if position_ids.ndim == 3 and position_ids.shape[0] == 3:
+            position_ids = torch.cat([position_ids[:1], position_ids], dim=0)
+
+        model_input["position_ids"] = position_ids
+
+    def _last_decoder_layer_output(self, model_input: dict[str, torch.Tensor]) -> torch.Tensor:
+        """Return the pre-final-norm decoder output consumed by the N1.7 action head.
+
+        Older Transformers releases exposed this tensor as ``hidden_states[-1]``.
+        Newer releases expose the post-final-norm tensor there instead. Capturing
+        the last decoder layer output directly keeps the N1.7 action head input
+        stable across Transformers versions.
+        """
+
+        captured: dict[str, torch.Tensor] = {}
+
+        def capture_output(_module: nn.Module, _inputs: tuple[Any, ...], output: Any) -> None:
+            if isinstance(output, torch.Tensor):
+                captured["features"] = output
+            elif isinstance(output, (tuple, list)) and output:
+                captured["features"] = output[0]
+            elif hasattr(output, "last_hidden_state"):
+                captured["features"] = output.last_hidden_state
+
+        hook = self.language_model.layers[-1].register_forward_hook(capture_output)
+        try:
+            outputs = self.model(**model_input, output_hidden_states=True)
+        finally:
+            hook.remove()
+
+        return captured.get("features", outputs.hidden_states[-1])
+
+    def forward(self, vl_input: BatchFeature) -> BatchFeature:
+        self.set_frozen_modules_to_eval_mode()
+        keys_to_use = ["input_ids", "attention_mask", "pixel_values", "image_grid_thw"]
+        optional_keys = ["mm_token_type_ids", "pixel_values_videos", "video_grid_thw"]
+        model_input = {key: vl_input[key] for key in keys_to_use}
+        model_input.update({key: vl_input[key] for key in optional_keys if key in vl_input})
+        self._ensure_mm_token_type_ids(model_input)
+        self._ensure_legacy_qwen3_position_ids(model_input)
+        features = self._last_decoder_layer_output(model_input)
+        image_mask = model_input["input_ids"] == self.model.config.image_token_id
+        attention_mask = model_input["attention_mask"] == 1
+        return BatchFeature(
+            data={
+                "backbone_features": features,
+                "backbone_attention_mask": attention_mask,
+                "image_mask": image_mask,
+            }
+        )
+
+
+class GR00TN17ActionHead(nn.Module):
+    supports_gradient_checkpointing = True
+
+    def __init__(self, config: GR00TN17Config):
+        require_package("diffusers", extra="groot")
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.input_embedding_dim = config.input_embedding_dim
+
+        if config.use_alternate_vl_dit:
+            self.model = AlternateVLDiT(
+                **config.diffusion_model_cfg,
+                cross_attention_dim=config.backbone_embedding_dim,
+                attend_text_every_n_blocks=config.attend_text_every_n_blocks,
+            )
+        else:
+            self.model = DiT(
+                **config.diffusion_model_cfg,
+                cross_attention_dim=config.backbone_embedding_dim,
+            )
+
+        self.action_dim = config.max_action_dim
+        self.action_horizon = config.action_horizon
+        self.num_inference_timesteps = config.num_inference_timesteps
+        self.state_encoder = CategorySpecificMLP(
+            num_categories=config.max_num_embodiments,
+            input_dim=config.max_state_dim * config.state_history_length,
+            hidden_dim=self.hidden_size,
+            output_dim=self.input_embedding_dim,
+        )
+        self.action_encoder = MultiEmbodimentActionEncoder(
+            action_dim=self.action_dim,
+            hidden_size=self.input_embedding_dim,
+            num_embodiments=config.max_num_embodiments,
+        )
+        self.action_decoder = CategorySpecificMLP(
+            num_categories=config.max_num_embodiments,
+            input_dim=self.hidden_size,
+            hidden_dim=self.hidden_size,
+            output_dim=self.action_dim,
+        )
+        self.vlln = nn.LayerNorm(config.backbone_embedding_dim) if config.use_vlln else nn.Identity()
+        vl_self_attention_cfg = getattr(config, "vl_self_attention_cfg", None)
+        if vl_self_attention_cfg and vl_self_attention_cfg.get("num_layers", 0) > 0:
+            self.vl_self_attention = SelfAttentionTransformer(**vl_self_attention_cfg)
+        else:
+            self.vl_self_attention = nn.Identity()
+        if config.add_pos_embed:
+            self.position_embedding = nn.Embedding(config.max_seq_len, self.input_embedding_dim)
+            nn.init.normal_(self.position_embedding.weight, mean=0.0, std=0.02)
+        self.state_dropout_prob = config.state_dropout_prob
+        self._noise_beta_alpha = config.noise_beta_alpha
+        self._noise_beta_beta = config.noise_beta_beta
+        self._beta_dist = None
+        self.num_timestep_buckets = config.num_timestep_buckets
+        self.set_trainable_parameters(config.tune_projector, config.tune_diffusion_model, config.tune_vlln)
+
+    def set_trainable_parameters(
+        self, tune_projector: bool, tune_diffusion_model: bool, tune_vlln: bool
+    ) -> None:
+        self.tune_projector = tune_projector
+        self.tune_diffusion_model = tune_diffusion_model
+        self.tune_vlln = tune_vlln
+        for parameter in self.parameters():
+            parameter.requires_grad = True
+        if not tune_projector:
+            self.state_encoder.requires_grad_(False)
+            self.action_encoder.requires_grad_(False)
+            self.action_decoder.requires_grad_(False)
+            if self.config.add_pos_embed:
+                self.position_embedding.requires_grad_(False)
+        if not tune_diffusion_model:
+            self.model.requires_grad_(False)
+        if not tune_vlln:
+            self.vlln.requires_grad_(False)
+            self.vl_self_attention.requires_grad_(False)
+
+    def set_frozen_modules_to_eval_mode(self) -> None:
+        if self.training:
+            if not self.tune_projector:
+                self.state_encoder.eval()
+                self.action_encoder.eval()
+                self.action_decoder.eval()
+                if self.config.add_pos_embed:
+                    self.position_embedding.eval()
+            if not self.tune_diffusion_model:
+                self.model.eval()
+            if not self.tune_vlln:
+                self.vlln.eval()
+                self.vl_self_attention.eval()
+
+    def sample_time(self, batch_size: int, device: torch.device, dtype: torch.dtype) -> torch.Tensor:
+        if self._beta_dist is None:
+            beta_alpha = torch.tensor(self._noise_beta_alpha, device="cpu", dtype=torch.float32)
+            beta_beta = torch.tensor(self._noise_beta_beta, device="cpu", dtype=torch.float32)
+            self._beta_dist = Beta(beta_alpha, beta_beta, validate_args=False)
+        sample = self._beta_dist.sample([batch_size]).to(device, dtype=dtype)
+        return (1 - sample) * self.config.noise_s
+
+    def process_backbone_output(self, backbone_output: BatchFeature) -> BatchFeature:
+        backbone_features = self.vlln(backbone_output["backbone_features"])
+        backbone_output["backbone_features"] = self.vl_self_attention(backbone_features)
+        return backbone_output
+
+    def forward(self, backbone_output: BatchFeature, action_input: BatchFeature) -> BatchFeature:
+        self.set_frozen_modules_to_eval_mode()
+        backbone_output = self.process_backbone_output(backbone_output)
+        vl_embeds = backbone_output.backbone_features
+        device = vl_embeds.device
+        embodiment_id = action_input.embodiment_id
+
+        if action_input.state.shape[1] != self.config.state_history_length:
+            raise ValueError("state history length does not match GR00T N1.7 config.")
+        state = action_input.state.view(action_input.state.shape[0], 1, -1)
+        state_features = self.state_encoder(state, embodiment_id)
+
+        if self.training and self.state_dropout_prob > 0:
+            do_dropout = (
+                torch.rand(state_features.shape[0], device=state_features.device) < self.state_dropout_prob
+            )
+            state_features = state_features * (1 - do_dropout[:, None, None].to(dtype=state_features.dtype))
+
+        actions = action_input.action
+        noise = torch.randn(actions.shape, device=actions.device, dtype=actions.dtype)
+        t = self.sample_time(actions.shape[0], device=actions.device, dtype=actions.dtype)
+        t = t[:, None, None]
+        noisy_trajectory = (1 - t) * noise + t * actions
+        velocity = actions - noise
+        t_discretized = (t[:, 0, 0] * self.num_timestep_buckets).long()
+        action_features = self.action_encoder(noisy_trajectory, t_discretized, embodiment_id)
+
+        if self.config.add_pos_embed:
+            pos_ids = torch.arange(action_features.shape[1], dtype=torch.long, device=device)
+            action_features = action_features + self.position_embedding(pos_ids).unsqueeze(0)
+
+        sa_embs = torch.cat((state_features, action_features), dim=1)
+        if self.config.use_alternate_vl_dit:
+            model_output, _ = self.model(
+                hidden_states=sa_embs,
+                encoder_hidden_states=vl_embeds,
+                encoder_attention_mask=backbone_output.backbone_attention_mask,
+                timestep=t_discretized,
+                return_all_hidden_states=True,
+                image_mask=backbone_output.image_mask,
+                backbone_attention_mask=backbone_output.backbone_attention_mask,
+            )
+        else:
+            model_output, _ = self.model(
+                hidden_states=sa_embs,
+                encoder_hidden_states=vl_embeds,
+                encoder_attention_mask=backbone_output.backbone_attention_mask,
+                timestep=t_discretized,
+                return_all_hidden_states=True,
+            )
+
+        pred = self.action_decoder(model_output, embodiment_id)
+        pred_actions = pred[:, -actions.shape[1] :]
+        action_mask = action_input.action_mask
+        action_loss = F.mse_loss(pred_actions, velocity, reduction="none") * action_mask
+        loss = action_loss.sum() / (action_mask.sum() + 1e-6)
+        return BatchFeature(
+            data={
+                "loss": loss,
+                "action_loss": action_loss,
+                "action_mask": action_mask,
+                "backbone_features": vl_embeds,
+                "state_features": state_features,
+            }
+        )
+
+    def _encode_features(self, backbone_output: BatchFeature, action_input: BatchFeature) -> BatchFeature:
+        backbone_output = self.process_backbone_output(backbone_output)
+        state = action_input.state
+        if state.shape[1] != self.config.state_history_length:
+            raise ValueError("state history length does not match GR00T N1.7 config.")
+        state = state.view(state.shape[0], 1, -1)
+        state_features = self.state_encoder(state, action_input.embodiment_id)
+        return BatchFeature(
+            data={"backbone_features": backbone_output.backbone_features, "state_features": state_features}
+        )
+
+    @torch.no_grad()
+    def get_action_with_features(
+        self,
+        backbone_features: torch.Tensor,
+        state_features: torch.Tensor,
+        embodiment_id: torch.Tensor,
+        backbone_output: BatchFeature,
+        action_input: BatchFeature,
+        options: dict[str, Any] | None = None,
+    ) -> BatchFeature:
+        vl_embeds = backbone_features
+        batch_size = vl_embeds.shape[0]
+        device = vl_embeds.device
+        actions = torch.randn(
+            size=(batch_size, self.config.action_horizon, self.action_dim),
+            dtype=vl_embeds.dtype,
+            device=device,
+        )
+        dt = 1.0 / self.num_inference_timesteps
+        vel_strength = torch.ones_like(actions)
+
+        if "action" in action_input:
+            if options is None:
+                raise ValueError("RTC options are required when action is provided to get_action.")
+            action_horizon_before_padding = options["action_horizon"]
+            actions[:, : options["rtc_overlap_steps"], :] = action_input["action"][
+                :,
+                action_horizon_before_padding - options["rtc_overlap_steps"] : action_horizon_before_padding,
+                :,
+            ]
+            vel_strength[:, : options["rtc_frozen_steps"], :] = 0.0
+            intermediate_steps = options["rtc_overlap_steps"] - options["rtc_frozen_steps"]
+            t = torch.linspace(0.0, 1.0, intermediate_steps + 2, device=device)
+            ramp = 1 - torch.exp(-options["rtc_ramp_rate"] * t)
+            ramp = ramp / ramp[-1].clamp_min(1e-8)
+            vel_strength[:, options["rtc_frozen_steps"] : options["rtc_overlap_steps"], :] = ramp[1:-1][
+                None, :, None
+            ].to(device)
+
+        for t_step in range(self.num_inference_timesteps):
+            t_cont = t_step / float(self.num_inference_timesteps)
+            t_discretized = int(t_cont * self.num_timestep_buckets)
+            timesteps_tensor = torch.full(size=(batch_size,), fill_value=t_discretized, device=device)
+            action_features = self.action_encoder(actions, timesteps_tensor, embodiment_id)
+            if self.config.add_pos_embed:
+                pos_ids = torch.arange(action_features.shape[1], dtype=torch.long, device=device)
+                action_features = action_features + self.position_embedding(pos_ids).unsqueeze(0)
+            sa_embs = torch.cat((state_features, action_features), dim=1)
+
+            if self.config.use_alternate_vl_dit:
+                model_output = self.model(
+                    hidden_states=sa_embs,
+                    encoder_hidden_states=vl_embeds,
+                    timestep=timesteps_tensor,
+                    image_mask=backbone_output.image_mask,
+                    backbone_attention_mask=backbone_output.backbone_attention_mask,
+                )
+            else:
+                model_output = self.model(
+                    hidden_states=sa_embs,
+                    encoder_hidden_states=vl_embeds,
+                    timestep=timesteps_tensor,
+                )
+            pred = self.action_decoder(model_output, embodiment_id)
+            actions = actions + dt * pred[:, -self.action_horizon :] * vel_strength
+
+        return BatchFeature(
+            data={
+                "action_pred": actions,
+                "backbone_features": vl_embeds,
+                "state_features": state_features,
+            }
+        )
+
+    @torch.no_grad()
+    def get_action(
+        self,
+        backbone_output: BatchFeature,
+        action_input: BatchFeature,
+        options: dict[str, Any] | None = None,
+    ) -> BatchFeature:
+        features = self._encode_features(backbone_output, action_input)
+        return self.get_action_with_features(
+            backbone_features=features.backbone_features,
+            state_features=features.state_features,
+            embodiment_id=action_input.embodiment_id,
+            backbone_output=backbone_output,
+            action_input=action_input,
+            options=options,
+        )
+
+    @property
+    def device(self) -> torch.device:
+        return next(iter(self.parameters())).device
+
+    @property
+    def dtype(self) -> torch.dtype:
+        return next(iter(self.parameters())).dtype
+
+    def prepare_input(self, batch: dict[str, Any]) -> BatchFeature:
+        return BatchFeature(data=batch)
+
+
+def _is_cosmos_reason2_backbone(model_name: str) -> bool:
+    return str(model_name).rstrip("/") == "nvidia/Cosmos-Reason2-2B"
+
+
+def _cosmos_reason2_qwen3_vl_config() -> PretrainedConfig:
+    return Qwen3VLConfig(
+        image_token_id=151655,
+        video_token_id=151656,
+        vision_start_token_id=151652,
+        vision_end_token_id=151653,
+        tie_word_embeddings=True,
+        text_config={
+            "attention_bias": False,
+            "attention_dropout": 0.0,
+            "bos_token_id": 151643,
+            "dtype": "bfloat16",
+            "eos_token_id": 151645,
+            "head_dim": 128,
+            "hidden_act": "silu",
+            "hidden_size": 2048,
+            "initializer_range": 0.02,
+            "intermediate_size": 6144,
+            "max_position_embeddings": 262144,
+            "model_type": "qwen3_vl_text",
+            "num_attention_heads": 16,
+            "num_hidden_layers": 28,
+            "num_key_value_heads": 8,
+            "rms_norm_eps": 1e-6,
+            "rope_scaling": {
+                "mrope_interleaved": True,
+                "mrope_section": [24, 20, 20],
+                "rope_type": "default",
+            },
+            "rope_theta": 5000000,
+            "tie_word_embeddings": True,
+            "use_cache": True,
+            "vocab_size": 151936,
+        },
+        vision_config={
+            "deepstack_visual_indexes": [5, 11, 17],
+            "depth": 24,
+            "hidden_act": "gelu_pytorch_tanh",
+            "hidden_size": 1024,
+            "in_channels": 3,
+            "initializer_range": 0.02,
+            "intermediate_size": 4096,
+            "model_type": "qwen3_vl",
+            "num_heads": 16,
+            "num_position_embeddings": 2304,
+            "out_hidden_size": 2048,
+            "patch_size": 16,
+            "spatial_merge_size": 2,
+            "temporal_patch_size": 2,
+        },
+    )
+
+
+def get_backbone_cls(config: GR00TN17Config):
+    if "nvidia/Cosmos-Reason2" in config.model_name or "Qwen/Qwen3-VL" in config.model_name:
+        return Qwen3Backbone
+    if config.backbone_model_type == "qwen":
+        logger.warning(
+            "Unrecognized GR00T N1.7 backbone model name '%s'; assuming a Qwen3-VL-compatible "
+            "backbone because backbone_model_type='qwen'.",
+            config.model_name,
+        )
+        return Qwen3Backbone
+    raise ValueError(f"Unsupported GR00T N1.7 backbone model: {config.model_name}")
+
+
+class GR00TN17(PreTrainedModel):
+    """GR00T N1.7 model with a Cosmos-Reason2/Qwen3-VL backbone."""
+
+    config_class = GR00TN17Config
+    supports_gradient_checkpointing = True
+
+    def __init__(
+        self,
+        config: GR00TN17Config,
+        transformers_loading_kwargs: dict[str, Any] | None = None,
+        load_backbone_weights: bool = True,
+    ):
+        _register_with_transformers()
+        super().__init__(config)
+        transformers_loading_kwargs = transformers_loading_kwargs or {"trust_remote_code": True}
+        self.config = config
+        backbone_cls = get_backbone_cls(config)
+        self.backbone = backbone_cls(
+            model_name=config.model_name,
+            tune_llm=config.tune_llm,
+            tune_visual=config.tune_visual,
+            select_layer=config.select_layer,
+            reproject_vision=config.reproject_vision,
+            use_flash_attention=config.use_flash_attention,
+            load_bf16=config.load_bf16,
+            tune_top_llm_layers=config.tune_top_llm_layers,
+            trainable_params_fp32=config.backbone_trainable_params_fp32,
+            transformers_loading_kwargs=transformers_loading_kwargs,
+            load_pretrained_weights=load_backbone_weights,
+        )
+        self.action_head = GR00TN17ActionHead(config)
+        self.post_init()
+
+    def prepare_input(self, inputs: dict[str, Any]) -> tuple[BatchFeature, BatchFeature]:
+        global tree
+        if tree is None:
+            require_package("dm-tree", extra="groot", import_name="tree")
+            tree = importlib.import_module("tree")
+        backbone_inputs = self.backbone.prepare_input(inputs)
+        action_inputs = self.action_head.prepare_input(inputs)
+
+        def to_device_with_dtype(x):
+            if not isinstance(x, torch.Tensor):
+                return x
+            if torch.is_floating_point(x):
+                return x.to(self.device, dtype=self.dtype)
+            return x.to(self.device)
+
+        return (
+            tree.map_structure(to_device_with_dtype, backbone_inputs),
+            tree.map_structure(to_device_with_dtype, action_inputs),
+        )
+
+    def forward(self, inputs: dict[str, Any]) -> BatchFeature:
+        backbone_inputs, action_inputs = self.prepare_input(inputs)
+        backbone_outputs = self.backbone(backbone_inputs)
+        return self.action_head(backbone_outputs, action_inputs)
+
+    def get_action(self, inputs: dict[str, Any], options: dict[str, Any] | None = None) -> BatchFeature:
+        backbone_inputs, action_inputs = self.prepare_input(inputs)
+        backbone_outputs = self.backbone(backbone_inputs)
+        return self.action_head.get_action(backbone_outputs, action_inputs, options)
+
+    @property
+    def device(self) -> torch.device:
+        return next(iter(self.parameters())).device
+
+    @property
+    def dtype(self) -> torch.dtype:
+        return next(iter(self.parameters())).dtype
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name_or_path: str, **kwargs):
+        tune_visual = kwargs.pop("tune_visual", True)
+        tune_llm = kwargs.pop("tune_llm", False)
+        tune_projector = kwargs.pop("tune_projector", True)
+        tune_diffusion_model = kwargs.pop("tune_diffusion_model", True)
+        tune_vlln = kwargs.pop("tune_vlln", True)
+        transformers_loading_kwargs = kwargs.pop("transformers_loading_kwargs", None) or {
+            "trust_remote_code": True
+        }
+        load_backbone_weights = kwargs.pop("load_backbone_weights", False)
+        for key in ("cache_dir", "local_files_only", "token"):
+            if key in kwargs:
+                transformers_loading_kwargs.setdefault(key, kwargs[key])
+
+        try:
+            local_model_path = snapshot_download(
+                pretrained_model_name_or_path,
+                repo_type="model",
+                revision=kwargs.get("revision"),
+                cache_dir=kwargs.get("cache_dir"),
+                local_files_only=kwargs.get("local_files_only", False),
+                token=kwargs.get("token"),
+            )
+        except (HFValidationError, RepositoryNotFoundError):
+            local_model_path = pretrained_model_name_or_path
+
+        pretrained_model = super().from_pretrained(
+            local_model_path,
+            transformers_loading_kwargs=transformers_loading_kwargs,
+            load_backbone_weights=load_backbone_weights,
+            **kwargs,
+        )
+        pretrained_model.backbone.set_trainable_parameters(
+            tune_visual=tune_visual,
+            tune_llm=tune_llm,
+            tune_top_llm_layers=pretrained_model.config.tune_top_llm_layers,
+        )
+        pretrained_model.action_head.set_trainable_parameters(
+            tune_projector=tune_projector,
+            tune_diffusion_model=tune_diffusion_model,
+            tune_vlln=tune_vlln,
+        )
+        return pretrained_model
+
+
+def _register_with_transformers() -> None:
+    """Register GR00T N1.7 with transformers' Auto* factories.
+
+    Idempotent: ``register(..., exist_ok=True)`` makes repeat calls no-ops (with a fallback that
+    suppresses the already-registered error on transformers builds whose ``register()`` predates
+    ``exist_ok``), so no run-once guard is needed.
+    """
+    if AutoConfig is None or AutoModel is None:
+        return
+    try:
+        AutoConfig.register(GR00TN17Config.model_type, GR00TN17Config, exist_ok=True)
+    except TypeError:
+        with suppress(ValueError):
+            AutoConfig.register(GR00TN17Config.model_type, GR00TN17Config)
+    try:
+        AutoModel.register(GR00TN17Config, GR00TN17, exist_ok=True)
+    except TypeError:
+        with suppress(ValueError):
+            AutoModel.register(GR00TN17Config, GR00TN17)

src/lerobot/policies/groot/modeling_groot.py

@@ -17,37 +17,43 @@
 """
 Groot Policy Wrapper for LeRobot Integration
 
-Minimal integration that delegates to Isaac-GR00T components where possible
-without porting their code. The intent is to:
-
-- Download and load the pretrained GR00T model via GR00TN15.from_pretrained
-- Optionally align action horizon similar to gr00t_finetune.py
-- Expose predict_action via GR00T model.get_action
-- Provide a training forward that can call the GR00T model forward if batch
-  structure matches.
-
-Notes:
-- Dataset loading and full training orchestration is handled by Isaac-GR00T
-  TrainRunner in their codebase. If you want to invoke that flow end-to-end
-  from LeRobot, see `GrootPolicy.finetune_with_groot_runner` below.
+Minimal integration that delegates to Isaac-GR00T N1.7 components where
+possible without porting their code. Dataset loading and training
+orchestration are handled by LeRobot's standard training stack.
 """
 
 import builtins
+import logging
 import os
 from collections import deque
 from pathlib import Path
 from typing import TypeVar
 
 import torch
+from huggingface_hub import hf_hub_download
+from huggingface_hub.constants import SAFETENSORS_SINGLE_FILE
+from huggingface_hub.errors import HfHubHTTPError
 from torch import Tensor
+from transformers.trainer_pt_utils import get_parameter_names
 
 from lerobot.configs import FeatureType, PolicyFeature
 from lerobot.utils.constants import ACTION, OBS_IMAGES
 from lerobot.utils.import_utils import require_package
 
 from ..pretrained import PreTrainedPolicy
-from .configuration_groot import GrootConfig
-from .groot_n1 import GR00TN15
+from ..utils import get_device_from_parameters
+from .configuration_groot import (
+    GROOT_N1_5,
+    GROOT_N1_5_REMOVAL_GUIDANCE,
+    GROOT_N1_7,
+    GrootConfig,
+    infer_groot_model_version,
+    infer_groot_n1_7_action_execution_horizon,
+    infer_groot_n1_7_action_horizon,
+)
+from .groot_n1_7 import GR00TN17, _tie_unused_qwen_lm_head
+
+logger = logging.getLogger(__name__)
 
 T = TypeVar("T", bound="GrootPolicy")
 
@@ -67,37 +73,77 @@ class GrootPolicy(PreTrainedPolicy):
 
         # Initialize GR00T model using ported components
         self._groot_model = self._create_groot_model()
+        self._action_queue_steps = self._resolve_action_queue_steps()
+        self._warned_native_relative_rtc_prefix_disabled = False
 
         self.reset()
 
     def _create_groot_model(self):
-        """Create and initialize the GR00T model using Isaac-GR00T API.
+        """Create and initialize the GR00T N1.7 model using the ported components."""
+        model_kwargs = {
+            "pretrained_model_name_or_path": self.config.base_model_path,
+            "tune_llm": self.config.tune_llm,
+            "tune_visual": self.config.tune_visual,
+            "tune_projector": self.config.tune_projector,
+            "tune_diffusion_model": self.config.tune_diffusion_model,
+            # Forwarded as a GR00TN17Config override; read back by set_trainable_parameters.
+            "tune_top_llm_layers": self.config.tune_top_llm_layers,
+            "use_flash_attention": self.config.use_flash_attention,
+        }
+        # Surface the inference-time knobs onto the model config only when the user set them; None
+        # leaves the value baked into the checkpoint untouched.
+        if self.config.num_inference_timesteps is not None:
+            model_kwargs["num_inference_timesteps"] = self.config.num_inference_timesteps
+        if self.config.rtc_ramp_rate is not None:
+            model_kwargs["rtc_ramp_rate"] = self.config.rtc_ramp_rate
 
-        This is only called when creating a NEW policy (not when loading from checkpoint).
-
-        Steps (delegating to Isaac-GR00T):
-        1) Download and load pretrained model via GR00TN15.from_pretrained
-        2) Align action horizon with data_config if provided
-        """
-        # Handle Flash Attention compatibility issues
-        self._handle_flash_attention_compatibility()
-
-        model = GR00TN15.from_pretrained(
-            pretrained_model_name_or_path=self.config.base_model_path,
-            tune_llm=self.config.tune_llm,
-            tune_visual=self.config.tune_visual,
-            tune_projector=self.config.tune_projector,
-            tune_diffusion_model=self.config.tune_diffusion_model,
+        model = GR00TN17.from_pretrained(
+            **model_kwargs,
+            tune_vlln=self.config.tune_vlln,
+            transformers_loading_kwargs={"trust_remote_code": True},
         )
-
-        model.compute_dtype = "bfloat16" if self.config.use_bf16 else model.compute_dtype
-        model.config.compute_dtype = model.compute_dtype
-
+        backbone = getattr(model, "backbone", None)
+        qwen_model = getattr(backbone, "model", None)
+        if qwen_model is not None:
+            _tie_unused_qwen_lm_head(qwen_model)
+        if self.config.model_params_fp32:
+            self._cast_model_parameters_to_fp32(model)
         return model
 
+    @staticmethod
+    def _cast_model_parameters_to_fp32(model: torch.nn.Module) -> None:
+        for parameter in model.parameters():
+            if parameter.is_floating_point():
+                parameter.data = parameter.data.to(torch.float32)
+
+    @staticmethod
+    def _build_weight_decay_parameter_groups(model: torch.nn.Module) -> list[dict[str, object]]:
+        forbidden_name_patterns = [
+            r"bias",
+            r"layernorm",
+            r"rmsnorm",
+            r"(?:^|\.)norm(?:$|\.)",
+            r"_norm(?:$|\.)",
+        ]
+        decay_names = set(get_parameter_names(model, [torch.nn.LayerNorm], forbidden_name_patterns))
+        decay_params = [
+            parameter
+            for name, parameter in model.named_parameters()
+            if parameter.requires_grad and name in decay_names
+        ]
+        no_decay_params = [
+            parameter
+            for name, parameter in model.named_parameters()
+            if parameter.requires_grad and name not in decay_names
+        ]
+        return [
+            {"params": decay_params},
+            {"params": no_decay_params, "weight_decay": 0.0},
+        ]
+
     def reset(self):
         """Reset policy state when environment resets."""
-        self._action_queue = deque([], maxlen=self.config.n_action_steps)
+        self._action_queue = deque([], maxlen=self._action_queue_steps)
 
     @classmethod
     def from_pretrained(
@@ -118,7 +164,7 @@ class GrootPolicy(PreTrainedPolicy):
         """Load Groot policy from pretrained model.
 
         Handles two cases:
-        1. Base GR00T models (e.g., 'nvidia/GR00T-N1.5-3B') - loads the raw model
+        1. Base GR00T N1.7 models - loads the raw model
         2. Fine-tuned LeRobot checkpoints - loads config and weights from safetensors
 
         Args:
@@ -137,13 +183,11 @@ class GrootPolicy(PreTrainedPolicy):
         Returns:
             Initialized GrootPolicy instance with loaded model
         """
-        from huggingface_hub import hf_hub_download
-        from huggingface_hub.constants import SAFETENSORS_SINGLE_FILE
-        from huggingface_hub.errors import HfHubHTTPError
-
-        print(
-            "The Groot policy is a wrapper around Nvidia's GR00T N1.5 model.\n"
-            f"Loading pretrained model from: {pretrained_name_or_path}"
+        requested_version = infer_groot_model_version(str(pretrained_name_or_path)) or GROOT_N1_7
+        logger.info(
+            "The Groot policy wraps NVIDIA's GR00T %s model. Loading pretrained model from: %s",
+            requested_version,
+            pretrained_name_or_path,
         )
 
         model_id = str(pretrained_name_or_path)
@@ -174,7 +218,7 @@ class GrootPolicy(PreTrainedPolicy):
 
         if is_finetuned_checkpoint:
             # This is a fine-tuned LeRobot checkpoint - use parent class loading
-            print("Detected fine-tuned LeRobot checkpoint, loading with state dict...")
+            logger.info("Detected fine-tuned LeRobot checkpoint, loading with state dict...")
             return super().from_pretrained(
                 pretrained_name_or_path=pretrained_name_or_path,
                 config=config,
@@ -190,11 +234,13 @@ class GrootPolicy(PreTrainedPolicy):
             )
 
         # This is a base GR00T model - load it fresh
-        print("Detected base GR00T model, loading from HuggingFace...")
+        logger.info("Detected base GR00T model, loading from HuggingFace...")
 
         if config is None:
             # Create default config with the pretrained path
-            config = GrootConfig(base_model_path=str(pretrained_name_or_path))
+            config = GrootConfig(
+                base_model_path=str(pretrained_name_or_path),
+            )
 
             # Add minimal visual feature required for validation
             # validate_features() will automatically add state and action features
@@ -215,6 +261,15 @@ class GrootPolicy(PreTrainedPolicy):
             if hasattr(config, key):
                 setattr(config, key, value)
 
+        inferred_version = infer_groot_model_version(config.base_model_path)
+        if inferred_version is not None and inferred_version != GROOT_N1_7:
+            message = (
+                f"GR00T model_version '{GROOT_N1_7}' does not match base_model_path "
+                f"'{config.base_model_path}', which looks like '{inferred_version}'."
+            )
+            if inferred_version == GROOT_N1_5:
+                message = f"{message} {GROOT_N1_5_REMOVAL_GUIDANCE}"
+            raise ValueError(message)
         # Create a fresh policy instance - this will automatically load the GR00T model
         # in __init__ via _create_groot_model()
         policy = cls(config)
@@ -222,24 +277,175 @@ class GrootPolicy(PreTrainedPolicy):
         policy.eval()
         return policy
 
-    def get_optim_params(self) -> dict:
-        return self.parameters()
+    def get_optim_params(self):  # type: ignore[override]
+        """Isaac-GR00T excludes biases and normalization parameters from weight decay."""
+        return self._build_weight_decay_parameter_groups(self)
+
+    def _resolve_action_queue_steps(self) -> int:
+        n_action_steps = int(self.config.n_action_steps)
+        checkpoint_action_horizon = infer_groot_n1_7_action_horizon(
+            self.config.base_model_path,
+            self.config.embodiment_tag,
+        )
+        execution_horizon = infer_groot_n1_7_action_execution_horizon(
+            self.config.base_model_path,
+            self.config.embodiment_tag,
+        )
+        horizons = [n_action_steps]
+        if checkpoint_action_horizon is not None:
+            horizons.append(checkpoint_action_horizon)
+        if execution_horizon is not None:
+            horizons.append(execution_horizon)
+        return min(horizons)
+
+    def _resolve_prediction_horizon(self, actions: Tensor) -> int:
+        """Return the policy-facing action horizon for a native GR00T prediction."""
+
+        horizons = [actions.shape[1]]
+        checkpoint_action_horizon = infer_groot_n1_7_action_horizon(
+            self.config.base_model_path,
+            self.config.embodiment_tag,
+        )
+        if checkpoint_action_horizon is not None:
+            horizons.append(checkpoint_action_horizon)
+
+        for horizon in (self.config.chunk_size, self.config.n_action_steps):
+            horizon = int(horizon)
+            if horizon > 0:
+                horizons.append(horizon)
+
+        return max(1, min(horizons))
+
+    def _filter_groot_inputs(self, batch: dict[str, Tensor], *, include_action: bool) -> dict[str, Tensor]:
+        allowed_base = {"state", "state_mask", "embodiment_id"}
+        if include_action:
+            allowed_base.update({"action", "action_mask"})
+
+        allowed_base.update(
+            {
+                "input_ids",
+                "attention_mask",
+                "pixel_values",
+                "image_grid_thw",
+                "mm_token_type_ids",
+                "pixel_values_videos",
+                "video_grid_thw",
+            }
+        )
+        allowed_base.add("action_mask")
+
+        return {
+            k: v for k, v in batch.items() if k in allowed_base and not (k.startswith("next.") or k == "info")
+        }
+
+    def _prepare_n1_7_rtc_inputs(
+        self,
+        inputs: dict[str, Tensor],
+        *,
+        inference_delay: object,
+        prev_chunk_left_over: object,
+    ) -> tuple[dict[str, Tensor], dict[str, object] | None]:
+        if prev_chunk_left_over is None:
+            return inputs, None
+        if getattr(self.config, "use_relative_actions", False):
+            # Generic RTC only provides normalized leftovers from the previous chunk. For
+            # native relative-action N1.7 checkpoints those rows are tied to the old
+            # observation state and old per-horizon stats row, so using them as the next
+            # prefix can push the policy in the wrong direction. Run without native RTC
+            # overlap guidance until a GROOT-specific RTC path can pass re-anchored
+            # absolute leftovers through.
+            if not getattr(self, "_warned_native_relative_rtc_prefix_disabled", False):
+                logger.info("Disabling native GR00T RTC prefix for relative-action policy")
+                self._warned_native_relative_rtc_prefix_disabled = True
+            return inputs, None
+        if not isinstance(prev_chunk_left_over, torch.Tensor):
+            raise TypeError("prev_chunk_left_over must be a torch.Tensor for GR00T N1.7 RTC.")
+        if prev_chunk_left_over.numel() == 0:
+            return inputs, None
+
+        prev_actions = prev_chunk_left_over
+        if prev_actions.ndim == 2:
+            prev_actions = prev_actions.unsqueeze(0)
+        elif prev_actions.ndim != 3:
+            raise ValueError("prev_chunk_left_over must have shape (T, A) or (B, T, A) for GR00T N1.7 RTC.")
+
+        state = inputs.get("state")
+        if state is None:
+            raise ValueError("GR00T N1.7 RTC requires `state` in the preprocessed batch.")
+        batch_size = state.shape[0]
+        if prev_actions.shape[0] == 1 and batch_size > 1:
+            prev_actions = prev_actions.expand(batch_size, -1, -1).clone()
+        elif prev_actions.shape[0] != batch_size:
+            raise ValueError("prev_chunk_left_over batch size must match the current GR00T N1.7 batch size.")
+
+        # The generic LeRobot RTC engine pads short leftovers with exact zero
+        # rows for fixed-shape policy calls. Native GR00T N1.7 RTC treats every
+        # provided prefix row as a real action constraint, so strip that padding
+        # before constructing the native overlap options.
+        valid_prefix_rows = prev_actions.detach().abs().sum(dim=(0, 2)) > 0
+        if valid_prefix_rows.any():
+            valid_prefix_steps = int(valid_prefix_rows.nonzero()[-1].item()) + 1
+            prev_actions = prev_actions[:, :valid_prefix_steps, :]
+        else:
+            return inputs, None
+
+        model_action_horizon = int(
+            getattr(self._groot_model.config, "action_horizon", self.config.chunk_size)
+        )
+        max_action_dim = int(getattr(self._groot_model.config, "max_action_dim", self.config.max_action_dim))
+        if prev_actions.shape[1] > model_action_horizon:
+            prev_actions = prev_actions[:, -model_action_horizon:, :]
+
+        action_horizon = int(prev_actions.shape[1])
+        if action_horizon <= 0:
+            return inputs, None
+
+        if prev_actions.shape[2] > max_action_dim:
+            prev_actions = prev_actions[:, :, :max_action_dim]
+        elif prev_actions.shape[2] < max_action_dim:
+            pad = torch.zeros(
+                prev_actions.shape[0],
+                prev_actions.shape[1],
+                max_action_dim - prev_actions.shape[2],
+                dtype=prev_actions.dtype,
+                device=prev_actions.device,
+            )
+            prev_actions = torch.cat([prev_actions, pad], dim=2)
+
+        prev_actions = prev_actions.to(device=state.device, dtype=state.dtype)
+
+        rtc_config = getattr(self.config, "rtc_config", None)
+        execution_horizon = int(getattr(rtc_config, "execution_horizon", action_horizon))
+        overlap_steps = max(0, min(action_horizon, execution_horizon))
+        if overlap_steps == 0:
+            return inputs, None
+
+        try:
+            frozen_steps = int(inference_delay or 0)
+        except (TypeError, ValueError):
+            frozen_steps = 0
+        frozen_steps = max(0, min(frozen_steps, overlap_steps))
+
+        options = {
+            "action_horizon": action_horizon,
+            "rtc_overlap_steps": overlap_steps,
+            "rtc_frozen_steps": frozen_steps,
+            "rtc_ramp_rate": float(getattr(self._groot_model.config, "rtc_ramp_rate", 6.0)),
+        }
+
+        inputs = dict(inputs)
+        inputs["action"] = prev_actions
+        return inputs, options
 
     def forward(self, batch: dict[str, Tensor]) -> tuple[Tensor, dict]:
         """Training forward pass.
 
         Delegates to Isaac-GR00T model.forward when inputs are compatible.
         """
-        # Build a clean input dict for GR00T: keep only tensors GR00T consumes
-        allowed_base = {"state", "state_mask", "action", "action_mask", "embodiment_id"}
-        groot_inputs = {
-            k: v
-            for k, v in batch.items()
-            if (k in allowed_base or k.startswith("eagle_")) and not (k.startswith("next.") or k == "info")
-        }
+        groot_inputs = self._filter_groot_inputs(batch, include_action=True)
 
         # Get device from model parameters
-        device = next(self.parameters()).device
+        device = get_device_from_parameters(self)
 
         # Run GR00T forward under bf16 autocast when enabled to reduce activation memory
         # Rationale: Matches original GR00T finetuning (bf16 compute, fp32 params) and avoids fp32 upcasts.
@@ -248,38 +454,52 @@ class GrootPolicy(PreTrainedPolicy):
 
         # Isaac-GR00T returns a BatchFeature; loss key is typically 'loss'
         loss = outputs.get("loss")
+        if loss is None:
+            raise RuntimeError(
+                "GR00T model.forward did not return a 'loss'. Training batches must include "
+                "'action' and 'action_mask'; check the preprocessor output."
+            )
 
         loss_dict = {"loss": loss.item()}
 
         return loss, loss_dict
 
     @torch.no_grad()
-    def predict_action_chunk(self, batch: dict[str, Tensor]) -> Tensor:
+    def predict_action_chunk(self, batch: dict[str, Tensor], **kwargs: object) -> Tensor:
         """Predict a chunk of actions for inference by delegating to Isaac-GR00T.
 
         Returns a tensor of shape (B, n_action_steps, action_dim).
+
+        For N1.7, LeRobot's RTC leftovers are converted into the native GR00T
+        action-overlap options before calling the underlying model.
         """
         self.eval()
 
-        # Build a clean input dict for GR00T: keep only tensors GR00T consumes
-        # Preprocessing is handled by the processor pipeline, so we just filter the batch
-        # NOTE: During inference, we should NOT pass action/action_mask (that's what we're predicting)
-        allowed_base = {"state", "state_mask", "embodiment_id"}
-        groot_inputs = {
-            k: v
-            for k, v in batch.items()
-            if (k in allowed_base or k.startswith("eagle_")) and not (k.startswith("next.") or k == "info")
-        }
+        # Preprocessing is handled by the processor pipeline, so we just filter the batch.
+        # During inference, we do not pass action because it is predicted.
+        # N1.7 still carries a 2-D action horizon mask from its checkpoint processor.
+        groot_inputs = self._filter_groot_inputs(batch, include_action=False)
+        groot_inputs, groot_options = self._prepare_n1_7_rtc_inputs(
+            groot_inputs,
+            inference_delay=kwargs.get("inference_delay"),
+            prev_chunk_left_over=kwargs.get("prev_chunk_left_over"),
+        )
 
         # Get device from model parameters
-        device = next(self.parameters()).device
+        device = get_device_from_parameters(self)
 
         # Use bf16 autocast for inference to keep memory low and match backbone dtype
         with torch.autocast(device_type=device.type, dtype=torch.bfloat16, enabled=self.config.use_bf16):
-            outputs = self._groot_model.get_action(groot_inputs)
+            if groot_options is not None:
+                outputs = self._groot_model.get_action(groot_inputs, options=groot_options)
+            else:
+                outputs = self._groot_model.get_action(groot_inputs)
 
         actions = outputs.get("action_pred")
 
+        prediction_horizon = self._resolve_prediction_horizon(actions)
+        actions = actions[:, :prediction_horizon]
+
         original_action_dim = self.config.output_features[ACTION].shape[0]
         actions = actions[:, :, :original_action_dim]
 
@@ -288,44 +508,17 @@ class GrootPolicy(PreTrainedPolicy):
     @torch.no_grad()
     def select_action(self, batch: dict[str, Tensor]) -> Tensor:
         """Select single action from action queue."""
+        if getattr(self.config, "use_relative_actions", False):
+            raise NotImplementedError(
+                "GrootPolicy.select_action does not support relative-action policies because cached "
+                "relative chunk actions can be decoded against newer observation states. Use "
+                "predict_action_chunk and postprocess the full chunk before queuing actions, or use "
+                "the RTC/chunked rollout inference path."
+            )
+
         self.eval()
 
         if len(self._action_queue) == 0:
             actions = self.predict_action_chunk(batch)
-            self._action_queue.extend(actions.transpose(0, 1))
+            self._action_queue.extend(actions[:, : self._action_queue_steps].transpose(0, 1))
         return self._action_queue.popleft()
-
-    # -------------------------
-    # Internal helpers
-    # -------------------------
-    def _handle_flash_attention_compatibility(self) -> None:
-        """Handle Flash Attention compatibility issues by setting environment variables.
-
-        This addresses the common 'undefined symbol' error that occurs when Flash Attention
-        is compiled against a different PyTorch version than what's currently installed.
-        """
-
-        # Set environment variables to handle Flash Attention compatibility
-        # These help with symbol resolution issues
-        os.environ.setdefault("FLASH_ATTENTION_FORCE_BUILD", "0")
-        os.environ.setdefault("FLASH_ATTENTION_SKIP_CUDA_BUILD", "0")
-
-        # Try to import flash_attn and handle failures gracefully
-        try:
-            import flash_attn
-
-            print(f"[GROOT] Flash Attention version: {flash_attn.__version__}")
-        except ImportError as e:
-            print(f"[GROOT] Flash Attention not available: {e}")
-            print("[GROOT] Will use fallback attention mechanism")
-        except Exception as e:
-            if "undefined symbol" in str(e):
-                print(f"[GROOT] Flash Attention compatibility issue detected: {e}")
-                print("[GROOT] This is likely due to PyTorch/Flash Attention version mismatch")
-                print("[GROOT] Consider reinstalling Flash Attention with compatible version:")
-                print("  pip uninstall flash-attn")
-                print("  pip install --no-build-isolation flash-attn==2.6.3")
-                print("[GROOT] Continuing with fallback attention mechanism")
-            else:
-                print(f"[GROOT] Flash Attention error: {e}")
-                print("[GROOT] Continuing with fallback attention mechanism")

src/lerobot/policies/groot/utils.py

@@ -1,47 +1,263 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Shared, side-effect-free utilities for the GR00T N1.7 policy.
+
+These helpers are consumed by both the config layer (checkpoint sidecar
+inspection) and the processor layer (stat flattening, action decoding, language
+and image packing). They are pure functions with no GR00T-specific state so they
+can be unit-tested in isolation and reused without importing the heavier
+config/processor modules.
+"""
+
+from __future__ import annotations
+
+import json
 from pathlib import Path
-from shutil import copytree
+from typing import Any
 
-from huggingface_hub import hf_hub_download
+import numpy as np
+import torch
 
 
-def ensure_eagle_cache_ready(vendor_dir: Path, cache_dir: Path, assets_repo: str) -> None:
-    """Populate the Eagle processor directory in cache and ensure tokenizer assets exist.
-
-    - Copies the vendored Eagle files into cache_dir (overwriting when needed).
-    - Downloads vocab.json and merges.txt into the same cache_dir if missing.
-    """
-    cache_dir = Path(cache_dir)
-    vendor_dir = Path(vendor_dir)
-
+def read_json(path: Path) -> dict[str, Any]:
+    """Read a JSON object from ``path``, returning ``{}`` on any read/parse error."""
     try:
-        # Populate/refresh cache with vendor files to ensure a complete processor directory
-        print(f"[GROOT] Copying vendor Eagle files to cache: {vendor_dir} -> {cache_dir}")
-        copytree(vendor_dir, cache_dir, dirs_exist_ok=True)
-    except Exception as exc:  # nosec: B110
-        print(f"[GROOT] Warning: Failed to copy vendor Eagle files to cache: {exc}")
+        with path.open() as f:
+            data = json.load(f)
+    except (OSError, json.JSONDecodeError):
+        return {}
+    return data if isinstance(data, dict) else {}
 
-    required_assets = [
-        "vocab.json",
-        "merges.txt",
-        "added_tokens.json",
-        "chat_template.json",
-        "special_tokens_map.json",
-        "config.json",
-        "generation_config.json",
-        "preprocessor_config.json",
-        "processor_config.json",
-        "tokenizer_config.json",
-    ]
 
-    print(f"[GROOT] Assets repo: {assets_repo} \n Cache dir: {cache_dir}")
+def as_int_pair(value: Any) -> list[int] | None:
+    if not isinstance(value, (list, tuple)) or len(value) != 2:
+        return None
+    try:
+        return [int(value[0]), int(value[1])]
+    except (TypeError, ValueError):
+        return None
 
-    for fname in required_assets:
-        dst = cache_dir / fname
-        if not dst.exists():
-            print(f"[GROOT] Fetching {fname}")
-            hf_hub_download(
-                repo_id=assets_repo,
-                filename=fname,
-                repo_type="model",
-                local_dir=str(cache_dir),
+
+def as_optional_int(value: Any) -> int | None:
+    if value is None:
+        return None
+    try:
+        return int(value)
+    except (TypeError, ValueError):
+        return None
+
+
+def as_optional_float(value: Any) -> float | None:
+    if value is None:
+        return None
+    try:
+        return float(value)
+    except (TypeError, ValueError):
+        return None
+
+
+def as_float_list(values: Any) -> list[float]:
+    if values is None:
+        return []
+    if isinstance(values, torch.Tensor):
+        return values.detach().cpu().reshape(-1).float().tolist()
+    if isinstance(values, np.ndarray):
+        return values.reshape(-1).astype(np.float32).tolist()
+    if isinstance(values, (list, tuple)):
+        flattened: list[float] = []
+        for value in values:
+            flattened.extend(as_float_list(value))
+        return flattened
+    return [float(values)]
+
+
+def config_value(value: Any) -> str:
+    if hasattr(value, "value"):
+        value = value.value
+    text = str(value).lower()
+    return {
+        "relative": "relative",
+        "absolute": "absolute",
+        "delta": "delta",
+        "eef": "eef",
+        "non_eef": "non_eef",
+        "default": "default",
+        "xyz_rot6d": "xyz+rot6d",
+        "xyz+rot6d": "xyz+rot6d",
+        "xyz_rotvec": "xyz+rotvec",
+        "xyz+rotvec": "xyz+rotvec",
+    }.get(text, text)
+
+
+def has_modality_stats(stats: dict[str, dict[str, Any]] | None) -> bool:
+    if not stats:
+        return False
+    return any(bool(modality_stats) for modality_stats in stats.values())
+
+
+def stat_dim_from_entry(entry: dict[str, Any]) -> int:
+    for stat_name in ("mean", "q01", "min", "max", "std"):
+        value = entry.get(stat_name)
+        if isinstance(value, torch.Tensor):
+            return int(value.shape[-1]) if value.ndim > 0 else 1
+        if isinstance(value, np.ndarray):
+            return int(value.shape[-1]) if value.ndim > 0 else 1
+        if isinstance(value, list) and len(value) > 0:
+            first = value[0]
+            if isinstance(first, (list, tuple)) and len(first) > 0:
+                return len(first)
+            return len(value)
+    return 0
+
+
+def flatten_n1_7_modality_stats(
+    *,
+    embodiment_stats: dict[str, Any],
+    embodiment_config: dict[str, Any],
+    modality: str,
+    use_percentiles: bool,
+    use_relative_action: bool,
+) -> dict[str, list[float]]:
+    """Flatten one N1.7 modality's grouped statistics in checkpoint order.
+
+    When checkpoints request percentile normalization, q01/q99 replace min/max
+    for regular groups. Relative action groups read from ``relative_action``
+    stats and keep min/max, matching Isaac-GR00T's processor override.
+    """
+
+    source_stats = embodiment_stats.get(modality, {})
+    modality_config = embodiment_config.get(modality, {})
+    if not isinstance(source_stats, dict) or not isinstance(modality_config, dict):
+        return {}
+    modality_keys = modality_config.get("modality_keys", [])
+    if not isinstance(modality_keys, list):
+        return {}
+
+    flattened: dict[str, list[float]] = {}
+    action_configs = modality_config.get("action_configs", []) if modality == "action" else []
+    if not isinstance(action_configs, list):
+        action_configs = []
+    relative_stats = embodiment_stats.get("relative_action", {})
+    if not isinstance(relative_stats, dict):
+        relative_stats = {}
+
+    for stat_name in ("min", "max", "mean", "std"):
+        values: list[float] = []
+        source_stat_name = stat_name
+        if use_percentiles and stat_name == "min":
+            source_stat_name = "q01"
+        elif use_percentiles and stat_name == "max":
+            source_stat_name = "q99"
+
+        for idx, modality_key in enumerate(modality_keys):
+            if not isinstance(modality_key, str):
+                continue
+            key_source_stats = source_stats
+            key_stat_name = source_stat_name
+            if modality == "action" and use_relative_action and idx < len(action_configs):
+                action_config = action_configs[idx]
+                if isinstance(action_config, dict) and config_value(action_config.get("rep")) == "relative":
+                    key_source_stats = relative_stats
+                    key_stat_name = stat_name
+            key_stats = key_source_stats.get(modality_key, {})
+            if not isinstance(key_stats, dict):
+                raise KeyError(f"Missing statistics for {modality}.{modality_key}")
+            raw_values = key_stats.get(key_stat_name)
+            if raw_values is None:
+                raise KeyError(f"Missing '{key_stat_name}' statistics for {modality}.{modality_key}")
+            values.extend(as_float_list(raw_values))
+        if values:
+            flattened[stat_name] = values
+
+    return flattened
+
+
+def rot6d_to_matrix(rot6d: np.ndarray) -> np.ndarray:
+    rows = rot6d.reshape(2, 3).astype(np.float64)
+    row1 = rows[0] / np.linalg.norm(rows[0])
+    row2 = rows[1] - np.dot(row1, rows[1]) * row1
+    row2 = row2 / np.linalg.norm(row2)
+    row3 = np.cross(row1, row2)
+    return np.vstack([row1, row2, row3])
+
+
+def xyz_rot6d_to_homogeneous(xyz_rot6d: np.ndarray) -> np.ndarray:
+    transform = np.eye(4, dtype=np.float64)
+    transform[:3, :3] = rot6d_to_matrix(xyz_rot6d[3:])
+    transform[:3, 3] = xyz_rot6d[:3]
+    return transform
+
+
+def homogeneous_to_xyz_rot6d(transform: np.ndarray) -> np.ndarray:
+    return np.concatenate([transform[:3, 3], transform[:2, :3].reshape(-1)], axis=0)
+
+
+def relative_eef_to_absolute(action: np.ndarray, reference_state: np.ndarray) -> np.ndarray:
+    """Convert relative EEF deltas in xyz+rot6d format to absolute EEF poses."""
+
+    out = np.empty_like(action, dtype=np.float64)
+    for batch_idx in range(action.shape[0]):
+        reference = xyz_rot6d_to_homogeneous(reference_state[batch_idx])
+        for timestep in range(action.shape[1]):
+            relative = xyz_rot6d_to_homogeneous(action[batch_idx, timestep])
+            out[batch_idx, timestep] = homogeneous_to_xyz_rot6d(reference @ relative)
+    return out.astype(np.float32)
+
+
+def infer_n1_7_batch_size_and_device(
+    obs: dict[str, Any], action: torch.Tensor | None
+) -> tuple[int, torch.device]:
+    for value in list(obs.values()) + [action]:
+        if isinstance(value, torch.Tensor):
+            return value.shape[0], value.device
+    video = obs.get("video")
+    if isinstance(video, np.ndarray):
+        return video.shape[0], torch.device("cpu")
+    return 1, torch.device("cpu")
+
+
+def prepare_n1_7_language_batch(
+    language: Any,
+    batch_size: int,
+    *,
+    formalize_language: bool,
+) -> list[str]:
+    default_language = "Perform the task."
+    if language is None or (isinstance(language, str) and language == ""):
+        languages = [default_language] * batch_size
+    elif isinstance(language, str):
+        languages = [language] * batch_size
+    elif isinstance(language, (list, tuple)):
+        languages = list(language)
+        if len(languages) == 0:
+            languages = [default_language] * batch_size
+        elif len(languages) == 1 and batch_size > 1:
+            languages = languages * batch_size
+        elif len(languages) != batch_size:
+            raise ValueError(
+                f"language batch has {len(languages)} entries, but GR00T N1.7 input batch has {batch_size}."
             )
+    else:
+        languages = [str(language)] * batch_size
+
+    formatted = []
+    for item in languages:
+        text = str(item) if item else default_language
+        if formalize_language:
+            text = text.lower()
+            text = "".join(ch for ch in text if ch.isalnum() or ch.isspace() or ch == "_")
+        formatted.append(text)
+    return formatted

src/lerobot/policies/utils.py

@@ -126,7 +126,8 @@ def prepare_observation_for_inference(
     for name in observation:
         observation[name] = torch.from_numpy(observation[name])
         if "image" in name:
-            observation[name] = observation[name].type(torch.float32) / 255
+            if observation[name].dtype == torch.uint8:
+                observation[name] = observation[name].type(torch.float32) / 255
             observation[name] = observation[name].permute(2, 0, 1).contiguous()
         observation[name] = observation[name].unsqueeze(0)
         observation[name] = observation[name].to(device)

src/lerobot/robots/hope_jr/hope_jr_arm.py

@@ -66,9 +66,14 @@ class HopeJrArm(Robot):
 
     @property
     def _cameras_ft(self) -> dict[str, tuple]:
-        return {
-            cam: (self.config.cameras[cam].height, self.config.cameras[cam].width, 3) for cam in self.cameras
-        }
+        features: dict[str, tuple] = {}
+        for cam in self.cameras:
+            cfg = self.config.cameras[cam]
+            if getattr(cfg, "use_rgb", True):
+                features[cam] = (cfg.height, cfg.width, 3)
+            if getattr(cfg, "use_depth", False):
+                features[f"{cam}_depth"] = (cfg.height, cfg.width, 1)
+        return features
 
     @cached_property
     def observation_features(self) -> dict[str, type | tuple]:
@@ -139,10 +144,17 @@ class HopeJrArm(Robot):
 
         # Capture images from cameras
         for cam_key, cam in self.cameras.items():
-            start = time.perf_counter()
-            obs_dict[cam_key] = cam.read_latest()
-            dt_ms = (time.perf_counter() - start) * 1e3
-            logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+            if getattr(cam, "use_rgb", True):
+                start = time.perf_counter()
+                obs_dict[cam_key] = cam.read_latest()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+
+            if getattr(cam, "use_depth", False):
+                start = time.perf_counter()
+                obs_dict[f"{cam_key}_depth"] = cam.read_latest_depth()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key} depth: {dt_ms:.1f}ms")
 
         return obs_dict
 

src/lerobot/robots/hope_jr/hope_jr_hand.py

@@ -102,9 +102,14 @@ class HopeJrHand(Robot):
 
     @property
     def _cameras_ft(self) -> dict[str, tuple]:
-        return {
-            cam: (self.config.cameras[cam].height, self.config.cameras[cam].width, 3) for cam in self.cameras
-        }
+        features: dict[str, tuple] = {}
+        for cam in self.cameras:
+            cfg = self.config.cameras[cam]
+            if getattr(cfg, "use_rgb", True):
+                features[cam] = (cfg.height, cfg.width, 3)
+            if getattr(cfg, "use_depth", False):
+                features[f"{cam}_depth"] = (cfg.height, cfg.width, 1)
+        return features
 
     @cached_property
     def observation_features(self) -> dict[str, type | tuple]:
@@ -170,10 +175,17 @@ class HopeJrHand(Robot):
 
         # Capture images from cameras
         for cam_key, cam in self.cameras.items():
-            start = time.perf_counter()
-            obs_dict[cam_key] = cam.read_latest()
-            dt_ms = (time.perf_counter() - start) * 1e3
-            logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+            if getattr(cam, "use_rgb", True):
+                start = time.perf_counter()
+                obs_dict[cam_key] = cam.read_latest()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+
+            if getattr(cam, "use_depth", False):
+                start = time.perf_counter()
+                obs_dict[f"{cam_key}_depth"] = cam.read_latest_depth()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key} depth: {dt_ms:.1f}ms")
 
         return obs_dict
 

src/lerobot/robots/koch_follower/koch_follower.py

@@ -68,9 +68,14 @@ class KochFollower(Robot):
 
     @property
     def _cameras_ft(self) -> dict[str, tuple]:
-        return {
-            cam: (self.config.cameras[cam].height, self.config.cameras[cam].width, 3) for cam in self.cameras
-        }
+        features: dict[str, tuple] = {}
+        for cam in self.cameras:
+            cfg = self.config.cameras[cam]
+            if getattr(cfg, "use_rgb", True):
+                features[cam] = (cfg.height, cfg.width, 3)
+            if getattr(cfg, "use_depth", False):
+                features[f"{cam}_depth"] = (cfg.height, cfg.width, 1)
+        return features
 
     @cached_property
     def observation_features(self) -> dict[str, type | tuple]:
@@ -192,10 +197,17 @@ class KochFollower(Robot):
 
         # Capture images from cameras
         for cam_key, cam in self.cameras.items():
-            start = time.perf_counter()
-            obs_dict[cam_key] = cam.read_latest()
-            dt_ms = (time.perf_counter() - start) * 1e3
-            logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+            if getattr(cam, "use_rgb", True):
+                start = time.perf_counter()
+                obs_dict[cam_key] = cam.read_latest()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+
+            if getattr(cam, "use_depth", False):
+                start = time.perf_counter()
+                obs_dict[f"{cam_key}_depth"] = cam.read_latest_depth()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key} depth: {dt_ms:.1f}ms")
 
         return obs_dict
 

src/lerobot/robots/lekiwi/lekiwi.py

@@ -72,6 +72,12 @@ class LeKiwi(Robot):
         )
         self.arm_motors = [motor for motor in self.bus.motors if motor.startswith("arm")]
         self.base_motors = [motor for motor in self.bus.motors if motor.startswith("base")]
+        depth_cameras = [name for name, cfg in config.cameras.items() if getattr(cfg, "use_depth", False)]
+        if depth_cameras:
+            raise NotImplementedError(
+                f"Depth cameras are not supported on LeKiwi (got depth-enabled cameras: {depth_cameras}). "
+                "The host/client transport only carries color frames."
+            )
         self.cameras = make_cameras_from_configs(config.cameras)
 
     @property

src/lerobot/robots/lekiwi/lekiwi_client.py

@@ -44,6 +44,13 @@ class LeKiwiClient(Robot):
         self.id = config.id
         self.robot_type = config.type
 
+        depth_cameras = [name for name, cfg in config.cameras.items() if getattr(cfg, "use_depth", False)]
+        if depth_cameras:
+            raise NotImplementedError(
+                f"Depth cameras are not supported on LeKiwi (got depth-enabled cameras: {depth_cameras}). "
+                "The host/client transport only carries color frames."
+            )
+
         self.remote_ip = config.remote_ip
         self.port_zmq_cmd = config.port_zmq_cmd
         self.port_zmq_observations = config.port_zmq_observations

src/lerobot/robots/omx_follower/omx_follower.py

@@ -68,9 +68,14 @@ class OmxFollower(Robot):
 
     @property
     def _cameras_ft(self) -> dict[str, tuple]:
-        return {
-            cam: (self.config.cameras[cam].height, self.config.cameras[cam].width, 3) for cam in self.cameras
-        }
+        features: dict[str, tuple] = {}
+        for cam in self.cameras:
+            cfg = self.config.cameras[cam]
+            if getattr(cfg, "use_rgb", True):
+                features[cam] = (cfg.height, cfg.width, 3)
+            if getattr(cfg, "use_depth", False):
+                features[f"{cam}_depth"] = (cfg.height, cfg.width, 1)
+        return features
 
     @cached_property
     def observation_features(self) -> dict[str, type | tuple]:
@@ -175,10 +180,17 @@ class OmxFollower(Robot):
 
         # Capture images from cameras
         for cam_key, cam in self.cameras.items():
-            start = time.perf_counter()
-            obs_dict[cam_key] = cam.read_latest()
-            dt_ms = (time.perf_counter() - start) * 1e3
-            logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+            if getattr(cam, "use_rgb", True):
+                start = time.perf_counter()
+                obs_dict[cam_key] = cam.read_latest()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+
+            if getattr(cam, "use_depth", False):
+                start = time.perf_counter()
+                obs_dict[f"{cam_key}_depth"] = cam.read_latest_depth()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key} depth: {dt_ms:.1f}ms")
 
         return obs_dict
 

src/lerobot/robots/openarm_follower/openarm_follower.py

@@ -101,9 +101,14 @@ class OpenArmFollower(Robot):
     @property
     def _cameras_ft(self) -> dict[str, tuple]:
         """Camera features for observation space."""
-        return {
-            cam: (self.config.cameras[cam].height, self.config.cameras[cam].width, 3) for cam in self.cameras
-        }
+        features: dict[str, tuple] = {}
+        for cam in self.cameras:
+            cfg = self.config.cameras[cam]
+            if getattr(cfg, "use_rgb", True):
+                features[cam] = (cfg.height, cfg.width, 3)
+            if getattr(cfg, "use_depth", False):
+                features[f"{cam}_depth"] = (cfg.height, cfg.width, 1)
+        return features
 
     @cached_property
     def observation_features(self) -> dict[str, type | tuple]:
@@ -242,10 +247,17 @@ class OpenArmFollower(Robot):
 
         # Capture images from cameras
         for cam_key, cam in self.cameras.items():
-            start = time.perf_counter()
-            obs_dict[cam_key] = cam.read_latest()
-            dt_ms = (time.perf_counter() - start) * 1e3
-            logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+            if getattr(cam, "use_rgb", True):
+                start = time.perf_counter()
+                obs_dict[cam_key] = cam.read_latest()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+
+            if getattr(cam, "use_depth", False):
+                start = time.perf_counter()
+                obs_dict[f"{cam_key}_depth"] = cam.read_latest_depth()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key} depth: {dt_ms:.1f}ms")
 
         dt_ms = (time.perf_counter() - start) * 1e3
         logger.debug(f"{self} get_observation took: {dt_ms:.1f}ms")

src/lerobot/robots/rebot_b601_follower/rebot_b601_follower.py

@@ -80,9 +80,14 @@ class RebotB601Follower(Robot):
 
     @property
     def _cameras_ft(self) -> dict[str, tuple]:
-        return {
-            cam: (self.config.cameras[cam].height, self.config.cameras[cam].width, 3) for cam in self.cameras
-        }
+        features: dict[str, tuple] = {}
+        for cam in self.cameras:
+            cfg = self.config.cameras[cam]
+            if getattr(cfg, "use_rgb", True):
+                features[cam] = (cfg.height, cfg.width, 3)
+            if getattr(cfg, "use_depth", False):
+                features[f"{cam}_depth"] = (cfg.height, cfg.width, 1)
+        return features
 
     @cached_property
     def observation_features(self) -> dict[str, type | tuple]:
@@ -213,10 +218,17 @@ class RebotB601Follower(Robot):
         logger.debug(f"{self} read state: {dt_ms:.1f}ms")
 
         for cam_key, cam in self.cameras.items():
-            start = time.perf_counter()
-            obs_dict[cam_key] = cam.read_latest()
-            dt_ms = (time.perf_counter() - start) * 1e3
-            logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+            if getattr(cam, "use_rgb", True):
+                start = time.perf_counter()
+                obs_dict[cam_key] = cam.read_latest()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+
+            if getattr(cam, "use_depth", False):
+                start = time.perf_counter()
+                obs_dict[f"{cam_key}_depth"] = cam.read_latest_depth()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key} depth: {dt_ms:.1f}ms")
 
         return obs_dict
 

src/lerobot/robots/so_follower/so_follower.py

@@ -68,9 +68,13 @@ class SOFollower(Robot):
 
     @property
     def _cameras_ft(self) -> dict[str, tuple]:
-        return {
-            cam: (self.config.cameras[cam].height, self.config.cameras[cam].width, 3) for cam in self.cameras
-        }
+        features: dict[str, tuple] = {}
+        for cam in self.cameras:
+            if getattr(self.cameras[cam], "use_rgb", True):
+                features[cam] = (self.cameras[cam].height, self.cameras[cam].width, 3)
+            if getattr(self.cameras[cam], "use_depth", False):
+                features[f"{cam}_depth"] = (self.cameras[cam].height, self.cameras[cam].width, 1)
+        return features
 
     @cached_property
     def observation_features(self) -> dict[str, type | tuple]:
@@ -185,10 +189,17 @@ class SOFollower(Robot):
 
         # Capture images from cameras
         for cam_key, cam in self.cameras.items():
-            start = time.perf_counter()
-            obs_dict[cam_key] = cam.read_latest()
-            dt_ms = (time.perf_counter() - start) * 1e3
-            logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+            if getattr(cam, "use_rgb", True):
+                start = time.perf_counter()
+                obs_dict[cam_key] = cam.read_latest()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key}: {dt_ms:.1f}ms")
+
+            if getattr(cam, "use_depth", False):
+                start = time.perf_counter()
+                obs_dict[f"{cam_key}_depth"] = cam.read_latest_depth()
+                dt_ms = (time.perf_counter() - start) * 1e3
+                logger.debug(f"{self} read {cam_key} depth: {dt_ms:.1f}ms")
 
         return obs_dict
 

src/lerobot/robots/unitree_g1/unitree_g1.py

@@ -222,9 +222,14 @@ class UnitreeG1(Robot):
 
     @property
     def _cameras_ft(self) -> dict[str, tuple]:
-        return {
-            cam: (self.config.cameras[cam].height, self.config.cameras[cam].width, 3) for cam in self.cameras
-        }
+        features: dict[str, tuple] = {}
+        for cam in self.cameras:
+            cfg = self.config.cameras[cam]
+            if getattr(cfg, "use_rgb", True):
+                features[cam] = (cfg.height, cfg.width, 3)
+            if getattr(cfg, "use_depth", False):
+                features[f"{cam}_depth"] = (cfg.height, cfg.width, 1)
+        return features
 
     @cached_property
     def observation_features(self) -> dict[str, type | tuple]:
@@ -458,7 +463,10 @@ class UnitreeG1(Robot):
 
         # Cameras - read images from ZMQ cameras
         for cam_name, cam in self._cameras.items():
-            obs[cam_name] = cam.read_latest()
+            if getattr(cam, "use_rgb", True):
+                obs[cam_name] = cam.read_latest()
+            if getattr(cam, "use_depth", False):
+                obs[f"{cam_name}_depth"] = cam.read_latest_depth()
 
         return obs
 

src/lerobot/rollout/context.py

@@ -332,7 +332,8 @@ def build_rollout_context(
                 cfg.dataset.repo_id,
                 root=cfg.dataset.root,
                 batch_encoding_size=cfg.dataset.video_encoding_batch_size,
-                camera_encoder=cfg.dataset.camera_encoder,
+                rgb_encoder=cfg.dataset.rgb_encoder,
+                depth_encoder=cfg.dataset.depth_encoder,
                 streaming_encoding=cfg.dataset.streaming_encoding,
                 encoder_queue_maxsize=cfg.dataset.encoder_queue_maxsize,
                 encoder_threads=cfg.dataset.encoder_threads,
@@ -367,7 +368,8 @@ def build_rollout_context(
                 image_writer_threads=cfg.dataset.num_image_writer_threads_per_camera
                 * len(robot.cameras if hasattr(robot, "cameras") else []),
                 batch_encoding_size=cfg.dataset.video_encoding_batch_size,
-                camera_encoder=cfg.dataset.camera_encoder,
+                rgb_encoder=cfg.dataset.rgb_encoder,
+                depth_encoder=cfg.dataset.depth_encoder,
                 streaming_encoding=cfg.dataset.streaming_encoding,
                 encoder_queue_maxsize=cfg.dataset.encoder_queue_maxsize,
                 encoder_threads=cfg.dataset.encoder_threads,

src/lerobot/scripts/lerobot_dataset_viz.py

@@ -77,15 +77,28 @@ from lerobot.utils.constants import ACTION, DONE, OBS_STATE, REWARD
 from lerobot.utils.utils import init_logging
 
 
+def check_chw_float32(frame: torch.Tensor) -> None:
+    """
+    Check if a frame is a channel-first, float32 tensor.
+    """
+    assert frame.dtype == torch.float32
+    assert frame.ndim == 3
+    c, h, w = frame.shape
+    assert c < h and c < w, f"expect channel first images, but instead {frame.shape}"
+
+
 def to_hwc_uint8_numpy(chw_float32_torch: torch.Tensor) -> np.ndarray:
-    assert chw_float32_torch.dtype == torch.float32
-    assert chw_float32_torch.ndim == 3
-    c, h, w = chw_float32_torch.shape
-    assert c < h and c < w, f"expect channel first images, but instead {chw_float32_torch.shape}"
+    check_chw_float32(chw_float32_torch)
     hwc_uint8_numpy = (chw_float32_torch * 255).type(torch.uint8).permute(1, 2, 0).numpy()
     return hwc_uint8_numpy
 
 
+def to_hwc_uint16_numpy(chw_float32_torch: torch.Tensor) -> np.ndarray:
+    check_chw_float32(chw_float32_torch)
+    hwc_uint16_numpy = chw_float32_torch.round().type(torch.uint16).permute(1, 2, 0).numpy()
+    return hwc_uint16_numpy
+
+
 def visualize_dataset(
     dataset: LeRobotDataset,
     episode_index: int,
@@ -138,6 +151,14 @@ def visualize_dataset(
 
     logging.info("Logging to Rerun")
 
+    # Use the dataset's q01/q99 depth statistics for robust depth range bounds
+    depth_ranges = {}
+    for key in dataset.meta.depth_keys:
+        stats = dataset.meta.stats[key]
+        lo = stats["q01"] if "q01" in stats else stats["min"]
+        hi = stats["q99"] if "q99" in stats else stats["max"]
+        depth_ranges[key] = (float(np.asarray(lo).item()), float(np.asarray(hi).item()))
+
     first_index = None
     for batch in tqdm.tqdm(dataloader, total=len(dataloader)):
         if first_index is None:
@@ -149,9 +170,18 @@ def visualize_dataset(
 
             # display each camera image
             for key in dataset.meta.camera_keys:
-                img = to_hwc_uint8_numpy(batch[key][i])
-                img_entity = rr.Image(img).compress() if display_compressed_images else rr.Image(img)
-                rr.log(key, entity=img_entity)
+                if key in dataset.meta.depth_keys:
+                    depth = to_hwc_uint16_numpy(batch[key][i])
+                    depth_entity = rr.DepthImage(
+                        depth,
+                        colormap=rr.components.Colormap.Viridis,
+                        depth_range=depth_ranges[key],
+                    )
+                    rr.log(key, entity=depth_entity)
+                else:
+                    img = to_hwc_uint8_numpy(batch[key][i])
+                    img_entity = rr.Image(img).compress() if display_compressed_images else rr.Image(img)
+                    rr.log(key, entity=img_entity)
 
             # display each dimension of action space (e.g. actuators command)
             if ACTION in batch:

src/lerobot/scripts/lerobot_edit_dataset.py

@@ -133,6 +133,15 @@ Convert image dataset to video format and save locally:
         --new_root /path/to/output/pusht_video \
         --operation.type convert_image_to_video
 
+Convert image dataset (with depth maps) to video format, customizing the depth encoder:
+    lerobot-edit-dataset \
+        --repo_id lerobot/pusht_image \
+        --new_root /path/to/output/pusht_video \
+        --operation.type convert_image_to_video \
+        --operation.depth_encoder.depth_min 0.01 \
+        --operation.depth_encoder.depth_max 10.0 \
+        --operation.depth_encoder.use_log true
+
 Convert image dataset to video format and save with new repo_id:
     lerobot-edit-dataset \
         --repo_id lerobot/pusht_image \
@@ -190,17 +199,17 @@ Re-encode all videos in a dataset (saves to lerobot/pusht_reencoded by default):
     lerobot-edit-dataset \
         --repo_id lerobot/pusht \
         --operation.type reencode_videos \
-        --operation.camera_encoder.vcodec h264 \
-        --operation.camera_encoder.pix_fmt yuv420p \
-        --operation.camera_encoder.crf 23
+        --operation.rgb_encoder.vcodec h264 \
+        --operation.rgb_encoder.pix_fmt yuv420p \
+        --operation.rgb_encoder.crf 23
 
 Re-encode videos into a new dataset using 4 parallel processes:
     lerobot-edit-dataset \
         --repo_id lerobot/pusht \
         --new_repo_id lerobot/pusht_h264 \
         --operation.type reencode_videos \
-        --operation.camera_encoder.vcodec h264 \
-        --operation.camera_encoder.crf 23 \
+        --operation.rgb_encoder.vcodec h264 \
+        --operation.rgb_encoder.crf 23 \
         --operation.num_workers 4
 
 Re-encode videos in-place (overwrites original dataset):
@@ -208,9 +217,16 @@ Re-encode videos in-place (overwrites original dataset):
         --repo_id lerobot/pusht \
         --new_repo_id lerobot/pusht \
         --operation.type reencode_videos \
-        --operation.camera_encoder.vcodec h264 \
+        --operation.rgb_encoder.vcodec h264 \
         --operation.overwrite true
 
+Re-encode both RGB and depth videos in a dataset (depth quantization params are preserved):
+    lerobot-edit-dataset \
+        --repo_id lerobot/pusht_depth \
+        --operation.type reencode_videos \
+        --operation.rgb_encoder.vcodec h264 \
+        --operation.depth_encoder.extra_options '{"x265-params": "lossless=1"}'
+
 Using JSON config file:
     lerobot-edit-dataset \
         --config_path path/to/edit_config.json
@@ -225,7 +241,13 @@ from pathlib import Path
 
 import draccus
 
-from lerobot.configs import VideoEncoderConfig, camera_encoder_defaults, parser
+from lerobot.configs import (
+    DepthEncoderConfig,
+    RGBEncoderConfig,
+    depth_encoder_defaults,
+    parser,
+    rgb_encoder_defaults,
+)
 from lerobot.datasets import (
     LeRobotDataset,
     convert_image_to_video_dataset,
@@ -287,7 +309,8 @@ class ModifyTasksConfig(OperationConfig):
 @dataclass
 class ConvertImageToVideoConfig(OperationConfig):
     output_dir: str | None = None
-    camera_encoder: VideoEncoderConfig = field(default_factory=camera_encoder_defaults)
+    rgb_encoder: RGBEncoderConfig = field(default_factory=rgb_encoder_defaults)
+    depth_encoder: DepthEncoderConfig = field(default_factory=depth_encoder_defaults)
     episode_indices: list[int] | None = None
     num_workers: int = 4
     max_episodes_per_batch: int | None = None
@@ -308,7 +331,8 @@ class RecomputeStatsConfig(OperationConfig):
 @OperationConfig.register_subclass("reencode_videos")
 @dataclass
 class ReencodeVideosConfig(OperationConfig):
-    camera_encoder: VideoEncoderConfig = field(default_factory=camera_encoder_defaults)
+    rgb_encoder: RGBEncoderConfig = field(default_factory=rgb_encoder_defaults)
+    depth_encoder: DepthEncoderConfig = field(default_factory=depth_encoder_defaults)
     num_workers: int = 0
     encoder_threads: int | None = None
     overwrite: bool = False
@@ -601,7 +625,8 @@ def handle_convert_image_to_video(cfg: EditDatasetConfig) -> None:
         dataset=dataset,
         output_dir=output_dir,
         repo_id=output_repo_id,
-        camera_encoder=getattr(cfg.operation, "camera_encoder", None) or camera_encoder_defaults(),
+        rgb_encoder=getattr(cfg.operation, "rgb_encoder", None) or rgb_encoder_defaults(),
+        depth_encoder=getattr(cfg.operation, "depth_encoder", None) or depth_encoder_defaults(),
         episode_indices=getattr(cfg.operation, "episode_indices", None),
         num_workers=getattr(cfg.operation, "num_workers", 4),
         max_episodes_per_batch=getattr(cfg.operation, "max_episodes_per_batch", None),
@@ -719,10 +744,14 @@ def handle_reencode_videos(cfg: EditDatasetConfig) -> None:
         shutil.copytree(input_root, output_root)
         dataset = LeRobotDataset(output_repo_id, root=output_root)
 
-    logging.info(f"Re-encoding videos in {output_repo_id} with {cfg.operation.camera_encoder}")
+    logging.info(
+        f"Re-encoding videos in {output_repo_id} with RGB encoder {cfg.operation.rgb_encoder} "
+        f"and depth encoder {cfg.operation.depth_encoder}"
+    )
     reencode_dataset(
         dataset,
-        camera_encoder=cfg.operation.camera_encoder,
+        rgb_encoder=cfg.operation.rgb_encoder,
+        depth_encoder=cfg.operation.depth_encoder,
         encoder_threads=cfg.operation.encoder_threads,
         num_workers=cfg.operation.num_workers,
     )

src/lerobot/scripts/lerobot_record.py

@@ -79,9 +79,9 @@ lerobot-record \\
     --dataset.single_task="Grab the cube" \\
     --dataset.streaming_encoding=true \\
     --dataset.encoder_threads=2 \\
-    --dataset.camera_encoder.vcodec=h264 \\
-    --dataset.camera_encoder.preset=fast \\
-    --dataset.camera_encoder.extra_options={"tune": "film", "profile:v": "high", "bf": 2} \\
+    --dataset.rgb_encoder.vcodec=h264 \\
+    --dataset.rgb_encoder.preset=fast \\
+    --dataset.rgb_encoder.extra_options={"tune": "film", "profile:v": "high", "bf": 2} \\
     --display_data=true
 ```
 """
@@ -400,7 +400,8 @@ def record(
                 cfg.dataset.repo_id,
                 root=cfg.dataset.root,
                 batch_encoding_size=cfg.dataset.video_encoding_batch_size,
-                camera_encoder=cfg.dataset.camera_encoder,
+                rgb_encoder=cfg.dataset.rgb_encoder,
+                depth_encoder=cfg.dataset.depth_encoder,
                 encoder_threads=cfg.dataset.encoder_threads,
                 streaming_encoding=cfg.dataset.streaming_encoding,
                 encoder_queue_maxsize=cfg.dataset.encoder_queue_maxsize,
@@ -429,7 +430,8 @@ def record(
                 image_writer_processes=cfg.dataset.num_image_writer_processes,
                 image_writer_threads=cfg.dataset.num_image_writer_threads_per_camera * len(robot.cameras),
                 batch_encoding_size=cfg.dataset.video_encoding_batch_size,
-                camera_encoder=cfg.dataset.camera_encoder,
+                rgb_encoder=cfg.dataset.rgb_encoder,
+                depth_encoder=cfg.dataset.depth_encoder,
                 encoder_threads=cfg.dataset.encoder_threads,
                 streaming_encoding=cfg.dataset.streaming_encoding,
                 encoder_queue_maxsize=cfg.dataset.encoder_queue_maxsize,
@@ -443,7 +445,7 @@ def record(
 
         if not cfg.dataset.streaming_encoding:
             logging.info(
-                "Streaming encoding is disabled. If you have capable hardware, consider enabling it for way faster episode saving. --dataset.streaming_encoding=true --dataset.encoder_threads=2 # --dataset.camera_encoder.vcodec=auto. More info in the documentation: https://huggingface.co/docs/lerobot/streaming_video_encoding"
+                "Streaming encoding is disabled. If you have capable hardware, consider enabling it for way faster episode saving. --dataset.streaming_encoding=true --dataset.encoder_threads=2 # --dataset.rgb_encoder.vcodec=auto. More info in the documentation: https://huggingface.co/docs/lerobot/streaming_video_encoding"
             )
 
         with VideoEncodingManager(dataset):

src/lerobot/scripts/lerobot_rollout.py

@@ -142,9 +142,9 @@ Usage examples
         --robot.port=/dev/ttyACM0 \\
         --task="pick up cube" --duration=60 \\
         --display_data=true \\
-        --dataset.camera_encoder.vcodec=h264 \\
-        --dataset.camera_encoder.preset=fast \\
-        --dataset.camera_encoder.extra_options={"tune": "film", "profile:v": "high", "bf": 2}
+        --dataset.rgb_encoder.vcodec=h264 \\
+        --dataset.rgb_encoder.preset=fast \\
+        --dataset.rgb_encoder.extra_options={"tune": "film", "profile:v": "high", "bf": 2}
 """
 
 import logging

src/lerobot/utils/feature_utils.py

@@ -51,7 +51,9 @@ def hw_to_dataset_features(
 
     This function takes a dictionary describing hardware outputs (like joint states
     or camera image shapes) and formats it into the standard LeRobot feature
-    specification.
+    specification. Single-channel cameras (shape ``(H, W, 1)``) are flagged as depth
+    maps via ``info["is_depth_map"] = True``; three-channel cameras ``(H, W, 3)`` are
+    treated as RGB.
 
     Args:
         hw_features (dict): Dictionary mapping feature names to their type (float for
@@ -61,7 +63,7 @@ def hw_to_dataset_features(
         use_video (bool): If True, image features are marked as "video", otherwise "image".
 
     Returns:
-        dict: A LeRobot features dictionary.
+        dict: A LeRobot features dictionary. Depth cameras carry ``info["is_depth_map"] = True``.
     """
     features = {}
     joint_fts = {
@@ -69,6 +71,7 @@ def hw_to_dataset_features(
         for key, ftype in hw_features.items()
         if ftype is float or (isinstance(ftype, PolicyFeature) and ftype.type != FeatureType.VISUAL)
     }
+    # TODO(CarolinePascal): we should not rely on the shape to determine if a feature is a camera !
     cam_fts = {key: shape for key, shape in hw_features.items() if isinstance(shape, tuple)}
 
     if joint_fts and prefix == ACTION:
@@ -86,11 +89,19 @@ def hw_to_dataset_features(
         }
 
     for key, shape in cam_fts.items():
-        features[f"{prefix}.images.{key}"] = {
-            "dtype": "video" if use_video else "image",
-            "shape": shape,
-            "names": ["height", "width", "channels"],
-        }
+        dtype = "video" if use_video else "image"
+        if len(shape) == 3 and shape[2] in (1, 3):
+            features[f"{prefix}.images.{key}"] = {
+                "dtype": dtype,
+                "shape": shape,
+                "names": ["height", "width", "channels"],
+                "info": {"is_depth_map": shape[2] == 1},
+            }
+        else:
+            raise ValueError(
+                f"Camera feature '{key}' has shape {shape}. "
+                f"Expected a 3-tuple (H, W, C), e.g. (480, 640, 3) for RGB or (480, 640, 1) for depth."
+            )
 
     _validate_feature_names(features)
     return features
@@ -149,11 +160,11 @@ def dataset_to_policy_features(features: dict[str, dict]) -> dict[str, PolicyFea
             type = FeatureType.VISUAL
             if len(shape) != 3:
                 raise ValueError(f"Number of dimensions of {key} != 3 (shape={shape})")
-
-            names = ft["names"]
-            # Backward compatibility for "channel" which is an error introduced in LeRobotDataset v2.0 for ported datasets.
-            if names[2] in ["channel", "channels"]:  # (h, w, c) -> (c, h, w)
-                shape = (shape[2], shape[0], shape[1])
+            else:
+                names = ft["names"]
+                # Backward compatibility for "channel" which is an error introduced in LeRobotDataset v2.0 for ported datasets.
+                if names[2] in ["channel", "channels"]:  # (h, w, c) -> (c, h, w)
+                    shape = (shape[2], shape[0], shape[1])
         elif key == OBS_ENV_STATE:
             type = FeatureType.ENV
         elif key.startswith(OBS_STR):

src/lerobot/utils/import_utils.py

@@ -129,6 +129,7 @@ _placo_available = is_package_available("placo")
 _hidapi_available = is_package_available("hidapi", import_name="hid")
 
 # Data / serialization
+_datasets_available = is_package_available("datasets")
 _pandas_available = is_package_available("pandas")
 _faker_available = is_package_available("faker")
 

src/lerobot/utils/visualization_utils.py

@@ -107,7 +107,10 @@ def log_rerun_data(
                     for i, vi in enumerate(arr):
                         rr.log(f"{key}_{i}", rr.Scalars(float(vi)))
                 else:
-                    img_entity = rr.Image(arr).compress() if compress_images else rr.Image(arr)
+                    if arr.shape[-1] == 1:
+                        img_entity = rr.DepthImage(arr, colormap=rr.components.Colormap.Viridis)
+                    else:
+                        img_entity = rr.Image(arr).compress() if compress_images else rr.Image(arr)
                     rr.log(key, entity=img_entity, static=True)
 
     if action:

tests/annotations/test_frames.py

@@ -47,6 +47,7 @@ class _FakeMeta:
     def __init__(self, video_keys: list[str], image_keys: list[str], video_path: Path | None = None) -> None:
         self.video_keys = video_keys
         self.camera_keys = [*video_keys, *image_keys]
+        self.depth_keys = []
         self._video_path = video_path
         self.episodes = {0: {f"videos/{key}/from_timestamp": 0.0 for key in video_keys}}
 
@@ -208,14 +209,14 @@ def test_episode_clip_path_trims_via_reencode_video(tmp_path: Path, monkeypatch)
     def fake_reencode(
         input_video_path,
         output_video_path,
-        camera_encoder=None,
+        video_encoder=None,
         overwrite=False,
         start_time_s=None,
         end_time_s=None,
     ):
         captured.update(
             src=Path(input_video_path),
-            encoder=camera_encoder,
+            encoder=video_encoder,
             start_time_s=start_time_s,
             end_time_s=end_time_s,
         )

tests/datasets/test_aggregate.py

@@ -29,7 +29,10 @@ from lerobot.configs import VIDEO_ENCODER_INFO_KEYS
 from lerobot.datasets.aggregate import aggregate_datasets
 from lerobot.datasets.feature_utils import features_equal_for_merge
 from lerobot.datasets.lerobot_dataset import LeRobotDataset
-from tests.fixtures.constants import DUMMY_REPO_ID
+from tests.fixtures.constants import (
+    DUMMY_CAMERA_FEATURES_WITH_DEPTH,
+    DUMMY_REPO_ID,
+)
 
 
 def assert_data_shards_one_row_group_per_episode(root):
@@ -211,6 +214,26 @@ def assert_dataset_iteration_works(aggr_ds):
         pass
 
 
+def assert_depth_keys_preserved(aggr_ds, ds_0, ds_1):
+    """Test that depth keys are correctly preserved after aggregation.
+
+    Ensures that the ``is_depth_map`` marker on visual features survives
+    aggregation, so that downstream consumers (e.g. the dataset reader's
+    depth decoding path) keep working on the merged dataset.
+    """
+    expected_depth_keys = set(ds_0.meta.depth_keys)
+    assert expected_depth_keys == set(ds_1.meta.depth_keys), (
+        "Source datasets disagree on depth_keys; test setup is inconsistent"
+    )
+    actual_depth_keys = set(aggr_ds.meta.depth_keys)
+    assert actual_depth_keys == expected_depth_keys, (
+        f"Expected depth_keys {expected_depth_keys}, got {actual_depth_keys}"
+    )
+    for key in expected_depth_keys:
+        info = aggr_ds.meta.info.features[key].get("info") or {}
+        assert info.get("is_depth_map") is True, f"Depth marker lost on feature {key!r} after aggregation"
+
+
 def assert_video_timestamps_within_bounds(aggr_ds):
     """Test that all video timestamps are within valid bounds for their respective video files.
 
@@ -260,7 +283,11 @@ def assert_video_timestamps_within_bounds(aggr_ds):
 
 
 def test_aggregate_datasets(tmp_path, lerobot_dataset_factory):
-    """Test basic aggregation functionality with standard parameters."""
+    """Test basic aggregation functionality with standard parameters.
+
+    Source datasets include both RGB and depth video features so the same
+    aggregation flow is exercised on the ``is_depth_map`` branch.
+    """
     ds_0_num_frames = 400
     ds_1_num_frames = 800
     ds_0_num_episodes = 10
@@ -272,14 +299,21 @@ def test_aggregate_datasets(tmp_path, lerobot_dataset_factory):
         repo_id=f"{DUMMY_REPO_ID}_0",
         total_episodes=ds_0_num_episodes,
         total_frames=ds_0_num_frames,
+        camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH,
     )
     ds_1 = lerobot_dataset_factory(
         root=tmp_path / "test_1",
         repo_id=f"{DUMMY_REPO_ID}_1",
         total_episodes=ds_1_num_episodes,
         total_frames=ds_1_num_frames,
+        camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH,
     )
 
+    # Confirm depth was actually wired into the source datasets so the
+    # rest of the assertions exercise the depth aggregation path.
+    assert len(ds_0.meta.depth_keys) > 0, "ds_0 should expose at least one depth key"
+    assert len(ds_1.meta.depth_keys) > 0, "ds_1 should expose at least one depth key"
+
     aggregate_datasets(
         repo_ids=[ds_0.repo_id, ds_1.repo_id],
         roots=[ds_0.root, ds_1.root],
@@ -306,6 +340,7 @@ def test_aggregate_datasets(tmp_path, lerobot_dataset_factory):
     assert_episode_indices_updated_correctly(aggr_ds, ds_0, ds_1)
     assert_video_frames_integrity(aggr_ds, ds_0, ds_1)
     assert_video_timestamps_within_bounds(aggr_ds)
+    assert_depth_keys_preserved(aggr_ds, ds_0, ds_1)
     assert_dataset_iteration_works(aggr_ds)
 
 
@@ -423,7 +458,11 @@ def test_aggregate_incomplete_video_encoder_info_warns_and_nuls_encoders(
 
 
 def test_aggregate_with_low_threshold(tmp_path, lerobot_dataset_factory):
-    """Test aggregation with small file size limits to force file rotation/sharding."""
+    """Test aggregation with small file size limits to force file rotation/sharding.
+
+    Depth video features are included to verify that file rotation/concat
+    correctly handles depth-marked features alongside regular RGB ones.
+    """
     ds_0_num_episodes = ds_1_num_episodes = 10
     ds_0_num_frames = ds_1_num_frames = 400
 
@@ -432,14 +471,19 @@ def test_aggregate_with_low_threshold(tmp_path, lerobot_dataset_factory):
         repo_id=f"{DUMMY_REPO_ID}_small_0",
         total_episodes=ds_0_num_episodes,
         total_frames=ds_0_num_frames,
+        camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH,
     )
     ds_1 = lerobot_dataset_factory(
         root=tmp_path / "small_1",
         repo_id=f"{DUMMY_REPO_ID}_small_1",
         total_episodes=ds_1_num_episodes,
         total_frames=ds_1_num_frames,
+        camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH,
     )
 
+    assert len(ds_0.meta.depth_keys) > 0, "ds_0 should expose at least one depth key"
+    assert len(ds_1.meta.depth_keys) > 0, "ds_1 should expose at least one depth key"
+
     # Use the new configurable parameters to force file rotation
     aggregate_datasets(
         repo_ids=[ds_0.repo_id, ds_1.repo_id],
@@ -470,6 +514,7 @@ def test_aggregate_with_low_threshold(tmp_path, lerobot_dataset_factory):
     assert_episode_indices_updated_correctly(aggr_ds, ds_0, ds_1)
     assert_video_frames_integrity(aggr_ds, ds_0, ds_1)
     assert_video_timestamps_within_bounds(aggr_ds)
+    assert_depth_keys_preserved(aggr_ds, ds_0, ds_1)
     assert_dataset_iteration_works(aggr_ds)
 
     # Check that multiple files were actually created due to small size limits
@@ -489,7 +534,8 @@ def test_video_timestamps_regression(tmp_path, lerobot_dataset_factory):
     """Regression test for video timestamp bug when merging datasets.
 
     This test specifically checks that video timestamps are correctly calculated
-    and accumulated when merging multiple datasets.
+    and accumulated when merging multiple datasets. Depth video features are
+    included so depth timestamps are also covered by the regression.
     """
     datasets = []
     for i in range(3):
@@ -498,9 +544,13 @@ def test_video_timestamps_regression(tmp_path, lerobot_dataset_factory):
             repo_id=f"{DUMMY_REPO_ID}_regression_{i}",
             total_episodes=2,
             total_frames=100,
+            camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH,
         )
         datasets.append(ds)
 
+    for i, ds in enumerate(datasets):
+        assert len(ds.meta.depth_keys) > 0, f"Dataset {i} should expose at least one depth key"
+
     aggregate_datasets(
         repo_ids=[ds.repo_id for ds in datasets],
         roots=[ds.root for ds in datasets],
@@ -517,12 +567,21 @@ def test_video_timestamps_regression(tmp_path, lerobot_dataset_factory):
         aggr_ds = LeRobotDataset(f"{DUMMY_REPO_ID}_regression_aggr", root=tmp_path / "regression_aggr")
 
     assert_video_timestamps_within_bounds(aggr_ds)
+    # Depth keys must survive the merge for the regression to cover the
+    # ``is_depth_map`` decoding branch.
+    assert set(aggr_ds.meta.depth_keys) == set(datasets[0].meta.depth_keys)
 
+    depth_keys = set(aggr_ds.meta.depth_keys)
     for i in range(len(aggr_ds)):
         item = aggr_ds[i]
         for key in aggr_ds.meta.video_keys:
             assert key in item, f"Video key {key} missing from item {i}"
-            assert item[key].shape[0] == 3, f"Expected 3 channels for video key {key}"
+            # Depth frames are single-channel (1, H, W) after dequantization;
+            # standard RGB frames keep the 3-channel layout.
+            expected_channels = 1 if key in depth_keys else 3
+            assert item[key].shape[0] == expected_channels, (
+                f"Expected {expected_channels} channels for video key {key}, got {item[key].shape}"
+            )
 
 
 def assert_image_schema_preserved(aggr_ds):
@@ -639,25 +698,31 @@ def test_aggregate_image_datasets(tmp_path, lerobot_dataset_factory):
     ds_0_num_episodes = 2
     ds_1_num_episodes = 3
 
-    # Create two image-based datasets (use_videos=False)
+    # Create two image-based datasets (use_videos=False) with a mix of RGB
+    # and depth-marked cameras so the depth path is exercised in image mode.
     ds_0 = lerobot_dataset_factory(
         root=tmp_path / "image_0",
         repo_id=f"{DUMMY_REPO_ID}_image_0",
         total_episodes=ds_0_num_episodes,
         total_frames=ds_0_num_frames,
-        use_videos=False,  # Image-based dataset
+        use_videos=False,
+        camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH,
     )
     ds_1 = lerobot_dataset_factory(
         root=tmp_path / "image_1",
         repo_id=f"{DUMMY_REPO_ID}_image_1",
         total_episodes=ds_1_num_episodes,
         total_frames=ds_1_num_frames,
-        use_videos=False,  # Image-based dataset
+        use_videos=False,
+        camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH,
     )
 
     # Verify source datasets have image keys
     assert len(ds_0.meta.image_keys) > 0, "ds_0 should have image keys"
     assert len(ds_1.meta.image_keys) > 0, "ds_1 should have image keys"
+    # And that the depth marker actually made it onto an image feature.
+    assert len(ds_0.meta.depth_keys) > 0, "ds_0 should expose at least one depth key"
+    assert len(ds_1.meta.depth_keys) > 0, "ds_1 should expose at least one depth key"
 
     # Aggregate the datasets
     aggregate_datasets(
@@ -692,6 +757,7 @@ def test_aggregate_image_datasets(tmp_path, lerobot_dataset_factory):
     # Image-specific assertions
     assert_image_schema_preserved(aggr_ds)
     assert_image_frames_integrity(aggr_ds, ds_0, ds_1)
+    assert_depth_keys_preserved(aggr_ds, ds_0, ds_1)
 
     # Verify images can be accessed and have correct shape
     sample_item = aggr_ds[0]

tests/datasets/test_compute_stats.py

@@ -35,7 +35,11 @@ from lerobot.utils.constants import OBS_IMAGE, OBS_STATE
 
 
 def mock_load_image_as_numpy(path, dtype, channel_first):
-    return np.ones((3, 32, 32), dtype=dtype) if channel_first else np.ones((32, 32, 3), dtype=dtype)
+    is_depth = "depth" in str(path)
+    channels = 1 if is_depth else 3
+    out_dtype = np.uint16 if is_depth else dtype
+    arr = np.arange(channels * 32 * 32, dtype=out_dtype).reshape(channels, 32, 32)
+    return arr if channel_first else arr.transpose(1, 2, 0)
 
 
 @pytest.fixture
@@ -168,22 +172,33 @@ def test_get_feature_stats_single_value():
 
 
 def test_compute_episode_stats():
+    depth_key = "observation.images.depth"
     episode_data = {
         OBS_IMAGE: [f"image_{i}.jpg" for i in range(100)],
+        depth_key: [f"depth_{i}.tiff" for i in range(100)],
         OBS_STATE: np.random.rand(100, 10),
     }
     features = {
         OBS_IMAGE: {"dtype": "image"},
+        depth_key: {"dtype": "image", "info": {"is_depth_map": True}},
         OBS_STATE: {"dtype": "numeric"},
     }
 
     with patch("lerobot.datasets.compute_stats.load_image_as_numpy", side_effect=mock_load_image_as_numpy):
         stats = compute_episode_stats(episode_data, features)
 
-    assert OBS_IMAGE in stats and OBS_STATE in stats
+    assert OBS_IMAGE in stats and depth_key in stats and OBS_STATE in stats
     assert stats[OBS_IMAGE]["count"].item() == 100
+    assert stats[depth_key]["count"].item() == 100
     assert stats[OBS_STATE]["count"].item() == 100
     assert stats[OBS_IMAGE]["mean"].shape == (3, 1, 1)
+    assert stats[depth_key]["mean"].shape == (1, 1, 1)
+    # Depth keeps raw values: max far exceeds 255, proving no /255 and no uint8 downcast.
+    assert stats[depth_key]["min"].item() == 0.0
+    assert stats[depth_key]["max"].item() == 1023.0
+    # RGB is normalized to [0, 1].
+    np.testing.assert_allclose(stats[OBS_IMAGE]["min"], 0.0)
+    np.testing.assert_allclose(stats[OBS_IMAGE]["max"], 1.0)
 
 
 def test_assert_type_and_shape_valid():
@@ -618,25 +633,31 @@ def test_compute_episode_stats_with_custom_quantiles():
 def test_compute_episode_stats_with_image_data():
     """Test quantile computation with image features."""
     image_paths = [f"image_{i}.jpg" for i in range(50)]
+    depth_paths = [f"depth_{i}.tiff" for i in range(50)]
     episode_data = {
         "observation.image": image_paths,
+        "observation.images.depth": depth_paths,
         "action": np.random.normal(0, 1, (50, 5)),
     }
     features = {
         "observation.image": {"dtype": "image"},
+        "observation.images.depth": {"dtype": "image", "info": {"is_depth_map": True}},
         "action": {"dtype": "float32", "shape": (5,)},
     }
 
     with patch("lerobot.datasets.compute_stats.load_image_as_numpy", side_effect=mock_load_image_as_numpy):
         stats = compute_episode_stats(episode_data, features)
 
-    # Image quantiles should be normalized and have correct shape
-    assert "q01" in stats["observation.image"]
-    assert "q50" in stats["observation.image"]
-    assert "q99" in stats["observation.image"]
-    assert stats["observation.image"]["q01"].shape == (3, 1, 1)
-    assert stats["observation.image"]["q50"].shape == (3, 1, 1)
-    assert stats["observation.image"]["q99"].shape == (3, 1, 1)
+    # RGB image quantiles should be normalized and per-channel.
+    for q in ("q01", "q50", "q99"):
+        assert stats["observation.image"][q].shape == (3, 1, 1)
+
+    # Depth quantiles are single-channel and kept in raw (un-normalized) units.
+    for q in ("q01", "q50", "q99"):
+        assert stats["observation.images.depth"][q].shape == (1, 1, 1)
+    # Depth max stays in raw units (not /255, not uint8-capped); RGB is normalized.
+    assert stats["observation.images.depth"]["max"].item() == 1023.0
+    np.testing.assert_allclose(stats["observation.image"]["max"], 1.0)
 
     # Action quantiles should have correct shape
     assert stats["action"]["q01"].shape == (5,)

tests/datasets/test_dataset_metadata.py

@@ -59,11 +59,13 @@ def _make_dummy_stats(features: dict) -> dict:
     stats = {}
     for key, ft in features.items():
         if ft["dtype"] in ("image", "video"):
+            channels = ft["shape"][-1]
+            stat_shape = (channels, 1, 1)
             stats[key] = {
-                "max": np.ones((3, 1, 1), dtype=np.float32),
-                "mean": np.full((3, 1, 1), 0.5, dtype=np.float32),
-                "min": np.zeros((3, 1, 1), dtype=np.float32),
-                "std": np.full((3, 1, 1), 0.25, dtype=np.float32),
+                "max": np.ones(stat_shape, dtype=np.float32),
+                "mean": np.full(stat_shape, 0.5, dtype=np.float32),
+                "min": np.zeros(stat_shape, dtype=np.float32),
+                "std": np.full(stat_shape, 0.25, dtype=np.float32),
                 "count": np.array([5]),
             }
         elif ft["dtype"] in ("float32", "float64", "int64"):
@@ -142,6 +144,45 @@ def test_create_without_videos_has_no_video_path(tmp_path):
     assert meta.video_keys == []
 
 
+@pytest.mark.parametrize(
+    ("marker_field", "marker_key"),
+    [
+        ("info", "is_depth_map"),
+        ("info", "video.is_depth_map"),
+        ("video_info", "video.is_depth_map"),
+    ],
+    ids=["info.is_depth_map", "info.video.is_depth_map_legacy", "video_info.video.is_depth_map_legacy"],
+)
+def test_depth_keys_property_filters_by_marker(tmp_path, marker_field, marker_key):
+    """``depth_keys`` recognises the canonical and the two legacy marker variants."""
+    depth_feature = {
+        "dtype": "video",
+        "shape": (64, 96, 1),
+        "names": ["height", "width", "channels"],
+        marker_field: {marker_key: True},
+    }
+    features = {
+        **VIDEO_FEATURES,
+        "observation.images.laptop_depth": depth_feature,
+    }
+    meta = LeRobotDatasetMetadata.create(
+        repo_id="test/depth_keys",
+        fps=DEFAULT_FPS,
+        features=features,
+        root=tmp_path / f"depth_keys_{marker_field}_{marker_key.replace('.', '_')}",
+    )
+
+    assert set(meta.video_keys) == {"observation.images.laptop", "observation.images.laptop_depth"}
+    assert meta.depth_keys == ["observation.images.laptop_depth"]
+
+
+def test_depth_keys_empty_when_no_marker(tmp_path):
+    meta = LeRobotDatasetMetadata.create(
+        repo_id="test/no_depth", fps=DEFAULT_FPS, features=VIDEO_FEATURES, root=tmp_path / "no_depth"
+    )
+    assert meta.depth_keys == []
+
+
 def test_create_raises_on_existing_directory(tmp_path):
     """create() raises if root directory already exists."""
     root = tmp_path / "existing"

tests/datasets/test_dataset_tools.py

@@ -24,7 +24,7 @@ import torch
 pytest.importorskip("datasets", reason="datasets is required (install lerobot[dataset])")
 
 
-from lerobot.configs import VideoEncoderConfig
+from lerobot.configs import DepthEncoderConfig, RGBEncoderConfig
 from lerobot.datasets.dataset_tools import (
     add_features,
     convert_image_to_video_dataset,
@@ -37,7 +37,9 @@ from lerobot.datasets.dataset_tools import (
     split_dataset,
 )
 from lerobot.datasets.io_utils import load_info
-from tests.datasets.test_video_encoding import _add_frames, require_h264, require_libsvtav1
+from tests.datasets.test_video_encoding import require_h264, require_hevc, require_libsvtav1
+from tests.fixtures.constants import DUMMY_DEPTH_FEATURES, DUMMY_DEPTH_KEY
+from tests.fixtures.dataset_factories import add_frames
 
 
 @pytest.fixture
@@ -1251,7 +1253,7 @@ def test_convert_image_to_video_dataset(tmp_path):
             dataset=source_dataset,
             output_dir=output_dir,
             repo_id="lerobot/pusht_video",
-            camera_encoder=VideoEncoderConfig(
+            rgb_encoder=RGBEncoderConfig(
                 vcodec="libsvtav1",
                 pix_fmt="yuv420p",
                 g=2,
@@ -1332,9 +1334,131 @@ def test_convert_image_to_video_dataset_subset_episodes(tmp_path):
             shutil.rmtree(output_dir)
 
 
+@require_libsvtav1
+@require_hevc
+def test_convert_image_to_video_dataset_depth(tmp_path, empty_lerobot_dataset_factory):
+    """Depth image features convert to depth videos using the depth encoder.
+
+    Mirrors :func:`test_convert_image_to_video_dataset` but with a small local
+    image dataset that mixes an RGB camera with a depth camera, so the
+    ``depth_keys`` → ``depth_encoder`` routing and ``is_depth_map`` preservation
+    are exercised end-to-end.
+    """
+    features = {
+        "action": {"dtype": "float32", "shape": (2,), "names": ["a", "b"]},
+        "observation.images.cam": {
+            "dtype": "image",
+            "shape": (64, 96, 3),
+            "names": ["height", "width", "channels"],
+        },
+        "observation.images.depth": {
+            "dtype": "image",
+            "shape": (64, 96, 1),
+            "names": ["height", "width", "channels"],
+            "info": {"is_depth_map": True},
+        },
+    }
+    source_dataset = empty_lerobot_dataset_factory(
+        root=tmp_path / "img_ds",
+        features=features,
+        use_videos=False,
+    )
+
+    add_frames(source_dataset, num_frames=4)
+    source_dataset.save_episode()
+    source_dataset.finalize()
+
+    # Source is an image dataset with the depth marker on the depth camera.
+    assert len(source_dataset.meta.video_keys) == 0
+    assert "observation.images.depth" in source_dataset.meta.depth_keys
+
+    output_dir = tmp_path / "video_ds"
+    with (
+        patch("lerobot.datasets.dataset_metadata.get_safe_version") as mock_get_safe_version,
+        patch("lerobot.datasets.dataset_metadata.snapshot_download") as mock_snapshot_download,
+    ):
+        mock_get_safe_version.return_value = "v3.0"
+        mock_snapshot_download.return_value = str(output_dir)
+
+        # Use non-default quantization params so the persisted metadata must
+        # come from the depth encoder (not RGB encoder defaults).
+        depth_encoder = DepthEncoderConfig(
+            vcodec="hevc",
+            pix_fmt="gray12le",
+            g=2,
+            crf=30,
+            depth_min=0.05,
+            depth_max=8.0,
+            shift=2.0,
+            use_log=False,
+        )
+        video_dataset = convert_image_to_video_dataset(
+            dataset=source_dataset,
+            output_dir=output_dir,
+            repo_id="dummy/depth_video",
+            rgb_encoder=RGBEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
+            depth_encoder=depth_encoder,
+            num_workers=1,
+        )
+
+    # Both cameras are now videos, and the depth marker survived the conversion.
+    assert "observation.images.cam" in video_dataset.meta.video_keys
+    assert "observation.images.depth" in video_dataset.meta.video_keys
+    assert "observation.images.depth" in video_dataset.meta.depth_keys
+    assert "observation.images.cam" not in video_dataset.meta.depth_keys
+
+    depth_path = video_dataset.root / video_dataset.meta.get_video_file_path(0, "observation.images.depth")
+    assert depth_path.exists(), f"Depth video file should exist: {depth_path}"
+
+    # The persisted depth-video metadata must carry the depth quantization params
+    # from the depth encoder (so frames dequantize correctly on read), and the RGB
+    # camera must not be marked as a depth map.
+    persisted_info = load_info(video_dataset.root)
+    depth_info = persisted_info.features["observation.images.depth"]["info"]
+    assert depth_info["is_depth_map"] is True
+    assert DepthEncoderConfig.from_video_info(depth_info) == depth_encoder
+
+    cam_info = persisted_info.features["observation.images.cam"]["info"]
+    assert cam_info.get("is_depth_map") is False
+    assert "video.codec" in cam_info
+
+
 # ─── reencode_dataset ─────────────────────────────────────────────────
 
 
+@require_hevc
+def test_reencode_dataset_depth_uses_depth_encoder(tmp_path, empty_lerobot_dataset_factory):
+    """Depth videos are re-encoded with the depth encoder and keep their depth metadata.
+
+    Depth-focused companion to :func:`test_reencode_dataset_multi_key_multiprocessing`.
+    """
+    initial_cfg = DepthEncoderConfig(vcodec="hevc", pix_fmt="gray12le", g=2, crf=30)
+    dataset = empty_lerobot_dataset_factory(
+        root=tmp_path / "ds",
+        features=DUMMY_DEPTH_FEATURES,
+        use_videos=True,
+        depth_encoder=initial_cfg,
+    )
+
+    add_frames(dataset, num_frames=4)
+    dataset.save_episode()
+    dataset.finalize()
+
+    assert DUMMY_DEPTH_KEY in dataset.meta.depth_keys
+
+    target_cfg = DepthEncoderConfig(vcodec="hevc", pix_fmt="gray12le", g=6, crf=23)
+    result = reencode_dataset(dataset, depth_encoder=target_cfg, num_workers=0)
+
+    assert result is dataset
+
+    persisted_info = load_info(dataset.root)
+    depth_info = persisted_info.features[DUMMY_DEPTH_KEY].get("info", {})
+    # Re-encode applied the new codec parameters to the depth video ...
+    assert DepthEncoderConfig.from_video_info(depth_info) == target_cfg
+    # ... while preserving the depth marker.
+    assert depth_info["is_depth_map"] is True
+
+
 @require_libsvtav1
 @require_h264
 def test_reencode_dataset_multi_key_multiprocessing(
@@ -1342,29 +1466,29 @@ def test_reencode_dataset_multi_key_multiprocessing(
 ):
     """Re-encode a two-camera dataset with num_workers=2 and verify metadata refresh."""
     features = features_factory(use_videos=True)
-    initial_cfg = VideoEncoderConfig(vcodec="libsvtav1", g=2, crf=30, preset=12)
+    initial_cfg = RGBEncoderConfig(vcodec="libsvtav1", g=2, crf=30, preset=12)
     dataset = empty_lerobot_dataset_factory(
         root=tmp_path / "ds",
         features=features,
         use_videos=True,
-        camera_encoder=initial_cfg,
+        rgb_encoder=initial_cfg,
     )
 
-    _add_frames(dataset, num_frames=4)
+    add_frames(dataset, num_frames=4)
     dataset.save_episode()
-    _add_frames(dataset, num_frames=4)
+    add_frames(dataset, num_frames=4)
     dataset.save_episode()
     dataset.finalize()
 
     assert len(dataset.meta.video_keys) == 2
 
-    target_cfg = VideoEncoderConfig(vcodec="h264", g=6, crf=23, pix_fmt="yuv420p")
+    target_cfg = RGBEncoderConfig(vcodec="h264", g=6, crf=23, pix_fmt="yuv420p")
 
-    result = reencode_dataset(dataset, camera_encoder=target_cfg, num_workers=2)
+    result = reencode_dataset(dataset, rgb_encoder=target_cfg, num_workers=2)
 
     assert result is dataset
 
     persisted_info = load_info(dataset.root)
     for vk in dataset.meta.video_keys:
-        persisted_encoder = VideoEncoderConfig.from_video_info(persisted_info.features[vk].get("info", {}))
+        persisted_encoder = RGBEncoderConfig.from_video_info(persisted_info.features[vk].get("info", {}))
         assert persisted_encoder == target_cfg

tests/datasets/test_dataset_writer.py

@@ -53,8 +53,8 @@ def _make_frame(features: dict, task: str = "Dummy task") -> dict:
 # ── Existing encode_video_worker tests ───────────────────────────────
 
 
-def test_encode_video_worker_forwards_camera_encoder(tmp_path):
-    """_encode_video_worker forwards camera_encoder to encode_video_frames."""
+def test_encode_video_worker_forwards_video_encoder(tmp_path):
+    """_encode_video_worker forwards video_encoder to encode_video_frames."""
     video_key = "observation.images.laptop"
     fpath = DEFAULT_IMAGE_PATH.format(image_key=video_key, episode_index=0, frame_index=0)
     img_dir = tmp_path / Path(fpath).parent
@@ -74,16 +74,16 @@ def test_encode_video_worker_forwards_camera_encoder(tmp_path):
             0,
             tmp_path,
             fps=30,
-            camera_encoder=VideoEncoderConfig(vcodec="h264", preset=None),
+            video_encoder=VideoEncoderConfig(vcodec="h264", preset=None),
             encoder_threads=4,
         )
 
-    assert captured_kwargs["camera_encoder"].vcodec == "h264"
+    assert captured_kwargs["video_encoder"].vcodec == "h264"
     assert captured_kwargs["encoder_threads"] == 4
 
 
-def test_encode_video_worker_default_camera_encoder(tmp_path):
-    """_encode_video_worker passes None camera_encoder which encode_video_frames defaults."""
+def test_encode_video_worker_default_video_encoder(tmp_path):
+    """_encode_video_worker passes None video_encoder which encode_video_frames defaults."""
     video_key = "observation.images.laptop"
     fpath = DEFAULT_IMAGE_PATH.format(image_key=video_key, episode_index=0, frame_index=0)
     img_dir = tmp_path / Path(fpath).parent
@@ -100,7 +100,7 @@ def test_encode_video_worker_default_camera_encoder(tmp_path):
     with patch("lerobot.datasets.dataset_writer.encode_video_frames", side_effect=mock_encode):
         _encode_video_worker(video_key, 0, tmp_path, fps=30)
 
-    assert captured_kwargs["camera_encoder"] is None
+    assert captured_kwargs["video_encoder"] is None
     assert captured_kwargs["encoder_threads"] is None
 
 

tests/datasets/test_datasets.py

@@ -1534,6 +1534,10 @@ def test_valid_video_codecs_constant():
     assert "auto" in VALID_VIDEO_CODECS
     assert "h264_videotoolbox" in VALID_VIDEO_CODECS
     assert "h264_nvenc" in VALID_VIDEO_CODECS
+    assert "h264_vaapi" in VALID_VIDEO_CODECS
+    assert "h264_qsv" in VALID_VIDEO_CODECS
+    assert "hevc_videotoolbox" in VALID_VIDEO_CODECS
+    assert "hevc_nvenc" in VALID_VIDEO_CODECS
     assert len(VALID_VIDEO_CODECS) == 10
 
 

tests/datasets/test_depth.py

@@ -0,0 +1,247 @@
+"""Tests for the depth-integration feature.
+
+Covers:
+- ``depth_utils`` quantize/dequantize round-trips and backend agreement.
+- Image-writer support for single-channel depth.
+- Hardware-feature → depth flag routing.
+- Feature-to-file-format routing through the dataset writer.
+
+Depth metadata detection on ``LeRobotDatasetMetadata.depth_keys`` lives in
+``test_dataset_metadata.py``. Depth video encoding/decoding lives in
+``test_video_encoding.py``.
+"""
+
+from pathlib import Path
+
+import pytest
+
+pytest.importorskip("av", reason="av is required (install lerobot[dataset])")
+
+import av
+import numpy as np
+import PIL.Image
+import torch
+
+from lerobot.configs import DepthEncoderConfig
+from lerobot.configs.video import (
+    DEFAULT_DEPTH_MAX,
+    DEFAULT_DEPTH_MIN,
+    DEPTH_METER_UNIT,
+    DEPTH_MILLIMETER_UNIT,
+    DEPTH_QMAX,
+)
+from lerobot.datasets.depth_utils import dequantize_depth, quantize_depth
+from lerobot.datasets.image_writer import image_array_to_pil_image, write_image
+from tests.fixtures.constants import (
+    DEFAULT_FPS,
+    DUMMY_CAMERA_FEATURES,
+    DUMMY_CAMERA_FEATURES_WITH_DEPTH,
+    DUMMY_CHW,
+    DUMMY_DEPTH_CAMERA_FEATURES,
+    DUMMY_REPO_ID,
+)
+from tests.fixtures.dataset_factories import add_frames
+
+_, H, W = DUMMY_CHW
+
+
+def _depth_metres_ramp() -> np.ndarray:
+    """Linearly-spaced float32 depth in metres covering the default range."""
+    return np.linspace(DEFAULT_DEPTH_MIN, DEFAULT_DEPTH_MAX, H * W, dtype=np.float32).reshape(H, W)
+
+
+# ── 1. Quantize / dequantize round-trips ──────────────────────────────
+
+
+class TestQuantizeDequantize:
+    """Numerical contract of ``quantize_depth`` / ``dequantize_depth``."""
+
+    @pytest.mark.parametrize("use_log", [False, True])
+    @pytest.mark.parametrize("output_unit", [DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT])
+    @pytest.mark.parametrize("output_channel_last", [False, True])
+    def test_roundtrip(self, use_log, output_unit, output_channel_last):
+        """quantize → dequantize recovers depth; layout and unit are honored."""
+        depth = _depth_metres_ramp()
+        quantized = quantize_depth(depth, use_log=use_log, video_backend=None)
+        recovered = dequantize_depth(
+            quantized,
+            use_log=use_log,
+            output_unit=output_unit,
+            output_tensor=False,
+            output_channel_last=output_channel_last,
+        )
+
+        expected_shape = (H, W, 1) if output_channel_last else (1, H, W)
+        assert recovered.shape == expected_shape
+
+        recovered_m = recovered.astype(np.float32)
+        if output_unit == DEPTH_MILLIMETER_UNIT:
+            recovered_m = recovered_m / 1000.0
+        recovered_2d = recovered_m[..., 0] if output_channel_last else recovered_m[0]
+
+        if use_log:
+            # Log mode: tighter near-range error than far-range (the whole point).
+            near = depth < 1.0
+            far = depth > 8.0
+            err_near = np.abs(recovered_2d[near] - depth[near])
+            err_far = np.abs(recovered_2d[far] - depth[far])
+            assert err_near.mean() < err_far.mean()
+        else:
+            # Linear mode: bounded by quant step + 1 mm of unit-conversion rounding.
+            tol = (DEFAULT_DEPTH_MAX - DEFAULT_DEPTH_MIN) / DEPTH_QMAX + 1e-3
+            np.testing.assert_allclose(recovered_2d, depth, atol=tol)
+
+    @pytest.mark.parametrize("use_log", [False, True])
+    @pytest.mark.parametrize("output_unit", [DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT])
+    def test_numpy_torch_agree(self, use_log, output_unit):
+        """Batched torch path produces the same values as the numpy path."""
+        batch_size = 3
+        per_frame = np.linspace(0, DEPTH_QMAX, H * W, dtype=np.uint16).reshape(H, W)
+        batch_np = np.broadcast_to(per_frame[None, None, ...], (batch_size, 1, H, W)).copy()
+        batch_t = torch.from_numpy(batch_np.astype(np.int32))  # torch.uint16 support is patchy.
+
+        ref = dequantize_depth(batch_np, use_log=use_log, output_unit=output_unit, output_tensor=False)
+        out = dequantize_depth(batch_t, use_log=use_log, output_unit=output_unit, output_tensor=True)
+
+        assert isinstance(out, torch.Tensor)
+        assert out.shape == (batch_size, 1, H, W)
+        # ``m``: float32 noise (~10 µm in log mode, after ``exp``) — still 200× below the ~2 mm quant step.
+        # ``mm`` + tensor stays in float32 (no uint16 round-trip), so allow 1 mm slop.
+        atol = 1e-5 if output_unit == DEPTH_METER_UNIT else 1.0
+        np.testing.assert_allclose(out.cpu().numpy().astype(np.float64), ref.astype(np.float64), atol=atol)
+
+    @pytest.mark.parametrize(
+        "input_shape,output_shape",
+        [
+            ((H, W), (1, H, W)),
+            ((1, H, W), (1, H, W)),
+            ((H, W, 1), (1, H, W)),
+            ((3, 1, H, W), (3, 1, H, W)),
+            ((3, H, W, 1), (3, 1, H, W)),
+        ],
+    )
+    def test_input_layouts_accepted(self, input_shape, output_shape):
+        """All documented input layouts decode to the channel-first default."""
+        quantized = np.full(input_shape, DEPTH_QMAX // 2, dtype=np.uint16)
+        out = dequantize_depth(quantized, output_unit=DEPTH_METER_UNIT, output_tensor=False)
+        assert out.shape == output_shape
+
+    def test_pyav_frame_roundtrip(self):
+        """quantize → av.VideoFrame → dequantize works."""
+        depth = _depth_metres_ramp()
+        frame = quantize_depth(depth, use_log=False, video_backend="pyav")
+        assert isinstance(frame, av.VideoFrame)
+
+        recovered = dequantize_depth(frame, use_log=False, output_unit=DEPTH_METER_UNIT, output_tensor=False)
+        assert recovered.shape == (1, H, W)
+        tol = (DEFAULT_DEPTH_MAX - DEFAULT_DEPTH_MIN) / DEPTH_QMAX + 1e-3
+        np.testing.assert_allclose(recovered[0], depth, atol=tol)
+
+    def test_invalid_log_params_raises(self):
+        with pytest.raises(ValueError, match=r"depth_min \+ shift must be positive"):
+            quantize_depth(_depth_metres_ramp(), depth_min=1.0, shift=-2.0, use_log=True, video_backend=None)
+
+
+# ── 2. Image writer depth support ─────────────────────────────────────
+
+
+class TestImageWriterDepth:
+    """``image_array_to_pil_image`` and ``write_image`` for depth maps."""
+
+    @pytest.mark.parametrize("dtype,expected_mode", [(np.uint16, "I;16"), (np.float32, "F")])
+    @pytest.mark.parametrize("shape", [(H, W), (H, W, 1), (1, H, W)])
+    def test_pil_depth_modes_and_squeeze(self, dtype, expected_mode, shape):
+        """Single-channel depth converts to PIL with the right mode and (W, H) size."""
+        arr = np.zeros(shape, dtype=dtype)
+        img = image_array_to_pil_image(arr)
+        assert img.mode == expected_mode
+        assert img.size == (W, H)
+
+    def test_write_image_tiff_roundtrip(self, tmp_path):
+        """uint16 depth round-trips through .tiff."""
+        arr = np.arange(H * W, dtype=np.uint16).reshape(H, W)
+        fpath = tmp_path / "depth.tiff"
+        write_image(arr, fpath)
+        with PIL.Image.open(fpath) as loaded:
+            recovered = np.array(loaded)
+        np.testing.assert_array_equal(recovered, arr)
+
+
+# ── 3. Hardware-feature → depth flag ──────────────────────────────────
+
+
+class TestHwToDatasetFeaturesDepth:
+    """``hw_to_dataset_features`` flags single-channel cameras as depth."""
+
+    @pytest.mark.parametrize("channels,is_depth", [(1, True), (3, False)])
+    def test_depth_marker_by_channels(self, channels, is_depth):
+        from lerobot.utils.feature_utils import hw_to_dataset_features
+
+        features = hw_to_dataset_features({"cam": (480, 640, channels)}, prefix="observation")
+        assert features["observation.images.cam"]["info"]["is_depth_map"] is is_depth
+
+    def test_invalid_channel_count_raises(self):
+        from lerobot.utils.feature_utils import hw_to_dataset_features
+
+        with pytest.raises(ValueError, match="Expected a 3-tuple"):
+            hw_to_dataset_features({"cam": (480, 640, 2)}, prefix="observation")
+
+
+# ── 4. Feature-to-file-format routing ────────────────────────────────
+
+
+# Keys derived from DUMMY_CAMERA_FEATURES_WITH_DEPTH; pick one RGB and the depth camera.
+RGB_KEY = next(iter(DUMMY_CAMERA_FEATURES))
+DEPTH_KEY = next(iter(DUMMY_DEPTH_CAMERA_FEATURES))
+
+
+class TestFeatureFileRouting:
+    """Depth vs RGB features route to the correct file format."""
+
+    NUM_FRAMES = 5
+
+    def test_image_mode_depth_tiff_rgb_png(self, tmp_path, features_factory):
+        """Without video encoding: depth → .tiff, RGB → .png."""
+        from lerobot.datasets.lerobot_dataset import LeRobotDataset
+
+        features = features_factory(camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH, use_videos=False)
+        dataset = LeRobotDataset.create(
+            repo_id=DUMMY_REPO_ID,
+            fps=DEFAULT_FPS,
+            features=features,
+            root=tmp_path / "ds",
+            use_videos=False,
+        )
+
+        add_frames(dataset, num_frames=self.NUM_FRAMES)
+
+        buf = dataset.writer.episode_buffer
+        assert all(Path(p).suffix == ".tiff" for p in buf[DEPTH_KEY])
+        assert all(Path(p).suffix == ".png" for p in buf[RGB_KEY])
+
+        dataset.save_episode()
+        dataset.finalize()
+
+    def test_video_mode_depth_uses_depth_encoder(self, tmp_path, features_factory):
+        """With streaming video encoding: depth → DepthEncoderConfig, RGB does not."""
+        from lerobot.datasets.lerobot_dataset import LeRobotDataset
+
+        features = features_factory(camera_features=DUMMY_CAMERA_FEATURES_WITH_DEPTH, use_videos=True)
+        dataset = LeRobotDataset.create(
+            repo_id=DUMMY_REPO_ID,
+            fps=DEFAULT_FPS,
+            features=features,
+            root=tmp_path / "ds",
+            use_videos=True,
+            streaming_encoding=True,
+        )
+
+        add_frames(dataset, num_frames=self.NUM_FRAMES)
+
+        encoder = dataset.writer._streaming_encoder
+        assert encoder is not None
+        assert isinstance(encoder._threads[DEPTH_KEY].video_encoder, DepthEncoderConfig)
+        assert not isinstance(encoder._threads[RGB_KEY].video_encoder, DepthEncoderConfig)
+
+        dataset.save_episode()
+        dataset.finalize()

tests/datasets/test_image_writer.py

@@ -94,7 +94,7 @@ def test_image_array_to_pil_image_pytorch_format(img_array_factory):
 
 def test_image_array_to_pil_image_single_channel(img_array_factory):
     img_array = img_array_factory(channels=1)
-    with pytest.raises(NotImplementedError):
+    with pytest.raises(ValueError, match="Unsupported single-channel image dtype"):
         image_array_to_pil_image(img_array)
 
 
@@ -344,7 +344,7 @@ def test_with_different_image_formats(tmp_path, img_array_factory):
     writer = AsyncImageWriter()
     try:
         image_array = img_array_factory()
-        formats = ["png", "jpeg", "bmp"]
+        formats = ["png", "tiff", "tif"]
         for fmt in formats:
             fpath = tmp_path / f"test_image.{fmt}"
             write_image(image_array, fpath)

tests/datasets/test_streaming_video_encoder.py

@@ -26,7 +26,7 @@ pytest.importorskip("av", reason="av is required (install lerobot[dataset])")
 
 import av  # noqa: E402
 
-from lerobot.configs import VideoEncoderConfig
+from lerobot.configs import RGBEncoderConfig
 from lerobot.datasets.pyav_utils import get_codec
 from lerobot.datasets.video_utils import (
     StreamingVideoEncoder,
@@ -57,13 +57,11 @@ class TestCameraEncoderThread:
         result_queue: queue.Queue = queue.Queue(maxsize=1)
         stop_event = threading.Event()
 
-        enc_cfg = VideoEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13)
+        enc_cfg = RGBEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13)
         encoder_thread = _CameraEncoderThread(
             video_path=video_path,
             fps=fps,
-            vcodec=enc_cfg.vcodec,
-            pix_fmt=enc_cfg.pix_fmt,
-            codec_options=enc_cfg.get_codec_options(as_strings=True),
+            video_encoder=enc_cfg,
             frame_queue=frame_queue,
             result_queue=result_queue,
             stop_event=stop_event,
@@ -108,13 +106,11 @@ class TestCameraEncoderThread:
         result_queue: queue.Queue = queue.Queue(maxsize=1)
         stop_event = threading.Event()
 
-        enc_cfg = VideoEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13)
+        enc_cfg = RGBEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13)
         encoder_thread = _CameraEncoderThread(
             video_path=video_path,
             fps=fps,
-            vcodec=enc_cfg.vcodec,
-            pix_fmt=enc_cfg.pix_fmt,
-            codec_options=enc_cfg.get_codec_options(as_strings=True),
+            video_encoder=enc_cfg,
             frame_queue=frame_queue,
             result_queue=result_queue,
             stop_event=stop_event,
@@ -142,13 +138,11 @@ class TestCameraEncoderThread:
         result_queue: queue.Queue = queue.Queue(maxsize=1)
         stop_event = threading.Event()
 
-        enc_cfg = VideoEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13)
+        enc_cfg = RGBEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13)
         encoder_thread = _CameraEncoderThread(
             video_path=video_path,
             fps=fps,
-            vcodec=enc_cfg.vcodec,
-            pix_fmt=enc_cfg.pix_fmt,
-            codec_options=enc_cfg.get_codec_options(as_strings=True),
+            video_encoder=enc_cfg,
             frame_queue=frame_queue,
             result_queue=result_queue,
             stop_event=stop_event,
@@ -171,15 +165,15 @@ class TestCameraEncoderThread:
 
 class TestStreamingVideoEncoder:
     def _make_encoder_config(self, **kwargs):
-        """Helper to build a VideoEncoderConfig."""
-        return VideoEncoderConfig(**kwargs)
+        """Helper to build an RGBEncoderConfig."""
+        return RGBEncoderConfig(**kwargs)
 
     def test_single_camera_episode(self, tmp_path):
         """Test encoding a single camera episode."""
         video_keys = [f"{OBS_IMAGES}.laptop"]
         encoder = StreamingVideoEncoder(
             fps=30,
-            camera_encoder=self._make_encoder_config(
+            rgb_encoder=self._make_encoder_config(
                 vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13
             ),
         )
@@ -211,7 +205,7 @@ class TestStreamingVideoEncoder:
         video_keys = [f"{OBS_IMAGES}.laptop", f"{OBS_IMAGES}.phone"]
         encoder = StreamingVideoEncoder(
             fps=30,
-            camera_encoder=self._make_encoder_config(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
+            rgb_encoder=self._make_encoder_config(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
         )
         encoder.start_episode(video_keys, tmp_path)
 
@@ -237,7 +231,7 @@ class TestStreamingVideoEncoder:
         video_keys = [f"{OBS_IMAGES}.cam"]
         encoder = StreamingVideoEncoder(
             fps=30,
-            camera_encoder=self._make_encoder_config(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
+            rgb_encoder=self._make_encoder_config(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
         )
 
         for ep in range(3):
@@ -263,7 +257,7 @@ class TestStreamingVideoEncoder:
         video_keys = [f"{OBS_IMAGES}.cam"]
         encoder = StreamingVideoEncoder(
             fps=30,
-            camera_encoder=self._make_encoder_config(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
+            rgb_encoder=self._make_encoder_config(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
         )
 
         encoder.start_episode(video_keys, tmp_path)
@@ -309,7 +303,7 @@ class TestStreamingVideoEncoder:
         video_keys = [f"{OBS_IMAGES}.cam"]
         encoder = StreamingVideoEncoder(
             fps=30,
-            camera_encoder=self._make_encoder_config(
+            rgb_encoder=self._make_encoder_config(
                 vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13
             ),
         )
@@ -346,7 +340,7 @@ class TestStreamingVideoEncoder:
         video_keys = [f"{OBS_IMAGES}.cam1", f"{OBS_IMAGES}.cam2"]
         encoder = StreamingVideoEncoder(
             fps=30,
-            camera_encoder=self._make_encoder_config(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
+            rgb_encoder=self._make_encoder_config(vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30),
         )
         encoder.start_episode(video_keys, tmp_path)
 
@@ -375,7 +369,7 @@ class TestStreamingVideoEncoder:
     def test_encoder_threads_passed_to_thread(self, tmp_path):
         """Test that encoder_threads is stored and passed through to encoder threads."""
         video_keys = [f"{OBS_IMAGES}.cam"]
-        cfg = VideoEncoderConfig(
+        cfg = RGBEncoderConfig(
             vcodec="libsvtav1",
             pix_fmt="yuv420p",
             g=2,
@@ -383,7 +377,7 @@ class TestStreamingVideoEncoder:
         )
         encoder = StreamingVideoEncoder(
             fps=30,
-            camera_encoder=cfg,
+            rgb_encoder=cfg,
             encoder_threads=2,
         )
         assert encoder._encoder_threads == 2
@@ -391,7 +385,8 @@ class TestStreamingVideoEncoder:
 
         # Verify codec options include thread tuning for libsvtav1 (lp=…)
         thread = encoder._threads[f"{OBS_IMAGES}.cam"]
-        assert "svtav1-params" in thread.codec_options or "threads" in thread.codec_options
+        codec_opts = thread.video_encoder.get_codec_options(encoder_threads=thread.encoder_threads)
+        assert "svtav1-params" in codec_opts or "threads" in codec_opts
 
         # Feed some frames and finish to ensure it works end-to-end
         num_frames = 10
@@ -422,7 +417,7 @@ class TestStreamingVideoEncoder:
         video_keys = [f"{OBS_IMAGES}.cam"]
         encoder = StreamingVideoEncoder(
             fps=30,
-            camera_encoder=self._make_encoder_config(
+            rgb_encoder=self._make_encoder_config(
                 vcodec="libsvtav1", pix_fmt="yuv420p", g=2, crf=30, preset=13
             ),
             queue_maxsize=1,

tests/datasets/test_video_encoding.py

@@ -26,7 +26,7 @@ pytest.importorskip("av", reason="av is required (install lerobot[dataset])")
 
 import av  # noqa: E402
 
-from lerobot.configs import VALID_VIDEO_CODECS, VideoEncoderConfig
+from lerobot.configs import VALID_VIDEO_CODECS, DepthEncoderConfig, RGBEncoderConfig, VideoEncoderConfig
 from lerobot.datasets.image_writer import write_image
 from lerobot.datasets.lerobot_dataset import LeRobotDataset
 from lerobot.datasets.pyav_utils import get_codec
@@ -37,7 +37,15 @@ from lerobot.datasets.video_utils import (
     get_video_info,
     reencode_video,
 )
-from tests.fixtures.constants import DUMMY_VIDEO_INFO
+from tests.fixtures.constants import (
+    DUMMY_DEPTH_FEATURES,
+    DUMMY_DEPTH_KEY,
+    DUMMY_DEPTH_VIDEO_INFO_FULL,
+    DUMMY_VIDEO_FEATURES,
+    DUMMY_VIDEO_INFO,
+    DUMMY_VIDEO_KEY,
+)
+from tests.fixtures.dataset_factories import add_frames
 
 
 # Per-codec skip markers — validation tests only fire when the codec is available
@@ -48,19 +56,74 @@ def _require_encoder(vcodec: str) -> pytest.MarkDecorator:
 
 require_libsvtav1 = _require_encoder("libsvtav1")
 require_h264 = _require_encoder("h264")
+require_hevc = _require_encoder("hevc")
 require_videotoolbox = _require_encoder("h264_videotoolbox")
 require_nvenc = _require_encoder("h264_nvenc")
 require_vaapi = _require_encoder("h264_vaapi")
 require_qsv = _require_encoder("h264_qsv")
 
 
-# ─── VideoEncoderConfig / codec options ──────────────────────────────
+TEST_ARTIFACTS_DIR = Path(__file__).parent.parent / "artifacts" / "encoded_videos"
+
+
+def _write_color_frames(imgs_dir: Path, num_frames: int = 4, height: int = 64, width: int = 96) -> None:
+    imgs_dir.mkdir(parents=True, exist_ok=True)
+    for i in range(num_frames):
+        arr = np.random.randint(0, 256, (height, width, 3), dtype=np.uint8)
+        write_image(arr, imgs_dir / f"frame-{i:06d}.png")
+
+
+def _write_depth_frames(imgs_dir: Path, num_frames: int = 4, height: int = 64, width: int = 96) -> None:
+    """Write synthetic uint16 depth TIFFs (millimetres) for depth encoder tests.
+
+    Uses a smooth linear ramp + per-frame offset (not white noise) so HEVC Main 12
+    on ``gray12le`` compresses well. Values span ~100 mm to 10 m, covering most
+    of the default ``[DEPTH_MIN, DEPTH_MAX]`` metres range after
+    ``quantize_depth(input_unit="auto"="mm")``.
+    """
+    imgs_dir.mkdir(parents=True, exist_ok=True)
+    base = np.linspace(100.0, 10_000.0, height * width, dtype=np.float32).reshape(height, width)
+    for i in range(num_frames):
+        arr = (base + 50.0 * i).clip(0, 65535).astype(np.uint16)
+        write_image(arr, imgs_dir / f"frame-{i:06d}.tiff")
+
+
+def _encode_video(
+    path: Path,
+    num_frames: int = 4,
+    fps: int = 30,
+    cfg: VideoEncoderConfig | None = None,
+    depth: bool = False,
+) -> Path:
+    """Write synthetic frames to a temp dir and encode them to ``path``.
+
+    ``depth=False`` writes uint8 RGB PNG noise and encodes with ``cfg``
+    (defaulting to the library default). ``depth=True`` writes synthetic uint16
+    depth TIFFs and encodes with ``cfg`` or a default :class:`DepthEncoderConfig`
+    (HEVC Main 12 / ``gray12le``).
+    """
+    imgs_dir = path.parent / f"imgs_{path.stem}"
+    if depth:
+        _write_depth_frames(imgs_dir, num_frames=num_frames)
+        cfg = cfg or DepthEncoderConfig()
+    else:
+        _write_color_frames(imgs_dir, num_frames=num_frames)
+    encode_video_frames(imgs_dir, path, fps=fps, video_encoder=cfg, overwrite=True)
+    return path
+
+
+def _read_feature_info(dataset: LeRobotDataset, key: str = DUMMY_VIDEO_KEY) -> dict:
+    info = json.loads((dataset.root / INFO_PATH).read_text())
+    return info["features"][key]["info"]
+
+
+# ─── RGBEncoderConfig / codec options ──────────────────────────────
 
 
 class TestCodecOptions:
     @require_libsvtav1
     def test_libsvtav1_defaults(self):
-        cfg = VideoEncoderConfig()
+        cfg = RGBEncoderConfig()
         opts = cfg.get_codec_options()
         assert opts["g"] == 2
         assert opts["crf"] == 30
@@ -68,12 +131,12 @@ class TestCodecOptions:
 
     @require_libsvtav1
     def test_libsvtav1_custom_preset(self):
-        cfg = VideoEncoderConfig(preset=8)
+        cfg = RGBEncoderConfig(preset=8)
         assert cfg.get_codec_options()["preset"] == 8
 
     @require_h264
     def test_h264_options(self):
-        cfg = VideoEncoderConfig(vcodec="h264", g=10, crf=23, preset=None)
+        cfg = RGBEncoderConfig(vcodec="h264", g=10, crf=23, preset=None)
         opts = cfg.get_codec_options()
         assert opts["g"] == 10
         assert opts["crf"] == 23
@@ -81,120 +144,120 @@ class TestCodecOptions:
 
     @require_videotoolbox
     def test_videotoolbox_options(self):
-        cfg = VideoEncoderConfig(vcodec="h264_videotoolbox", g=2, crf=30, preset=None)
+        cfg = RGBEncoderConfig(vcodec="h264_videotoolbox", g=2, crf=30, preset=None)
         opts = cfg.get_codec_options()
         assert opts["g"] == 2
         assert opts["q:v"] == 40
         assert "crf" not in opts
 
-    @_require_encoder("h264_nvenc")
+    @require_nvenc
     def test_nvenc_options(self):
-        cfg = VideoEncoderConfig(vcodec="h264_nvenc", g=2, crf=25, preset=None)
+        cfg = RGBEncoderConfig(vcodec="h264_nvenc", g=2, crf=25, preset=None)
         opts = cfg.get_codec_options()
         assert opts["rc"] == 0
         assert opts["qp"] == 25
         assert "crf" not in opts
         assert opts["g"] == 2
 
-    @_require_encoder("h264_vaapi")
+    @require_vaapi
     def test_vaapi_options(self):
-        cfg = VideoEncoderConfig(vcodec="h264_vaapi", crf=28, preset=None)
+        cfg = RGBEncoderConfig(vcodec="h264_vaapi", crf=28, preset=None)
         assert cfg.get_codec_options()["qp"] == 28
 
-    @_require_encoder("h264_qsv")
+    @require_qsv
     def test_qsv_options(self):
-        cfg = VideoEncoderConfig(vcodec="h264_qsv", crf=25, preset=None)
+        cfg = RGBEncoderConfig(vcodec="h264_qsv", crf=25, preset=None)
         assert cfg.get_codec_options()["global_quality"] == 25
 
     @require_h264
     def test_no_g_no_crf(self):
-        cfg = VideoEncoderConfig(vcodec="h264", g=None, crf=None, preset=None)
+        cfg = RGBEncoderConfig(vcodec="h264", g=None, crf=None, preset=None)
         opts = cfg.get_codec_options()
         assert "g" not in opts
         assert "crf" not in opts
 
     @require_libsvtav1
     def test_encoder_threads_libsvtav1(self):
-        cfg = VideoEncoderConfig(fast_decode=0)
+        cfg = RGBEncoderConfig(fast_decode=0)
         opts = cfg.get_codec_options(encoder_threads=4)
         assert "lp=4" in opts.get("svtav1-params", "")
 
     @require_h264
     def test_encoder_threads_h264(self):
-        cfg = VideoEncoderConfig(vcodec="h264", preset=None)
+        cfg = RGBEncoderConfig(vcodec="h264", preset=None)
         assert cfg.get_codec_options(encoder_threads=2)["threads"] == 2
 
     @require_libsvtav1
     def test_fast_decode_libsvtav1(self):
-        cfg = VideoEncoderConfig(fast_decode=1)
+        cfg = RGBEncoderConfig(fast_decode=1)
         opts = cfg.get_codec_options()
         assert "fast-decode=1" in opts.get("svtav1-params", "")
 
     @require_libsvtav1
     def test_libsvtav1_fast_decode_clamped_to_svt_range(self):
         """Out-of-range fast_decode is clamped to [0, 2] in svtav1-params (SVT-AV1 FastDecode)."""
-        cfg = VideoEncoderConfig(fast_decode=100)
+        cfg = RGBEncoderConfig(fast_decode=100)
         assert "fast-decode=2" in cfg.get_codec_options().get("svtav1-params", "")
-        cfg_neg = VideoEncoderConfig(fast_decode=-5)
+        cfg_neg = RGBEncoderConfig(fast_decode=-5)
         assert "fast-decode=0" in cfg_neg.get_codec_options().get("svtav1-params", "")
 
     @require_h264
     def test_fast_decode_h264(self):
-        cfg = VideoEncoderConfig(vcodec="h264", fast_decode=1, preset=None)
+        cfg = RGBEncoderConfig(vcodec="h264", fast_decode=1, preset=None)
         assert cfg.get_codec_options()["tune"] == "fastdecode"
 
     @require_libsvtav1
     def test_pix_fmt_unsupported_raises(self):
         """Passing an unsupported pix_fmt is a hard error."""
         with pytest.raises(ValueError, match="pix_fmt"):
-            VideoEncoderConfig(pix_fmt="yuv444p")  # libsvtav1 only supports yuv420p variants
+            RGBEncoderConfig(pix_fmt="yuv444p")  # libsvtav1 only supports yuv420p variants
 
     @require_libsvtav1
     @require_h264
     def test_preset_default_behaviour(self):
         """Empty constructor picks preset=12 (libsvtav1 path); other codecs stay None."""
-        assert VideoEncoderConfig().preset == 12
-        assert VideoEncoderConfig(vcodec="libsvtav1").preset == 12
-        assert VideoEncoderConfig(vcodec="h264").preset is None
-        assert VideoEncoderConfig(vcodec="h264", preset=None).preset is None
+        assert RGBEncoderConfig().preset == 12
+        assert RGBEncoderConfig(vcodec="libsvtav1").preset == 12
+        assert RGBEncoderConfig(vcodec="h264").preset is None
+        assert RGBEncoderConfig(vcodec="h264", preset=None).preset is None
 
     @require_h264
     def test_preset_string_on_h264(self):
         """h264 accepts string presets and forwards them to FFmpeg."""
-        cfg = VideoEncoderConfig(vcodec="h264", preset="slow")
+        cfg = RGBEncoderConfig(vcodec="h264", preset="slow")
         assert cfg.get_codec_options()["preset"] == "slow"
 
     @require_videotoolbox
     def test_preset_on_videotoolbox_not_set(self):
         """videotoolbox has no preset option at all."""
-        cfg = VideoEncoderConfig(vcodec="h264_videotoolbox", preset="slow")
+        cfg = RGBEncoderConfig(vcodec="h264_videotoolbox", preset="slow")
         assert "preset" not in cfg.get_codec_options()
 
     @require_libsvtav1
     def test_libsvtav1_preset_out_of_range_raises(self):
         """libsvtav1 preset must sit in [-2, 13] as exposed by PyAV."""
         with pytest.raises(ValueError, match="out of range"):
-            VideoEncoderConfig(vcodec="libsvtav1", preset=100)
+            RGBEncoderConfig(vcodec="libsvtav1", preset=100)
         with pytest.raises(ValueError, match="out of range"):
-            VideoEncoderConfig(vcodec="libsvtav1", preset=-3)
+            RGBEncoderConfig(vcodec="libsvtav1", preset=-3)
 
     @require_libsvtav1
     def test_libsvtav1_crf_out_of_range_raises(self):
         """libsvtav1 crf must sit in [0, 63]."""
         with pytest.raises(ValueError, match="crf.*out of range"):
-            VideoEncoderConfig(vcodec="libsvtav1", crf=64)
+            RGBEncoderConfig(vcodec="libsvtav1", crf=64)
 
     @require_libsvtav1
     def test_libsvtav1_crf_rejects_python_float(self):
         """libsvtav1 exposes ``crf`` as an INT AVOption; Python float must not pass validation."""
         with pytest.raises(ValueError, match="float values are not allowed"):
-            VideoEncoderConfig(vcodec="libsvtav1", crf=2.5)
+            RGBEncoderConfig(vcodec="libsvtav1", crf=2.5)
 
     @require_libsvtav1
     def test_libsvtav1_extra_crf_rejects_fractional_string(self):
         """INT options reject fractional values even when supplied only via ``extra_options``."""
         with pytest.raises(ValueError, match="float values are not allowed"):
-            VideoEncoderConfig(
+            RGBEncoderConfig(
                 vcodec="libsvtav1",
                 crf=None,
                 extra_options={"crf": "2.5"},
@@ -203,7 +266,7 @@ class TestCodecOptions:
     @require_libsvtav1
     def test_libsvtav1_extra_crf_rejects_float(self):
         with pytest.raises(ValueError, match="float values are not allowed"):
-            VideoEncoderConfig(
+            RGBEncoderConfig(
                 vcodec="libsvtav1",
                 crf=None,
                 extra_options={"crf": 2.5},
@@ -212,13 +275,13 @@ class TestCodecOptions:
     @require_h264
     def test_h264_crf_accepts_float_and_int(self):
         """x264 exposes crf as a FLOAT option, so both int and float are accepted."""
-        assert VideoEncoderConfig(vcodec="h264", crf=23).get_codec_options()["crf"] == 23
-        assert VideoEncoderConfig(vcodec="h264", crf=23.5).get_codec_options()["crf"] == 23.5
+        assert RGBEncoderConfig(vcodec="h264", crf=23).get_codec_options()["crf"] == 23
+        assert RGBEncoderConfig(vcodec="h264", crf=23.5).get_codec_options()["crf"] == 23.5
 
     @require_libsvtav1
     def test_validate_is_rerunnable(self):
         """After mutating a field, validate() re-checks and surfaces new issues."""
-        cfg = VideoEncoderConfig(vcodec="libsvtav1")
+        cfg = RGBEncoderConfig(vcodec="libsvtav1")
         cfg.preset = 100  # now out of range
         with pytest.raises(ValueError, match="out of range"):
             cfg.validate()
@@ -227,58 +290,58 @@ class TestCodecOptions:
 class TestExtraOptions:
     @require_libsvtav1
     def test_default_is_empty_dict(self):
-        cfg = VideoEncoderConfig()
+        cfg = RGBEncoderConfig()
         assert cfg.extra_options == {}
 
     @require_libsvtav1
     def test_unknown_key_passes_through(self):
         """Keys not published as AVOptions are forwarded to FFmpeg."""
-        cfg = VideoEncoderConfig(extra_options={"totally_made_up_option": "value"})
+        cfg = RGBEncoderConfig(extra_options={"totally_made_up_option": "value"})
         assert cfg.extra_options == {"totally_made_up_option": "value"}
 
     @require_libsvtav1
     def test_numeric_value_in_range_ok(self):
         """libsvtav1 exposes ``qp`` as INT in [0, 63]."""
-        cfg = VideoEncoderConfig(extra_options={"qp": 30})
+        cfg = RGBEncoderConfig(extra_options={"qp": 30})
         assert cfg.extra_options == {"qp": 30}
 
     @require_libsvtav1
     def test_numeric_out_of_range_raises(self):
         with pytest.raises(ValueError, match=r"qp=.*out of range"):
-            VideoEncoderConfig(extra_options={"qp": 999})
+            RGBEncoderConfig(extra_options={"qp": 999})
 
     @require_libsvtav1
     def test_numeric_string_accepted_in_range(self):
         """Numeric strings are accepted for numeric options (mirrors FFmpeg)."""
-        cfg = VideoEncoderConfig(extra_options={"qp": "18"})
+        cfg = RGBEncoderConfig(extra_options={"qp": "18"})
         assert cfg.extra_options == {"qp": "18"}
 
     @require_libsvtav1
     def test_numeric_string_out_of_range_raises(self):
         with pytest.raises(ValueError, match=r"qp=.*out of range"):
-            VideoEncoderConfig(extra_options={"qp": "999"})
+            RGBEncoderConfig(extra_options={"qp": "999"})
 
     @require_libsvtav1
     def test_non_numeric_string_on_numeric_option_raises(self):
         with pytest.raises(ValueError, match=r"qp=.*not numeric"):
-            VideoEncoderConfig(extra_options={"qp": "medium"})
+            RGBEncoderConfig(extra_options={"qp": "medium"})
 
     @require_libsvtav1
     def test_bool_on_numeric_option_raises(self):
         """``bool`` is explicitly rejected for numeric options."""
         with pytest.raises(ValueError, match=r"qp=.*not numeric"):
-            VideoEncoderConfig(extra_options={"qp": True})
+            RGBEncoderConfig(extra_options={"qp": True})
 
     @require_h264
     def test_string_option_passes_through_unchecked(self):
         """String-typed AVOptions are NOT enum-checked (too many accept freeform)."""
-        cfg = VideoEncoderConfig(vcodec="h264", preset=None, extra_options={"tune": "some-future-tune"})
+        cfg = RGBEncoderConfig(vcodec="h264", preset=None, extra_options={"tune": "some-future-tune"})
         assert cfg.extra_options == {"tune": "some-future-tune"}
 
     @require_libsvtav1
     def test_merged_into_codec_options_and_stringified(self):
         """Typed merge by default; ``as_strings=True`` matches FFmpeg option dict."""
-        cfg = VideoEncoderConfig(extra_options={"qp": 20})
+        cfg = RGBEncoderConfig(extra_options={"qp": 20})
         opts = cfg.get_codec_options()
         assert opts["qp"] == 20
         assert isinstance(opts["qp"], int)
@@ -287,25 +350,25 @@ class TestExtraOptions:
     @require_libsvtav1
     def test_structured_fields_win_on_collision(self):
         """A colliding extra_options key is discarded; the structured field wins."""
-        cfg = VideoEncoderConfig(crf=30, extra_options={"crf": 18})
+        cfg = RGBEncoderConfig(crf=30, extra_options={"crf": 18})
         assert cfg.get_codec_options()["crf"] == 30
 
 
 class TestEncoderDetection:
     @require_h264
     def test_explicit_codec_kept_when_available(self):
-        cfg = VideoEncoderConfig(vcodec="h264")
+        cfg = RGBEncoderConfig(vcodec="h264")
         assert cfg.vcodec == "h264"
 
     @require_videotoolbox
     def test_auto_picks_videotoolbox_when_available(self):
         """``h264_videotoolbox`` sits at the top of ``HW_VIDEO_CODECS`` so it wins when present."""
-        cfg = VideoEncoderConfig(vcodec="auto")
+        cfg = RGBEncoderConfig(vcodec="auto")
         assert cfg.vcodec == "h264_videotoolbox"
 
     def test_invalid_codec_raises(self):
         with pytest.raises(ValueError, match="Invalid vcodec"):
-            VideoEncoderConfig(vcodec="not_a_real_codec")
+            RGBEncoderConfig(vcodec="not_a_real_codec")
 
     def test_hw_encoder_names_listed_as_valid(self):
         assert "auto" in VALID_VIDEO_CODECS
@@ -313,59 +376,6 @@ class TestEncoderDetection:
         assert "h264_nvenc" in VALID_VIDEO_CODECS
 
 
-TEST_ARTIFACTS_DIR = Path(__file__).parent.parent / "artifacts" / "encoded_videos"
-
-# Default video feature set used by persistence tests.
-VIDEO_FEATURES = {
-    "observation.images.cam": {
-        "dtype": "video",
-        "shape": (64, 96, 3),
-        "names": ["height", "width", "channels"],
-    },
-    "action": {"dtype": "float32", "shape": (2,), "names": ["a", "b"]},
-}
-VIDEO_KEY = "observation.images.cam"
-
-
-def _write_frames(imgs_dir: Path, num_frames: int = 4, height: int = 64, width: int = 96) -> None:
-    imgs_dir.mkdir(parents=True, exist_ok=True)
-    for i in range(num_frames):
-        arr = np.random.randint(0, 256, (height, width, 3), dtype=np.uint8)
-        write_image(arr, imgs_dir / f"frame-{i:06d}.png")
-
-
-def _encode_video(
-    path: Path, num_frames: int = 4, fps: int = 30, cfg: VideoEncoderConfig | None = None
-) -> Path:
-    imgs_dir = path.parent / f"imgs_{path.stem}"
-    _write_frames(imgs_dir, num_frames=num_frames)
-    encode_video_frames(imgs_dir, path, fps=fps, camera_encoder=cfg, overwrite=True)
-    return path
-
-
-def _read_feature_info(dataset: LeRobotDataset) -> dict:
-    info = json.loads((dataset.root / INFO_PATH).read_text())
-    return info["features"][VIDEO_KEY]["info"]
-
-
-def _add_frames(dataset: LeRobotDataset, num_frames: int, video_keys: list[str] | None = None) -> None:
-    from lerobot.utils.constants import DEFAULT_FEATURES
-
-    if video_keys is None:
-        video_keys = dataset.meta.video_keys
-    for _ in range(num_frames):
-        frame: dict = {"task": "test"}
-        for key, ft in dataset.meta.features.items():
-            if key in DEFAULT_FEATURES:
-                continue
-            shape = ft["shape"]
-            if key in video_keys:
-                frame[key] = np.random.randint(0, 256, shape, dtype=np.uint8)
-            else:
-                frame[key] = np.zeros(shape, dtype=np.float32)
-        dataset.add_frame(frame)
-
-
 class TestGetVideoInfo:
     def test_returns_all_stream_fields(self):
         info = get_video_info(TEST_ARTIFACTS_DIR / "clip_4frames.mp4")
@@ -375,7 +385,7 @@ class TestGetVideoInfo:
         assert info["video.pix_fmt"] == "yuv420p"
         assert info["video.fps"] == 30
         assert info["video.channels"] == 3
-        assert info["video.is_depth_map"] is False
+        assert info["is_depth_map"] is False
         assert info["has_audio"] is False
         assert "video.g" not in info
         assert "video.crf" not in info
@@ -383,9 +393,9 @@ class TestGetVideoInfo:
 
     @require_libsvtav1
     def test_merges_encoder_config_as_video_prefixed_entries(self):
-        cfg = VideoEncoderConfig(vcodec="libsvtav1", g=2, crf=30, preset=12)
+        cfg = RGBEncoderConfig(vcodec="libsvtav1", g=2, crf=30, preset=12)
 
-        info = get_video_info(TEST_ARTIFACTS_DIR / "clip_4frames.mp4", camera_encoder=cfg)
+        info = get_video_info(TEST_ARTIFACTS_DIR / "clip_4frames.mp4", video_encoder=cfg)
 
         assert info["video.g"] == 2
         assert info["video.crf"] == 30
@@ -396,13 +406,18 @@ class TestGetVideoInfo:
 
     @require_libsvtav1
     def test_stream_derived_keys_take_precedence_over_config(self):
-        cfg = VideoEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p")
+        cfg = RGBEncoderConfig(vcodec="libsvtav1", pix_fmt="yuv420p")
 
-        info = get_video_info(TEST_ARTIFACTS_DIR / "clip_4frames.mp4", camera_encoder=cfg)
+        info = get_video_info(TEST_ARTIFACTS_DIR / "clip_4frames.mp4", video_encoder=cfg)
 
         assert info["video.codec"]  # populated from stream, not from config's vcodec
         assert info["video.pix_fmt"] == "yuv420p"
 
+    def test_depth_encoder_config_sets_is_depth_map_true(self):
+        """A ``DepthEncoderConfig`` causes ``get_video_info`` to mark the stream as depth."""
+        info = get_video_info(TEST_ARTIFACTS_DIR / "clip_4frames.mp4", video_encoder=DepthEncoderConfig())
+        assert info["is_depth_map"] is True
+
 
 class TestEncodeVideoFrames:
     @require_libsvtav1
@@ -434,7 +449,7 @@ class TestEncodeVideoFrames:
 
     def test_overwrite_false_skips_existing_file(self, tmp_path):
         imgs_dir = tmp_path / "imgs"
-        _write_frames(imgs_dir)
+        _write_color_frames(imgs_dir)
         video_path = tmp_path / "out.mp4"
         sentinel = b"pre-existing content"
         video_path.write_bytes(sentinel)
@@ -446,7 +461,7 @@ class TestEncodeVideoFrames:
     @require_libsvtav1
     def test_overwrite_true_replaces_existing_file(self, tmp_path):
         imgs_dir = tmp_path / "imgs"
-        _write_frames(imgs_dir)
+        _write_color_frames(imgs_dir)
         video_path = tmp_path / "out.mp4"
         video_path.write_bytes(b"stale content")
 
@@ -458,10 +473,10 @@ class TestEncodeVideoFrames:
     @require_libsvtav1
     def test_custom_encoder_config_fields_stored_in_info(self, tmp_path):
         """All stream-derived and encoder config fields are present after encoding."""
-        cfg = VideoEncoderConfig(vcodec="libsvtav1", g=4, crf=25, preset=10)
+        cfg = RGBEncoderConfig(vcodec="libsvtav1", g=4, crf=25, preset=10)
         video_path = _encode_video(tmp_path / "out.mp4", num_frames=4, fps=30, cfg=cfg)
 
-        info = get_video_info(video_path, camera_encoder=cfg)
+        info = get_video_info(video_path, video_encoder=cfg)
 
         # Stream-derived
         assert info["video.height"] == 64
@@ -470,7 +485,7 @@ class TestEncodeVideoFrames:
         assert info["video.codec"] == "av1"
         assert info["video.pix_fmt"] == "yuv420p"
         assert info["video.fps"] == 30
-        assert info["video.is_depth_map"] is False
+        assert info["is_depth_map"] is False
         assert info["has_audio"] is False
         # Encoder config
         assert info["video.g"] == 4
@@ -487,15 +502,15 @@ class TestReencodeVideo:
     def test_reencode_video(self, tmp_path):
         src = TEST_ARTIFACTS_DIR / "clip_4frames.mp4"
         out = tmp_path / "reencoded.mp4"
-        cfg = VideoEncoderConfig(vcodec="h264", g=6, crf=23, pix_fmt="yuv444p")
-        reencode_video(src, out, camera_encoder=cfg, overwrite=True)
+        cfg = RGBEncoderConfig(vcodec="h264", g=6, crf=23, pix_fmt="yuv444p")
+        reencode_video(src, out, video_encoder=cfg, overwrite=True)
 
         assert out.exists()
         with av.open(str(out)) as container:
             n_frames = sum(1 for _ in container.decode(video=0))
         assert n_frames == 4
 
-        info = get_video_info(out, camera_encoder=cfg)
+        info = get_video_info(out, video_encoder=cfg)
         assert info["video.codec"] == "h264"
         assert info["video.pix_fmt"] == "yuv444p"
         assert info["video.height"] == 64
@@ -508,8 +523,8 @@ class TestReencodeVideo:
     def test_reencode_video_trim_window(self, tmp_path):
         src = TEST_ARTIFACTS_DIR / "clip_6frames.mp4"
         out = tmp_path / "trim_window.mp4"
-        cfg = VideoEncoderConfig(vcodec="h264")
-        reencode_video(src, out, camera_encoder=cfg, start_time_s=0.05, end_time_s=0.12, overwrite=True)
+        cfg = RGBEncoderConfig(vcodec="h264")
+        reencode_video(src, out, video_encoder=cfg, start_time_s=0.05, end_time_s=0.12, overwrite=True)
 
         with av.open(str(out)) as container:
             frames = list(container.decode(video=0))
@@ -578,12 +593,12 @@ class TestEncoderConfigPersistence:
 
     @require_libsvtav1
     def test_first_episode_save_persists_encoder_config(self, tmp_path, empty_lerobot_dataset_factory):
-        cfg = VideoEncoderConfig(vcodec="libsvtav1", g=2, crf=30, preset=12)
+        cfg = RGBEncoderConfig(vcodec="libsvtav1", g=2, crf=30, preset=12)
         dataset = empty_lerobot_dataset_factory(
-            root=tmp_path / "ds", features=VIDEO_FEATURES, use_videos=True, camera_encoder=cfg
+            root=tmp_path / "ds", features=DUMMY_VIDEO_FEATURES, use_videos=True, rgb_encoder=cfg
         )
 
-        _add_frames(dataset, num_frames=4)
+        add_frames(dataset, num_frames=4)
         dataset.save_episode()
         dataset.finalize()
 
@@ -601,16 +616,16 @@ class TestEncoderConfigPersistence:
 
     @require_libsvtav1
     def test_second_episode_does_not_overwrite_encoder_fields(self, tmp_path, empty_lerobot_dataset_factory):
-        cfg = VideoEncoderConfig(vcodec="libsvtav1", g=2, crf=30, preset=12)
+        cfg = RGBEncoderConfig(vcodec="libsvtav1", g=2, crf=30, preset=12)
         dataset = empty_lerobot_dataset_factory(
-            root=tmp_path / "ds", features=VIDEO_FEATURES, use_videos=True, camera_encoder=cfg
+            root=tmp_path / "ds", features=DUMMY_VIDEO_FEATURES, use_videos=True, rgb_encoder=cfg
         )
 
-        _add_frames(dataset, num_frames=4)
+        add_frames(dataset, num_frames=4)
         dataset.save_episode()
         first_info = dict(_read_feature_info(dataset))
 
-        _add_frames(dataset, num_frames=4)
+        add_frames(dataset, num_frames=4)
         dataset.save_episode()
         dataset.finalize()
 
@@ -618,13 +633,13 @@ class TestEncoderConfigPersistence:
 
 
 class TestFromVideoInfo:
-    """``VideoEncoderConfig.from_video_info`` reconstructs an encoder config
+    """``RGBEncoderConfig.from_video_info`` reconstructs an encoder config
     from the ``video.*`` keys persisted in a dataset's ``info.json``.
     """
 
     @require_libsvtav1
     def test_reconstructs_from_dummy_video_info(self):
-        cfg = VideoEncoderConfig.from_video_info(DUMMY_VIDEO_INFO)
+        cfg = RGBEncoderConfig.from_video_info(DUMMY_VIDEO_INFO)
 
         # Canonical stream codec ``"av1"`` is aliased to the encoder name.
         assert cfg.vcodec == "libsvtav1"
@@ -636,4 +651,220 @@ class TestFromVideoInfo:
         assert cfg.video_backend == DUMMY_VIDEO_INFO["video.video_backend"]
         # ``{}`` placeholder (typical after a merge with disagreeing sources)
         # must not leak into the reconstructed config.
-        assert cfg.extra_options == VideoEncoderConfig().extra_options
+        assert cfg.extra_options == RGBEncoderConfig().extra_options
+
+
+# ─── Depth-specific encoding tests ────────────────────────────────────
+
+
+class TestEncodeDepthVideoFrames:
+    """Depth mirror of :class:`TestEncodeVideoFrames`.
+
+    Exercises ``encode_video_frames`` end-to-end through
+    :class:`DepthEncoderConfig` (HEVC Main 12 / ``gray12le``) on synthetic
+    uint16 depth TIFFs.
+    """
+
+    @require_hevc
+    def test_produces_readable_file(self, tmp_path):
+        video_path = _encode_video(tmp_path / "out.mp4", depth=True)
+
+        assert video_path.exists()
+        info = get_video_info(video_path, video_encoder=DepthEncoderConfig())
+        assert info["video.height"] == 64
+        assert info["video.width"] == 96
+        assert info["video.codec"] == "hevc"
+        assert info["video.pix_fmt"] == "gray12le"
+        assert info["video.channels"] == 1
+        assert info["is_depth_map"] is True
+
+    @require_hevc
+    def test_frame_count_and_duration_match_input(self, tmp_path):
+        num_frames = 10
+        fps = 30
+        video_path = _encode_video(tmp_path / "out.mp4", num_frames=num_frames, fps=fps, depth=True)
+
+        with av.open(str(video_path)) as container:
+            stream = container.streams.video[0]
+            actual_frames = sum(1 for _ in container.decode(stream))
+            duration = (
+                float(stream.duration * stream.time_base)
+                if stream.duration is not None
+                else float(container.duration / av.time_base)
+            )
+
+        assert actual_frames == num_frames
+        assert abs(duration - num_frames / fps) < 0.1
+
+    def test_overwrite_false_skips_existing_file(self, tmp_path):
+        """Codec-agnostic: file-system semantics must hold even without an HEVC encoder."""
+        imgs_dir = tmp_path / "imgs"
+        _write_depth_frames(imgs_dir)
+        video_path = tmp_path / "out.mp4"
+        sentinel = b"pre-existing depth content"
+        video_path.write_bytes(sentinel)
+
+        encode_video_frames(imgs_dir, video_path, fps=30, video_encoder=DepthEncoderConfig(), overwrite=False)
+
+        assert video_path.read_bytes() == sentinel
+
+    @require_hevc
+    def test_overwrite_true_replaces_existing_file(self, tmp_path):
+        imgs_dir = tmp_path / "imgs"
+        _write_depth_frames(imgs_dir)
+        video_path = tmp_path / "out.mp4"
+        video_path.write_bytes(b"stale content")
+
+        encode_video_frames(imgs_dir, video_path, fps=30, video_encoder=DepthEncoderConfig(), overwrite=True)
+
+        info = get_video_info(video_path, video_encoder=DepthEncoderConfig())
+        assert info["video.height"] == 64
+        assert info["video.pix_fmt"] == "gray12le"
+        assert info["is_depth_map"] is True
+
+    @require_hevc
+    def test_custom_encoder_config_fields_stored_in_info(self, tmp_path):
+        """All stream-derived and depth-encoder config fields are present after encoding."""
+        cfg = DepthEncoderConfig(
+            vcodec="hevc",
+            pix_fmt="gray12le",
+            g=4,
+            crf=25,
+            extra_options={},
+            depth_min=0.05,
+            depth_max=8.0,
+            shift=2.5,
+            use_log=False,
+        )
+        video_path = _encode_video(tmp_path / "out.mp4", num_frames=4, fps=30, cfg=cfg, depth=True)
+
+        info = get_video_info(video_path, video_encoder=cfg)
+
+        # Stream-derived
+        assert info["video.height"] == 64
+        assert info["video.width"] == 96
+        assert info["video.channels"] == 1
+        assert info["video.codec"] == "hevc"
+        assert info["video.pix_fmt"] == "gray12le"
+        assert info["video.fps"] == 30
+        assert info["is_depth_map"] is True
+        assert info["has_audio"] is False
+        # Base encoder config
+        assert info["video.g"] == 4
+        assert info["video.crf"] == 25
+        assert info["video.fast_decode"] == 0
+        assert info["video.video_backend"] == "pyav"
+        assert info["video.extra_options"] == {}
+        # Depth-specific tuning
+        assert info["video.depth_min"] == 0.05
+        assert info["video.depth_max"] == 8.0
+        assert info["video.shift"] == 2.5
+        assert info["video.use_log"] is False
+
+
+class TestDepthEncoderConfigPersistence:
+    """Depth mirror of :class:`TestEncoderConfigPersistence`.
+
+    ``DepthEncoderConfig`` must be stored as ``video.<field>`` entries
+    (including the depth-specific ``depth_min`` / ``depth_max`` / ``shift`` /
+    ``use_log``) under ``info["features"][<depth_key>]["info"]`` when the
+    first episode is saved.
+    """
+
+    @require_hevc
+    def test_first_episode_save_persists_depth_encoder_config(self, tmp_path, empty_lerobot_dataset_factory):
+        cfg = DepthEncoderConfig(
+            vcodec="hevc",
+            pix_fmt="gray12le",
+            g=2,
+            crf=30,
+            extra_options={},
+            depth_min=0.05,
+            depth_max=8.0,
+            shift=2.5,
+            use_log=False,
+        )
+        dataset = empty_lerobot_dataset_factory(
+            root=tmp_path / "ds", features=DUMMY_DEPTH_FEATURES, use_videos=True, depth_encoder=cfg
+        )
+
+        add_frames(dataset, num_frames=4)
+        dataset.save_episode()
+        dataset.finalize()
+
+        info = _read_feature_info(dataset, key=DUMMY_DEPTH_KEY)
+
+        # Stream-derived
+        assert info["video.height"] == 64
+        assert info["video.width"] == 96
+        assert info["video.fps"] == 30
+        assert info["video.codec"] == "hevc"
+        assert info["video.pix_fmt"] == "gray12le"
+        assert info["is_depth_map"] is True
+        # Base encoder config
+        assert info["video.g"] == 2
+        assert info["video.crf"] == 30
+        assert info["video.fast_decode"] == 0
+        assert info["video.video_backend"] == "pyav"
+        assert info["video.extra_options"] == {}
+        # Depth-specific tuning
+        assert info["video.depth_min"] == 0.05
+        assert info["video.depth_max"] == 8.0
+        assert info["video.shift"] == 2.5
+        assert info["video.use_log"] is False
+
+    @require_hevc
+    def test_second_episode_does_not_overwrite_depth_encoder_fields(
+        self, tmp_path, empty_lerobot_dataset_factory
+    ):
+        cfg = DepthEncoderConfig(
+            vcodec="hevc",
+            pix_fmt="gray12le",
+            g=2,
+            crf=30,
+            depth_min=0.05,
+            depth_max=8.0,
+            shift=2.5,
+            use_log=False,
+        )
+        dataset = empty_lerobot_dataset_factory(
+            root=tmp_path / "ds", features=DUMMY_DEPTH_FEATURES, use_videos=True, depth_encoder=cfg
+        )
+
+        add_frames(dataset, num_frames=4)
+        dataset.save_episode()
+        first_info = dict(_read_feature_info(dataset, key=DUMMY_DEPTH_KEY))
+
+        add_frames(dataset, num_frames=4)
+        dataset.save_episode()
+        dataset.finalize()
+
+        assert _read_feature_info(dataset, key=DUMMY_DEPTH_KEY) == first_info
+
+
+class TestDepthFromVideoInfo:
+    """``DepthEncoderConfig.from_video_info`` reconstructs a depth encoder
+    config from the ``video.*`` keys persisted in a dataset's ``info.json``.
+
+    Depth mirror of :class:`TestFromVideoInfo`.
+    """
+
+    @require_hevc
+    def test_reconstructs_from_dummy_depth_video_info(self):
+        cfg = DepthEncoderConfig.from_video_info(DUMMY_DEPTH_VIDEO_INFO_FULL)
+
+        # No alias for ``"hevc"``; the canonical stream codec is reused as-is.
+        assert cfg.vcodec == "hevc"
+        assert cfg.pix_fmt == DUMMY_DEPTH_VIDEO_INFO_FULL["video.pix_fmt"]
+        assert cfg.g == DUMMY_DEPTH_VIDEO_INFO_FULL["video.g"]
+        assert cfg.crf == DUMMY_DEPTH_VIDEO_INFO_FULL["video.crf"]
+        assert cfg.fast_decode == DUMMY_DEPTH_VIDEO_INFO_FULL["video.fast_decode"]
+        assert cfg.video_backend == DUMMY_DEPTH_VIDEO_INFO_FULL["video.video_backend"]
+        # ``{}`` placeholder (typical after a merge with disagreeing sources)
+        # must not leak into the reconstructed config.
+        assert cfg.extra_options == DepthEncoderConfig().extra_options
+        # Depth-specific tuning round-trips through ``info.json``.
+        assert cfg.depth_min == DUMMY_DEPTH_VIDEO_INFO_FULL["video.depth_min"]
+        assert cfg.depth_max == DUMMY_DEPTH_VIDEO_INFO_FULL["video.depth_max"]
+        assert cfg.shift == DUMMY_DEPTH_VIDEO_INFO_FULL["video.shift"]
+        assert cfg.use_log == DUMMY_DEPTH_VIDEO_INFO_FULL["video.use_log"]

tests/fixtures/constants.py

@@ -39,12 +39,56 @@ DUMMY_VIDEO_INFO = {
     "video.crf": 30,
     "video.preset": 12,
     "video.fast_decode": 0,
-    "video.is_depth_map": False,
+    "is_depth_map": False,
     "has_audio": False,
 }
 DUMMY_CAMERA_FEATURES = {
     "laptop": {"shape": (64, 96, 3), "names": ["height", "width", "channels"], "info": DUMMY_VIDEO_INFO},
     "phone": {"shape": (64, 96, 3), "names": ["height", "width", "channels"], "info": DUMMY_VIDEO_INFO},
 }
+DUMMY_DEPTH_VIDEO_INFO = {
+    **DUMMY_VIDEO_INFO,
+    "is_depth_map": True,
+}
+DUMMY_DEPTH_VIDEO_INFO_FULL = {
+    **{k: v for k, v in DUMMY_VIDEO_INFO.items() if k != "video.preset"},
+    "video.codec": "hevc",
+    "video.pix_fmt": "gray12le",
+    "is_depth_map": True,
+    "video.depth_min": 0.05,
+    "video.depth_max": 8.0,
+    "video.shift": 2.5,
+    "video.use_log": True,
+}
+DUMMY_DEPTH_CAMERA_FEATURES = {
+    "laptop_depth": {
+        "shape": (64, 96, 1),
+        "names": ["height", "width", "channels"],
+        "info": DUMMY_DEPTH_VIDEO_INFO,
+    },
+}
+DUMMY_CAMERA_FEATURES_WITH_DEPTH = {**DUMMY_CAMERA_FEATURES, **DUMMY_DEPTH_CAMERA_FEATURES}
 DUMMY_CHW = (3, 96, 128)
 DUMMY_HWC = (96, 128, 3)
+
+# Default video feature set used by video-encoding persistence tests.
+DUMMY_VIDEO_FEATURES = {
+    "observation.images.cam": {
+        "dtype": "video",
+        "shape": (64, 96, 3),
+        "names": ["height", "width", "channels"],
+    },
+    "action": {"dtype": "float32", "shape": (2,), "names": ["a", "b"]},
+}
+DUMMY_VIDEO_KEY = "observation.images.cam"
+
+DUMMY_DEPTH_FEATURES = {
+    "observation.images.depth": {
+        "dtype": "video",
+        "shape": (64, 96, 1),
+        "names": ["height", "width", "channels"],
+        "info": {"is_depth_map": True},
+    },
+    "action": {"dtype": "float32", "shape": (2,), "names": ["a", "b"]},
+}
+DUMMY_DEPTH_KEY = "observation.images.depth"

tests/fixtures/dataset_factories.py

@@ -49,6 +49,39 @@ from tests.fixtures.constants import (
 )
 
 
+def add_frames(dataset: LeRobotDataset, num_frames: int) -> None:
+    """Append ``num_frames`` synthetic frames to ``dataset``.
+
+    Generates per-feature payloads from ``dataset.meta``: uint16 depth ramps for
+    keys in ``dataset.meta.depth_keys``, uint8 random noise for video/image keys,
+    and float32 zeros for everything else. ``DEFAULT_FEATURES`` (timestamp,
+    frame_index, ...) are auto-populated by ``add_frame`` and skipped here.
+    """
+    video_keys = dataset.meta.video_keys
+    depth_keys = dataset.meta.depth_keys
+    # Smooth gradient base reused per (H, W) to keep depth frames cheap to
+    # encode (HEVC Main 12 hates white noise).
+    _depth_base_cache: dict[tuple[int, int], np.ndarray] = {}
+    for i in range(num_frames):
+        frame: dict = {"task": "test"}
+        for key, ft in dataset.meta.features.items():
+            if key in DEFAULT_FEATURES:
+                continue
+            shape = ft["shape"]
+            if key in depth_keys:
+                h, w, _ = shape
+                base = _depth_base_cache.setdefault(
+                    (h, w),
+                    np.linspace(100.0, 10_000.0, h * w, dtype=np.float32).reshape(h, w, 1),
+                )
+                frame[key] = (base + 50.0 * i).clip(0, 65535).astype(np.uint16)
+            elif key in video_keys:
+                frame[key] = np.random.randint(0, 256, shape, dtype=np.uint8)
+            else:
+                frame[key] = np.zeros(shape, dtype=np.float32)
+        dataset.add_frame(frame)
+
+
 class LeRobotDatasetFactory(Protocol):
     def __call__(self, *args, **kwargs) -> LeRobotDataset: ...
 
@@ -485,10 +518,14 @@ def lerobot_dataset_factory(
         hf_dataset: datasets.Dataset | None = None,
         data_files_size_in_mb: float = DEFAULT_DATA_FILE_SIZE_IN_MB,
         chunks_size: int = DEFAULT_CHUNK_SIZE,
+        camera_features: dict | None = None,
         **kwargs,
     ) -> LeRobotDataset:
         # Instantiate objects
         if info is None:
+            info_kwargs = {}
+            if camera_features is not None:
+                info_kwargs["camera_features"] = camera_features
             info = info_factory(
                 total_episodes=total_episodes,
                 total_frames=total_frames,
@@ -496,6 +533,7 @@ def lerobot_dataset_factory(
                 use_videos=use_videos,
                 data_files_size_in_mb=data_files_size_in_mb,
                 chunks_size=chunks_size,
+                **info_kwargs,
             )
         if stats is None:
             stats = stats_factory(features=info.features)

tests/policies/groot/artifacts/.gitignore

@@ -0,0 +1,2 @@
+# Local-only parity artifacts (regenerated via dump_original_n1_7.py); never committed.
+*.npz

tests/policies/groot/test_groot_lerobot.py

@@ -14,7 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""Test script for LeRobot's Groot policy forward and inference passes."""
+"""Test script for LeRobot's GR00T N1.7 policy forward and inference passes."""
 
 import gc
 import os
@@ -41,13 +41,20 @@ pytestmark = pytest.mark.skipif(
 )
 
 
-# Define constants for dummy data
+# Define constants for dummy data (GR00T N1.7 native conventions).
+# N1.7 internally uses a 40-step action chunk, 132-dim state/action, and 256px images
+# (see GrootConfig.__post_init__). Use a chunk-sized action horizon so the dummy batch
+# matches the model's native action space.
 DUMMY_STATE_DIM = 44
 DUMMY_ACTION_DIM = 44
-DUMMY_ACTION_HORIZON = 16
+DUMMY_ACTION_HORIZON = 40
 IMAGE_SIZE = 256
 DEVICE = auto_select_torch_device()
-MODEL_PATH = "aractingi/bimanual-handover-groot-10k"
+# GR00T N1.7 checkpoint (N1.5 is no longer supported). The N1.7-3B base model loads
+# via GrootPolicy.from_pretrained with root-level sharded safetensors.
+MODEL_PATH = "nvidia/GR00T-N1.7-3B"
+# Valid N1.7 embodiment tag carried by the checkpoint metadata.
+EMBODIMENT_TAG = "gr1_unified"
 
 
 def cleanup_memory():
@@ -88,13 +95,13 @@ def instantiate_lerobot_groot(
     PolicyProcessorPipeline[dict[str, Any], dict[str, Any]],
     PolicyProcessorPipeline[PolicyAction, PolicyAction],
 ]:
-    """Instantiate LeRobot Groot policy with preprocessor and postprocessor."""
+    """Instantiate LeRobot GR00T N1.7 policy with preprocessor and postprocessor."""
     if from_pretrained:
         policy = GrootPolicy.from_pretrained(
             pretrained_name_or_path=model_path,
             strict=False,
         )
-        policy.config.embodiment_tag = "gr1"
+        policy.config.embodiment_tag = EMBODIMENT_TAG
     else:
         config = GrootConfig(
             base_model_path=model_path,
@@ -102,7 +109,7 @@ def instantiate_lerobot_groot(
             chunk_size=DUMMY_ACTION_HORIZON,
             image_size=[IMAGE_SIZE, IMAGE_SIZE],
             device=DEVICE,
-            embodiment_tag="gr1",
+            embodiment_tag=EMBODIMENT_TAG,
         )
         policy = GrootPolicy(config)
 
@@ -148,8 +155,8 @@ def create_dummy_data(device=DEVICE):
 
 @require_cuda
 def test_lerobot_groot_inference():
-    """Test the inference pass (select_action) of LeRobot's Groot policy."""
-    print("Test: LeRobot Groot Inference Pass")
+    """Test the inference pass (select_action) of LeRobot's GR00T N1.7 policy."""
+    print("Test: LeRobot GR00T N1.7 Inference Pass")
 
     set_seed_all(42)
 
@@ -181,9 +188,9 @@ def test_lerobot_groot_inference():
 
 @require_cuda
 def test_lerobot_groot_forward_pass():
-    """Test the forward pass of LeRobot's Groot policy."""
+    """Test the forward pass of LeRobot's GR00T N1.7 policy."""
     print("\n" + "=" * 50)
-    print("Test: LeRobot Groot Forward Pass (Training Mode)")
+    print("Test: LeRobot GR00T N1.7 Forward Pass (Training Mode)")
 
     set_seed_all(42)
 

tests/policies/groot/test_groot_n1_7_oss_parity.py

@@ -0,0 +1,259 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import hashlib
+import os
+from pathlib import Path
+
+import numpy as np
+import pytest
+import torch
+
+from lerobot.policies.groot.action_head.cross_attention_dit import AlternateVLDiT
+from lerobot.policies.groot.groot_n1_7 import GR00TN17
+from lerobot.policies.groot.processor_groot import (
+    GrootN17ActionDecodeStep,
+    GrootN17PackInputsStep,
+    GrootN17VLMEncodeStep,
+    _transform_n1_7_image_for_vlm_albumentations,
+)
+from lerobot.types import TransitionKey
+from lerobot.utils.constants import OBS_STATE
+
+OSS_REFERENCE_COMMIT = "ab88b50c718f6528e1df9dcbaf75865d1b604760"
+
+
+def _fixture_path(filename: str) -> Path:
+    fixture_dir = os.environ.get("GROOT_N17_OSS_PARITY_FIXTURE_DIR")
+    if fixture_dir is None:
+        pytest.skip("Set GROOT_N17_OSS_PARITY_FIXTURE_DIR to run external OSS parity fixtures.")
+    path = Path(fixture_dir) / filename
+    if not path.is_file():
+        pytest.skip(f"External OSS parity fixture not found: {path}")
+    return path
+
+
+def test_groot_n1_7_eval_image_transform_matches_oss_reference():
+    """Match the native N1.7 eval transform for a non-square SO-101 frame."""
+
+    y, x = np.indices((480, 640), dtype=np.uint16)
+    image = np.stack(
+        ((x + 3 * y) % 256, (2 * x + y) % 256, (x + 5 * y) % 256),
+        axis=-1,
+    ).astype(np.uint8)
+    actual = _transform_n1_7_image_for_vlm_albumentations(
+        image,
+        image_crop_size=[230, 230],
+        image_target_size=[256, 256],
+        shortest_image_edge=256,
+        crop_fraction=0.95,
+    )
+
+    assert actual.shape == (256, 340, 3)
+    assert hashlib.sha256(actual.tobytes()).hexdigest() == (
+        "c17e47af68a812aa79db3bb7b64b549ddf10148ac1b204a9686095018561ae9e"
+    )
+
+
+def test_groot_n1_7_vlm_chat_content_order_matches_oss_reference():
+    """Native OSS places all image items before the language item."""
+
+    class RecordingProcessor:
+        def __init__(self):
+            self.content_types = None
+
+        def apply_chat_template(self, conversation, tokenize, add_generation_prompt):
+            assert tokenize is False
+            assert add_generation_prompt is False
+            self.content_types = [item["type"] for item in conversation[0]["content"]]
+            return "rendered"
+
+        def __call__(self, **kwargs):
+            return {}
+
+    processor = RecordingProcessor()
+    step = GrootN17VLMEncodeStep(
+        image_crop_size=[230, 230],
+        image_target_size=[256, 256],
+        shortest_image_edge=256,
+        crop_fraction=0.95,
+        use_albumentations=True,
+        device="cpu",
+    )
+    step._proc = processor
+    transition = {
+        TransitionKey.OBSERVATION: {
+            "video": np.zeros((1, 1, 2, 480, 640, 3), dtype=np.uint8),
+        },
+        TransitionKey.COMPLEMENTARY_DATA: {"language": ["pick up the vial"]},
+    }
+
+    step(transition)
+
+    assert processor.content_types == ["image", "image", "text"]
+
+
+def test_groot_n1_7_alternate_vl_dit_matches_oss_reference():
+    """Run the LeRobot DiT with native OSS weights and identical inputs."""
+
+    pytest.importorskip("diffusers")
+
+    fixture = torch.load(_fixture_path("alternate_vl_dit_small.pt"), map_location="cpu", weights_only=True)
+    model = AlternateVLDiT(
+        output_dim=8,
+        num_attention_heads=2,
+        attention_head_dim=4,
+        num_layers=4,
+        dropout=0.0,
+        final_dropout=False,
+        max_num_positional_embeddings=16,
+        compute_dtype=torch.float32,
+        interleave_self_attention=True,
+        cross_attention_dim=6,
+    ).eval()
+    model.load_state_dict(fixture["state_dict"], strict=True)
+
+    actual = model(
+        hidden_states=fixture["hidden_states"],
+        encoder_hidden_states=fixture["encoder_hidden_states"],
+        timestep=fixture["timestep"],
+        image_mask=fixture["image_mask"],
+        backbone_attention_mask=fixture["backbone_attention_mask"],
+    )
+
+    torch.testing.assert_close(actual, fixture["output"], atol=1e-6, rtol=1e-6)
+
+
+def _state_decode_reference():
+    fixture = np.load(_fixture_path("state_and_action_decode.npz"))
+    raw_stats = {
+        "state": {
+            "single_arm": {"q01": fixture["state_single_arm_q01"], "q99": fixture["state_single_arm_q99"]},
+            "gripper": {"q01": fixture["state_gripper_q01"], "q99": fixture["state_gripper_q99"]},
+        },
+        "action": {
+            "single_arm": {"q01": fixture["action_single_arm_q01"], "q99": fixture["action_single_arm_q99"]},
+            "gripper": {"q01": fixture["action_gripper_q01"], "q99": fixture["action_gripper_q99"]},
+        },
+        "relative_action": {
+            "single_arm": {
+                "min": fixture["relative_single_arm_min"],
+                "max": fixture["relative_single_arm_max"],
+            },
+        },
+    }
+    for modality_stats in raw_stats.values():
+        for entry in modality_stats.values():
+            for key, value in entry.items():
+                if isinstance(value, np.ndarray):
+                    entry[key] = value.tolist()
+    modality_config = {
+        "state": {"modality_keys": ["single_arm", "gripper"]},
+        "action": {
+            "delta_indices": list(range(16)),
+            "modality_keys": ["single_arm", "gripper"],
+            "action_configs": [
+                {"rep": "RELATIVE", "type": "NON_EEF", "format": "DEFAULT", "state_key": None},
+                {"rep": "ABSOLUTE", "type": "NON_EEF", "format": "DEFAULT", "state_key": None},
+            ],
+        },
+    }
+    state_min = np.concatenate((fixture["state_single_arm_q01"], fixture["state_gripper_q01"]))
+    state_max = np.concatenate((fixture["state_single_arm_q99"], fixture["state_gripper_q99"]))
+    pack_step = GrootN17PackInputsStep(
+        normalize_min_max=True,
+        stats={OBS_STATE: {"min": state_min, "max": state_max}},
+        raw_stats=raw_stats,
+        modality_config=modality_config,
+        use_percentiles=True,
+    )
+    raw_state = np.concatenate((fixture["state_single_arm"], fixture["state_gripper"]), axis=-1)
+    transition = {
+        TransitionKey.OBSERVATION: {OBS_STATE: torch.from_numpy(raw_state)},
+        TransitionKey.COMPLEMENTARY_DATA: {},
+    }
+    packed = pack_step(transition)
+    return fixture, raw_stats, modality_config, pack_step, packed
+
+
+def test_groot_n1_7_state_normalization_matches_oss_checkpoint_reference():
+    fixture, _raw_stats, _modality_config, _pack_step, packed = _state_decode_reference()
+    expected = np.concatenate(
+        (fixture["normalized_state_single_arm"], fixture["normalized_state_gripper"]), axis=-1
+    )
+
+    actual = packed[TransitionKey.OBSERVATION]["state"][:, 0, :6]
+
+    torch.testing.assert_close(actual, torch.from_numpy(expected), atol=1e-6, rtol=1e-6)
+
+
+def test_groot_n1_7_relative_action_decode_matches_oss_checkpoint_reference():
+    fixture, raw_stats, modality_config, pack_step, _packed = _state_decode_reference()
+    decode_step = GrootN17ActionDecodeStep(
+        env_action_dim=6,
+        raw_stats=raw_stats,
+        modality_config=modality_config,
+        use_percentiles=True,
+        use_relative_action=True,
+        pack_step=pack_step,
+    )
+    decoded = decode_step({TransitionKey.ACTION: torch.from_numpy(fixture["normalized_action"])})[
+        TransitionKey.ACTION
+    ]
+    expected = np.concatenate((fixture["decoded_single_arm"], fixture["decoded_gripper"]), axis=-1).astype(
+        np.float32
+    )
+
+    torch.testing.assert_close(decoded, torch.from_numpy(expected), atol=1e-5, rtol=1e-5)
+
+
+def test_groot_n1_7_qwen_backbone_matches_oss_checkpoint_reference():
+    """Compare the actual 3B checkpoint backbone when explicitly enabled."""
+
+    checkpoint = os.environ.get("GROOT_N17_PARITY_CHECKPOINT")
+    if checkpoint is None:
+        pytest.skip("Set GROOT_N17_PARITY_CHECKPOINT to run the 3B OSS Qwen parity test.")
+    if not torch.cuda.is_available():
+        pytest.skip("The 3B OSS Qwen parity test requires CUDA.")
+
+    pytest.importorskip("transformers")
+
+    from transformers.feature_extraction_utils import BatchFeature
+
+    fixture = torch.load(_fixture_path("qwen_backbone_so101.pt"), map_location="cpu", weights_only=True)
+    model = GR00TN17.from_pretrained(checkpoint).to(device="cuda", dtype=torch.bfloat16).eval()
+    backbone_input = BatchFeature(
+        data={
+            key.removeprefix("input."): value.to("cuda")
+            for key, value in fixture.items()
+            if key.startswith("input.")
+        }
+    )
+
+    with torch.inference_mode():
+        actual = model.backbone(backbone_input)
+
+    feature_error = (
+        actual.backbone_features.cpu().float() - fixture["output.backbone_features"].float()
+    ).abs()
+    # Native OSS and LeRobot use different Torch/Transformers/Flash-Attention releases.
+    # Require the measured BF16 accumulation envelope while rejecting structural drift.
+    assert feature_error.mean().item() <= 0.04
+    assert feature_error.max().item() <= 2.0
+    torch.testing.assert_close(
+        actual.backbone_attention_mask.cpu(), fixture["output.backbone_attention_mask"]
+    )
+    torch.testing.assert_close(actual.image_mask.cpu(), fixture["output.image_mask"])

tests/policies/groot/test_groot_state_dropout.py

@@ -0,0 +1,100 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Isaac-GR00T N1.7 raw-state dropout training contract.
+
+Isaac-GR00T zeroes the entire proprioceptive state of a sample with probability
+``state_dropout_prob`` (configured in the checkpoint's processor sidecar) during
+training only. Baseline LeRobot kept the processor deterministic, so this
+regularization never activated. These tests pin the train/eval split.
+"""
+
+import torch
+
+from lerobot.policies.groot.processor_groot import GrootN17PackInputsStep
+from lerobot.types import TransitionKey
+from lerobot.utils.constants import OBS_STATE
+
+
+def _make_transition():
+    return {
+        TransitionKey.OBSERVATION: {OBS_STATE: torch.tensor([[1.0, 2.0], [3.0, 4.0]])},
+        TransitionKey.COMPLEMENTARY_DATA: {"task": ["Move", "Move"]},
+    }
+
+
+def test_groot_n1_7_training_applies_raw_state_dropout_before_encoder():
+    step = GrootN17PackInputsStep(
+        max_state_dim=4,
+        max_action_dim=4,
+        normalize_min_max=False,
+        training=True,
+        state_dropout_prob=1.0,
+    )
+
+    output = step(_make_transition())
+
+    expected = torch.zeros(2, 1, 4)
+    torch.testing.assert_close(output[TransitionKey.OBSERVATION]["state"], expected)
+
+
+def test_groot_n1_7_training_state_dropout_is_disabled_under_no_grad():
+    step = GrootN17PackInputsStep(
+        max_state_dim=4,
+        max_action_dim=4,
+        normalize_min_max=False,
+        training=True,
+        state_dropout_prob=1.0,
+    )
+
+    with torch.no_grad():
+        output = step(_make_transition())
+
+    expected = torch.tensor([[[1.0, 2.0, 0.0, 0.0]], [[3.0, 4.0, 0.0, 0.0]]])
+    torch.testing.assert_close(output[TransitionKey.OBSERVATION]["state"], expected)
+
+
+def test_groot_n1_7_eval_mode_state_dropout_is_inactive():
+    step = GrootN17PackInputsStep(
+        max_state_dim=4,
+        max_action_dim=4,
+        normalize_min_max=False,
+        training=False,
+        state_dropout_prob=1.0,
+    )
+
+    output = step(_make_transition())
+
+    expected = torch.tensor([[[1.0, 2.0, 0.0, 0.0]], [[3.0, 4.0, 0.0, 0.0]]])
+    torch.testing.assert_close(output[TransitionKey.OBSERVATION]["state"], expected)
+
+
+def test_groot_n1_7_pack_step_serializes_dropout_prob_but_not_training_mode():
+    step = GrootN17PackInputsStep(
+        max_state_dim=4,
+        max_action_dim=4,
+        normalize_min_max=False,
+        training=True,
+        state_dropout_prob=0.2,
+    )
+
+    serialized = step.get_config()
+    restored = GrootN17PackInputsStep(**serialized)
+
+    assert "training" not in serialized
+    assert serialized["state_dropout_prob"] == 0.2
+    assert restored.training is False
+    assert restored.state_dropout_prob == 0.2

tests/policies/groot/test_groot_train_random_crop.py

@@ -0,0 +1,156 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Isaac-GR00T N1.7 train-time random crop contract (crop geometry only).
+
+Isaac-GR00T crops a random ``crop_fraction`` window during training and the
+deterministic center window at eval, replaying the sampled window across all
+camera views of a sample (gr00t/data/transform/video.py, n1.5-release onward:
+"If mode is 'train', return a random crop transform. If mode is 'eval', return
+a center crop transform."). This mirrors LeRobot's own Diffusion/VQBeT
+``crop_is_random`` pattern. Color jitter is intentionally out of scope here.
+"""
+
+import random
+
+import numpy as np
+import torch
+
+from lerobot.policies.groot.processor_groot import (
+    GrootN17VLMEncodeStep,
+    _transform_n1_7_image_for_vlm_albumentations,
+)
+
+
+def _structured_image(h=480, w=640):
+    yy, xx = np.mgrid[0:h, 0:w]
+    return np.stack(
+        [(xx * 255 / w), (yy * 255 / h), ((xx + yy) * 255 / (h + w))], axis=-1
+    ).astype(np.uint8)
+
+
+def test_crop_position_none_is_bitexact_center_crop():
+    """crop_position=None must remain byte-identical to the pre-change eval path."""
+    img = _structured_image()
+    ref = _transform_n1_7_image_for_vlm_albumentations(
+        img, image_crop_size=None, image_target_size=[256, 256],
+        shortest_image_edge=256, crop_fraction=0.95,
+    )
+    out = _transform_n1_7_image_for_vlm_albumentations(
+        img, image_crop_size=None, image_target_size=[256, 256],
+        shortest_image_edge=256, crop_fraction=0.95, crop_position=None,
+    )
+    np.testing.assert_array_equal(ref, out)
+
+
+def test_crop_position_center_matches_center_crop():
+    img = _structured_image()
+    center = _transform_n1_7_image_for_vlm_albumentations(
+        img, image_crop_size=None, image_target_size=[256, 256],
+        shortest_image_edge=256, crop_fraction=0.95, crop_position=None,
+    )
+    explicit = _transform_n1_7_image_for_vlm_albumentations(
+        img, image_crop_size=None, image_target_size=[256, 256],
+        shortest_image_edge=256, crop_fraction=0.95, crop_position=(0.5, 0.5),
+    )
+    # int-floor center vs rounded positional center may differ by <=1 px of grid
+    assert center.shape == explicit.shape
+    diff = np.abs(center.astype(np.int16) - explicit.astype(np.int16))
+    assert diff.mean() < 3.0
+
+
+def test_crop_position_corners_differ_from_center():
+    img = _structured_image()
+
+    def crop_at(position):
+        return _transform_n1_7_image_for_vlm_albumentations(
+            img,
+            image_crop_size=None,
+            image_target_size=[256, 256],
+            shortest_image_edge=256,
+            crop_fraction=0.95,
+            crop_position=position,
+        )
+
+    center = crop_at(None)
+    tl = crop_at((0.0, 0.0))
+    br = crop_at((1.0, 1.0))
+    assert not np.array_equal(center, tl)
+    assert not np.array_equal(tl, br)
+
+
+def _video(img, views=2):
+    return np.stack([img] * views, axis=0).reshape(1, 1, views, *img.shape)
+
+
+def _step(training):
+    return GrootN17VLMEncodeStep(
+        image_target_size=[256, 256],
+        shortest_image_edge=256,
+        crop_fraction=0.95,
+        use_albumentations=True,
+        training=training,
+    )
+
+
+def test_training_crop_replays_one_window_across_views():
+    video = _video(_structured_image())
+    frames = _step(training=True)._build_sample_images(video, batch_size=1, target_device=None)[0]
+    np.testing.assert_array_equal(np.asarray(frames[0]), np.asarray(frames[1]))
+
+
+def test_training_crop_differs_from_eval_center_crop():
+    video = _video(_structured_image())
+    random.seed(3)  # a draw that is not the exact center
+    train_frame = np.asarray(
+        _step(training=True)._build_sample_images(video, batch_size=1, target_device=None)[0][0]
+    )
+    eval_frame = np.asarray(
+        _step(training=False)._build_sample_images(video, batch_size=1, target_device=None)[0][0]
+    )
+    assert not np.array_equal(train_frame, eval_frame)
+
+
+def test_training_crop_is_disabled_under_no_grad():
+    video = _video(_structured_image())
+    with torch.no_grad():
+        no_grad_frame = np.asarray(
+            _step(training=True)._build_sample_images(video, batch_size=1, target_device=None)[0][0]
+        )
+    eval_frame = np.asarray(
+        _step(training=False)._build_sample_images(video, batch_size=1, target_device=None)[0][0]
+    )
+    np.testing.assert_array_equal(no_grad_frame, eval_frame)
+
+
+def test_training_mode_is_not_serialized():
+    step = _step(training=True)
+    serialized = step.get_config()
+    assert "training" not in serialized
+    restored = GrootN17VLMEncodeStep(**serialized)
+    assert restored.training is False
+
+
+def test_training_crop_respects_global_seed():
+    video = _video(_structured_image())
+
+    def draw():
+        random.seed(11)
+        return np.asarray(
+            _step(training=True)._build_sample_images(video, batch_size=1, target_device=None)[0][0]
+        )
+
+    np.testing.assert_array_equal(draw(), draw())

tests/policies/groot/test_groot_training_optim_contract.py

@@ -0,0 +1,121 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Isaac-GR00T N1.7 optimizer/scheduler/precision training contract.
+
+Pins the LeRobot GR00T fine-tuning recipe to the native Isaac-GR00T contract:
+AdamW(lr=1e-4, betas=(0.9, 0.999), eps=1e-8, weight_decay=1e-5, grad clip 1.0),
+HF cosine schedule with ~5% warmup over the actual update count, FP32 master
+parameters under BF16 autocast, transformers-style weight-decay grouping, the
+frozen LM-head weight tie, and episode-tail exclusion for incomplete chunks.
+"""
+
+import pytest
+import torch
+
+from lerobot.optim.schedulers import DiffuserSchedulerConfig
+from lerobot.policies.groot.configuration_groot import GrootConfig
+from lerobot.policies.groot.groot_n1_7 import _tie_unused_qwen_lm_head
+from lerobot.policies.groot.modeling_groot import GrootPolicy
+
+
+def test_groot_n1_7_optimizer_matches_isaac_training_contract():
+    optimizer = GrootConfig().get_optimizer_preset()
+
+    assert optimizer.lr == pytest.approx(1e-4)
+    assert optimizer.betas == pytest.approx((0.9, 0.999))
+    assert optimizer.eps == pytest.approx(1e-8)
+    assert optimizer.weight_decay == pytest.approx(1e-5)
+    assert optimizer.grad_clip_norm == pytest.approx(1.0)
+
+
+def test_groot_n1_7_sampler_excludes_incomplete_action_tails():
+    config = GrootConfig(chunk_size=16, n_action_steps=16)
+
+    assert len(config.action_delta_indices) == 16
+    assert config.drop_n_last_frames == 15
+
+
+def test_groot_n1_7_scheduler_matches_isaac_hf_cosine_contract():
+    config = GrootConfig(max_steps=20_000)
+    scheduler_config = config.get_scheduler_preset()
+
+    assert isinstance(scheduler_config, DiffuserSchedulerConfig)
+    assert scheduler_config.name == "cosine"
+    assert scheduler_config.num_warmup_steps == 1_000
+
+    parameter = torch.nn.Parameter(torch.ones(()))
+    optimizer = torch.optim.AdamW([parameter], lr=config.optimizer_lr)
+    scheduler = scheduler_config.build(optimizer, num_training_steps=20_000)
+    lr_factor = scheduler.lr_lambdas[0]
+
+    assert lr_factor(0) == pytest.approx(0.0)
+    assert lr_factor(1_000) == pytest.approx(1.0)
+    assert lr_factor(10_500) == pytest.approx(0.5)
+    assert lr_factor(20_000) == pytest.approx(0.0, abs=1e-12)
+
+
+def test_groot_n1_7_scheduler_rounds_fractional_warmup_up_like_transformers():
+    scheduler_config = GrootConfig(max_steps=777).get_scheduler_preset()
+
+    assert scheduler_config.num_warmup_steps == 39
+
+
+def test_groot_n1_7_model_parameters_use_fp32_checkpoint_and_optimizer_precision():
+    module = torch.nn.Module()
+    module.trainable = torch.nn.Parameter(torch.ones(3, dtype=torch.bfloat16))
+    module.frozen = torch.nn.Parameter(torch.ones(3, dtype=torch.bfloat16), requires_grad=False)
+
+    GrootPolicy._cast_model_parameters_to_fp32(module)
+
+    assert module.trainable.dtype == torch.float32
+    assert module.frozen.dtype == torch.float32
+
+
+def test_groot_n1_7_ties_unused_qwen_lm_head_to_frozen_input_embeddings():
+    class DummyQwen(torch.nn.Module):
+        def __init__(self):
+            super().__init__()
+            self.embed_tokens = torch.nn.Embedding(7, 3)
+            self.lm_head = torch.nn.Linear(3, 7, bias=False)
+
+        def get_input_embeddings(self):
+            return self.embed_tokens
+
+    model = DummyQwen()
+    _tie_unused_qwen_lm_head(model)
+
+    assert model.lm_head.weight is model.embed_tokens.weight
+    assert len(list(model.parameters())) == 1
+
+
+def test_groot_n1_7_optimizer_groups_match_transformers_weight_decay_rules():
+    module = torch.nn.Module()
+    module.linear = torch.nn.Linear(3, 2)
+    module.norm = torch.nn.LayerNorm(2)
+    module.frozen = torch.nn.Parameter(torch.ones(1), requires_grad=False)
+
+    groups = GrootPolicy._build_weight_decay_parameter_groups(module)
+
+    assert len(groups) == 2
+    assert "weight_decay" not in groups[0]
+    assert groups[1]["weight_decay"] == 0.0
+    assert groups[0]["params"] == [module.linear.weight]
+    assert {id(parameter) for parameter in groups[1]["params"]} == {
+        id(module.linear.bias),
+        id(module.norm.weight),
+        id(module.norm.bias),
+    }

tests/policies/groot/test_groot_vs_original.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 
-# Copyright 2025 The HuggingFace Inc. team. All rights reserved.
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,431 +14,194 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""Test script to verify Groot policy integration with LeRobot vs the original implementation, only meant to be run locally!"""
+"""Parity test: original NVIDIA GR00T N1.7 vs the GR00T N1.7 integration in LeRobot.
+
+Verifies that the self-contained LeRobot reimplementation of the GR00T N1.7 action
+head + Qwen3-VL backbone produces the SAME raw model output (``action_pred``, the
+normalized flow-matching prediction before any action decoding) as NVIDIA's original
+``gr00t`` package, given byte-identical pre-processed inputs and the same
+flow-matching seed. The comparison is parametrized over every embodiment tag present
+in the checkpoint.
+
+To keep the comparison fair, the original outputs + the exact collated inputs are
+produced once per embodiment in the original ``gr00t`` env via the companion script
+``utils/dump_original_n1_7.py`` (in the ``utils`` package next to this file) and saved
+to per-tag ``.npz`` files.
+This test discovers those artifacts, replays the identical inputs through the LeRobot
+model, and compares.
+
+This test is LOCAL-only and skips on CI, when ``gr00t``-side prerequisites are not
+present, or when no artifact has been generated. By default it looks for artifacts in
+``<this dir>/artifacts/``; override with ``GROOT_N1_7_PARITY_DIR``. See the
+"Original-vs-LeRobot parity test" section of ``src/lerobot/policies/groot/README.md``
+for the full run procedure.
+"""
 
-import gc
 import os
-from copy import deepcopy
-from typing import Any
+from pathlib import Path
 
 import numpy as np
 import pytest
 import torch
 
-from lerobot.policies.groot.configuration_groot import GrootConfig
-from lerobot.policies.groot.modeling_groot import GrootPolicy
-from lerobot.policies.groot.processor_groot import make_groot_pre_post_processors
-from lerobot.processor import PolicyProcessorPipeline
-from lerobot.types import PolicyAction
-
-pytest.importorskip("gr00t")
-pytest.importorskip("transformers")
-
 pytestmark = pytest.mark.skipif(
     os.environ.get("CI") == "true" or os.environ.get("GITHUB_ACTIONS") == "true",
-    reason="This test requires local Groot installation and is not meant for CI",
+    reason="Requires a local GR00T N1.7 checkpoint + pre-generated artifacts; not for CI.",
 )
 
+from lerobot.policies.groot.configuration_groot import GROOT_N1_7  # noqa: E402,F401
 
-from gr00t.data.dataset import ModalityConfig  # noqa: E402
-from gr00t.data.embodiment_tags import EmbodimentTag  # noqa: E402
-from gr00t.data.transform.base import ComposedModalityTransform  # noqa: E402
-from gr00t.model.policy import Gr00tPolicy  # noqa: E402
+SEED = 42
+DEVICE = os.environ.get("GROOT_PARITY_DEVICE", "cuda" if torch.cuda.is_available() else "cpu")
+ATOL = float(os.environ.get("GROOT_PARITY_ATOL", "1e-3"))
+RTOL = float(os.environ.get("GROOT_PARITY_RTOL", "1e-3"))
 
-# GR1 humanoid dimensions (from pretrained model metadata)
-# The actual GR1 robot has 44 dimensions for both state and action
-# GR00TTransform will pad state to 64 and truncate action to 32
-DUMMY_STATE_DIM = 44
-DUMMY_ACTION_DIM = 44
-DUMMY_ACTION_HORIZON = 16
-IMAGE_SIZE = 256
-DEVICE = "cpu"
-MODEL_PATH = "nvidia/GR00T-N1.5-3B"
-
-GR1_BODY_PARTS = {
-    "left_arm": 7,
-    "left_hand": 6,
-    "left_leg": 6,
-    "neck": 3,
-    "right_arm": 7,
-    "right_hand": 6,
-    "right_leg": 6,
-    "waist": 3,
-}
+# Artifact filenames are original_n1_7_<embodiment_tag>.npz
+_ARTIFACT_PREFIX = "original_n1_7_"
+_ARTIFACT_SUFFIX = ".npz"
 
 
-def cleanup_memory():
-    """Clean up GPU/MPS memory to prevent OOM errors between tests."""
-    print("\nCleaning up memory...")
-    gc.collect()
-    if torch.cuda.is_available():
-        torch.cuda.empty_cache()
-        torch.cuda.synchronize()
-    if torch.backends.mps.is_available():
-        torch.mps.empty_cache()
-    print("Memory cleanup complete.")
+def _artifact_dir() -> Path:
+    """Directory holding the per-embodiment .npz artifacts.
 
-
-def set_seed_all(seed: int):
-    """Set random seed for all RNG sources to ensure reproducibility."""
-    import random
-
-    random.seed(seed)
-    np.random.seed(seed)
-    torch.manual_seed(seed)
-
-    if torch.cuda.is_available():
-        torch.cuda.manual_seed(seed)
-        torch.cuda.manual_seed_all(seed)
-
-    # Set deterministic behavior
-    torch.backends.cudnn.deterministic = True
-    torch.backends.cudnn.benchmark = False
-    torch.use_deterministic_algorithms(True, warn_only=True)
-
-
-def instantiate_lerobot_groot(
-    from_pretrained: bool = False,
-    model_path: str = MODEL_PATH,
-) -> tuple[
-    GrootPolicy,
-    PolicyProcessorPipeline[dict[str, Any], dict[str, Any]],
-    PolicyProcessorPipeline[PolicyAction, PolicyAction],
-]:
-    """Instantiate LeRobot Groot policy with preprocessor and postprocessor."""
-    if from_pretrained:
-        policy = GrootPolicy.from_pretrained(
-            pretrained_name_or_path=model_path,
-            strict=False,
-        )
-        policy.config.embodiment_tag = "gr1"
-    else:
-        config = GrootConfig(
-            base_model_path=model_path,
-            n_action_steps=DUMMY_ACTION_HORIZON,
-            chunk_size=DUMMY_ACTION_HORIZON,
-            image_size=[IMAGE_SIZE, IMAGE_SIZE],
-            device=DEVICE,
-            embodiment_tag="gr1",
-        )
-        policy = GrootPolicy(config)
-
-    policy.to(DEVICE)
-    policy.config.device = DEVICE
-
-    preprocessor, postprocessor = make_groot_pre_post_processors(
-        config=policy.config,
-        dataset_stats=None,  # Pass None for dataset_stats to disable normalization (original GR00T doesn't normalize)
-    )
-
-    return (policy, preprocessor, postprocessor)
-
-
-def instantiate_original_groot(
-    from_pretrained: bool = False,
-    model_path: str = MODEL_PATH,
-):
-    """Instantiate original Groot policy from NVIDIA's implementation."""
-    from gr00t.data.transform.concat import ConcatTransform
-    from gr00t.data.transform.state_action import StateActionToTensor
-    from gr00t.data.transform.video import VideoToNumpy, VideoToTensor
-    from gr00t.model.transforms import GR00TTransform
-
-    video_keys = ["video.ego_view"]
-    state_keys = [
-        "state"
-    ]  # Important: Use single concatenated "state" key (not split body parts) to match preprocessing
-    action_keys = [
-        "action.left_arm",
-        "action.right_arm",
-        "action.left_hand",
-        "action.right_hand",
-        "action.left_leg",
-        "action.right_leg",
-        "action.neck",
-        "action.waist",
-    ]
-    language_keys = ["annotation.human.action.task_description"]
-
-    modality_config = {
-        "video": ModalityConfig(
-            delta_indices=[0],  # Current frame only
-            modality_keys=video_keys,
-        ),
-        "state": ModalityConfig(
-            delta_indices=[0],
-            modality_keys=state_keys,
-        ),
-        "action": ModalityConfig(
-            delta_indices=list(range(DUMMY_ACTION_HORIZON)),
-            modality_keys=action_keys,
-        ),
-        "language": ModalityConfig(
-            delta_indices=[0],
-            modality_keys=language_keys,
-        ),
-    }
-
-    modality_transform = ComposedModalityTransform(
-        transforms=[
-            VideoToTensor(apply_to=video_keys),
-            VideoToNumpy(apply_to=video_keys),  # Convert to numpy (GR00TTransform expects numpy arrays)
-            # State is already a single concatenated key, so no StateActionToTensor needed
-            # Convert action from numpy to tensor
-            StateActionToTensor(apply_to=action_keys),
-            # Concatenate only video and actions (state is already single key)
-            ConcatTransform(
-                video_concat_order=video_keys,
-                state_concat_order=[],  # Empty:state is already single key
-                action_concat_order=action_keys,
-            ),
-            GR00TTransform(
-                max_state_dim=64,
-                max_action_dim=32,
-                state_horizon=1,
-                action_horizon=DUMMY_ACTION_HORIZON,
-                training=False,
-            ),
-        ]
-    )
-
-    policy = Gr00tPolicy(
-        model_path=model_path,
-        embodiment_tag=EmbodimentTag.GR1,
-        modality_config=modality_config,
-        modality_transform=modality_transform,
-        device=DEVICE,
-    )
-
-    return policy, modality_config, modality_transform
-
-
-def create_dummy_data(device=DEVICE):
-    """Create dummy data for testing both implementations."""
-    batch_size = 2
-    prompt = "Pick up the red cube and place it in the bin"
-    state = torch.randn(batch_size, DUMMY_STATE_DIM, dtype=torch.float32, device=device)
-
-    batch = {
-        "observation.state": state,
-        "action": torch.randn(
-            batch_size,
-            DUMMY_ACTION_HORIZON,
-            DUMMY_ACTION_DIM,
-            dtype=torch.float32,
-            device=device,  # Action ground truth (for training)
-        ),
-        "observation.images.ego_view": torch.rand(
-            batch_size,
-            3,
-            IMAGE_SIZE,
-            IMAGE_SIZE,
-            dtype=torch.float32,
-            device=device,  # Images in [0, 1] range as expected by LeRobot
-        ),
-        "task": [prompt for _ in range(batch_size)],
-    }
-
-    return batch
-
-
-def convert_lerobot_to_original_format(batch, modality_config):
-    """Convert LeRobot batch format to original Groot format.
-
-    The original Groot expects observations in this format:
-    {
-        "video.<camera_name>": np.ndarray (T, H, W, C) or (B, T, H, W, C)
-        "state.<state_component>": np.ndarray (T, D) or (B, T, D)
-        "action.<action_component>": np.ndarray (T, D) or (B, T, D)
-        "annotation.<annotation_type>": str or list[str]
-    }
+    Self-contained by default: a sibling ``artifacts/`` directory next to this test.
+    Override with ``GROOT_N1_7_PARITY_DIR`` (e.g. to point at a scratch location).
+    The directory is read-only here -- it is populated by ``utils/dump_original_n1_7.py``
+    run in the original gr00t environment; the test never creates it.
     """
-    # Original Groot expects (T, H, W, C) format for images
-    # LeRobot has (B, C, H, W) format, so we need to convert
-    observation = {}
-
-    for img_key in ["ego_view"]:
-        lerobot_key = f"observation.images.{img_key}"
-        if lerobot_key in batch:
-            img = batch[lerobot_key]
-            # Convert from (B, C, H, W) to (B, T=1, H, W, C)
-            img_np = img.permute(0, 2, 3, 1).unsqueeze(1).cpu().numpy()
-            # Convert [0, 1] to [0, 255] uint8 as expected by original
-            img_np = (img_np * 255).astype(np.uint8)
-            observation[f"video.{img_key}"] = img_np
-
-    # Important: The Original's GR00TTransform expects "state" as (B, T, D), not split body parts
-    if "observation.state" in batch:
-        state = batch["observation.state"]
-        state_np = state.unsqueeze(1).cpu().numpy()  # (B, 1, D)
-        observation["state"] = state_np
-
-    if "action" in batch:
-        action = batch["action"]
-        action_np = action.cpu().numpy()
-
-        start_idx = 0
-        for part_name, part_dim in GR1_BODY_PARTS.items():
-            end_idx = start_idx + part_dim
-            observation[f"action.{part_name}"] = action_np[:, :, start_idx:end_idx]
-            start_idx = end_idx
-
-    if "task" in batch:
-        task_list = batch["task"]
-        # GR00TTransform expects language with (B, T) shape for batched data
-        # Create a (B, T=1) array where each element is the string directly
-        bsz = len(task_list)
-        task_array = np.empty((bsz, 1), dtype=object)
-        for i in range(bsz):
-            task_array[i, 0] = task_list[i]  # Assign string directly to each (i, 0) position
-        observation["annotation.human.action.task_description"] = task_array
-
-    return observation
+    env = os.environ.get("GROOT_N1_7_PARITY_DIR")
+    if env:
+        return Path(env)
+    return Path(__file__).resolve().parent / "artifacts"
 
 
-def test_groot_original_vs_lerobot_pretrained():
-    """Test Groot original implementation vs LeRobot implementation with pretrained weights."""
-    print("Test: Groot Original vs LeRobot with Pretrained Weights (Inference)")
+def _discover_artifacts() -> list[tuple[str, Path]]:
+    """Return [(embodiment_tag, npz_path), ...] for every dumped artifact."""
+    d = _artifact_dir()
+    if not d.is_dir():
+        return []
+    out = []
+    for p in sorted(d.glob(f"{_ARTIFACT_PREFIX}*{_ARTIFACT_SUFFIX}")):
+        tag = p.name[len(_ARTIFACT_PREFIX) : -len(_ARTIFACT_SUFFIX)]
+        out.append((tag, p))
+    return out
 
-    set_seed_all(42)
 
-    lerobot_policy, lerobot_preprocessor, lerobot_postprocessor = instantiate_lerobot_groot(
-        from_pretrained=True
+def _resolve_checkpoint() -> str:
+    env = os.environ.get("GROOT_N1_7_LIBERO_CKPT")
+    if env:
+        if not Path(env).exists():
+            pytest.skip(f"GROOT_N1_7_LIBERO_CKPT={env} does not exist")
+        return env
+    try:
+        from huggingface_hub import snapshot_download
+
+        root = snapshot_download(
+            "nvidia/GR00T-N1.7-LIBERO",
+            local_files_only=True,
+            allow_patterns=["libero_10/*"],
+        )
+    except Exception as exc:  # noqa: BLE001
+        pytest.skip(f"GR00T N1.7 LIBERO checkpoint not available locally: {exc}")
+    ckpt = Path(root) / "libero_10"
+    if not (ckpt / "config.json").exists():
+        pytest.skip(f"GR00T N1.7 LIBERO checkpoint incomplete at {ckpt}")
+    return str(ckpt)
+
+
+def _load_artifact(path: Path):
+    data = np.load(path, allow_pickle=True)
+    original_action = torch.from_numpy(data["action_pred"]).float()
+    dtypes = dict(zip(data["meta_keys"].tolist(), data["meta_dtypes"].tolist(), strict=False))
+    inputs = {}
+    for key in data.files:
+        if not key.startswith("in::"):
+            continue
+        name = key[4:]
+        arr = data[key]
+        t = torch.from_numpy(np.asarray(arr))
+        declared = dtypes.get(key, "")
+        if "int" in declared or "long" in declared:
+            t = t.long()
+        inputs[name] = t
+    return original_action, inputs
+
+
+def _unflatten(inputs: dict[str, torch.Tensor]) -> dict:
+    """Rebuild the nested model-input dict from dot-prefixed flat keys."""
+    nested: dict = {}
+    for dotted, value in inputs.items():
+        parts = dotted.split(".")
+        cur = nested
+        for p in parts[:-1]:
+            cur = cur.setdefault(p, {})
+        cur[parts[-1]] = value
+    return nested.get("inputs", nested)
+
+
+@pytest.fixture(scope="module")
+def lerobot_model():
+    """Load the LeRobot GR00T N1.7 model once (fp32 + SDPA) and reuse across tags."""
+    ckpt = _resolve_checkpoint()
+    from lerobot.policies.groot.groot_n1_7 import GR00TN17
+
+    model = GR00TN17.from_pretrained(
+        ckpt,
+        tune_llm=False,
+        tune_visual=False,
+        tune_projector=False,
+        tune_diffusion_model=False,
+        tune_vlln=False,
+        transformers_loading_kwargs={"trust_remote_code": True},
     )
-    original_policy, modality_config, modality_transform = instantiate_original_groot(from_pretrained=True)
+    # fp32 + SDPA on both sides: bf16 + differing attention kernels otherwise introduce
+    # ~1e-2 numerical noise unrelated to the implementations.
+    model.compute_dtype = "float32"
+    model.config.compute_dtype = model.compute_dtype
+    model.to(device=DEVICE, dtype=torch.float32)
+    model.eval()
+    return model
 
-    batch = create_dummy_data()
-    batch_lerobot = deepcopy(batch)
 
-    print("\n[LeRobot] Running inference...")
-    lerobot_policy.eval()
-    batch_lerobot_processed = lerobot_preprocessor(batch_lerobot)
+_ARTIFACTS = _discover_artifacts()
 
-    # Important: Reset seed immediately before inference to ensure identical RNG state
-    torch.manual_seed(42)
 
-    with torch.no_grad():
-        lerobot_actions = lerobot_policy.select_action(batch_lerobot_processed)
+@pytest.mark.skipif(
+    not _ARTIFACTS,
+    reason=(
+        "No GR00T N1.7 parity artifacts found. Generate them first in the original gr00t "
+        "env:\n  .venv-original/bin/python tests/policies/groot/utils/dump_original_n1_7.py "
+        "--ckpt <ckpt> --out-dir tests/policies/groot/artifacts --device cuda"
+    ),
+)
+@pytest.mark.parametrize("embodiment_tag,artifact", _ARTIFACTS, ids=[t for t, _ in _ARTIFACTS])
+def test_groot_get_action_parity(embodiment_tag, artifact, lerobot_model):
+    """Raw model.get_action(action_pred) parity per embodiment: original vs LeRobot."""
+    original_action, flat_inputs = _load_artifact(artifact)
+    model_inputs = _unflatten(flat_inputs)
 
-    print("\n[Original] Running inference...")
-    original_policy.model.eval()
-    observation = convert_lerobot_to_original_format(batch, modality_config)
-    original_obs_transformed = modality_transform(deepcopy(observation))
+    # Align the flow-matching RNG exactly as the producer did (seed right before sampling).
+    torch.manual_seed(SEED)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed_all(SEED)
+    with torch.inference_mode():
+        out = lerobot_model.get_action(model_inputs)
+    lerobot_action = out["action_pred"].float().cpu()
 
-    # Important: Reset seed immediately before inference to ensure identical RNG state
-    torch.manual_seed(42)
+    t = min(original_action.shape[1], lerobot_action.shape[1])
+    d = min(original_action.shape[2], lerobot_action.shape[2])
+    original_action = original_action[:, :t, :d]
+    lerobot_action = lerobot_action[:, :t, :d]
 
-    with torch.no_grad():
-        original_model_output = original_policy.model.get_action(original_obs_transformed)
-        original_actions_raw = original_model_output["action_pred"]  # [2, 16, 32]
-    # Take first timestep
-    original_actions = original_actions_raw[:, 0, :].to(lerobot_actions.device).to(lerobot_actions.dtype)
-
-    print("Action Comparison:")
-    diff = lerobot_actions - original_actions
-    abs_diff = torch.abs(diff)
-
-    for batch_idx in range(lerobot_actions.shape[0]):
-        print(f"\n{'=' * 60}")
-        print(f"Batch {batch_idx}")
-        print(f"{'=' * 60}")
-        print(f"{'Idx':<5} {'LeRobot':<14} {'Original':<14} {'Difference':<14}")
-        print("-" * 60)
-        for action_idx in range(lerobot_actions.shape[1]):
-            lr_val = lerobot_actions[batch_idx, action_idx].item()
-            orig_val = original_actions[batch_idx, action_idx].item()
-            diff_val = abs(lr_val - orig_val)
-            sign = "+" if (lr_val - orig_val) > 0 else "-"
-            print(f"{action_idx:<5} {lr_val:>13.6f} {orig_val:>13.6f} {sign}{diff_val:>12.6f}")
-
-    max_diff = abs_diff.max().item()
-    tolerance = 0.001
-    assert torch.allclose(lerobot_actions, original_actions, atol=tolerance), (
-        f"Actions differ by more than tolerance ({tolerance}): max diff = {max_diff:.6f}"
+    diff = torch.abs(lerobot_action - original_action)
+    max_diff = diff.max().item()
+    print(
+        f"\n[{embodiment_tag}] shapes lerobot={tuple(lerobot_action.shape)} "
+        f"original={tuple(original_action.shape)}  "
+        f"max|diff|={max_diff:.6e}  mean|diff|={diff.mean().item():.6e}"
     )
-    print(f"\nSuccess: Actions match within tolerance ({tolerance})!")
 
-    del lerobot_policy, lerobot_preprocessor, lerobot_postprocessor
-    del original_policy, modality_config, modality_transform
-    del batch, batch_lerobot, observation
-    cleanup_memory()
-
-
-def test_groot_forward_pass_comparison():
-    """Test forward pass comparison between LeRobot and Original Groot implementations."""
-    print("Test: Forward Pass Comparison (Training Mode)")
-
-    set_seed_all(42)
-
-    lerobot_policy, lerobot_preprocessor, lerobot_postprocessor = instantiate_lerobot_groot(
-        from_pretrained=True
+    assert torch.allclose(lerobot_action, original_action, atol=ATOL, rtol=RTOL), (
+        f"GR00T N1.7 raw action_pred differs for embodiment '{embodiment_tag}' beyond "
+        f"atol={ATOL}, rtol={RTOL}: max|diff|={max_diff:.6e}"
     )
-    original_policy, modality_config, modality_transform = instantiate_original_groot(from_pretrained=True)
-
-    batch = create_dummy_data()
-    lerobot_policy.eval()
-    original_policy.model.eval()
-
-    print("\n[LeRobot] Running forward pass...")
-    batch_lerobot = deepcopy(batch)
-    batch_lerobot_processed = lerobot_preprocessor(batch_lerobot)
-
-    set_seed_all(42)
-    with torch.no_grad():
-        lerobot_loss, lerobot_metrics = lerobot_policy.forward(batch_lerobot_processed)
-
-    print(f"  Loss: {lerobot_loss.item():.6f}")
-
-    print("\n[Original] Running forward pass...")
-    observation = convert_lerobot_to_original_format(batch, modality_config)
-    transformed_obs = modality_transform(observation)
-
-    if "action" not in transformed_obs:
-        action_for_forward = batch_lerobot_processed["action"]
-        action_mask_for_forward = batch_lerobot_processed["action_mask"]
-
-        # Match action horizon if needed
-        if action_for_forward.shape[1] != original_policy.model.action_horizon:
-            if action_for_forward.shape[1] < original_policy.model.action_horizon:
-                pad_size = original_policy.model.action_horizon - action_for_forward.shape[1]
-                last_action = action_for_forward[:, -1:, :]
-                padding = last_action.repeat(1, pad_size, 1)
-                action_for_forward = torch.cat([action_for_forward, padding], dim=1)
-
-                mask_padding = torch.zeros(
-                    action_mask_for_forward.shape[0],
-                    pad_size,
-                    action_mask_for_forward.shape[2],
-                    dtype=action_mask_for_forward.dtype,
-                    device=action_mask_for_forward.device,
-                )
-                action_mask_for_forward = torch.cat([action_mask_for_forward, mask_padding], dim=1)
-            else:
-                action_for_forward = action_for_forward[:, : original_policy.model.action_horizon, :]
-                action_mask_for_forward = action_mask_for_forward[
-                    :, : original_policy.model.action_horizon, :
-                ]
-
-        transformed_obs["action"] = action_for_forward
-        transformed_obs["action_mask"] = action_mask_for_forward
-
-    set_seed_all(42)
-    with torch.no_grad():
-        original_outputs = original_policy.model.forward(transformed_obs)
-
-    original_loss = original_outputs["loss"]
-    print(f"  Loss: {original_loss.item():.6f}")
-
-    loss_diff = abs(lerobot_loss.item() - original_loss.item())
-    loss_rel_diff = loss_diff / (abs(original_loss.item()) + 1e-8) * 100
-
-    print("\nLoss Values:")
-    print(f"  LeRobot: {lerobot_loss.item():.6f}")
-    print(f"  Original: {original_loss.item():.6f}")
-    print(f"  Absolute difference: {loss_diff:.6f}")
-    print(f"  Relative difference: {loss_rel_diff:.2f}%")
-
-    del lerobot_policy, lerobot_preprocessor, lerobot_postprocessor
-    del original_policy, modality_config, modality_transform
-    del batch, batch_lerobot, observation, transformed_obs
-    cleanup_memory()

tests/policies/groot/utils/__init__.py

@@ -0,0 +1 @@
+"""Utilities shared by GR00T policy tests."""

tests/policies/groot/utils/dump_original_n1_7.py

@@ -0,0 +1,200 @@
+#!/usr/bin/env python
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+# Licensed under the Apache License, Version 2.0 (the "License").
+"""Producer (run in the ORIGINAL gr00t env): dump original GR00T N1.7 outputs + inputs.
+
+The original NVIDIA ``gr00t`` package pins ``transformers==4.57.3`` (py3.10) and its
+model-config dataclasses are incompatible with the ``transformers==5.x`` that the
+LeRobot GR00T N1.7 integration requires. The two implementations therefore cannot be
+imported in the same Python process. To keep the parity comparison FAIR, we run the
+original model in its native env here and serialize, PER EMBODIMENT TAG:
+
+  * the exact pre-processed/collated model inputs (so the LeRobot side consumes the
+    byte-identical tensors -- same image preprocessing, tokenization, normalization),
+  * the random seed used right before the flow-matching sampler,
+  * the raw ``action_pred`` tensor returned by ``model.get_action`` (normalized space,
+    before any per-implementation action decoding).
+
+Inputs are built GENERICALLY from the checkpoint metadata (no per-tag hardcoding):
+state keys + dims come from ``statistics.json``; video + language keys come from the
+processor's per-embodiment modality configs. This lets us test many embodiment tags
+from the SAME checkpoint and confirm the LeRobot integration is not overfit to
+``libero_sim``.
+
+The companion pytest (run in the LeRobot env) loads each .npz, replays the identical
+inputs + seed through the LeRobot GR00T N1.7 model, and asserts the outputs match.
+
+Usage:
+    .venv-original/bin/python tests/policies/groot/utils/dump_original_n1_7.py \
+        --ckpt <path-to-GR00T-N1.7-LIBERO/libero_10> \
+        --out-dir tests/policies/groot/artifacts \
+        [--tags libero_sim,oxe_droid_relative_eef_relative_joint,...] \
+        [--device cuda] [--seed 42]
+
+If --tags is omitted, every embodiment present in the checkpoint statistics is dumped.
+"""
+
+import argparse
+import json
+import os
+from pathlib import Path
+
+import numpy as np
+import torch
+
+IMAGE_SIZE = 256
+BATCH_SIZE = 2
+PROMPT = "pick up the black bowl and place it on the plate"
+
+
+def load_statistics(ckpt: str) -> dict:
+    with open(os.path.join(ckpt, "statistics.json")) as f:
+        return json.load(f)
+
+
+def make_observation(seed: int, video_keys, lang_key, state_spec):
+    """Build a dummy observation dict generically from the embodiment metadata."""
+    rng = np.random.default_rng(seed)
+    video = {
+        k: rng.integers(0, 256, (BATCH_SIZE, 1, IMAGE_SIZE, IMAGE_SIZE, 3), dtype=np.uint8)
+        for k in video_keys
+    }
+    # One ndarray per state key, shape (B, T=1, key_dim); dim taken from statistics.
+    # Keys with dim 0 (e.g. disabled eef on some embodiments) are still emitted as
+    # present-but-empty so the processor's state transform finds every expected key.
+    state = {k: rng.standard_normal((BATCH_SIZE, 1, dim)).astype(np.float32) for k, dim in state_spec}
+    language = {lang_key: [[PROMPT] for _ in range(BATCH_SIZE)]}
+    return {"video": video, "state": state, "language": language}
+
+
+def dump_one_tag(policy, fair_model, tag, modality_cfg, state_spec, args, out_path):
+    from gr00t.data.types import MessageType
+
+    video_keys = modality_cfg["video"].modality_keys
+    lang_key = modality_cfg["language"].modality_keys[0]
+    observation = make_observation(args.seed, video_keys, lang_key, state_spec)
+
+    # Point the policy preprocessing at this embodiment (mirrors Gr00tPolicy.__init__).
+    policy.embodiment_tag = type(policy.embodiment_tag)(tag)
+    policy.modality_configs = {
+        k: v for k, v in policy.processor.get_modality_configs()[tag].items() if k != "rl_info"
+    }
+    policy.language_key = policy.modality_configs["language"].modality_keys[0]
+
+    torch.manual_seed(args.seed)
+    np.random.seed(args.seed)
+
+    unbatched = policy._unbatch_observation(observation)
+    processed = []
+    for obs in unbatched:
+        vla = policy._to_vla_step_data(obs)
+        processed.append(policy.processor([{"type": MessageType.EPISODE_STEP.value, "content": vla}]))
+    collated = policy.collate_fn(processed)
+
+    def to_dev(x):
+        if isinstance(x, torch.Tensor) and torch.is_floating_point(x):
+            return x.to(args.device, torch.float32)
+        if isinstance(x, torch.Tensor):
+            return x.to(args.device)
+        if isinstance(x, dict):
+            return {k: to_dev(v) for k, v in x.items()}
+        return x
+
+    collated = {k: to_dev(v) for k, v in collated.items()}
+
+    torch.manual_seed(args.seed)
+    with torch.inference_mode():
+        out = fair_model.get_action(**collated)
+    action_pred = out["action_pred"].float().cpu().numpy()
+
+    flat, meta = {}, {}
+
+    def flatten(prefix, obj):
+        if isinstance(obj, torch.Tensor):
+            arr = obj.float().cpu().numpy() if torch.is_floating_point(obj) else obj.cpu().numpy()
+            flat[f"in::{prefix}"] = arr
+            meta[f"in::{prefix}"] = str(obj.dtype)
+        elif isinstance(obj, dict):
+            for k, v in obj.items():
+                flatten(f"{prefix}.{k}" if prefix else k, v)
+        elif isinstance(obj, (list, tuple)):
+            flat[f"in::{prefix}"] = np.array(obj, dtype=object)
+        else:
+            flat[f"in::{prefix}"] = np.array(obj)
+
+    flatten("", collated)
+
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    np.savez(
+        out_path,
+        action_pred=action_pred,
+        seed=np.array(args.seed),
+        device=np.array(args.device),
+        embodiment_tag=np.array(tag),
+        meta_keys=np.array(list(meta.keys()), dtype=object),
+        meta_dtypes=np.array(list(meta.values()), dtype=object),
+        **flat,
+    )
+    print(f"[{tag}] action_pred {action_pred.shape} -> {out_path.name} ({os.path.getsize(out_path)} B)")
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--ckpt", required=True)
+    ap.add_argument("--out-dir", required=True, help="directory for per-tag .npz files")
+    ap.add_argument("--tags", default="", help="comma-separated embodiment tags (default: all in stats)")
+    ap.add_argument("--device", default="cuda")
+    ap.add_argument("--seed", type=int, default=42)
+    args = ap.parse_args()
+
+    from gr00t.policy.gr00t_policy import Gr00tPolicy
+    from transformers import AutoConfig, AutoModel
+
+    stats = load_statistics(args.ckpt)
+    requested = [t.strip() for t in args.tags.split(",") if t.strip()] or list(stats.keys())
+
+    # Load the policy once (for its processor/preprocessing) on any valid tag.
+    bootstrap_tag = "libero_sim" if "libero_sim" in stats else requested[0]
+    policy = Gr00tPolicy(embodiment_tag=bootstrap_tag, model_path=args.ckpt, device=args.device)
+    all_modality = policy.processor.get_modality_configs()
+
+    # Load a FAIR model (SDPA + fp32) once and reuse across tags. Otherwise the
+    # original checkpoint default (flash_attention_2 + bf16) introduces kernel/rounding
+    # noise vs the LeRobot env (which has no flash_attn and runs SDPA).
+    cfg = AutoConfig.from_pretrained(args.ckpt, trust_remote_code=True)
+    cfg.use_flash_attention = False
+    cfg.load_bf16 = False
+    fair_model = AutoModel.from_pretrained(args.ckpt, config=cfg, trust_remote_code=True)
+    fair_model.to(device=args.device, dtype=torch.float32)
+    fair_model.eval()
+
+    out_dir = Path(args.out_dir)
+    done, skipped = [], []
+    for tag in requested:
+        if tag not in stats or tag not in all_modality:
+            print(f"[skip] {tag}: not present in checkpoint statistics/modality configs")
+            skipped.append(tag)
+            continue
+        state_spec = [(k, len(v["min"])) for k, v in stats[tag]["state"].items()]
+        try:
+            dump_one_tag(
+                policy,
+                fair_model,
+                tag,
+                all_modality[tag],
+                state_spec,
+                args,
+                out_dir / f"original_n1_7_{tag}.npz",
+            )
+            done.append(tag)
+        except Exception as exc:  # noqa: BLE001
+            print(f"[fail] {tag}: {type(exc).__name__}: {exc}")
+            skipped.append(tag)
+
+    print(f"\nDumped {len(done)} tags: {done}")
+    if skipped:
+        print(f"Skipped/failed {len(skipped)} tags: {skipped}")
+
+
+if __name__ == "__main__":
+    main()

tests/scripts/test_edit_dataset_parsing.py

@@ -27,6 +27,7 @@ from lerobot.scripts.lerobot_edit_dataset import (
     MergeConfig,
     ModifyTasksConfig,
     OperationConfig,
+    ReencodeVideosConfig,
     RemoveFeatureConfig,
     SplitConfig,
     _validate_config,
@@ -103,3 +104,47 @@ class TestOperationTypeParsing:
         )
         resolved_name = OperationConfig.get_choice_name(type(cfg.operation))
         assert resolved_name == type_name
+
+
+class TestDepthEncoderParsing:
+    """Test that the depth encoder is exposed and parsed for video operations."""
+
+    def test_reencode_has_default_depth_encoder(self):
+        cfg = parse_cfg(["--repo_id", "test/repo", "--operation.type", "reencode_videos"])
+        assert isinstance(cfg.operation, ReencodeVideosConfig)
+        # A depth encoder is configured by default so depth videos are re-encoded too.
+        assert cfg.operation.depth_encoder is not None
+        assert hasattr(cfg.operation.depth_encoder, "depth_min")
+
+    def test_reencode_parses_depth_encoder_overrides(self):
+        cfg = parse_cfg(
+            [
+                "--repo_id",
+                "test/repo",
+                "--operation.type",
+                "reencode_videos",
+                "--operation.depth_encoder.extra_options",
+                '{"x265-params": "lossless=1"}',
+                "--operation.depth_encoder.depth_max",
+                "12.0",
+                "--operation.depth_encoder.use_log",
+                "false",
+            ]
+        )
+        assert cfg.operation.depth_encoder.extra_options == {"x265-params": "lossless=1"}
+        assert cfg.operation.depth_encoder.depth_max == 12.0
+        assert cfg.operation.depth_encoder.use_log is False
+
+    def test_convert_image_to_video_parses_depth_encoder_overrides(self):
+        cfg = parse_cfg(
+            [
+                "--repo_id",
+                "test/repo",
+                "--operation.type",
+                "convert_image_to_video",
+                "--operation.depth_encoder.depth_min",
+                "0.05",
+            ]
+        )
+        assert isinstance(cfg.operation, ConvertImageToVideoConfig)
+        assert cfg.operation.depth_encoder.depth_min == 0.05

tests/utils/test_visualization_utils.py

@@ -43,6 +43,11 @@ def mock_rerun(monkeypatch):
         def __init__(self, arr):
             self.arr = arr
 
+    class DummyDepthImage:
+        def __init__(self, arr, colormap=None):
+            self.arr = arr
+            self.colormap = colormap
+
     def dummy_log(key, obj=None, **kwargs):
         # Accept either positional `obj` or keyword `entity` and record remaining kwargs.
         if obj is None and "entity" in kwargs:
@@ -55,6 +60,8 @@ def mock_rerun(monkeypatch):
         __spec__=SimpleNamespace(name="rerun", submodule_search_locations=None),
         Scalars=DummyScalar,
         Image=DummyImage,
+        DepthImage=DummyDepthImage,
+        components=SimpleNamespace(Colormap=SimpleNamespace(Viridis="viridis")),
         log=dummy_log,
         init=lambda *a, **k: None,
         spawn=lambda *a, **k: None,
@@ -225,7 +232,7 @@ def test_log_rerun_data_kwargs_only(mock_rerun):
     assert temp.value == pytest.approx(10.0)
 
     img = _obj_for(calls, "observation.gray")
-    assert type(img).__name__ == "DummyImage"
+    assert type(img).__name__ == "DummyDepthImage"  # single-channel -> DepthImage
     assert img.arr.shape == (8, 8, 1)  # remains HWC
     assert _kwargs_for(calls, "observation.gray").get("static", False) is True