Keep the pin comment dependabot-compatible

* docs(prettier): making video encoding parameters docs prettier

* chore(format): formatting code

* chore(contrast): removing poor contrast elements

.dockerignore

@@ -22,6 +22,10 @@ outputs
 rl
 media
 
+# Local virtualenvs (the image provides its own)
+.venv
+venv
+
 
 # Logging
 logs

.github/workflows/benchmark_tests.yml

@@ -167,9 +167,9 @@ jobs:
 
       # ── LIBERO TRAIN+EVAL SMOKE ──────────────────────────────────────────────
       # Train SmolVLA for 1 step (batch_size=1, dataset episode 0 only) then
-      # immediately runs eval inside the training loop (eval_freq=1, 1 episode).
+      # immediately runs eval inside the training loop (env_eval_freq=1, 1 episode).
       # Tests the full train→eval-within-training pipeline end-to-end.
-      - name: Run Libero train+eval smoke (1 step, eval_freq=1)
+      - name: Run Libero train+eval smoke (1 step, env_eval_freq=1)
         if: env.HF_USER_TOKEN != ''
         run: |
           docker run --name libero-train-smoke --gpus all \
@@ -196,7 +196,7 @@ jobs:
                 --output_dir=/tmp/train-smoke \
                 --steps=1 \
                 --batch_size=1 \
-                --eval_freq=1 \
+                --env_eval_freq=1 \
                 --eval.n_episodes=1 \
                 --eval.batch_size=1 \
                 --eval.use_async_envs=false \

.github/workflows/documentation.yml

@@ -55,7 +55,7 @@ jobs:
       github.repository == 'huggingface/lerobot'
     permissions:
       contents: read
-    uses: huggingface/doc-builder/.github/workflows/build_main_documentation.yml@2430c1ec91d04667414e2fa31ecfc36c153ea391  # main
+    uses: huggingface/doc-builder/.github/workflows/build_main_documentation.yml@e60a538eea9817ab312196d0d233604b01697265  # main
     with:
       commit_sha: ${{ github.sha }}
       package: lerobot
@@ -78,7 +78,7 @@ jobs:
     permissions:
       contents: read
       pull-requests: write
-    uses: huggingface/doc-builder/.github/workflows/build_pr_documentation.yml@2430c1ec91d04667414e2fa31ecfc36c153ea391  # main
+    uses: huggingface/doc-builder/.github/workflows/build_pr_documentation.yml@e60a538eea9817ab312196d0d233604b01697265  # main
     with:
       commit_sha: ${{ github.event.pull_request.head.sha }}
       pr_number: ${{ github.event.number }}

.pre-commit-config.yaml

@@ -65,6 +65,9 @@ repos:
         name: Format Markdown with Prettier
         types_or: [markdown, mdx]
         args: [--prose-wrap=preserve]
+        # Jinja2 model-card templates use a .md extension but contain {% ... %} /
+        # {{ ... }} tags that prettier's Markdown formatter mangles (e.g. table loops).
+        exclude: ^src/lerobot/templates/.*\.md$
 
   ##### Security #####
   - repo: https://github.com/gitleaks/gitleaks

AGENT_GUIDE.md

@@ -138,7 +138,7 @@ lerobot-replay --robot.type=so101_follower --robot.port=<FOLLOWER_PORT> --robot.
   --dataset.repo_id=${HF_USER}/my_task --dataset.episode=0
 ```
 
-**4.9 Train** (default: ACT — fastest, lowest memory). Apple silicon: `--policy.device=mps`. See §6/§7 for policy and duration.
+**4.9 Train** (default: ACT — fastest, lowest memory). Apple silicon: `--policy.device=mps`. No local GPU? Add `--job.target=<flavor>` (e.g. `a10g-small`, list them with `hf jobs hardware`) to run on Hugging Face Jobs instead. See §6/§7 for policy and duration.
 
 ```bash
 lerobot-train \

Makefile

@@ -58,7 +58,7 @@ test-act-ete-train:
 		--dataset.episodes="[0]" \
 		--batch_size=2 \
 		--steps=4 \
-		--eval_freq=2 \
+		--env_eval_freq=2 \
 		--eval.n_episodes=1 \
 		--eval.batch_size=1 \
 		--save_freq=2 \
@@ -96,7 +96,7 @@ test-diffusion-ete-train:
 		--dataset.episodes="[0]" \
 		--batch_size=2 \
 		--steps=2 \
-		--eval_freq=2 \
+		--env_eval_freq=2 \
 		--eval.n_episodes=1 \
 		--eval.batch_size=1 \
 		--save_checkpoint=true \
@@ -126,7 +126,7 @@ test-tdmpc-ete-train:
 		--dataset.episodes="[0]" \
 		--batch_size=2 \
 		--steps=2 \
-		--eval_freq=2 \
+		--env_eval_freq=2 \
 		--eval.n_episodes=1 \
 		--eval.batch_size=1 \
 		--save_checkpoint=true \
@@ -161,7 +161,7 @@ test-smolvla-ete-train:
 		--dataset.episodes="[0]" \
 		--batch_size=2 \
 		--steps=4 \
-		--eval_freq=2 \
+		--env_eval_freq=2 \
 		--eval.n_episodes=1 \
 		--eval.batch_size=1 \
 		--save_freq=2 \
@@ -178,3 +178,9 @@ test-smolvla-ete-eval:
 		--env.episode_length=5 \
 		--eval.n_episodes=1 \
 		--eval.batch_size=1
+
+# E2E annotation pipeline smoke test against a tiny in-memory fixture
+# dataset. Opt-in (not part of `make test-end-to-end`) and uses a stub VLM
+# backend, so it does not require a real model checkpoint or GPU.
+annotation-e2e:
+	uv run python -m tests.annotations.run_e2e_smoke

README.md

@@ -87,7 +87,7 @@ Learn more about it in the [LeRobotDataset Documentation](https://huggingface.co
 
 ## SoTA Models
 
-LeRobot implements state-of-the-art policies in pure PyTorch, covering Imitation Learning, Reinforcement Learning, and Vision-Language-Action (VLA) models, with more coming soon. It also provides you with the tools to instrument and inspect your training process.
+LeRobot implements state-of-the-art policies in pure PyTorch, covering Imitation Learning, Reinforcement Learning, Vision-Language-Action (VLA) models, World Models, and Reward Models, with more coming soon. It also provides you with the tools to instrument and inspect your training process.
 
 <p align="center">
   <img alt="Gr00t Architecture" src="./media/readme/VLA_architecture.jpg" width="640px">
@@ -97,17 +97,17 @@ Training a policy is as simple as running a script configuration:
 
 ```bash
 lerobot-train \
-  --policy=act \
+  --policy.type=act \
   --dataset.repo_id=lerobot/aloha_mobile_cabinet
 ```
 
-| Category                   | Models                                                                                                                                                                                                                                                                                                                                                     |
-| -------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| **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) |
-| **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)                                                                                                                                                                                                                                         |
+| Category                   | Models                                                                                                                                                                                                                                                                                                                                                                                     |
+| -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| **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 N1.7](./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), [EVO1](./docs/source/evo1.mdx) |
+| **World Models**           | [VLA-JEPA](./docs/source/vla_jepa.mdx), [LingBot-VA](./docs/source/lingbot_va.mdx), [FastWAM](./docs/source/fastwam.mdx)                                                                                                                                                                                                                                                                   |
+| **Reward Models**          | [SARM](./docs/source/sarm.mdx), [TOPReward](./docs/source/topreward.mdx), [Robometer](./docs/source/robometer.mdx)                                                                                                                                                                                                                                                                         |
 
 Similarly to the hardware, you can easily implement your own policy & leverage LeRobot's data collection, training, and visualization tools, and share your model to the HF Hub
 
@@ -136,6 +136,7 @@ Learn how to implement your own simulation environment or benchmark and distribu
 - **[X](https://x.com/LeRobotHF):** Follow us on X to stay up-to-date with the latest developments.
 - **[Robot Learning Tutorial](https://huggingface.co/spaces/lerobot/robot-learning-tutorial):** A free, hands-on course to learn robot learning using LeRobot.
 - **[T-Shirt Folding Experiment](https://huggingface.co/spaces/lerobot/robot-folding):** An end-to-end demonstration of folding t-shirts with LeRobot.
+- **[LeLab](https://github.com/huggingface/leLab):** A web interface for LeRobot — teleoperate, calibrate, record datasets, replay, and train your SO arm from the browser, no CLI required.
 
 ## Citation
 

docs/source/_toctree.yml

@@ -45,6 +45,8 @@
     title: Language Columns and Recipes
   - local: tools
     title: Tools
+  - local: annotation_pipeline
+    title: Annotation Pipeline
   - local: video_encoding_parameters
     title: Video encoding parameters
   - local: streaming_video_encoding
@@ -67,8 +69,14 @@
     title: VLA-JEPA
   - local: eo1
     title: EO-1
+  - local: lingbot_va
+    title: LingBot-VA
+  - local: fastwam
+    title: FastWAM
+  - local: evo1
+    title: EVO1
   - local: groot
-    title: NVIDIA GR00T N1.5
+    title: NVIDIA GR00T
   - local: xvla
     title: X-VLA
   - local: multi_task_dit
@@ -161,6 +169,8 @@
 - sections:
   - local: phone_teleop
     title: Phone
+  - local: isaac_teleop
+    title: Isaac Teleop
   title: "Teleoperators"
 - sections:
   - local: cameras

docs/source/annotation_pipeline.mdx

@@ -0,0 +1,291 @@
+# Annotation Pipeline
+
+`lerobot-annotate` watches each episode's video with a vision-language
+model (VLM) and writes natural-language annotations back into your
+dataset. It fills the two language columns from the
+[Language Columns and Recipes](./language_and_recipes) page —
+`language_persistent` and `language_events` — straight into
+`data/chunk-*/file-*.parquet`.
+
+In short: point it at a LeRobot dataset, and it adds subtasks, plans,
+memory, interjections, speech, and visual Q&A that a policy can be
+trained on.
+
+## How it fits together
+
+```text
+  your dataset                  lerobot-annotate
+  (LeRobot v3.1)
+        │
+        ▼
+  ┌─────────────────────────────────────────────────────┐
+  │                    read episodes                     │
+  └──────────────────────────┬──────────────────────────┘
+                             │
+        ┌────────────────────┼────────────────────┐
+        ▼                    ▼                     ▼
+  ┌──────────┐      ┌───────────────┐        ┌──────────┐       one shared Qwen-VL
+  │   plan   │      │ interjections │        │   vqa    │  ◀──   server (vLLM, OpenAI
+  └────┬─────┘      └───────┬───────┘        └────┬─────┘        API) drives all three
+       └────────────────────┼─────────────────────┘
+                            │   each module stages raw JSONL
+                            ▼   into .annotate_staging/
+                  ┌─────────────────┐
+                  │    validator    │  ◀──  checks everything
+                  └────────┬────────┘
+                           ▼
+                  ┌─────────────────┐
+                  │     writer      │
+                  └────────┬────────┘
+                           ▼
+              data/chunk-*/file-*.parquet
+              (+ meta/info.json tools)
+```
+
+Three modules (`plan`, `interjections`, `vqa`) all talk to **one** shared
+VLM. Each module stages its output to disk, a validator checks it, and a
+single writer rewrites the dataset shards in place.
+
+## What the pipeline produces
+
+Each module emits a few kinds of annotation ("styles"), routed to one of
+the two language columns:
+
+| Style / atom                                | Column                | Module          |
+| ------------------------------------------- | --------------------- | --------------- |
+| `subtask` (Pi0.7-style "how, not what")     | `language_persistent` | `plan`          |
+| `plan` (initial + refresh on interjection)  | `language_persistent` | `plan`          |
+| `memory` (MEM-style compression)            | `language_persistent` | `plan`          |
+| `task_aug` (rephrasings of the task)        | `language_persistent` | `plan`          |
+| `interjection`                              | `language_events`     | `interjections` |
+| speech tool-call atom (`style=null`, `say`) | `language_events`     | `interjections` |
+| `vqa` (user / assistant pair)               | `language_events`     | `vqa`           |
+
+### How subtasks are generated
+
+The `plan` module doesn't ask the VLM for subtasks in one shot. Instead
+it uses a two-step **describe → segment** flow:
+
+1. **Describe** — the VLM narrates only what it actually sees in the
+   chosen camera (no guessing about the task).
+2. **Segment** — that description is fed back in, and the VLM splits the
+   episode into consecutive atomic subtasks.
+
+Both passes see the episode as **timestamped contact sheets** — frames
+sampled at `frames_per_second` (0.5s by default) and packed into JPEG
+grids with each frame's time burned into its corner, so the VLM cites
+exact boundary times directly. This is far cheaper in vision tokens than
+one image per frame, so the sampling can stay dense; episodes longer than
+`max_frames_per_prompt` are split into windows at the same density and
+merged. Both prompts also carry a causal **event-boundary** definition (a
+new event starts when an object becomes held / is released / reaches a new
+location / a lid changes state / contents move) to sharpen where cuts land.
+
+The resulting spans are then stitched into a gap-free, full-episode
+cover, so **every frame has exactly one active subtask**. See
+[`run_hf_job.py`](https://github.com/huggingface/lerobot/blob/main/examples/annotations/run_hf_job.py)
+for the production settings (single camera, timestamped contact sheets,
+auto-windowed subtask generation).
+
+### Tools
+
+The writer does **not** add a `tools` column to the parquet. The tool
+catalog lives in `meta/info.json["tools"]` instead (see [Tools](./tools)).
+After every run, the pipeline makes sure the canonical `say` schema is in
+that list, keeping any tools you declared beforehand.
+
+Want to add your own tool? Edit `meta/info.json["tools"]` directly — the
+pipeline preserves whatever is already there. That makes the tool visible
+to the chat template, so the model can learn to _generate_ the call. The
+runtime layer that actually _executes_ a generated call (the `Tool`
+protocol / `TOOL_REGISTRY` under `src/lerobot/tools/`) is not part of
+this PR — the [Tools](./tools) doc marks those pieces as
+not-yet-implemented.
+
+## Running on Hugging Face Jobs
+
+Annotation runs on [Hugging Face Jobs](https://huggingface.co/docs/hub/en/jobs).
+The repo ships a launcher script you copy and tweak for your dataset:
+
+```bash
+HF_TOKEN=hf_... uv run python examples/annotations/run_hf_job.py
+```
+
+[`run_hf_job.py`](https://github.com/huggingface/lerobot/blob/main/examples/annotations/run_hf_job.py)
+starts a single-GPU `h200` job (bump it to `h200x4` for big datasets)
+that:
+
+1. installs `lerobot` (from `main`) plus the annotation extras,
+2. boots one vLLM server per GPU (using the `vllm/vllm-openai` image) and
+   drives it over the OpenAI-compatible API,
+3. runs the `plan` / `interjections` / `vqa` modules across the dataset
+   with `lerobot-annotate`,
+4. with `--push_to_hub=true`, uploads the result to `--new_repo_id` (or
+   back to `--repo_id` in place if you leave that unset).
+
+To use a different dataset, model, or hub repo, edit the `CMD` block in
+the script. Every flag there maps directly to a `lerobot-annotate` flag
+(run `lerobot-annotate --help` for the full list).
+
+## Key options
+
+These are the flags you'll reach for most often. Run
+`lerobot-annotate --help` for everything else; the defaults are tuned for
+short manipulation episodes.
+
+### Dataset in / out
+
+| Flag              | Default | What it does                                                            |
+| ----------------- | ------- | ----------------------------------------------------------------------- |
+| `--repo_id`       | —       | Hub dataset to annotate (downloaded if `--root` unset).                 |
+| `--root`          | —       | Annotate a local dataset directory instead.                             |
+| `--new_repo_id`   | —       | Push the result to a new repo (leaves the source repo untouched).       |
+| `--push_to_hub`   | `false` | Upload after annotating (to `--new_repo_id`, else back to `--repo_id`). |
+| `--only_episodes` | all     | Annotate just these episode indices (handy for a test run).             |
+| `--seed`          | `1729`  | Seeds the RNGs that pick interjection timestamps + VQA question types.  |
+
+### Which modules run
+
+Every module is on by default and can be toggled independently (set to
+`false` to skip it, e.g. to iterate on one module at a time):
+
+| Flag                      | Default | Turns off                           |
+| ------------------------- | ------- | ----------------------------------- |
+| `--plan.enabled`          | `true`  | subtasks + plan + memory + task_aug |
+| `--interjections.enabled` | `true`  | interjections + speech atoms        |
+| `--vqa.enabled`           | `true`  | the VQA pairs                       |
+
+### The VLM (`--vlm.*`)
+
+| Flag                       | Default            | What it does                                                                        |
+| -------------------------- | ------------------ | ----------------------------------------------------------------------------------- |
+| `--vlm.model_id`           | `Qwen/Qwen3.6-27B` | The model to serve and prompt.                                                      |
+| `--vlm.camera_key`         | first `images.*`   | Which camera every prompt is grounded on.                                           |
+| `--vlm.serve_command`      | auto               | The exact `vllm serve …` command (set TP size, GPU memory, `--max-model-len` here). |
+| `--vlm.parallel_servers`   | `1`                | Independent servers for round-robin routing (one per GPU).                          |
+| `--vlm.num_gpus`           | `0`                | GPUs per server (`0` = one each).                                                   |
+| `--vlm.client_concurrency` | `16`               | In-flight requests across all servers.                                              |
+| `--vlm.max_new_tokens`     | `512`              | Generation cap per call.                                                            |
+| `--vlm.temperature`        | `0.2`              | Sampling temperature.                                                               |
+
+### Subtasks / plan / memory (`--plan.*`)
+
+| Flag                            | Default    | What it does                                                                                                              |
+| ------------------------------- | ---------- | ------------------------------------------------------------------------------------------------------------------------- |
+| `--plan.frames_per_second`      | `2.0`      | Frame sampling rate for the contact sheets (`2.0` = one frame every 0.5s).                                                |
+| `--plan.max_frames_per_prompt`  | `60`       | Frame budget per VLM call. Episodes whose sampling exceeds this are auto-windowed at the same density, then stitched.     |
+| `--plan.contact_sheet_columns`  | `5`        | Columns per contact-sheet grid (`contact_sheet_frames_per_sheet` tiles, time row-major).                                  |
+| `--plan.plan_max_steps`         | `8`        | Upper bound on subtasks per episode.                                                                                      |
+| `--plan.subtask_describe_first` | `true`     | Run the describe→segment grounding pass (best subtask quality; +1 call/episode).                                          |
+| `--plan.emit_plan`              | `true`     | Emit the numbered `plan` rows (`false` = subtasks + memory only).                                                         |
+| `--plan.emit_memory`            | `true`     | Emit the `memory` rows (`false` = subtasks + plan only); symmetric to `emit_plan`.                                        |
+| `--plan.n_task_rephrasings`     | `10`       | How many `task_aug` rephrasings to emit (`0` disables).                                                                   |
+| `--plan.derive_task_from_video` | `if_short` | Use the dataset task as-is (`off`), only when it's missing/short (`if_short`), or always re-derive from video (`always`). |
+
+### Interjections + VQA
+
+| Flag                                            | Default | What it does                                               |
+| ----------------------------------------------- | ------- | ---------------------------------------------------------- |
+| `--interjections.max_interjections_per_episode` | `3`     | Cap on interjection/speech pairs per episode.              |
+| `--vqa.vqa_emission_hz`                         | `1.0`   | How often VQA pairs are emitted.                           |
+| `--vqa.restrict_to_default_camera`              | `false` | Ground VQA only on `--vlm.camera_key` (else every camera). |
+| `--executor.episode_parallelism`                | `16`    | Episodes processed concurrently within each phase.         |
+
+## Contributing new modules
+
+The pipeline is built to grow, and **contributions are very welcome** —
+a brand-new module (say, trajectory traces or affordances), a new prompt
+template, a smarter grounding flow, or quality fixes to the existing
+`plan` / `interjections` / `vqa` modules.
+
+Every module lives under
+`src/lerobot/annotations/steerable_pipeline/modules/`, shares the VLM
+client and the keyframe cache, writes its raw output to the staging
+tree, and plugs into the executor as its own phase. Got an idea? Open an
+issue or PR on [the repo](https://github.com/huggingface/lerobot).
+
+## How recipes consume the output
+
+The annotations are meant to be read by recipes (see
+[Language Columns and Recipes](./language_and_recipes)). Typically:
+
+- low-level / high-level / memory-update branches read
+  `subtask` / `plan` / `memory` from `language_persistent`.
+- an interjection-response branch reads `interjection` events plus the
+  paired speech atom (merged into one assistant turn via `tool_calls_from`)
+  and the matching `plan` refresh at the same timestamp.
+- a VQA branch reads the `(vqa, user)` and `(vqa, assistant)` pairs from
+  `language_events`.
+
+## Why state and events are split
+
+Two ideas shape the design:
+
+1. **Persistent state vs. exact events.** Persistent rows (`subtask`,
+   `plan`, `memory`) apply to the whole episode and answer "what's true
+   right now?". Event rows (`interjection`, `vqa`, speech) appear only on
+   the one frame whose timestamp matches. Timestamps are copied straight
+   from the source parquet — never recomputed in floating point.
+2. **One VLM pass.** All three modules share a single VLM client (the
+   OpenAI-compatible client talking to the job's vLLM server), so you pay
+   for one model load per dataset, not three.
+
+## Re-running a single module
+
+Each module stages its raw output to
+`<root>/.annotate_staging/episode_{N:06d}/<module>.jsonl`. This makes
+prompt iteration cheap: re-running one module overwrites only its own
+JSONL, then the writer recomposes the final parquet. Disable modules you
+don't want with `--plan.enabled=false` (and likewise
+`--interjections.enabled` / `--vqa.enabled`) to test one at a time.
+
+## What the validator checks
+
+Before the writer runs, `StagingValidator` confirms:
+
+- every event row lands exactly on a real frame timestamp;
+- no speech / interjection pairs are left orphaned;
+- `plan` is refreshed at every interjection timestamp;
+- `memory` rows fall on subtask boundaries (a warning, not an error);
+- each VQA assistant `content` is valid JSON in one of the
+  bbox / keypoint / count / attribute / spatial shapes;
+- every row goes to the column chosen by `column_for_style(style)`.
+
+Any error aborts the writer. Pass `--skip_validation=true` to override
+while debugging.
+
+## Where each module's ideas come from
+
+- **`plan` — subtasks.** Hi Robot ([Shi 2025](https://arxiv.org/abs/2502.19417))
+  for atom granularity ("pick up one piece of lettuce", "place bowl to
+  box"); Pi0.7 ([Physical Intelligence 2025](https://pi.website/pi07))
+  for "how, not what" detail.
+- **`plan` — memory.** MEM ([Torne 2026](https://arxiv.org/abs/2603.03596)):
+  keep only the minimal relevant information — preserve outcomes, drop
+  specific attributes.
+- **`interjections`.** Hi Robot's scenario taxonomy: negative task,
+  situated correction, specific constraint, preference. Speech is a
+  tool-call-only atom
+  (`tool_calls=[{type:function, function:{name:"say", arguments:{text:...}}}]`).
+- **`vqa`.** ECoT ([Zawalski 2024](https://arxiv.org/abs/2407.08693)) for
+  grounded features (pixel bounding boxes `[x_min, y_min, x_max, y_max]`,
+  keypoints) and Steerable VLA Policies
+  ([Zhao 2025](https://arxiv.org/abs/2509.07626)) for multi-abstraction
+  grounding. Pi0.7 also grounds answers across abstraction levels.
+
+When improving a module, tweak its prompt template in
+`src/lerobot/annotations/steerable_pipeline/prompts/` rather than
+rewriting from scratch.
+
+## Roughly how much it costs
+
+Per episode, the pipeline makes about `max_steps` plan calls,
+`max_interjections_per_episode` interjection calls, and
+`vqa_emission_hz × episode_seconds` VQA calls. With the defaults (8
+subtasks, 1 interjection, 1 Hz × 3 pairs) on a 30-second episode, that's
+~50 VLM calls.
+
+Storage stays small: `language_persistent` is at most tens of KB per
+episode (parquet dictionary-encodes the one entry that repeats across
+frames), and `language_events` is empty on most frames — its size scales
+with the number of emissions, not `num_frames × num_emissions`.

docs/source/bring_your_own_policies.mdx

@@ -295,11 +295,12 @@ The file names are load-bearing: the factory does lazy imports by name, and the
 
 ### Wiring
 
-Three places need to know about your policy. All by name.
+Four places need to know about your policy. All by name.
 
 1. **`policies/__init__.py`** — re-export `MyPolicyConfig` and add it to `__all__`. **Don't** re-export the modeling class; it loads lazily through the factory (so `import lerobot` stays fast).
 2. **`factory.py:get_policy_class`** — add a branch returning `MyPolicy` from a lazy import.
 3. **`factory.py:make_policy_config`** and **`factory.py:make_pre_post_processors`** — same idea, two more branches.
+4. **`templates/lerobot_modelcard_template.md` and the root `README.md`** — the template is what `push_model_to_hub` renders into the model card of every checkpoint trained with your policy: add a one-line description of your policy in the `model_name` branches, map it in `policy_docs` so cards link to your MDX guide, and optionally add an architecture image to `diagrams`. Then add your policy to the models table in the root `README.md`, under the right category, linking to your doc page.
 
 Mirror an existing policy that's structurally similar to yours; the diff is small.
 
@@ -371,6 +372,8 @@ The general expectations are in [`CONTRIBUTING.md`](https://github.com/huggingfa
 - [ ] Optional deps live behind a `[project.optional-dependencies]` extra and the `TYPE_CHECKING + require_package` guard.
 - [ ] `tests/policies/` updated; backward-compat artifact committed & policy-specific tests.
 - [ ] `src/lerobot/policies/<name>/README.md` symlinked into `docs/source/policy_<name>_README.md`; user-facing `docs/source/<name>.mdx` written and added to `_toctree.yml`.
+- [ ] `templates/lerobot_modelcard_template.md` has a description entry and a `policy_docs` link for your policy.
+- [ ] The models table in the root `README.md` lists your policy in the right category, linking to your doc page.
 - [ ] At least one reproducible benchmark eval in the policy MDX with a published checkpoint (sim benchmark, or real-robot dataset + checkpoint).
 
 The fastest way to get a clean PR is to copy the directory of the existing policy closest to yours, rename, and replace contents method by method. Don't wait until everything is polished — open a draft PR early and iterate with us; reviewers would much rather give feedback on a half-finished branch than a fully-merged one.

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:
@@ -120,6 +150,14 @@ lerobot-train \
     --steps=20000
 ```
 
+No local GPU? Add `--job.target=<flavor>` (e.g. `a10g-small`) to either command and `lerobot-train` runs it on [Hugging Face Jobs](https://huggingface.co/docs/hub/jobs) instead — it uploads a local-only dataset for you and pushes the trained model. List flavors with `hf jobs hardware`.
+
+To resume, point `--config_path` at a checkpoint and add `--resume=true`. It accepts a local path or a Hub repo id (the latest checkpoint is fetched), and works locally or on a job by adding `--job.target=<flavor>`:
+
+```bash
+lerobot-train --config_path=${HF_USER}/policy_test --resume=true --job.target=a10g-small
+```
+
 ### Inference
 
 Inference means running the trained policy/model on a robot. For that we use `lerobot-rollout`. You will need to provide a path to your policy. It can be a local path or a path to Hugging Face for example "lerobot/folding_latest". Your cameras configuration needs to match what was used when collecting the dataset. Duration is in seconds if unspecified, it will run forever.

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/evo1.mdx

@@ -0,0 +1,191 @@
+# EVO1
+
+EVO1 is a Vision-Language-Action policy for robot control built around an InternVL3 backbone and a continuous flow-matching action head. This LeRobot integration exposes EVO1 as a standard policy type so it can be trained and evaluated with the usual LeRobot dataset, checkpoint, and processor APIs.
+
+## Model Overview
+
+The policy embeds one or more camera images and the language task prompt with InternVL3, pads robot state/action vectors to fixed maximum dimensions, and predicts future action chunks with a flow-matching action head. During inference, the policy samples an action chunk and returns `n_action_steps` actions from that chunk before sampling again.
+
+### What the LeRobot Integration Covers
+
+- Standard `policy.type=evo1` configuration through LeRobot
+- InternVL3 image/text embedding with optional FlashAttention fallback
+- Stage-based finetuning controls for action-head-only and VLM finetuning runs
+- Continuous flow-matching action prediction
+- Checkpoint save/load through LeRobot policy APIs
+- Training with `lerobot-train` and evaluation with standard policy inference APIs
+
+The broader EVO1 project may include additional training scripts and dataset tooling. This page focuses on the LeRobot robot-control policy path.
+
+## Installation Requirements
+
+1. Install LeRobot by following the [Installation Guide](./installation).
+2. Install EVO1 dependencies:
+
+   ```bash
+   pip install -e ".[evo1]"
+   ```
+
+   For LIBERO evaluation, install the LIBERO extra as well:
+
+   ```bash
+   pip install -e ".[evo1,libero]"
+   ```
+
+3. Install a `flash-attn` wheel only if it is compatible with your Python, PyTorch, CUDA, and GPU stack. EVO1 falls back to standard attention when `flash_attn` is not available.
+
+EVO1 uses the native Hugging Face `transformers` InternVL implementation, so `policy.vlm_model_name` must point to a natively converted checkpoint such as `OpenGVLab/InternVL3-1B-hf` (note the `-hf` suffix). The first run may download the configured VLM checkpoint unless `policy.vlm_model_name` points to a local model directory.
+
+## Data Requirements
+
+EVO1 expects a LeRobot dataset with:
+
+- One to `policy.max_views` visual observations, for example `observation.images.image`
+- `observation.state`
+- `action`
+- A language task instruction in the dataset `task` field, or another field configured with `policy.task_field`
+
+State and action vectors are padded to `policy.max_state_dim` and `policy.max_action_dim`. Predictions are cropped back to the dataset action dimension before being returned.
+
+## Usage
+
+To use EVO1 in a LeRobot configuration, specify:
+
+```python
+policy.type=evo1
+```
+
+By default, a new EVO1 policy initializes its VLM from:
+
+```python
+policy.vlm_model_name=OpenGVLab/InternVL3-1B-hf
+```
+
+Once a LeRobot-format EVO1 checkpoint is available, load it with:
+
+```python
+policy.path=your-org/your-evo1-checkpoint
+```
+
+## Training
+
+### Stage 1
+
+Stage 1 freezes the VLM and trains the action head:
+
+```bash
+lerobot-train \
+  --dataset.repo_id=your_org/your_dataset \
+  --policy.type=evo1 \
+  --policy.training_stage=stage1 \
+  --policy.vlm_model_name=OpenGVLab/InternVL3-1B-hf \
+  --policy.device=cuda \
+  --policy.chunk_size=50 \
+  --policy.n_action_steps=50 \
+  --policy.max_state_dim=24 \
+  --policy.max_action_dim=24 \
+  --policy.optimizer_lr=1e-5 \
+  --batch_size=4 \
+  --steps=5000 \
+  --output_dir=./outputs/evo1_stage1
+```
+
+### Stage 2
+
+Stage 2 finetunes the VLM branches and action head. A common workflow starts from a Stage 1 checkpoint:
+
+```bash
+lerobot-train \
+  --dataset.repo_id=your_org/your_dataset \
+  --policy.path=./outputs/evo1_stage1/checkpoints/005000/pretrained_model \
+  --policy.training_stage=stage2 \
+  --policy.vlm_model_name=OpenGVLab/InternVL3-1B-hf \
+  --policy.device=cuda \
+  --policy.chunk_size=50 \
+  --policy.n_action_steps=50 \
+  --policy.max_state_dim=24 \
+  --policy.max_action_dim=24 \
+  --policy.optimizer_lr=1e-5 \
+  --batch_size=4 \
+  --steps=80000 \
+  --output_dir=./outputs/evo1_stage2
+```
+
+By default, `policy.training_stage` reapplies the finetuning defaults for that stage. This is important when
+starting Stage 2 from a Stage 1 checkpoint, because the Stage 1 checkpoint config stores the VLM finetuning
+flags as disabled. These stage defaults take precedence over saved or manually supplied `policy.finetune_*`
+flags unless `policy.apply_training_stage_defaults=false`, so set that flag only when manually controlling
+every finetuning flag.
+
+### Key Training Parameters
+
+| Parameter                                     | Default                     | Description                                                       |
+| --------------------------------------------- | --------------------------- | ----------------------------------------------------------------- |
+| `policy.vlm_model_name`                       | `OpenGVLab/InternVL3-1B-hf` | Natively converted InternVL3 checkpoint or local model directory  |
+| `policy.training_stage`                       | `stage1`                    | `stage1` trains the action head; `stage2` finetunes VLM branches  |
+| `policy.apply_training_stage_defaults`        | `true`                      | Reapplies stage finetuning defaults after loading a checkpoint    |
+| `policy.vlm_num_layers`                       | `14`                        | Number of InternVL3 language layers kept for the policy           |
+| `policy.vlm_dtype`                            | `bfloat16`                  | Requested VLM dtype                                               |
+| `policy.use_flash_attn`                       | `true`                      | Requests FlashAttention when installed; otherwise falls back      |
+| `policy.enable_gradient_checkpointing`        | `true`                      | Enables checkpointing on supported InternVL3 modules              |
+| `policy.gradient_checkpointing_use_reentrant` | `false`                     | Reentrant setting passed to gradient checkpointing when supported |
+| `policy.chunk_size`                           | `50`                        | Number of future actions predicted per chunk                      |
+| `policy.n_action_steps`                       | `50`                        | Number of actions consumed from a sampled chunk                   |
+| `policy.max_state_dim`                        | `24`                        | State padding dimension                                           |
+| `policy.max_action_dim`                       | `24`                        | Action padding dimension                                          |
+| `policy.postprocess_action_dim`               | `null`                      | Optional action dimension returned after EVO1 postprocessing      |
+| `policy.binarize_gripper`                     | `false`                     | Binarizes the postprocessed gripper channel for LIBERO-style eval |
+| `policy.task_field`                           | `task`                      | Batch field used as the language prompt                           |
+
+## Inference
+
+Try it out with a trained EVO1 checkpoint:
+
+```bash
+lerobot-rollout \
+  --policy.path=your-org/your-evo1-checkpoint \
+  --inference.type=rtc \ # optional
+  ...
+```
+
+## Results
+
+### LIBERO Evaluation
+
+> [!NOTE]
+> Benchmark results for a `lerobot`-hosted LIBERO checkpoint trained with this implementation
+> will be added once training completes.
+
+The official EVO1 LIBERO rollout protocol uses the raw LIBERO camera feature names
+(`observation.images.agentview_image` and `observation.images.robot0_eye_in_hand_image`), replans every
+14 actions, and binarizes the gripper command before stepping the simulator. The EVO1 policy postprocessor
+can crop the padded 24D action back to the 7D LIBERO action space and apply that gripper binarization. To
+evaluate a LIBERO checkpoint under the same one-episode-per-task setting, keep the raw camera names instead
+of the default `image`/`image2` mapping and set the LIBERO action postprocessing flags:
+
+```bash
+lerobot-eval \
+  --policy.path=your-org/your-evo1-libero-checkpoint \
+  --policy.vlm_model_name=OpenGVLab/InternVL3-1B-hf \
+  --policy.device=cuda \
+  --policy.use_flash_attn=true \
+  --policy.n_action_steps=14 \
+  --policy.postprocess_action_dim=7 \
+  --policy.binarize_gripper=true \
+  --env.type=libero \
+  --env.task=libero_object \
+  --env.camera_name_mapping="{agentview_image: agentview_image, robot0_eye_in_hand_image: robot0_eye_in_hand_image}" \
+  --env.observation_height=448 \
+  --env.observation_width=448 \
+  --eval.batch_size=1 \
+  --eval.n_episodes=1
+```
+
+## References
+
+- [EVO1 repository](https://github.com/MINT-SJTU/Evo-1)
+- [InternVL3-1B-hf](https://huggingface.co/OpenGVLab/InternVL3-1B-hf)
+
+## License
+
+This LeRobot integration follows the Apache 2.0 License used by LeRobot. Check the upstream EVO1 and InternVL3 model pages for the licenses of released checkpoints and data.

docs/source/fastwam.mdx

@@ -0,0 +1,167 @@
+# FastWAM
+
+FastWAM is a World Action Model policy for robot control. The LeRobot integration exposes FastWAM through the standard policy API so it can be configured with `policy.type=fastwam`, trained with `lerobot-train`, and loaded through the LeRobot pretrained policy interface.
+
+## Model Overview
+
+FastWAM keeps video modeling during training, but uses direct action prediction at inference time instead of iteratively generating future observations. This LeRobot policy wraps the FastWAM action model, adapts LeRobot batches to FastWAM training samples, and provides the standard processor pipeline for normalization and action postprocessing.
+
+The implementation initializes the visual world-model components from `Wan-AI/Wan2.2-TI2V-5B` by default and predicts action chunks with shape `[batch, action_horizon, action_dim]`.
+
+### What the LeRobot Integration Covers
+
+- Standard `policy.type=fastwam` configuration through LeRobot
+- Image, state, action, and language-task batch adaptation
+- Action chunk inference through `select_action` and `predict_action_chunk`
+- Checkpoint save/load through the LeRobot policy APIs
+- Configurable LIBERO gripper action postprocessing
+
+## Installation Requirements
+
+Install LeRobot from source, then install FastWAM dependencies:
+
+```bash
+pip install -e ".[fastwam]"
+```
+
+This installs the FastWAM policy extra from `pyproject.toml`: `transformers`,
+`diffusers`, `ftfy`, and `regex`, plus LeRobot's base dependencies.
+
+For LIBERO evaluation, install the benchmark dependencies too:
+
+```bash
+pip install -e ".[fastwam,libero]"
+```
+
+This installs both extras. In addition to the FastWAM dependencies above, the
+`libero` extra installs LeRobot dataset dependencies, `hf-libero` on Linux, and
+`scipy`.
+
+FastWAM uses the Wan2.2 TI2V backbone. The default model id is:
+
+```python
+policy.model_id=Wan-AI/Wan2.2-TI2V-5B
+```
+
+## Data Requirements
+
+FastWAM expects a LeRobot dataset with:
+
+- one or more visual observations whose widths concatenate to `policy.image_size[1]`
+- `observation.state` when `policy.proprio_dim` is not `None`
+- `action`
+- a language task instruction through the dataset task field, or precomputed `context` and `context_mask` tensors
+
+The default visual setup is one image feature named `observation.images.image` with shape `(3, 224, 448)`. If the dataset uses two cameras, configure `policy.input_features` so their heights match `224` and their widths sum to `448`.
+
+## Usage
+
+Create a new FastWAM policy with:
+
+```bash
+lerobot-train \
+  --dataset.repo_id=your-org/your-dataset \
+  --policy.type=fastwam \
+  --policy.action_dim=7 \
+  --policy.proprio_dim=8 \
+  --policy.action_horizon=32 \
+  --policy.n_action_steps=10 \
+  --policy.image_size='[224,448]' \
+  --output_dir=./outputs/fastwam_training \
+  --job_name=fastwam_training \
+  --steps=300000 \
+  --batch_size=8 \
+  --policy.device=cuda
+```
+
+Evaluate an existing LeRobot-format checkpoint on LIBERO-10 with:
+
+```bash
+lerobot-eval \
+  --policy.path=ZibinDong/fastwam_libero_uncond_2cam224 \
+  --policy.device=cuda \
+  --policy.torch_dtype=float32 \
+  --policy.n_action_steps=10 \
+  --env.type=libero \
+  --env.task=libero_10 \
+  --env.observation_height=224 \
+  --env.observation_width=224 \
+  --eval.batch_size=1 \
+  --eval.n_episodes=50 \
+  --seed=0 \
+  --env.episode_length=600
+```
+
+For `libero_goal`, `libero_spatial`, and `libero_object`, use
+`--env.episode_length=300`.
+
+For real-robot rollout, use the same checkpoint path:
+
+```bash
+lerobot-rollout \
+  --robot.type=so101_follower \
+  --robot.port=/dev/ttyACM0 \
+  --policy.path=your-org/fastwam-real-robot
+```
+
+## Configuration Notes
+
+### Image Features
+
+`policy.image_size` is the size of the concatenated FastWAM image tensor as `(height, width)`. Each configured image feature must have shape `(3, height, camera_width)`, and all camera widths must sum to the configured width.
+
+### Action Chunking
+
+`policy.action_horizon` controls the number of future actions supervised during training and predicted during inference. `policy.n_action_steps` controls how many actions are consumed before the policy predicts a fresh chunk. `policy.n_action_steps` must be less than or equal to `policy.action_horizon`.
+
+### Wan Components
+
+FastWAM loads the Wan VAE, video DiT, text encoder, and tokenizer from the configured Wan model directory or Hugging Face Hub model id. LeRobot-format FastWAM checkpoints saved by `save_pretrained` also copy the local Wan component files needed by `from_pretrained`.
+
+### Attention Backend
+
+FastWAM's DiT uses PyTorch's `scaled_dot_product_attention` (SDPA) for all attention. It does **not** use FlashAttention: its Mixture-of-Transformers (MoT) routing needs arbitrary boolean `[query, key]` attention masks, which the FlashAttention varlen API cannot express. Installing the `flash-attn` package therefore has no effect on the FastWAM path. (Note that SDPA itself may still select PyTorch's own flash / memory-efficient / math kernel internally — this is unrelated to the `flash-attn` package.)
+
+### LIBERO Action Toggle
+
+FastWAM LIBERO checkpoints use `policy.toggle_action_dimensions=[-1]` by
+default to match the gripper action convention used by the original FastWAM
+evaluation pipeline:
+
+```bash
+--policy.toggle_action_dimensions='[-1]'
+```
+
+## Results
+
+Evaluated on LIBERO with [`ZibinDong/fastwam_libero_uncond_2cam224`](https://huggingface.co/ZibinDong/fastwam_libero_uncond_2cam224):
+
+| Suite          | Success rate | n_episodes |
+| -------------- | -----------: | ---------: |
+| libero_spatial |        97.6% |        500 |
+| libero_object  |        99.0% |        500 |
+| libero_goal    |        95.0% |        500 |
+| libero_10      |        94.0% |        500 |
+| **average**    |    **96.4%** |       2000 |
+
+Reproduce: `lerobot-eval --policy.path=ZibinDong/fastwam_libero_uncond_2cam224 --policy.device=cuda --policy.torch_dtype=float32 --policy.n_action_steps=10 --env.type=libero --env.task=libero_spatial --env.observation_height=256 --env.observation_width=256 --eval.batch_size=1 --eval.n_episodes=50 --seed=0 --env.episode_length=300` (1x H20 140 GB).
+
+## References
+
+- [Fast-WAM paper](https://arxiv.org/abs/2603.16666)
+- [Fast-WAM project page](https://yuantianyuan01.github.io/FastWAM/)
+- [Fast-WAM code](https://github.com/yuantianyuan01/FastWAM)
+- [Released upstream checkpoints](https://huggingface.co/yuanty/fastwam)
+- [Wan2.2 TI2V 5B](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B)
+
+## Citation
+
+```bibtex
+@article{yuan2026fastwam,
+  title = {Fast-WAM: Do World Action Models Need Test-time Future Imagination?},
+  author = {Tianyuan Yuan and Zibin Dong and Yicheng Liu and Hang Zhao},
+  journal = {arXiv preprint arXiv:2603.16666},
+  year = {2026},
+  url = {https://arxiv.org/abs/2603.16666}
+}
+```

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 +57,171 @@ 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="$IMAGE_TRANSFORMS" \
+  --policy.type=groot \
+  --policy.base_model_path=nvidia/GR00T-N1.7-3B \
+  --policy.embodiment_tag=libero_sim \
+  --policy.push_to_hub=false \
+  --policy.use_relative_actions=false \
+  --policy.max_steps=20000 \
+  --batch_size=320 \
+  --steps=20000 \
+  --save_freq=2000 \
+  --env_eval_freq=0 \
+  --eval_steps=0 \
+  --log_freq=10 \
+  --wandb.enable=true \
+  --wandb.project=lerobot \
+  --wandb.mode=online \
+  --wandb.disable_artifact=true \
+  --num_workers=4 \
+  --prefetch_factor=2 \
+  --persistent_workers=true \
+  --output_dir=$OUTPUT_DIR \
+  --job_name=$JOB_NAME
+```
+
+This will follow the recipe found [here](https://github.com/NVIDIA/Isaac-GR00T/blob/main/examples/LIBERO/README.md).
+
+### GR00T N1.7 LIBERO Results
+
+Preliminary LeRobot integration results (GR00T-LeRobot, `eval.n_episodes >= 50` per suite):
+
+| Suite            | Success rate | Checkpoint                                                                                                    |
+| ---------------- | -----------: | ------------------------------------------------------------------------------------------------------------- |
+| LIBERO Spatial   |          91% | [nvidia/gr00t17-lerobot-libero_spatial-640](https://huggingface.co/nvidia/gr00t17-lerobot-libero_spatial-640) |
+| LIBERO Object    |          81% | [nvidia/gr00t17-lerobot-libero_object-640](https://huggingface.co/nvidia/gr00t17-lerobot-libero_object-640)   |
+| LIBERO Goal      |          97% | [nvidia/gr00t17-lerobot-libero_goal-640](https://huggingface.co/nvidia/gr00t17-lerobot-libero_goal-640)       |
+| LIBERO 10 (Long) |          84% | [nvidia/gr00t17-lerobot-libero_10-640](https://huggingface.co/nvidia/gr00t17-lerobot-libero_10-640)           |
+| **Average**      |   **88.25%** |                                                                                                               |
+
+```bash
+export MODEL_ID=your_trained_model_on_huggingface
+
+lerobot-eval \
+  --policy.type=groot \
+  --policy.base_model_path=$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=True \ # set to False if it causes inference instability
+  --inference.rtc.execution_horizon=8 \
+  --inference.queue_threshold=0
 ```
 
+> [!NOTE]
+> Value of `inference.queue_threshold` should not exceed 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/hardware_guide.mdx

@@ -82,17 +82,18 @@ VRAM is the first filter. Within a tier, pick by budget and availability — the
 
 ### Hugging Face Jobs
 
-[Hugging Face Jobs](https://huggingface.co/docs/hub/jobs) lets you run training on managed HF infrastructure, billed by the second. The repo publishes a ready-to-use image: **`huggingface/lerobot-gpu:latest`**, rebuilt **every night at 02:00 UTC from `main`** ([`docker_publish.yml`](https://github.com/huggingface/lerobot/blob/main/.github/workflows/docker_publish.yml)) — so it tracks the current state of the repo, not a tagged release.
+[Hugging Face Jobs](https://huggingface.co/docs/hub/jobs) lets you run training on managed HF infrastructure, billed by the second, without owning a GPU. `lerobot-train` submits and streams the job for you — just add `--job.target=<flavor>` to a normal training command:
 
 ```bash
-hf jobs run --flavor a10g-large huggingface/lerobot-gpu:latest \
-  bash -c "nvidia-smi && lerobot-train \
-    --policy.type=act --dataset.repo_id=<USER>/<DATASET> \
-    --policy.repo_id=<USER>/act_<task> --batch_size=8 --steps=50000"
+lerobot-train \
+  --policy.type=act --dataset.repo_id=<USER>/<DATASET> \
+  --policy.repo_id=<USER>/act_<task> \
+  --job.target=a10g-large
 ```
 
 Notes:
 
-- The leading `nvidia-smi` is a quick sanity check that CUDA is visible inside the container — useful to fail fast if the flavor or driver mismatched.
-- The default Job timeout is 30 minutes; pass `--timeout 4h` (or longer) for real training.
-- `--flavor` maps onto the table above: `t4-small`/`t4-medium` (T4, ACT only), `l4x1`/`l4x4` (L4 24 GB), `a10g-small/large/largex2/largex4` (A10G 24 GB scaled out), `a100-large` (A100). For the current full catalogue + pricing see [https://huggingface.co/docs/hub/jobs](https://huggingface.co/docs/hub/jobs).
+- Run `hf auth login` once before submitting, the job runs under your token.
+- `--job.target` maps onto the table above: `t4-small`/`t4-medium` (T4, ACT only), `l4x1`/`l4x4` (L4 24 GB), `a10g-small/large/largex2/largex4` (A10G 24 GB scaled out), `a100-large` (A100). List the current catalogue with pricing via `hf jobs hardware`, or see [https://huggingface.co/docs/hub/jobs](https://huggingface.co/docs/hub/jobs).
+- The job defaults to a `2d` (48h) timeout. Override it with `--job.timeout=4h` (or any other valid duration string) to shorten or extend the timeout. The job automatically stops when the command completes.
+- For the full walkthrough — dataset upload, checkpoint streaming, resuming a run on a job — see the [imitation-learning training guide](./il_robots#train-using-hugging-face-jobs).

docs/source/hil_data_collection.mdx

@@ -57,11 +57,11 @@ The `lerobot-rollout --strategy.type=dagger` mode requires **teleoperators with
 
 **Compatible teleoperators:**
 
-- `openarm_mini` - OpenArm Mini
+- `bi_openarm_mini` - Bimanual OpenArm Mini
 - `so_leader` - SO100 / SO101 leader arm
 
 > [!IMPORTANT]
-> The provided commands default to `bi_openarm_follower` + `openarm_mini`.
+> The provided commands default to `bi_openarm_follower` + `bi_openarm_mini`.
 > `so_follower` + `so_leader` configs are also registered and can be used via CLI flags.
 
 ---
@@ -104,9 +104,9 @@ lerobot-rollout --strategy.type=dagger \
     --robot.right_arm_config.port=can0 \
     --robot.right_arm_config.side=right \
     --robot.cameras='{left_wrist: {type: opencv, index_or_path: "/dev/video0", width: 1280, height: 720, fps: 30}, right_wrist: {type: opencv, index_or_path: "/dev/video4", width: 1280, height: 720, fps: 30}, base: {type: opencv, index_or_path: "/dev/video2", width: 640, height: 480, fps: 30}}' \
-    --teleop.type=openarm_mini \
-    --teleop.port_left=/dev/ttyACM0 \
-    --teleop.port_right=/dev/ttyACM1 \
+    --teleop.type=bi_openarm_mini \
+    --teleop.left_arm_config.port=/dev/ttyACM0 \
+    --teleop.right_arm_config.port=/dev/ttyACM1 \
     --policy.path=outputs/pretrain/checkpoints/last/pretrained_model \
     --dataset.repo_id=your-username/rollout_hil_dataset \
     --dataset.single_task="Fold the T-shirt properly" \
@@ -131,9 +131,9 @@ lerobot-rollout --strategy.type=dagger \
     --robot.right_arm_config.port=can0 \
     --robot.right_arm_config.side=right \
     --robot.cameras='{left_wrist: {type: opencv, index_or_path: "/dev/video0", width: 1280, height: 720, fps: 30}, right_wrist: {type: opencv, index_or_path: "/dev/video4", width: 1280, height: 720, fps: 30}, base: {type: opencv, index_or_path: "/dev/video2", width: 640, height: 480, fps: 30}}' \
-    --teleop.type=openarm_mini \
-    --teleop.port_left=/dev/ttyACM0 \
-    --teleop.port_right=/dev/ttyACM1 \
+    --teleop.type=bi_openarm_mini \
+    --teleop.left_arm_config.port=/dev/ttyACM0 \
+    --teleop.right_arm_config.port=/dev/ttyACM1 \
     --policy.path=outputs/pretrain/checkpoints/last/pretrained_model \
     --dataset.repo_id=your-username/rollout_hil_rtc_dataset \
     --dataset.single_task="Fold the T-shirt properly" \

docs/source/hilserl.mdx

@@ -719,7 +719,7 @@ Example configuration for training the [reward classifier](https://huggingface.c
   "num_workers": 4,
   "steps": 5000,
   "log_freq": 10,
-  "eval_freq": 1000,
+  "env_eval_freq": 1000,
   "save_freq": 1000,
   "save_checkpoint": true,
   "seed": 2,

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

@@ -126,7 +126,7 @@ import time
 from lerobot.teleoperators.so_leader import SO101Leader, SO101LeaderConfig
 from lerobot.robots.so_follower import SO101Follower, SO101FollowerConfig
 from lerobot.cameras.opencv import OpenCVCameraConfig
-from lerobot.utils.visualization_utils import init_rerun, log_rerun_data, shutdown_rerun
+from lerobot.utils.visualization_utils import init_visualization, log_visualization_data, shutdown_visualization
 
 robot_config = SO101FollowerConfig(
     port="/dev/tty.usbmodem5AB90687491",
@@ -142,7 +142,7 @@ teleop_config = SO101LeaderConfig(
     id="my_leader_arm",
 )
 
-init_rerun(session_name="teleoperation")
+init_visualization("rerun", session_name="teleoperation")  # pass "foxglove" to stream to Foxglove instead
 
 robot = SO101Follower(robot_config)
 teleop_device = SO101Leader(teleop_config)
@@ -158,7 +158,7 @@ while True:
     observation = robot.get_observation()
     action = teleop_device.get_action()
     robot.send_action(action)
-    log_rerun_data(observation=observation, action=action)
+    log_visualization_data("rerun", observation=observation, action=action)
 
     elapsed_time = time.perf_counter() - start_time
     sleep_time = TIME_PER_FRAME - elapsed_time
@@ -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>
@@ -223,7 +223,7 @@ from lerobot.teleoperators.so_leader.config_so_leader import SO101LeaderConfig
 from lerobot.teleoperators.so_leader.so_leader import SO101Leader
 from lerobot.common.control_utils import init_keyboard_listener
 from lerobot.utils.utils import log_say
-from lerobot.utils.visualization_utils import init_rerun
+from lerobot.utils.visualization_utils import init_visualization
 from lerobot.scripts.lerobot_record import record_loop
 from lerobot.processor import make_default_processors
 
@@ -270,7 +270,7 @@ def main():
 
     # Initialize the keyboard listener and rerun visualization
     _, events = init_keyboard_listener()
-    init_rerun(session_name="recording")
+    init_visualization("rerun", session_name="recording")
 
     # Connect the robot and teleoperator
     robot.connect()
@@ -390,9 +390,17 @@ Set the flow of data recording using command-line arguments:
 
 Control the data recording flow using keyboard shortcuts:
 
-- Press **Right Arrow (`→`)**: Early stop the current episode or reset time and move to the next.
-- Press **Left Arrow (`←`)**: Cancel the current episode and re-record it.
-- Press **Escape (`ESC`)**: Immediately stop the session, encode videos, and upload the dataset.
+- Press **Right Arrow (`→`)** or **`n`**: Early stop the current episode or reset time and move to the next.
+- Press **Left Arrow (`←`)** or **`r`**: Cancel the current episode and re-record it.
+- Press **Escape (`ESC`)** or **`q`**: Immediately stop the session, encode videos, and upload the dataset.
+
+<Tip>
+
+These control-flow shortcuts work on **X11, Wayland, and headless/SSH** sessions. When a global keyboard backend isn't available (Wayland, a headless machine, or macOS without Accessibility permission), `lerobot-record` automatically reads the same keys from the terminal — launch it from an interactive terminal and keep it focused. You can also use the letter equivalents **`n`** (next, same as `→`), **`r`** (re-record, same as `←`) and **`q`** (quit, same as `ESC`). No `$DISPLAY` setup is required.
+
+This applies to the recording control flow only. Keyboard **teleoperation** (driving the robot with the keyboard) still needs a global key backend, so it works only on an X11 session, a Windows desktop, or macOS with Accessibility/Input Monitoring granted — not on Wayland or headless sessions.
+
+</Tip>
 
 #### Tips for gathering data
 
@@ -406,7 +414,7 @@ If you want to dive deeper into this important topic, you can check out the [blo
 
 #### Troubleshooting:
 
-- On Linux, if the left and right arrow keys and escape key don't have any effect during data recording, make sure you've set the `$DISPLAY` environment variable. See [pynput limitations](https://pynput.readthedocs.io/en/latest/limitations.html#linux).
+- On Linux, the recording control-flow keys (arrow keys, Escape) work on X11, Wayland, and headless/SSH sessions as long as `lerobot-record` runs in an interactive terminal — no `$DISPLAY` setup is needed. If the keys have no effect, make sure you are in an interactive (TTY) terminal, not a piped/non-TTY session, and that it is focused; the letter equivalents `n` / `r` / `q` also work. Keyboard _teleoperation_ (as opposed to the recording control flow) still requires a global key backend — an X11 session, a Windows desktop, or macOS with Accessibility/Input Monitoring granted — and is unavailable on Wayland or headless machines. See [pynput limitations](https://pynput.readthedocs.io/en/latest/limitations.html#linux).
 
 ## Visualize a dataset
 
@@ -506,6 +514,12 @@ lerobot-train \
   --resume=true
 ```
 
+`--config_path` also accepts a **Hub repo id**: if a run pushed its checkpoints to the Hub (with `--save_checkpoint_to_hub=true`), you can resume straight from the repo — its latest checkpoint is downloaded and training continues, restoring the optimizer, scheduler, step counter and data order:
+
+```bash
+lerobot-train --config_path=${HF_USER}/my_policy --resume=true
+```
+
 If you do not want to push your model to the hub after training use `--policy.push_to_hub=false`.
 
 Additionally you can provide extra `tags` or specify a `license` for your model or make the model repo `private` by adding this: `--policy.private=true --policy.tags=\[ppo,rl\] --policy.license=mit`
@@ -518,78 +532,48 @@ If your local computer doesn't have a powerful GPU you could utilize Google Cola
 
 Hugging Face jobs let's you easily select hardware and run the training in the cloud. So if you don't have a powerful GPU or you need more VRAM or just want to train a model much faster use HF Jobs! It's pay as you go and you simply pay for each second of use, you can see the pricing and additional information [here](https://huggingface.co/docs/hub/jobs).
 
-To run the training use this command:
+`lerobot-train` runs locally by default. To run on a HuggingFace GPU, pass `--job.target` with a hardware flavor name:
 
-<hfoptions id="train_with_hf_jobs">
-<hfoption id="Command">
 ```bash
-hf jobs run \
-  --flavor a10g-small \
-  --timeout 4h \
-  --secrets HF_TOKEN \
-  huggingface/lerobot-gpu:latest \
-  -- \
-  python -m lerobot.scripts.lerobot_train \
-    --dataset.repo_id=username/dataset \
-    --policy.type=act \
-    --steps=5000 \
-    --batch_size=16 \
-    --policy.device=cuda \
-    --policy.repo_id=username/your_policy \
-    --log_freq=100
+lerobot-train \
+  --dataset.repo_id=${HF_USER}/so101_test \
+  --policy.type=act \
+  --policy.repo_id=${HF_USER}/my_policy \
+  --job.target=a10g-small
 ```
-</hfoption>
-<hfoption id="API example">
 
-<!-- prettier-ignore-start -->
-```python
-from huggingface_hub import run_job, get_token
+List available flavors and pricing with `hf jobs hardware`. The run streams its logs to your terminal; press Ctrl-C to detach (the job keeps running in the cloud). Re-attach or cancel with:
 
-run_name = "act_so101_hf_jobs"
-dataset_id = "username/dataset"
-user_hub_id = "username"
-
-command_args = [
-    "python", "-m", "lerobot.scripts.lerobot_train",
-    "--dataset.repo_id", dataset_id,
-    "--policy.type", "act",
-    "--steps", "5000",
-    "--batch_size", "16",
-    "--num_workers", "4",
-    "--policy.device", "cuda",
-    "--log_freq", "100",
-    "--save_freq", "1000",
-    "--save_checkpoint", "true",
-    "--wandb.enable", "false",
-    "--policy.repo_id", f"{user_hub_id}/{run_name}"
-]
-
-print(f"Submitting job '{run_name}' to Hugging Face Infrastructure...")
-
-job_info = run_job(
-    image="huggingface/lerobot-gpu:latest",
-    command=command_args,
-    flavor="a10g-small",
-    timeout="4h",
-    secrets={"HF_TOKEN": get_token()}
-)
-
-print("\n🚀 Job successfully launched!")
-print(f"🔹 Job ID: {job_info.id}")
-print(f"🔗 Live UI Dashboard & Logs: {job_info.url}")
+```bash
+hf jobs logs <job-id>
+hf jobs cancel <job-id>
 ```
-<!-- prettier-ignore-end -->
 
-</hfoption>
-</hfoptions>
+If your dataset exists only locally (not yet on the Hub), it is automatically pushed to a **private** Hub repo so the job can download it by `repo_id` (nothing is made public). The trained model is pushed to the model repo at the end of the run. To also push every intermediate checkpoint to the Hub as it is saved (so you can monitor progress mid-run), add `--save_checkpoint_to_hub=true` — this requires a runtime image that includes this feature.
 
-You can modify the `--flavor` to use different hardware, for example: `t4-small`, `a100-large`, `h200`. Use `hf jobs hardware` to see the full list with pricing.
-Depending on the model you want to train and the hardware you selected you can also modify the `--batch_size` and `--number_of_workers`.
-For longer training sessions increase the timeout.
+Every job (and any dataset pushed by the run) is tagged `lerobot` so it's easy to find on the Hub. Add your own with `--job.tags '["my-tag"]'`.
 
-Once the training is started you can go to [Jobs](https://huggingface.co/settings/jobs) and see if your jobs is running as well as all the outputs. Sometimes it takes a few minutes to schedule your job so be patient.
+By default the job is capped at `2d` (48h) of wall-clock. Override it with an HF Jobs duration string, e.g. `--job.timeout=4h` to fail faster or `--job.timeout=7d` for a longer run.
 
-After training the model will be pushed to hub and you can use it as any other model with LeRobot.
+> **Note:** the model repo is created up front (it holds the staged training config the job runs from). If a run fails before the model is pushed, that repo is left on the Hub so you can inspect it — it is not deleted automatically, so repeated failures can leave empty repos behind. Remove one with `hf repo delete <repo-id>`.
+
+**Prerequisites:** run `hf auth login` before submitting. For Weights & Biases integration, run `wandb login` or set `WANDB_API_KEY` on your machine — the key is forwarded to the job automatically.
+
+**Resuming on a job.** Adding `--job.target` to a resume command runs the resume in the cloud — the same command works locally or remotely. The checkpoint repo is the source of truth, and new checkpoints continue the lineage in the same repo:
+
+```bash
+# resume a Hub run on a job (its checkpoints are already on the Hub)
+lerobot-train --config_path=${HF_USER}/my_policy --resume=true --job.target=a10g-small
+
+# resume a LOCAL run on a job — the checkpoint is uploaded to a private Hub repo first,
+# then the job resumes from it (a local-only dataset is uploaded the same way)
+lerobot-train \
+  --config_path=outputs/train/act_so101_test/checkpoints/last/pretrained_model/train_config.json \
+  --resume=true \
+  --job.target=a10g-small
+```
+
+Job settings come from the current command, so override `--job.target`, `--job.timeout`, etc. as needed; for the resumed run to itself be resumable later, keep `--save_checkpoint_to_hub=true`.
 
 #### Upload policy checkpoints
 
@@ -612,6 +596,8 @@ hf upload ${HF_USER}/act_so101_test${CKPT} \
 
 Use `lerobot-rollout` to deploy a trained policy on your robot. You can choose different strategies depending on your needs:
 
+The examples below load the model from `--policy.path`. To pin a specific pushed version — useful once `--save_checkpoint_to_hub=true` has committed several checkpoints — add `--policy.pretrained_revision` with a commit hash, branch, or tag. Each pushed checkpoint is tagged with its step (e.g. `--policy.pretrained_revision=010000`), so you can recover a checkpoint by step without looking up its commit sha.
+
 <hfoptions id="eval">
 <hfoption id="Base mode (no recording)">
 ```bash

docs/source/inference.mdx

@@ -117,7 +117,7 @@ lerobot-rollout \
     --strategy.num_episodes=20 \
     --policy.path=outputs/pretrain/checkpoints/last/pretrained_model \
     --robot.type=bi_openarm_follower \
-    --teleop.type=openarm_mini \
+    --teleop.type=bi_openarm_mini \
     --dataset.repo_id=${HF_USER}/rollout_hil_data \
     --dataset.single_task="Fold the T-shirt"
 ```

docs/source/isaac_teleop.mdx

@@ -0,0 +1,397 @@
+# Isaac Teleop
+
+Control your robot with NVIDIA [Isaac Teleop](https://github.com/NVIDIA/IsaacTeleop), a
+multi-modal teleoperation framework. Isaac Teleop drives a single `TeleopSession` from a range
+of input devices — XR (VR) controllers, hand tracking, full-body tracking, Manus gloves, foot
+pedals, and more.
+
+In LeRobot, Isaac Teleop ships as a self-contained example under
+[`examples/isaac_teleop_to_so101/`](https://github.com/huggingface/lerobot/tree/main/examples/isaac_teleop_to_so101).
+Each Isaac Teleop input device is its own `Teleoperator` subclass in the example's
+`isaac_teleop` package, sharing one session lifecycle (see `IsaacTeleopTeleoperator`). The
+devices available today are the **XR controller** (`XRController`) and a back-drivable
+**SO-101 leader arm** (`SO101LeaderArm`); Manus gloves and hand/full-body tracking are the
+natural next devices. This guide focuses on the XR controller; the SO-101 leader is summarized
+under [Run the example](#step-3-run-the-example).
+
+**In this guide you'll learn:**
+
+- How an Isaac Teleop device drives a robot end‑effector (EE) target
+- How the _clutch_ (squeeze/grip on the XR controller) engages teleoperation without jerking the arm
+- How to run the SO‑101 teleoperation example and tune motion / gripper / IK
+
+## Installation
+
+The example lives in the LeRobot repository (it is not part of the `lerobot` pip package), so
+clone the repo and install from source. The canonical, always-up-to-date install and usage
+reference is the example's
+[`README.md`](https://github.com/huggingface/lerobot/tree/main/examples/isaac_teleop_to_so101/README.md);
+in short:
+
+```bash
+git clone https://github.com/huggingface/lerobot.git
+cd lerobot
+uv pip install -e ".[feetech,kinematics,dataset]" "huggingface_hub>=1.5"
+uv pip install "isaacteleop[cloudxr,retargeters-lite]~=1.3.131" "scipy>=1.14"
+```
+
+`isaacteleop` is published on public PyPI (Linux only). The `cloudxr` extra brings the CloudXR
+runtime bindings; `retargeters-lite` is the scipy-based retargeter path that resolves on both
+x86_64 and ARM (on aarch64 — e.g. a DGX Spark — the full `retargeters` extra does not resolve
+because of its `dex-retargeting`/`nlopt` pins, which is why it is not the default here). On
+x86_64 you can additionally install the full retargeter stack:
+
+```bash
+uv pip install "isaacteleop[retargeters]~=1.3.131"
+```
+
+### Set up CloudXR and connect a headset
+
+Isaac Teleop streams the headset to your machine over **NVIDIA CloudXR**, which provides the
+OpenXR runtime the session connects to. By default LeTeleop **auto-launches the CloudXR runtime
+for you** when you call `teleop_device.connect()` — you no longer have to run `python -m
+isaacteleop.cloudxr` and `source cloudxr.env` in a separate shell. All you need is a supported
+headset connected and the CloudXR firewall ports open. Follow the Isaac Teleop
+[Quick Start](https://nvidia.github.io/IsaacTeleop/main/getting_started/quick_start.html) for the
+headset-pairing and firewall details.
+
+**First run (EULA).** The very first launch must accept the NVIDIA CloudXR EULA. The auto-launch
+prompts for it **on stdin**, so on a headless machine it will hang waiting for input. Bootstrap
+the EULA once, interactively, with:
+
+```bash
+python -m isaacteleop.cloudxr --accept-eula   # one-time: accept the CloudXR EULA
+```
+
+After that, `connect()` launches the runtime non-interactively. The launch **blocks for ~30s**
+while the runtime comes up.
+
+**Configuration.** Two fields on `IsaacTeleopConfig` (shared by every device) control this:
+
+- `auto_launch_cloudxr` (default `True`) — whether `connect()` starts the runtime. Set `False`
+  when CloudXR is already running externally.
+- `cloudxr_env_file` (default `None`) — an optional CloudXR device-profile `.env` selecting the
+  headset transport (e.g. an Apple Vision Pro profile). This is launcher **input**; it is not the
+  `~/.cloudxr/run/cloudxr.env` **output** file the old manual flow told you to `source`. `None`
+  keeps the default auto-WebRTC profile — though the SO-101 example overrides it to the
+  `default.env` shipped next to `teleoperate.py` unless you pass `--teleop.cloudxr_env_file`.
+
+**Opting out.** To skip the auto-launch (CloudXR already running), either set
+`auto_launch_cloudxr=False` or export:
+
+```bash
+export LEROBOT_CLOUDXR_SKIP_AUTOLAUNCH=1
+```
+
+The **env var takes precedence over the config field**: if `LEROBOT_CLOUDXR_SKIP_AUTOLAUNCH=1` is
+set, the auto-launch is skipped even when `auto_launch_cloudxr=True`. This variable is
+**independent** of Isaac Lab's `ISAACLAB_CXR_SKIP_AUTOLAUNCH` — setting one does not affect the
+other.
+
+**One teleoperator per process.** The CloudXR runtime configures the environment process-wide (a
+singleton), so run a single Isaac Teleop teleoperator per process.
+
+**Shutting down.** Always call `teleop_device.disconnect()` on exit — including on Ctrl-C. Wrap
+your teleoperation loop in `try/finally` and call `disconnect()` in the `finally`. This tears down
+the OpenXR session **before** the CloudXR runtime, which is the required order; the launcher's
+`atexit` hook only reaps the runtime and does not run the session's `__exit__`, so without an
+explicit `disconnect()` an interrupted run shuts down in the wrong order.
+
+```python
+teleop_device.connect()
+try:
+    while True:
+        action = teleop_device.get_action()
+        # ... drive the robot ...
+finally:
+    teleop_device.disconnect()
+```
+
+See [System Requirements](https://nvidia.github.io/IsaacTeleop/main/references/requirements.html)
+for supported OS / GPU / CloudXR versions and headsets.
+
+## How it works
+
+The XR controller is one Isaac Teleop **input** device. `XRController` is a deliberately thin
+reader: it exposes the **raw** controller grip pose — already statically rebased into the robot
+base frame — plus the squeeze and trigger analog values. It has **no** retargeters and **no**
+clutch logic of its own. The clutch (engage latch + delta rebasing onto the EE) and the gripper
+mapping live downstream in the example loop, which then feeds LeRobot's existing closed‑loop
+Cartesian IK pipeline — the same one the phone teleoperator uses. The device‑specific pieces are
+`XRController`, the loop's `Clutch`, and `MapXRControllerActionToRobotAction`; everything downstream
+(`EEBoundsAndSafety`, `InverseKinematicsEEToJoints`) is shared, and a future device (e.g. Manus
+gloves) would swap in its own `teleop_<device>.py` + processor while reusing the rest.
+
+`XRController._build_pipeline` wires Isaac Teleop's `ControllersSource` — statically rebased into
+the robot base frame by the native `ControllerTransform` (`base_T_anchor`) — and exposes the
+transformed controller stream verbatim. `get_action()` reads the grip pose, squeeze, and trigger
+straight off it; the session is always stepped `RUNNING` (there is no clutch retargeter to gate).
+
+The `Clutch` class (in `examples/isaac_teleop_to_so101/isaac_teleop/clutch.py`, driven by the
+loop in `common.py`) mirrors Isaac Teleop's `SO101ClutchRetargeter`, but lives in-loop so the
+device can stay a thin reader:
+
+- It latches its engage origin on the squeeze **engage edge** (the frame the squeeze first crosses
+  `clutch_threshold`) and rebases both position and orientation around it, so engaging does not
+  teleport the arm. `Clutch.rebase` returns the absolute base-frame target as a `(pos, quat)`
+  pair, which the loop concatenates into the 7D `ee_pose` fed to the processor.
+- The analog trigger becomes a gripper `closedness` in `[0, 1]` (0 = open, 1 = closed),
+  proportional to the trigger pull, which `MapXRControllerActionToRobotAction` maps to a jaw target.
+
+See the Isaac Teleop
+[Retargeting interface](https://nvidia.github.io/IsaacTeleop/main/references/retargeting/index.html)
+and [architecture overview](https://nvidia.github.io/IsaacTeleop/main/overview/architecture.html)
+for how source nodes and retargeters compose.
+
+```text
+  VR controller (OpenXR)
+        │
+        ▼
+  XRController.get_action()          ── raw base-frame grip_pos / grip_quat + squeeze + trigger
+        │                                (TeleopSession always stepped RUNNING; clutch lives downstream)
+        ▼
+  Clutch.rebase(grip_pos, grip_quat) ── engage-relative delta applied to the EE home (pos + orient)
+        │  ee_pose (7) / closedness       → absolute ee_pose; closedness = trigger
+        ▼
+  MapXRControllerActionToRobotAction ── absolute ee.x/y/z; ee.w* = orientation rotvec target;
+        │  ee.x/y/z / ee.w* / ee.gripper_pos   ee.gripper_pos = (1 - closedness) * 100
+        ▼
+  EEBoundsAndSafety                ── workspace clip + per-frame step clamp (clamp+warn)
+        │
+        ▼
+  InverseKinematicsEEToJoints      ── closed-loop Placo IK; position + soft-orientation
+        │  (orientation_weight=0.01)   (passes ee.gripper_pos → gripper.pos)
+        ▼
+  SO-101 follower joint targets
+```
+
+### The clutch: owned by the example loop
+
+Unlike the phone pipeline (which splits the clutch across `MapPhoneActionToRobotAction` and
+`EEReferenceAndDelta`), the XR clutch lives entirely in the example loop's `Clutch` class. It emits
+an **absolute** EE pose, so there is no `EEReferenceAndDelta` stage and no delta accumulation in the
+processor — `MapXRControllerActionToRobotAction` is a pure, stateless per‑frame mapping.
+
+The clutch latches its engage origin on the squeeze **engage edge** (the moment the squeeze crosses
+`clutch_threshold`) and drives the EE from the motion _relative_ to that origin, so the arm does not
+teleport on engage. On **every** engage — startup and mid‑task re‑clutch alike — the home
+_position_ is latched from forward kinematics on the arm's **measured joints**, so the home equals
+where the arm physically is even if it moved while disengaged, and the engage is jump‑free. The
+home _orientation_ keeps the last commanded rotation: the 5‑DOF arm tracks orientation only
+softly, so latching the measured wrist orientation would inject its tracking offset into the
+command on every re‑clutch.
+
+## Controls
+
+- **Squeeze / grip** — the **clutch** (deadman). Hold it past `clutch_threshold` to engage
+  teleoperation; release to pause. Each engage re‑captures the origin, so you can reposition
+  your hand while paused and re‑engage without the arm jumping (index/clutch style).
+- **Trigger** — the **gripper**, controlled **analog**. The jaw tracks the trigger
+  proportionally — a half‑pressed trigger leaves the jaw half‑closed — via a closedness in
+  `[0, 1]` (0 = open, 1 = closed) that maps to an absolute gripper joint target.
+- **Controller orientation** — the **wrist**. The clutch rebases the controller orientation
+  (engage‑relative, base‑frame) into a soft IK orientation target the wrist tracks alongside
+  position. On the 5‑DOF SO‑101 the wrist follows the hand only partially by design — see
+  `orientation_weight` below.
+
+## Get started
+
+### Step 1: Create the teleoperator
+
+```python
+# Run from the repo root so the `examples` package is importable.
+from examples.isaac_teleop_to_so101.isaac_teleop import XRController, XRControllerConfig
+
+teleop_config = XRControllerConfig(
+    hand_side="right",      # "left" or "right" controller
+    clutch_threshold=0.5,   # squeeze value above which the clutch engages
+)
+teleop_device = XRController(teleop_config)
+```
+
+`XRController.get_action()` returns the **raw** base‑frame controller pose, not a clutch‑rebased
+target: `grip_pos` (3,) `[x, y, z]` [m] and `grip_quat` (4,) `[qx, qy, qz, qw]` in the robot base
+frame, plus scalar `squeeze` and `trigger` analog values in `[0, 1]`. The example loop's `Clutch`
+turns these into the absolute `ee_pose`, and the squeeze is thresholded by the loop against
+`clutch_threshold` to engage.
+
+### Step 2: Connect
+
+Calling `teleop_device.connect()` first auto-launches the CloudXR runtime (unless you opted out —
+see [Set up CloudXR and connect a headset](#set-up-cloudxr-and-connect-a-headset); this blocks for
+~30s and on the first run prompts for the EULA on stdin), then starts the Isaac Teleop
+[`TeleopSession`](https://nvidia.github.io/IsaacTeleop/main/getting_started/teleop_session.html)
+(opens the OpenXR session and discovers the controllers). XR controllers are self‑calibrating, so
+there is no manual calibration step — the clutch handles re‑centering each time you engage. Pair
+`connect()` with a `try/finally` that calls `disconnect()` so the session tears down before the
+runtime on exit/Ctrl-C.
+
+### Step 3: Run the example
+
+The example assumes you configured your robot (SO‑101 follower) and set the correct serial port.
+
+The **robot URDF and its meshes are fetched automatically** on first run: the XR device downloads
+the SO-101 URDF from the
+[`lerobot/robot-urdfs` Hugging Face bucket](https://huggingface.co/buckets/lerobot/robot-urdfs/tree/so101)
+into the LeRobot cache (`HF_LEROBOT_HOME/robot-urdfs/so101/`) and reuses it after, so there is no
+separate download step :
+
+```bash
+python -m examples.isaac_teleop_to_so101.teleoperate --robot.type=so101_follower --robot.port=/dev/ttyACM0 \
+    --robot.id=so101_follower_arm --teleop.type=xr_controller
+```
+
+The CLI is `lerobot-teleoperate`-style (draccus): `--robot.*` configures the SO-101 follower and
+`--teleop.type` selects the Isaac input device (`xr_controller` | `so101_leader`), with
+`--teleop.*` its device knobs. `--teleop.type=xr_controller` runs the XR-controller path described
+above. The startup safety contract: by default it slews all joints to a default reset pose over
+`--reset_duration` seconds (`--reset_to_origin=false` keeps the arm where it is), then seeds the
+clutch home from the arm's measured pose so the first engage is jump-free; the follower is
+commanded only while the clutch is engaged.
+
+**Customizing the reset pose.** The reset pose ships as a built-in default (a comfortable mid-range
+pose) and works out of the box — you do **not** need to record anything. To tailor it to your setup,
+back-drive the arm to the pose you want and run
+`python -m examples.isaac_teleop_to_so101.override_reset_pose --id <robot.id>`; it writes the
+current joints to a per-arm file in the LeRobot cache
+(`HF_LEROBOT_HOME/reset_poses/<robot.name>/<robot.id>.json`, keyed like calibration), which then takes
+priority over the built-in default on the next run. Because it lives in the user-local cache (not
+the repo), your override stays on your machine, and both `teleoperate` and `record` honor it
+when launched with the same `--robot.id`.
+
+The other device, `--teleop.type=so101_leader`, mirrors the follower 1:1 from a back-drivable
+SO-101 _leader arm_ whose joints are streamed by Isaac Teleop's native `so101_leader` plugin (no
+clutch, no IK — the leader and follower share the SO-101 kinematics).
+
+The `so101_leader_plugin` binary is a C++ plugin that is **not** part of the `isaacteleop` pip
+package — you build it from the Isaac Teleop source tree. Follow
+[Build Isaac Teleop from source](https://nvidia.github.io/IsaacTeleop/main/getting_started/build_from_source/index.html)
+(in short, from your Isaac Teleop checkout: `cmake -B build && cmake --build build --parallel &&
+cmake --install build`); the build installs the plugins under `<IsaacTeleop>/install/plugins/`, so
+the binary lands at `install/plugins/so101_leader/so101_leader_plugin` — the `--launch_plugin` path
+below. See the plugin's own `README.md` (next to the binary) for its serial/calibration details.
+
+Point `--teleop.port` at the physical leader's serial port and `--launch_plugin` at that plugin
+binary to have the script spawn it after CloudXR is up:
+
+```bash
+python -m examples.isaac_teleop_to_so101.teleoperate --robot.type=so101_follower --robot.port=/dev/ttyACM0 \
+    --robot.id=so101_follower_arm --teleop.type=so101_leader \
+    --teleop.port=/dev/ttyACM1 --teleop.id=so101_leader_arm \
+    --launch_plugin=/code/Teleop/install/plugins/so101_leader/so101_leader_plugin
+```
+
+(Note `so101_leader` here is the _Isaac_ leader, resolved against the Isaac Teleop device
+registry, distinct from `lerobot-teleoperate`'s serial `so101_leader`.) When a `--teleop.port` is
+set, the plugin's tick→radian calibration is inferred from `--teleop.id` and passed to the plugin
+as its third positional arg — the LeRobot-format JSON at
+`HF_LEROBOT_CALIBRATION/teleoperators/so_leader/<id>.json`, the same file the serial SO-101 leader
+uses (`lerobot-calibrate --teleop.type=so101_leader --teleop.id=<id>`). If it is missing the script
+warns and the plugin uses built-in defaults. Run `python -m examples.isaac_teleop_to_so101.teleoperate --help` for all flags. Its
+startup safety contract: by default the follower is
+slewed to the leader's first reading over `--align_duration` seconds (`--align=false` to skip) so
+the arm does not snap when the mirror begins, and while the leader stream is stale the follower is
+held at its measured pose.
+
+The URDF fetch uses `huggingface_hub` (already a LeRobot dependency) against the public
+`lerobot/robot-urdfs` bucket, so it needs no login. It is cached under
+`HF_LEROBOT_HOME/robot-urdfs/so101/`; delete that folder to force a re‑download.
+
+Then, in your headset: squeeze and hold the grip to engage, move the controller to drive the
+arm, twist/tilt it to orient the wrist, and press the trigger to close the gripper
+(proportionally — release to open).
+
+To record a dataset (not just teleoperate), use `record.py` in the same folder. It dispatches on
+`--teleop.type` (`xr_controller` | `so101_leader`) exactly like `teleoperate.py`, so either device
+can drive the follower, and it saves the commanded joints to a LeRobot dataset (`lerobot-record`-style
+`--dataset.*` flags). See its module docstring for the full CLI and the keyboard recording shortcuts.
+
+## Important pipeline steps and options
+
+The clutch already produces an absolute base‑frame pose, so the processor side is a thin
+**absolute‑pose** path — there is no frame remap, no delta accumulation, and no
+`EEReferenceAndDelta` stage.
+
+- `MapXRControllerActionToRobotAction` is a stateless per‑frame mapping from the device output to
+  the IK input contract. It writes the absolute base‑frame position, encodes the absolute
+  orientation as a rotvec target, and inverts the closedness into a motor gripper target:
+
+  ```python
+  action["ee.x"], action["ee.y"], action["ee.z"] = ee_pose[:3]   # absolute, base frame [m]
+  action["ee.wx"], action["ee.wy"], action["ee.wz"] = orient_rotvec  # orientation target (rotvec)
+  action["ee.gripper_pos"] = (1 - closedness) * 100   # motor units; SO-101 calibrates 100 = open
+  ```
+
+  The gripper polarity (`100 = open, 0 = closed`) is a hardware‑calibration convention in the source — flip it there if the jaw opens when it should close.
+
+- `EEBoundsAndSafety` clamps the EE to a workspace and rate‑limits per‑frame jumps. The clutch's
+  no‑teleport keeps frames small, so `max_ee_step_m` mostly catches transient controller tracking
+  glitches. The z floor is `0.0` (the table plane) so a stray target cannot drive the EE below the
+  table; x/y stay at the loose `[-1, 1]` m box. Set `raise_on_jump=False` so an over‑limit frame is
+  **clamped and warned** instead of raising — a crash mid‑loop would leave the arm uncontrolled:
+
+  ```python
+  EEBoundsAndSafety(
+      end_effector_bounds={"min": [-1.0, -1.0, 0.0], "max": [1.0, 1.0, 1.0]},
+      max_ee_step_m=0.10,
+      raise_on_jump=False,
+  )
+  ```
+
+- `InverseKinematicsEEToJoints(initial_guess_current_joints=False, orientation_weight=0.01)` solves
+  closed‑loop Placo IK. SO‑101 is a 5‑DOF arm, so the IK is position‑dominant; the small
+  `orientation_weight` lets it softly track the orientation target carried in `ee.w*` so the wrist
+  follows the hand, while the under‑determined roll stays partial by design. There is **no**
+  `GripperVelocityToJoint`: the absolute `ee.gripper_pos` is passed straight to `gripper.pos`.
+  `initial_guess_current_joints=False` warm‑starts each solve from the **previous IK solution**
+  rather than re‑seeding from the measured joints, so the joint trajectory stays continuous
+  frame‑to‑frame. Tune `orientation_weight` on hardware — too high fights position tracking, too
+  low ignores the orientation command.
+
+The example also gates safety at the loop level: after the startup reset slew (on by default —
+pass `--reset_to_origin=false` to keep the arm where it is), it commands the robot **only while
+the clutch is engaged**, and re‑sends the measured joints while disengaged, so releasing the
+clutch freezes the arm in place.
+
+See the [Processors for Robots and Teleoperators](./processors_robots_teleop) guide for more on
+adapting the pipeline to other robots.
+
+## Troubleshooting
+
+- **`ModuleNotFoundError: isaacteleop`** — the `isaacteleop` package is not installed in the
+  active environment. Re-run the install command at the top of this guide:
+  `uv pip install "isaacteleop[cloudxr,retargeters-lite]~=1.3.131"`.
+- **No controllers found** — make sure the CloudXR runtime is running, the firewall ports are
+  whitelisted, and the headset is connected (see
+  [Set up CloudXR and connect a headset](#set-up-cloudxr-and-connect-a-headset) and the Isaac
+  Teleop [Quick Start](https://nvidia.github.io/IsaacTeleop/main/getting_started/quick_start.html)).
+- **CloudXR auto-launch failed** — `connect()` raises a `RuntimeError` if the runtime does not
+  come up within its startup timeout. Check the launcher logs under `~/.cloudxr/logs`. Common
+  causes: the EULA was never accepted (run `python -m isaacteleop.cloudxr --accept-eula` once,
+  interactively — the auto-launch prompts on stdin and hangs headless), or the runtime is already
+  running externally (set `LEROBOT_CLOUDXR_SKIP_AUTOLAUNCH=1` or `auto_launch_cloudxr=False` to
+  skip the auto-launch).
+- **Arm does not move** — the clutch is a deadman: you must hold the squeeze/grip past
+  `clutch_threshold`. Lower the threshold if your controller's squeeze is reported softly.
+- **Motion feels misaligned** — confirm the headset/play space orientation. The controller stream
+  is rebased into the robot base frame by the `base_T_anchor` transform on `XRControllerConfig`
+  (default: standard OpenXR → robot axis convention); adjust it if your anchor frame differs.
+
+## Learn more
+
+NVIDIA Isaac Teleop documentation ([docs home](https://nvidia.github.io/IsaacTeleop/),
+[GitHub](https://github.com/NVIDIA/IsaacTeleop)):
+
+- [Quick Start](https://nvidia.github.io/IsaacTeleop/main/getting_started/quick_start.html) —
+  install, run the CloudXR server, connect a headset, run a teleop example.
+- [TeleopSession](https://nvidia.github.io/IsaacTeleop/main/getting_started/teleop_session.html) —
+  the session API `XRController` wraps.
+- [Retargeting interface](https://nvidia.github.io/IsaacTeleop/main/references/retargeting/index.html)
+  and [architecture overview](https://nvidia.github.io/IsaacTeleop/main/overview/architecture.html) —
+  how source nodes and retargeters compose into a pipeline.
+- [Build from source](https://nvidia.github.io/IsaacTeleop/main/getting_started/build_from_source/index.html) —
+  build `isaacteleop` (and its C++ plugins, including the `so101_leader` plugin used above) from a
+  local checkout.
+- [System Requirements](https://nvidia.github.io/IsaacTeleop/main/references/requirements.html) and
+  the [CloudXR SDK docs](https://docs.nvidia.com/cloudxr-sdk) — supported platforms, GPUs,
+  CloudXR/OpenXR runtime versions, and headsets.

docs/source/lekiwi.mdx

@@ -319,7 +319,7 @@ If you want to dive deeper into this important topic, you can check out the [blo
 
 #### Troubleshooting:
 
-- On Linux, if the left and right arrow keys and escape key don't have any effect during data recording, make sure you've set the `$DISPLAY` environment variable. See [pynput limitations](https://pynput.readthedocs.io/en/latest/limitations.html#linux).
+- On Linux, the recording control-flow keys (arrow keys, Escape) work on X11, Wayland, and headless/SSH sessions as long as you run the recording from an interactive terminal (keep it focused) — no `$DISPLAY` setup is needed; the letter equivalents `n` / `r` / `q` also work. Note that **keyboard teleoperation of the LeKiwi base** is different: it relies on a global key backend and therefore works only on an X11 session, a Windows desktop, or macOS with Accessibility/Input Monitoring granted — not on Wayland or headless machines. See [pynput limitations](https://pynput.readthedocs.io/en/latest/limitations.html#linux).
 
 ## Replay an episode
 

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/libero.mdx

@@ -143,7 +143,7 @@ lerobot-train \
   --batch_size=4 \
   --eval.batch_size=1 \
   --eval.n_episodes=1 \
-  --eval_freq=1000
+  --env_eval_freq=1000
 ```
 
 ## Reproducing published results

docs/source/libero_plus.mdx

@@ -173,7 +173,7 @@ lerobot-train \
     --batch_size=4 \
     --eval.batch_size=1 \
     --eval.n_episodes=1 \
-    --eval_freq=1000
+    --env_eval_freq=1000
 ```
 
 ## Relationship to LIBERO

docs/source/lingbot_va.mdx

@@ -0,0 +1,187 @@
+# LingBot-VA
+
+LingBot-VA is an **autoregressive video-action world-model policy** built on the **Wan2.2**
+video-diffusion stack. It interleaves, in one autoregressive sequence, the prediction of
+future **video latents** and **robot actions** ("VA" = Video-Action). The LeRobot
+integration wires LingBot-VA into the standard training, evaluation and processor
+interfaces.
+
+## Model Overview
+
+LingBot-VA is a **dual-stream "mixture-of-transformers"**: a video/latent stream
+(`patch_embedding_mlp → blocks → proj_out`) and an action stream
+(`action_embedder → blocks → action_proj_out`) share the same 30 transformer blocks and
+text conditioning.
+
+| Component                | Class                   | Role                                                        |
+| ------------------------ | ----------------------- | ----------------------------------------------------------- |
+| DiT backbone (trainable) | `WanTransformer3DModel` | ~5B-param dual-stream transformer.                          |
+| VAE (frozen)             | `AutoencoderKLWan`      | Wan2.2 VAE, `z_dim=48`. Lazy-pulled from the source repo.   |
+| Text encoder (frozen)    | `UMT5EncoderModel`      | UMT5-XXL, `d_model=4096`. Lazy-pulled from the source repo. |
+
+At inference the policy runs an autoregressive loop per chunk: it denoises the video-latent
+stream (CFG, ~20 steps) and the action stream (~50 steps) with two independent
+flow-matching schedulers, maintaining a KV cache across chunks. Real observed keyframes are
+fed back into the KV cache as the chunk is executed (closed-loop world modeling).
+
+### What the LeRobot Integration Covers
+
+- Standard `policy.type=lingbot_va` configuration through LeRobot.
+- Ready-to-use LeRobot-format checkpoints on the Hub (converted from the released upstream ones).
+- Autoregressive dual-stream inference behind the standard `select_action` interface
+  (single-environment eval, `--eval.batch_size=1`).
+- Opt-in saving of the policy's **predicted (imagined) videos** during eval / training.
+- Evaluation with `lerobot-eval` on LIBERO and RoboTwin.
+- Training / fine-tuning via the dual-stream flow-matching loss (`policy.forward`), see below.
+
+## Installation
+
+1. Install LeRobot by following the [Installation Guide](./installation).
+2. Install the LingBot-VA extra:
+
+```bash
+pip install -e ".[lingbot_va]"
+```
+
+## Checkpoints
+
+The released upstream checkpoints have been converted to LeRobot format and pushed to the Hub:
+
+| Variant                | LeRobot checkpoint               |
+| ---------------------- | -------------------------------- |
+| LIBERO-Long post-train | `lerobot/lingbot_va_libero_long` |
+| RoboTwin post-train    | `lerobot/lingbot_va_robotwin`    |
+| Pretrained base        | `lerobot/lingbot_va_base`        |
+
+Only the trainable ~5B transformer is stored in the LeRobot
+`model.safetensors`. The frozen VAE + UMT5 + tokenizer (~20 GB) are pulled from
+`config.wan_pretrained_path` at load time (defaults to the source `robbyant/*` repo). The
+UMT5-XXL text encoder runs on CPU by default (`config.text_encoder_device`) so the 5B
+transformer + VAE fit on a single 24–32 GB GPU.
+
+## Evaluation (LIBERO)
+
+```bash
+lerobot-eval \
+    --policy.path=lerobot/lingbot_va_libero_long \
+    --policy.device=cuda \
+    --env.type=libero --env.task=libero_10 \
+    --env.observation_height=128 --env.observation_width=128 \
+    --eval.n_episodes=50 --eval.batch_size=1 \
+    --output_dir=outputs/eval/lingbot_va_libero
+```
+
+LingBot-VA's streaming inference (KV cache + observed-keyframe feedback) is implemented for
+single-environment eval; use `--eval.batch_size=1`.
+
+## Evaluation (RoboTwin)
+
+RoboTwin 2.0 needs the SAPIEN + CuRobo simulator stack. You can use the benchmark Docker image
+(`docker/Dockerfile.benchmark.robotwin`, which also needs `warp-lang==1.3.1` and CuRobo built
+with the GPU's compute capability in `TORCH_CUDA_ARCH_LIST`). RoboTwin uses **end-effector-pose
+control**, so run with `--env.action_mode=ee`: the policy predicts per-arm `xyz+quaternion+gripper`
+deltas (`robotwin_tshape` latent layout) that are composed onto the episode's initial eef pose and
+executed via CuRobo IK.
+
+```bash
+lerobot-eval \
+    --policy.path=lerobot/lingbot_va_robotwin \
+    --policy.device=cuda \
+    --env.type=robotwin --env.task=beat_block_hammer --env.action_mode=ee \
+    --eval.n_episodes=10 --eval.batch_size=1 \
+    --output_dir=outputs/eval/lingbot_va_robotwin
+```
+
+### Saving predicted (imagined) videos
+
+Set `--policy.save_predicted_video=true` to additionally VAE-decode the predicted video
+latents and write `pred_episode_*.mp4` next to the env-rendered `eval_episode_*.mp4` videos.
+The same flag works for the periodic eval during `lerobot-train`.
+
+## Training / fine-tuning
+
+`LingBotVAPolicy.forward(batch)` implements the dual-stream **flow-matching** loss
+(`latent_loss + action_loss`, timestep-weighted, action-masked) from the paper: it VAE-encodes
+the camera clips into video latents, UMT5-encodes the task, noises both streams, runs the
+transformer's block-causal training pass and returns `(loss, metrics)`. Optimizer preset is AdamW
+with a linear-warmup-then-constant schedule (matching upstream).
+
+Requirements:
+
+- The block-causal masks use PyTorch **flex-attention**, so build the policy with
+  `--policy.attn_mode=flex` for training (the default `torch` SDPA is inference-only).
+- The full 5B DiT does not fit a single 24–32 GB GPU under AdamW; fine-tune with **LoRA**
+  (`--policy.use_peft=true`) and/or optimizer offload. `get_optim_params` returns only the
+  trainable (e.g. adapter) parameters; the VAE + UMT5 text encoder stay frozen.
+
+```bash
+lerobot-train \
+  --policy.path=lerobot/lingbot_va_libero_long --policy.attn_mode=flex \
+  --policy.use_peft=true \
+  --dataset.repo_id=<your LeRobot-format dataset> \
+  --batch_size=1 --steps=... --output_dir=outputs/train/lingbot_va
+```
+
+The dataset must provide camera clips (a temporal window per camera, VAE-encoded to
+`frame_chunk_size` latent frames) and `frame_chunk_size * action_per_frame` action steps per item.
+
+## Data format (action channels & camera order)
+
+LingBot-VA is an **end-effector (Cartesian) pose** policy, it predicts EEF poses + gripper, not
+joint positions. Actions live in a fixed multi-embodiment **30-dim** layout; map your robot's
+action dimensions into these channels and pad the rest with `0` (`used_action_channel_ids` selects
+the channels a given checkpoint actually uses):
+
+| channels | meaning                                               |
+| -------- | ----------------------------------------------------- |
+| 0–6      | Left-arm end-effector pose                            |
+| 7–13     | Right-arm end-effector pose                           |
+| 14–20    | Left-arm joints (unused by the released checkpoints)  |
+| 21–27    | Right-arm joints (unused by the released checkpoints) |
+| 28       | Left gripper                                          |
+| 29       | Right gripper                                         |
+
+- **LIBERO** uses channels `0–6`: a 6-DoF EEF delta (xyz + rotation) + gripper (single arm).
+- **RoboTwin** uses channels `[0–6, 28, 7–13, 29]`: left EEF (xyz + quaternion) + left gripper +
+  right EEF + right gripper (16 dims). The env converts these poses to joint trajectories via
+  CuRobo IK — joints are never predicted.
+
+Joint-space datasets (or a different EEF convention) must be remapped into this schema before
+fine-tuning these checkpoints.
+
+**Camera order is fixed and order-sensitive**, per-camera latents are concatenated spatially in
+`obs_cam_keys` order, so the physical camera→slot mapping must match training:
+
+| benchmark | `obs_cam_keys` (in order)                                                                             | `camera_layout`                                                     |
+| --------- | ----------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------- |
+| LIBERO    | `observation.images.image` (agentview / 3rd-person), `observation.images.image2` (eye-in-hand wrist)  | `width_concat` (latents concatenated on width)                      |
+| RoboTwin  | `observation.images.head_camera`, `observation.images.left_camera`, `observation.images.right_camera` | `robotwin_tshape` (full-res head below, two half-res wrists on top) |
+
+The first camera is the exterior/head view and the rest are wrist views.
+
+## Inference Hyperparameters (LIBERO)
+
+| Key                                    | Value                                                                             |
+| -------------------------------------- | --------------------------------------------------------------------------------- |
+| height × width                         | 128 × 128                                                                         |
+| cameras                                | `observation.images.image` (agentview), `observation.images.image2` (eye-in-hand) |
+| action channels used                   | 0–6 (7-DoF arm + gripper)                                                         |
+| action_per_frame / frame_chunk_size    | 4 / 4                                                                             |
+| attn_window                            | 30                                                                                |
+| video / action denoising steps         | 20 / 50                                                                           |
+| guidance_scale / action_guidance_scale | 5 / 1                                                                             |
+| snr_shift / action_snr_shift           | 5.0 / 0.05                                                                        |
+
+These are the defaults of `LingBotVAConfig`; override any of them via `--policy.<name>=...`.
+
+## Notes
+
+- **Attention backend:** inference uses the `torch` SDPA backend (always available). The
+  `flashattn` and `flex` backends are optional; `flex` is only needed for training.
+- **Model size:** the DiT is ~5B params and the frozen VAE+UMT5 add ~20 GB; inference needs
+  roughly 18–24 GB of VRAM.
+
+## License
+
+LingBot-VA is released under Apache-2.0. See the
+[upstream repository](https://github.com/Robbyant/lingbot-va).

docs/source/metaworld.mdx

@@ -120,11 +120,11 @@ lerobot-train \
   --batch_size=4 \
   --eval.batch_size=1 \
   --eval.n_episodes=1 \
-  --eval_freq=1000
+  --env_eval_freq=1000
 ```
 
 ## Practical tips
 
 - Use the one-hot task conditioning for multi-task training (MT10/MT50 conventions) so policies have explicit task context.
 - Inspect the dataset task descriptions and the `info["is_success"]` keys when writing post-processing or logging so your success metrics line up with the benchmark.
-- Adjust `batch_size`, `steps`, and `eval_freq` to match your compute budget.
+- Adjust `batch_size`, `steps`, and `env_eval_freq` to match your compute budget.

docs/source/molmoact2.mdx

@@ -17,7 +17,7 @@ the paper, see [allenai/molmoact2](https://github.com/allenai/molmoact2).
 Install LeRobot with the MolmoAct2 optional dependencies:
 
 ```bash
-pip install -e ".[molmoact2]"
+uv sync --locked --extra molmoact2
 ```
 
 To run the models in this repository, you need an NVIDIA GPU. The measurements
@@ -46,8 +46,8 @@ The repo has been tested with Ubuntu 22.04.
 
 To use MolmoAct2 in a LeRobot training config, set:
 
-```python
-policy.type=molmoact2
+```bash
+--policy.type=molmoact2
 ```
 
 ## Training
@@ -103,7 +103,7 @@ accelerate launch \
   --batch_size=32 \
   --num_workers=4 \
   --log_freq=20 \
-  --eval_freq=-1 \
+  --env_eval_freq=-1 \
   --save_checkpoint=true \
   --save_freq=2000
 ```
@@ -142,7 +142,7 @@ accelerate launch \
   --batch_size=32 \
   --num_workers=4 \
   --log_freq=20 \
-  --eval_freq=-1 \
+  --env_eval_freq=-1 \
   --save_checkpoint=true \
   --save_freq=2000
 ```
@@ -386,6 +386,68 @@ These results demonstrate MolmoAct2's strong performance across diverse robotic
 manipulation tasks. To reproduce them, follow the instructions in the LIBERO
 evaluation section.
 
+## Hardware Deployment (lerobot-rollout)
+
+LeRobot-format checkpoints are available on the Hub for direct use with
+`lerobot-rollout`. Each checkpoint uses specific camera names that must
+match your robot's camera configuration.
+
+### Camera naming convention
+
+Each checkpoint expects specific `observation.images.*` keys.
+If your robot cameras have different names, use `--rename_map` to map them:
+
+| Checkpoint                    | Camera keys            | Description              |
+| ----------------------------- | ---------------------- | ------------------------ |
+| MolmoAct2-LIBERO-LeRobot      | `image`, `wrist_image` | LIBERO sim cameras       |
+| MolmoAct2-BimanualYAM-LeRobot | `top`, `left`, `right` | YAM 3-camera setup       |
+| MolmoAct2-DROID-LeRobot       | `cam0`, `cam1`         | External + wrist         |
+| MolmoAct2-SO100_101-LeRobot   | `cam0`, `cam1`         | Primary + secondary view |
+
+Example with an SO-100 robot using top and side cameras:
+
+```bash
+lerobot-rollout \
+  --policy.path=lerobot/MolmoAct2-SO100_101-LeRobot \
+  --rename_map='{"observation.images.top": "observation.images.cam0", "observation.images.side": "observation.images.cam1"}' \
+  --robot.type=so100_follower \
+  --robot.port=/dev/ttyACM0 \
+  --robot.cameras='{
+      top: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30},
+      side: {type: opencv, index_or_path: 2, width: 640, height: 480, fps: 30}
+  }' \
+  --task="pick up the red cube" --duration=30
+```
+
+To use a wrist camera instead, just change the rename mapping:
+
+```bash
+--rename_map='{"observation.images.top": "observation.images.cam0", "observation.images.wrist": "observation.images.cam1"}'
+```
+
+### Joint frame transform (SO-100/101 zero-shot)
+
+<Tip warning={true}>
+The MolmoAct2-SO100_101 checkpoint was trained on data that uses a different
+joint calibration convention than LeRobot >= 0.5.0. Without a frame
+correction, the arm may move in the wrong direction.
+
+This affects both **zero-shot deployment** and **fine-tuning** from the
+original checkpoint. The pretrained weights expect the old convention, so
+all joint data (observations and actions) must be transformed to match.
+
+The converted LeRobot checkpoint (`lerobot/MolmoAct2-SO100_101-LeRobot`)
+already includes this correction in its processor pipeline. If you convert
+or fine-tune the checkpoint yourself, set the following in the policy config (`configuration_molmoact2.py`):
+
+- `joint_signs`: `[1, -1, 1, 1, 1, 1]` (flips shoulder_lift direction)
+- `joint_offsets`: `[0, 90, 90, 0, 0, 0]` (shifts shoulder_lift and elbow_flex by 90°)
+
+See the [backward compatibility guide](./backwardcomp) for details on the
+calibration change.
+
+</Tip>
+
 ## Differences From the Original Implementation
 
 This LeRobot port is intended to match MolmoAct2 behavior while using LeRobot's

docs/source/multi_gpu_training.mdx

@@ -95,7 +95,7 @@ If you want to scale your hyperparameters when using multiple GPUs, you should d
 accelerate launch --num_processes=2 $(which lerobot-train) \
   --optimizer.lr=2e-4 \
   --dataset.repo_id=lerobot/pusht \
-  --policy=act
+  --policy.type=act
 ```
 
 **Training Steps Scaling:**
@@ -110,9 +110,64 @@ accelerate launch --num_processes=2 $(which lerobot-train) \
   --batch_size=8 \
   --steps=50000 \
   --dataset.repo_id=lerobot/pusht \
-  --policy=act
+  --policy.type=act
 ```
 
+## Training Large Models with FSDP
+
+DDP replicates the full model on every GPU, so a model that doesn't fit on one GPU won't fit under
+DDP either. For large models, use **FSDP** (Fully Sharded Data Parallel), which shards parameters,
+gradients, and optimizer state across GPUs. See the [accelerate FSDP guide](https://huggingface.co/docs/accelerate/usage_guides/fsdp) for background.
+
+An example on how to launch LeRobot training with FSDP across 4 GPUs (1 machine):
+
+```bash
+accelerate launch --config_file fsdp.yaml --num_processes=4 $(which lerobot-train) \
+  --dataset.repo_id=${HF_USER}/my_dataset \
+  --policy.type=<your_policy> \
+  --output_dir=outputs/train/my_policy_fsdp
+```
+
+A minimal `fsdp.yaml` (FSDP1; shards params/grads/optimizer — ZeRO-3-equivalent):
+
+```yaml
+compute_environment: LOCAL_MACHINE
+distributed_type: FSDP
+mixed_precision: bf16
+num_machines: 1
+num_processes: 4
+fsdp_config:
+  fsdp_version: 1
+  fsdp_sharding_strategy: FULL_SHARD # params + grads + optimizer (ZeRO-3)
+  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  fsdp_transformer_layer_cls_to_wrap: <YourTransformerBlock> # repeated block class to shard
+  fsdp_use_orig_params: true # required: optimizer is built pre-prepare
+  fsdp_state_dict_type: FULL_STATE_DICT
+```
+
+Set `fsdp_transformer_layer_cls_to_wrap` to your model's repeated transformer-block class so each
+block is sharded as its own unit. `fsdp_use_orig_params: true` is required because LeRobot builds the
+optimizer before `accelerator.prepare()`.
+
+### FSDP checkpoints
+
+LeRobot gathers the full state dict across all ranks and the main process writes it as a single
+`model.safetensors`, loadable as usual with `Policy.from_pretrained(...)`. Two things to look out for:
+
+- **Checkpoints store fp32 weights.** Under mixed precision (`bf16`/`fp16`) FSDP keeps an fp32 master
+  copy, and the checkpoint saves it (~2× the bf16 size on disk) so training can resume consistently
+  with the fp32 optimizer state; `from_pretrained` casts back to the policy dtype on load. FSDP-specific
+  caveat: an fp32 checkpoint is materialized in full precision on the target device _before_ casting,
+  so loading it for inference on a tight GPU can OOM even when the bf16 model would fit — load on CPU
+  first, or cast `model.safetensors` to the deployment dtype offline.
+- The sharded optimizer state is gathered into a full (world-size-independent) state dict and saved
+  alongside the model in the same `optimizer_state.safetensors` / `optimizer_param_groups.json`
+  format as single-GPU training, so **resume-from-checkpoint is supported** with `--resume=true`.
+  Resume reshards both the model and the optimizer state to the _current_ FSDP topology, so you can
+  resume an FSDP checkpoint on a different number of GPUs. Note that the data sampler is only
+  sample-exact when the world size and batch size match the original run (a warning is logged
+  otherwise); the optimizer/model state itself is unaffected.
+
 ## Notes
 
 - The `--policy.use_amp` flag in `lerobot-train` is only used when **not** running with accelerate. When using accelerate, mixed precision is controlled by accelerate's configuration.

docs/source/multi_task_dit.mdx

@@ -314,7 +314,7 @@ lerobot-train \
   --steps=30000 \
   --save_freq=1000 \
   --log_freq=100 \
-  --eval_freq=1000 \
+  --env_eval_freq=1000 \
   --policy.type=multi_task_dit \
   --policy.device=cuda \
   --policy.horizon=32 \

docs/source/pi0fast.mdx

@@ -96,7 +96,7 @@ lerobot-train \
     --policy.type=pi0_fast \
     --output_dir=./outputs/pi0fast_training \
     --job_name=pi0fast_training \
-    --policy.pretrained_path=lerobot/pi0_fast_base \
+    --policy.pretrained_path=lerobot/pi0fast-base \
     --policy.dtype=bfloat16 \
     --policy.gradient_checkpointing=true \
     --policy.chunk_size=10 \
@@ -187,7 +187,7 @@ lerobot-train \
   --dataset.repo_id=lerobot/libero \
   --output_dir=outputs/libero_pi0fast \
   --job_name=libero_pi0fast \
-  --policy.path=lerobot/pi0fast_base \
+  --policy.path=lerobot/pi0fast-base \
   --policy.dtype=bfloat16 \
   --steps=100000 \
   --save_freq=20000 \

docs/source/policy_evo1_README.md

@@ -0,0 +1,18 @@
+# EVO1
+
+EVO1 is a Vision-Language-Action policy for robot control. The LeRobot
+integration uses an InternVL3 vision-language backbone with a flow-matching
+action head, and supports staged training through the standard LeRobot policy
+APIs.
+
+The upstream EVO1 project is available at
+[MINT-SJTU/Evo-1](https://github.com/MINT-SJTU/Evo-1).
+
+```bibtex
+@misc{evo1,
+  title = {EVO1},
+  author = {{MINT-SJTU}},
+  year = {2025},
+  howpublished = {\url{https://github.com/MINT-SJTU/Evo-1}},
+}
+```

docs/source/policy_fastwam_README.md

@@ -0,0 +1,56 @@
+## Research Paper
+
+Paper: https://arxiv.org/abs/2603.16666
+
+## Repository
+
+Code: https://github.com/yuantianyuan01/FastWAM
+
+Project page: https://yuantianyuan01.github.io/FastWAM/
+
+## Citation
+
+```bibtex
+@article{yuan2026fastwam,
+  title = {Fast-WAM: Do World Action Models Need Test-time Future Imagination?},
+  author = {Tianyuan Yuan and Zibin Dong and Yicheng Liu and Hang Zhao},
+  journal = {arXiv preprint arXiv:2603.16666},
+  year = {2026},
+  url = {https://arxiv.org/abs/2603.16666}
+}
+```
+
+## Additional Resources
+
+Base video model: https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B
+
+Released upstream checkpoints: https://huggingface.co/yuanty/fastwam
+
+## Results
+
+Evaluated on LIBERO with [`ZibinDong/fastwam_libero_uncond_2cam224`](https://huggingface.co/ZibinDong/fastwam_libero_uncond_2cam224):
+
+| Suite          | Success rate | n_episodes |
+| -------------- | -----------: | ---------: |
+| libero_spatial |        97.6% |        500 |
+| libero_object  |        99.0% |        500 |
+| libero_goal    |        95.0% |        500 |
+| libero_10      |        94.0% |        500 |
+| **average**    |    **96.4%** |       2000 |
+
+Reproduce: `lerobot-eval --policy.path=ZibinDong/fastwam_libero_uncond_2cam224 --policy.device=cuda --policy.torch_dtype=float32 --policy.n_action_steps=10 --env.type=libero --env.task=libero_spatial --env.observation_height=256 --env.observation_width=256 --eval.batch_size=1 --eval.n_episodes=50 --seed=0 --env.episode_length=300`.
+
+For LIBERO-10, use `--env.task=libero_10 --env.episode_length=600`:
+
+```bash
+lerobot-eval \
+    --policy.path=ZibinDong/fastwam_libero_uncond_2cam224 \
+    --policy.device=cuda \
+    --policy.torch_dtype=float32 \
+    --policy.n_action_steps=10 \
+    --env.type=libero \
+    --env.task=libero_10 --env.observation_height=256 --env.observation_width=256 \
+    --eval.batch_size=1 \
+    --eval.n_episodes=50 \
+    --seed=0 --env.episode_length=600
+```

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 +31,108 @@ 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
+
+<details>
+<summary><b>Original-vs-LeRobot parity test</b></summary>
+
+## 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                  |
+
+</details>

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/robocasa.mdx

@@ -166,7 +166,7 @@ lerobot-train \
   --output_dir=./outputs/smolvla_robocasa_CloseFridge \
   --steps=100000 \
   --batch_size=4 \
-  --eval_freq=5000 \
+  --env_eval_freq=5000 \
   --eval.batch_size=1 \
   --eval.n_episodes=5 \
   --save_freq=10000

docs/source/so101.mdx

@@ -122,7 +122,7 @@ The video below shows the sequence of steps for setting the motor ids.
 
 #### Follower
 
-Connect the usb cable from your computer and the power supply to the follower arm's controller board. Then, run the following command or run the API example with the port you got from the previous step. You'll also need to give your leader arm a name with the `id` parameter.
+Connect the usb cable from your computer and the power supply to the follower arm's controller board. Then, run the following command or run the API example with the port you got from the previous step. You'll also need to give your follower arm a name with the `id` parameter.
 
 <hfoptions id="setup_motors">
 <hfoption id="Command">

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
 
@@ -224,6 +265,8 @@ lerobot-dataset-viz \
 
 Once executed, the tool opens `rerun.io` and displays the camera streams, robot states, and actions for the selected episode.
 
+To use [Foxglove](https://foxglove.dev) instead of Rerun, install the extra add `--display-mode foxglove`. This starts a WebSocket server (connect the Foxglove app to `ws://127.0.0.1:8765`) that serves the episode as a seekable timeline you can play/pause and scrub.
+
 For advanced usage—including visualizing datasets stored on a remote server—run:
 
 ```bash

docs/source/video_encoding_parameters.mdx

@@ -2,16 +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 —
-  `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.
-</Tip>
+> [!TIP]
+> 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 `rgb_encoder` is
+> ignored.
 
 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 +32,9 @@ 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
 ```
 
@@ -43,14 +42,12 @@ lerobot-record \
 
 ## Tuning parameters
 
-<Tip warning={true}>
-The defaults are tuned to balance **compression ratio**, **visual quality**, and **decoding/seek speed** for typical robotics datasets. Changing them can affect both recording (CPU load, frame drops) and training (decoding throughput, image quality).
+> [!WARNING]
+> The defaults are tuned to balance **compression ratio**, **visual quality**, and **decoding/seek speed** for typical robotics datasets. Changing them can affect both recording (CPU load, frame drops) and training (decoding throughput, image quality).
+>
+> Only override these parameters if you have a specific reason to, and measure the impact on your pipeline before relying on the new settings.
 
-Only override these parameters if you have a specific reason to, and measure the impact on your pipeline before relying on the new settings.
-
-</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 +62,144 @@ 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.
+
+<div style="margin:28px 0;padding:14px 0;">
+  <div style="margin:0 auto;display:flex;flex-wrap:wrap;justify-content:center;align-items:stretch;gap:6px;font-family:'Source Sans 3',ui-sans-serif,system-ui,sans-serif;font-size:14px;font-weight:600;color:#1B1B1D;">
+    <span style="display:flex;flex-direction:column;justify-content:center;align-items:center;text-align:center;gap:2px;background:#DBEAFE;color:#1D4ED8;border-radius:9px;padding:8px 12px;">
+      <span>Raw depth</span>
+      <span style="font-size:11px;font-weight:400;color:#3B6FD4;white-space:nowrap;">
+        uint16 mm
+        <br />
+        float32 m
+      </span>
+    </span>
+    <span style="display:flex;align-items:center;font-size:16px;color:#C3CBD9;">
+      →
+    </span>
+    <div style="border:2px dashed #C4B5FD;border-radius:13px;padding:18px 12px 12px;position:relative;display:flex;align-items:stretch;gap:6px;">
+      <span style="position:absolute;top:-10px;left:12px;background:#fff;padding:0 6px;font-size:11px;font-weight:700;color:#7E22CE;text-transform:uppercase;letter-spacing:0.5px;white-space:nowrap;">
+        Record time
+      </span>
+      <span style="display:flex;flex-direction:column;justify-content:center;align-items:center;text-align:center;gap:2px;background:#F3E8FF;color:#7E22CE;border-radius:9px;padding:8px 12px;">
+        <span>Clip</span>
+        <span style="font-size:11px;font-weight:400;color:#9061C2;white-space:nowrap;">
+          to [depth_min,
+          <br />
+          depth_max]
+        </span>
+      </span>
+      <span style="display:flex;align-items:center;font-size:16px;color:#C3CBD9;">
+        →
+      </span>
+      <span style="display:flex;flex-direction:column;justify-content:center;align-items:center;text-align:center;gap:2px;background:#F3E8FF;color:#7E22CE;border-radius:9px;padding:8px 12px;">
+        <span>Quantize</span>
+        <span style="font-size:11px;font-weight:400;color:#9061C2;white-space:nowrap;">
+          12-bit codes 0–4095
+          <br />
+          log (default) or linear
+        </span>
+      </span>
+      <span style="display:flex;align-items:center;font-size:16px;color:#C3CBD9;">
+        →
+      </span>
+      <span style="display:flex;flex-direction:column;justify-content:center;align-items:center;text-align:center;gap:2px;background:#F3E8FF;color:#7E22CE;border-radius:9px;padding:8px 12px;">
+        <span>Pack</span>
+        <span style="font-size:11px;font-weight:400;color:#9061C2;white-space:nowrap;">
+          into gray12le
+          <br />
+          plane
+        </span>
+      </span>
+      <span style="display:flex;align-items:center;font-size:16px;color:#C3CBD9;">
+        →
+      </span>
+      <span style="display:flex;flex-direction:column;justify-content:center;align-items:center;text-align:center;gap:2px;background:#F3E8FF;color:#7E22CE;border-radius:9px;padding:8px 12px;">
+        <span>Encode</span>
+        <span style="font-size:11px;font-weight:400;color:#9061C2;white-space:nowrap;">
+          HEVC
+          <br />
+          Main 12
+        </span>
+      </span>
+    </div>
+    <span style="display:flex;align-items:center;font-size:16px;color:#C3CBD9;">
+      →
+    </span>
+    <span style="display:flex;flex-direction:column;justify-content:center;align-items:center;text-align:center;gap:2px;background:#FEF3C7;color:#B45309;border-radius:9px;padding:8px 12px;">
+      <span>MP4</span>
+      <span style="font-size:11px;font-weight:400;color:#C77D18;white-space:nowrap;">
+        stored
+        <br />
+        stream
+      </span>
+    </span>
+    <span style="display:flex;align-items:center;font-size:16px;color:#34A06B;">
+      →
+    </span>
+    <div style="border:2px dashed #6EE7B7;border-radius:13px;padding:18px 12px 12px;position:relative;display:flex;align-items:center;gap:6px;">
+      <span style="position:absolute;top:-10px;left:12px;background:#fff;padding:0 6px;font-size:11px;font-weight:700;color:#047857;text-transform:uppercase;letter-spacing:0.5px;white-space:nowrap;">
+        Load time
+      </span>
+      <span style="display:flex;flex-direction:column;justify-content:center;align-items:center;text-align:center;gap:2px;background:#D1FAE5;color:#047857;border-radius:9px;padding:8px 12px;">
+        <span>Dequantize</span>
+        <span style="font-size:11px;font-weight:400;color:#059669;white-space:nowrap;">
+          to mm / m
+        </span>
+      </span>
+    </div>
+  </div>
+</div>
+
+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 +217,7 @@ 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",
@@ -97,15 +232,16 @@ 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`.
+| Source              | Where it comes from                                   | Fields                                                                                                                  |
+| ------------------- | ----------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------- |
+| **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`, `audio.*` |
+| **Encoder-derived** | Taken from `RGBEncoderConfig` / `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
-  encoder settings partway through a recording is not supported — the
-  `info.json` will only reflect the parameters used for the first episode.
-</Tip>
+> [!IMPORTANT]
+> This block is populated **once**, from the **first** episode. It assumes every
+> 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.
 
 ---
 
@@ -113,5 +249,7 @@ Two sources contribute to the `info` block:
 
 When aggregating datasets with `merge_datasets`, video files are concatenated as-is (no re-encoding), and encoder fields in `info.json` are merged per-key:
 
-- **Stream-derived fields must match** across sources: `video.codec`, `video.pix_fmt`, `video.height`, `video.width`, `video.fps`. Otherwise FFmpeg's concat demuxer fails.
-- **Encoder-tuning fields are merged loosely**: `video.g`, `video.crf`, `video.preset`, `video.fast_decode`, `video.extra_options`. If every source agrees, the value is kept; if not, it's set to `null` (or `{}` for `video.extra_options`) and a warning is logged.
+| Merge rule         | Fields                                                                             | Behaviour                                                                                                                                                 |
+| ------------------ | ---------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| **Must match**     | `video.codec`, `video.pix_fmt`, `video.height`, `video.width`, `video.fps`         | Stream-derived fields must match across sources, otherwise FFmpeg's concat demuxer fails.                                                                 |
+| **Merged loosely** | `video.g`, `video.crf`, `video.preset`, `video.fast_decode`, `video.extra_options` | Encoder-tuning fields. If every source agrees, the value is kept; if not, it's set to `null` (or `{}` for `video.extra_options`) and a warning is logged. |

docs/source/vlabench.mdx

@@ -165,7 +165,7 @@ lerobot-train \
   --output_dir=./outputs/smolvla_vlabench_primitive \
   --steps=100000 \
   --batch_size=4 \
-  --eval_freq=5000 \
+  --env_eval_freq=5000 \
   --eval.batch_size=1 \
   --eval.n_episodes=1 \
   --save_freq=10000

examples/annotations/run_hf_job.py

@@ -0,0 +1,77 @@
+#!/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.
+"""Launch ``lerobot-annotate`` on a Hugging Face job (vllm + Qwen3.6-27B VLM).
+
+Spawns one single-GPU ``h200`` job that:
+
+  1. installs ``lerobot`` from ``main`` plus the annotation extras,
+  2. boots one vllm server with Qwen3.6-27B (dense VLM),
+  3. runs the plan / interjections / vqa modules across the dataset
+     in free-form mode (each episode generates its own subtasks +
+     memory),
+  4. uploads the annotated dataset to ``--new_repo_id`` (when set)
+     or back to ``--repo_id``.
+
+Usage:
+
+    HF_TOKEN=hf_... uv run python examples/annotations/run_hf_job.py
+
+Adjust ``CMD`` (dataset, model, hub repo) and ``flavor`` below for your
+run. For larger datasets, scale to ``h200x4`` and raise
+``--vlm.parallel_servers`` / ``--vlm.num_gpus`` to match.
+"""
+
+import os
+
+from huggingface_hub import get_token, run_job
+
+token = os.environ.get("HF_TOKEN") or get_token()
+if not token:
+    raise RuntimeError("No HF token. Run `huggingface-cli login` or `export HF_TOKEN=hf_...`")
+
+CMD = (
+    "apt-get update -qq && apt-get install -y -qq git ffmpeg && "
+    "pip install --no-deps "
+    "'lerobot @ git+https://github.com/huggingface/lerobot.git@main' && "
+    "pip install --upgrade-strategy only-if-needed "
+    "datasets pyarrow av jsonlines draccus gymnasium torchcodec mergedeep pyyaml-include toml typing-inspect "
+    "openai && "
+    "export VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 && "
+    "export VLLM_VIDEO_BACKEND=pyav && "
+    "lerobot-annotate "
+    "--repo_id=pepijn223/robocasa_pretrain_human300_v4 "
+    "--new_repo_id=pepijn223/robocasa_pretrain_human300_v4_annotated "
+    "--push_to_hub=true "
+    "--vlm.backend=openai "
+    "--vlm.model_id=Qwen/Qwen3.6-27B "
+    "--vlm.num_gpus=1 "
+    '--vlm.serve_command="vllm serve Qwen/Qwen3.6-27B '
+    "--tensor-parallel-size 1 --max-model-len 32768 "
+    '--gpu-memory-utilization 0.8 --uvicorn-log-level warning --port {port}" '
+    "--vlm.serve_ready_timeout_s=1800 "
+    # Qwen3.6 ships with thinking on; annotation wants plain JSON answers.
+    "--vlm.chat_template_kwargs='{\"enable_thinking\": false}'"
+)
+
+job = run_job(
+    image="vllm/vllm-openai:latest",
+    command=["bash", "-c", CMD],
+    flavor="h200",
+    secrets={"HF_TOKEN": token},
+    timeout="2h",
+)
+print(f"Job URL: {job.url}")
+print(f"Job ID:  {job.id}")

examples/isaac_teleop_to_so101/README.md

@@ -0,0 +1,131 @@
+# Isaac Teleop → SO-101
+
+Teleoperate an SO-101/SO-100 follower arm — and record LeRobot datasets — with NVIDIA
+[Isaac Teleop](https://github.com/NVIDIA/IsaacTeleop). Two input devices ship today:
+
+- **XR (VR) controller** (`--teleop.type=xr_controller`) — the controller's grip pose drives the
+  end-effector through a squeeze-to-engage clutch and LeRobot's Cartesian IK pipeline; the analog
+  trigger drives the gripper.
+- **SO-101 leader arm** (`--teleop.type=so101_leader`) — a back-drivable leader arm mirrored 1:1
+  onto the follower via Isaac Teleop's native `so101_leader` plugin (no clutch, no IK).
+
+The full narrative guide (how the clutch works, CloudXR setup, headset pairing, tuning, and
+troubleshooting) is in the [LeRobot docs](https://huggingface.co/docs/lerobot/isaac_teleop)
+(source: `docs/source/isaac_teleop.mdx`). This README is the canonical install and usage
+reference.
+
+## Requirements
+
+- Linux workstation (see NVIDIA's
+  [system requirements](https://nvidia.github.io/IsaacTeleop/main/references/requirements.html)
+  for supported OS/GPU/headset combinations; `isaacteleop` publishes Linux wheels only).
+- An SO-101 (or SO-100) follower arm, calibrated with `lerobot-calibrate`.
+- For the XR device: a CloudXR-capable headset (e.g. Quest 3, Pico 4, Apple Vision Pro) on the
+  same network.
+- For the leader device: a second, back-drivable SO-101 leader arm and the `so101_leader` plugin
+  binary built from the Isaac Teleop source tree (see
+  [Build from source](https://nvidia.github.io/IsaacTeleop/main/getting_started/build_from_source/index.html)).
+
+## Installation
+
+This example lives in the LeRobot repository and is not part of the `lerobot` pip package, so
+work from a source checkout. From the repo root:
+
+```bash
+# LeRobot with the extras this example uses:
+#   feetech    - SO-101 serial motor bus
+#   kinematics - Placo IK solver (XR controller path)
+#   dataset    - dataset recording (record.py)
+# huggingface_hub >= 1.5 is needed by the automatic URDF fetch (Buckets API).
+uv pip install -e ".[feetech,kinematics,dataset]" "huggingface_hub>=1.5"
+
+# Isaac Teleop from public PyPI. `cloudxr` brings the CloudXR runtime bindings;
+# `retargeters-lite` is the scipy-based retargeter path that resolves on both
+# x86_64 and ARM (the full `retargeters` extra does not resolve on aarch64).
+uv pip install "isaacteleop[cloudxr,retargeters-lite]~=1.3.131" "scipy>=1.14"
+
+# Optional, x86_64 only: the full retargeter stack.
+uv pip install "isaacteleop[retargeters]~=1.3.131"
+```
+
+One-time CloudXR EULA (the auto-launch prompts on stdin and would hang on a headless machine):
+
+```bash
+python -m isaacteleop.cloudxr --accept-eula
+```
+
+## Usage
+
+Run everything from the repo root with `python -m` so the `examples` package resolves.
+
+### Teleoperate — XR controller
+
+```bash
+python -m examples.isaac_teleop_to_so101.teleoperate \
+    --robot.type=so101_follower \
+    --robot.port=/dev/ttyACM0 \
+    --robot.id=so101_follower_arm \
+    --teleop.type=xr_controller
+```
+
+On startup the script launches the CloudXR runtime (~30 s), prints the workstation IP to enter in
+the headset's CloudXR web client, waits for the controllers to stream, slews the arm to a reset
+pose (`--reset_to_origin=false` to skip), and then: **hold the squeeze/grip** to engage, move the
+controller to drive the arm, pull the trigger to close the gripper. Releasing the squeeze freezes
+the arm. The SO-101 URDF is fetched automatically from the `lerobot/robot-urdfs` Hugging Face
+bucket into the LeRobot cache on first run.
+
+To customize the reset pose: back-drive the arm to the pose you want, then
+
+```bash
+python -m examples.isaac_teleop_to_so101.override_reset_pose --port /dev/ttyACM0 --id so101_follower_arm
+```
+
+which writes it to `HF_LEROBOT_HOME/reset_poses/<robot.name>/<robot.id>.json`; runs with the same
+`--robot.id` use it automatically.
+
+### Teleoperate — SO-101 leader arm
+
+```bash
+python -m examples.isaac_teleop_to_so101.teleoperate \
+    --robot.type=so101_follower --robot.port=/dev/ttyACM0 --robot.id=so101_follower_arm \
+    --teleop.type=so101_leader --teleop.port=/dev/ttyACM1 --teleop.id=so101_leader_arm \
+    --launch_plugin=/path/to/IsaacTeleop/install/plugins/so101_leader/so101_leader_plugin
+```
+
+The follower is first slewed to the leader's pose over `--align_duration` seconds
+(`--align=false` to skip), then mirrors it 1:1. The plugin reuses the serial leader's calibration
+(`HF_LEROBOT_CALIBRATION/teleoperators/so_leader/<teleop.id>.json`).
+
+### Record a dataset
+
+`record.py` takes the same `--robot.*`/`--teleop.*`/loop flags plus `lerobot-record`-style
+`--dataset.*` flags:
+
+```bash
+python -m examples.isaac_teleop_to_so101.record \
+    --robot.type=so101_follower --robot.port=/dev/ttyACM0 --robot.id=so101_follower_arm \
+    --teleop.type=xr_controller \
+    --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30}}" \
+    --dataset.repo_id=<hf_user>/<dataset_name> \
+    --dataset.single_task="Pick up the cube" \
+    --dataset.num_episodes=3 --dataset.episode_time_s=20 --dataset.reset_time_s=5
+```
+
+Keyboard shortcuts (terminal-first, so they work over SSH): **Right/n** end episode early,
+**Left/r** re-record, **Esc/q** stop after the current episode.
+
+Run either script with `--help` for all flags.
+
+## Layout
+
+```
+isaac_teleop/            device library: session lifecycle (base.py), XRController,
+                         SO101LeaderArm, Clutch, configs, and the XR→IK processor step
+common.py                shared loop infra: device bundles, clutch/IK pipeline wiring,
+                         reset/align slews, URDF fetch, keyboard listener
+teleoperate.py           teleoperation CLI (device selected via --teleop.type)
+record.py                dataset-recording CLI (same device selection + --dataset.*)
+override_reset_pose.py   save the current joints as the per-arm reset pose
+default.env              CloudXR device-profile overrides passed to the launcher
+```

examples/isaac_teleop_to_so101/__init__.py

@@ -0,0 +1,17 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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 Teleop -> SO-101 example package."""

examples/isaac_teleop_to_so101/common.py

@@ -0,0 +1,650 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""Shared device + control-loop infrastructure for the Isaac Teleop -> SO-101 examples.
+
+Consumed by ``teleoperate.py`` and ``record.py``, which both build a per-device
+:class:`Device` bundle and run the same loop: read -> (maybe command) -> hold-when-idle ->
+sleep. A :class:`Device` bundles three closures: ``compute(obs) -> RobotAction | None``
+(``None`` = hold at the measured pose while idle), ``startup``, and ``cleanup``. The devices:
+
+* ``xr_controller`` — a thin :class:`XRController` whose raw grip pose an in-loop
+  :class:`Clutch` turns into an EE target for LeRobot's Cartesian IK pipeline.
+* ``so101_leader`` — a back-drivable leader arm mirrored 1:1 into the follower.
+
+Requires the ``isaacteleop`` package and an OpenXR runtime (install instructions in this
+folder's ``README.md``). User-facing guide: ``docs/source/isaac_teleop.mdx``.
+"""
+
+import json
+import logging
+import socket
+import subprocess
+import sys
+import time
+from collections.abc import Callable
+from contextlib import suppress
+from dataclasses import dataclass
+from importlib.resources import files
+from pathlib import Path
+from typing import Protocol
+
+import numpy as np
+
+from lerobot.model.kinematics import RobotKinematics
+from lerobot.processor import (
+    RobotProcessorPipeline,
+    robot_action_observation_to_transition,
+    transition_to_robot_action,
+)
+from lerobot.robots import RobotConfig, make_robot_from_config
+from lerobot.robots.so_follower import SOFollowerConfig  # noqa: F401  (registers so101_follower)
+from lerobot.robots.so_follower.robot_kinematic_processor import (
+    EEBoundsAndSafety,
+    InverseKinematicsEEToJoints,
+)
+from lerobot.types import RobotAction, RobotObservation
+from lerobot.utils.constants import HF_LEROBOT_CALIBRATION, HF_LEROBOT_HOME, TELEOPERATORS
+from lerobot.utils.robot_utils import precise_sleep
+
+from .isaac_teleop import (
+    Clutch,
+    IsaacTeleopConfig,
+    MapXRControllerActionToRobotAction,
+    SO101LeaderArm,
+    SO101LeaderArmConfig,
+    XRController,
+)
+
+# Fixed rate [Hz] for the teleoperate loop and the pre-loop slews / connect-wait poll sleeps.
+FPS = 30
+
+# CloudXR device-profile env file passed to the launcher (see default.env in this package).
+CLOUDXR_ENV_FILE = str(files(__package__) / "default.env")
+
+
+class LoopConfig(Protocol):
+    """Structural type for the loop/launch knobs ``build_device`` and the ``setup_*`` read.
+
+    Both ``TeleoperateConfig`` and ``RecordConfig`` satisfy it, keeping ``common`` decoupled
+    from either entry point's concrete config.
+    """
+
+    teleop: IsaacTeleopConfig
+    robot: RobotConfig
+    launch_plugin: str | None
+    reset_to_origin: bool
+    reset_duration: float
+    align: bool
+    align_duration: float
+
+
+# Per-device bundle consumed by the shared loop. ``compute`` returns None to mean
+# "idle -> hold at the measured pose"; ``startup`` warms up; ``cleanup`` reaps/disconnects.
+@dataclass(frozen=True)
+class Device:
+    compute: Callable[[RobotObservation | None], RobotAction | None]
+    startup: Callable[[], None]
+    cleanup: Callable[[], None]
+
+
+def hold_action(obs: RobotObservation, motor_names: list[str]) -> dict[str, float]:
+    """Re-send the measured joints — the explicit hold when a device is idle."""
+    return {f"{name}.pos": float(obs[f"{name}.pos"]) for name in motor_names}
+
+
+class HoldLatch:
+    """Resolve the per-frame action, holding one LATCHED pose while the device is idle.
+
+    Re-sending the freshly measured joints on every idle frame would ratchet the arm
+    downward: under gravity the P-only servo settles below its goal by a steady-state
+    error, so each re-command of the measurement lowers the goal by that error again.
+    Latching the target once on the active->idle transition holds a fixed pose instead.
+    """
+
+    def __init__(self, motor_names: list[str]):
+        self._motor_names = motor_names
+        self._held: dict[str, float] | None = None
+
+    def resolve(self, action: RobotAction | None, obs: RobotObservation) -> RobotAction:
+        """Pass through an active action (clearing the latch); latch + hold when idle."""
+        if action is not None:
+            self._held = None
+            return action
+        if self._held is None:
+            self._held = hold_action(obs, self._motor_names)
+        return self._held
+
+
+def slew(
+    robot,
+    motor_names: list[str],
+    target_fn: Callable[[], dict[str, float]],
+    duration_s: float,
+) -> None:
+    """Linearly slew all joints from their current measured pose toward a target.
+
+    ``target_fn`` is called EACH step, so the leader can pass a live re-read (landing on its
+    current pose at ``alpha == 1`` for a continuous handoff) while XR passes a constant.
+    """
+    obs = robot.get_observation()
+    start = {name: float(obs[f"{name}.pos"]) for name in motor_names}
+    n_steps = max(1, int(duration_s * FPS))
+    for step in range(1, n_steps + 1):
+        alpha = step / n_steps
+        target = target_fn()
+        action = {f"{name}.pos": start[name] + alpha * (target[name] - start[name]) for name in motor_names}
+        robot.send_action(action)
+        precise_sleep(1.0 / FPS)
+
+
+# ============================================================================
+# XR controller device
+# ============================================================================
+
+# Per-frame EE rate limit [m]. With raise_on_jump=False, EEBoundsAndSafety clamps an
+# over-limit step instead of raising, absorbing a tracking glitch as one slow frame. At
+# FPS=30, 0.1 m/frame caps EE speed at ~3 m/s. (end_effector_bounds clips the absolute target.)
+MAX_EE_STEP_M = 0.1
+
+# Soft-orientation IK weight: small but nonzero so the wrist follows the hand while position
+# dominates (the 5-DOF SO-101 cannot realize an arbitrary orientation). 0.0 = position-only.
+IK_ORIENTATION_WEIGHT = 0.01
+
+
+def _ensure_so101_urdf() -> str:
+    """Return the cached SO-101 URDF path, fetching the ``so101`` folder (URDF + meshes) from
+    the public ``lerobot/robot-urdfs`` HF bucket into the LeRobot cache on first use."""
+    dest_dir = HF_LEROBOT_HOME / "robot-urdfs" / "so101"
+    urdf_path = dest_dir / "so101_new_calib.urdf"
+    # Completeness marker written only after a FULL sync: the URDF file alone is not a
+    # completeness signal (an interrupted first sync can leave the meshes it references
+    # missing, which the URDF's mere existence would then hide forever). Re-syncing is
+    # idempotent and repairs a partial cache; delete the folder to force a re-download.
+    marker = dest_dir / ".sync_complete"
+    if not marker.exists():
+        from huggingface_hub import sync_bucket
+
+        sync_bucket("hf://buckets/lerobot/robot-urdfs/so101", str(dest_dir), quiet=True)
+        marker.touch()
+    return str(urdf_path)
+
+
+# Default duration [s] for the startup reset-to-origin slew.
+RESET_DURATION_S = 5.0
+
+# Optional cached file written by override_reset_pose.py. When present it takes priority over RESET_ORIGIN_DEG.
+RESET_POSE_FILE = str(HF_LEROBOT_HOME / "reset_poses" / "{robot_name}" / "{robot_id}.json")
+
+# Reset target in each motor's native units (arm joints in degrees, gripper RANGE_0_100,
+# 100 = open). An empirically comfortable pose (elbow/wrist bent) avoiding the singularity of
+# a fully-extended arm; assumes standard calibration. Override per-arm via override_reset_pose.py.
+RESET_ORIGIN_DEG: dict[str, float] = {
+    "shoulder_pan": -4.0,
+    "shoulder_lift": -103.0,
+    "elbow_flex": 97.0,
+    "wrist_flex": 78.0,
+    "wrist_roll": -65.0,
+    "gripper": 0.0,
+}
+
+
+def _load_reset_target(reset_pose_file: Path, motor_names: list[str]) -> dict[str, float]:
+    """Return reset targets: the saved reset pose if present, else RESET_ORIGIN_DEG."""
+    if reset_pose_file.exists():
+        saved = json.loads(reset_pose_file.read_text())
+        # Fill any missing motors from the fallback dict.
+        return {name: float(saved.get(name, RESET_ORIGIN_DEG.get(name, 0.0))) for name in motor_names}
+    return {name: RESET_ORIGIN_DEG.get(name, 0.0) for name in motor_names}
+
+
+# CloudXR web client URL opened in the headset (Isaac Teleop quick start, step 5).
+_CLOUDXR_WEB_CLIENT_URL = "https://nvidia.github.io/IsaacTeleop/client"
+# WSS-proxy / self-signed-cert port the operator accepts in-browser before connecting.
+_CLOUDXR_WSS_PORT = 48322
+# How often to re-print the connection hint while waiting for the headset [s].
+_XR_CONNECT_REMINDER_S = 15.0
+# Virtual / bridge / USB-gadget interfaces a headset can't reach over the network — skip
+# by name prefix (``docker0``, compose ``br-*``, ``veth*``, libvirt ``virbr*``, and the
+# Tegra USB device-mode bridge ``l4tbr0``).
+_SKIP_IFACE_PREFIXES = ("docker", "br-", "veth", "virbr", "l4tbr")
+
+
+def _primary_ipv4() -> str | None:
+    """The workstation's primary outbound IPv4, via the UDP-socket trick (``connect()`` on a
+    datagram socket selects the egress interface without sending packets)."""
+    with socket.socket(socket.AF_INET, socket.SOCK_DGRAM) as s:
+        try:
+            s.connect(("8.8.8.8", 80))
+            return s.getsockname()[0]
+        except OSError:
+            return None
+
+
+def _candidate_ipv4s() -> list[tuple[str, str]]:
+    """Return ``[(interface, ipv4), ...]`` the headset might reach this workstation at.
+
+    Lists each interface's IPv4 via ``psutil`` (dropping loopback, link-local, and the
+    virtual/bridge interfaces in ``_SKIP_IFACE_PREFIXES``), primary outbound first. Falls
+    back to just the primary IP when ``psutil`` is unavailable.
+    """
+    primary = _primary_ipv4()
+    found: list[tuple[str, str]] = []
+    try:
+        import psutil
+
+        for iface, addrs in psutil.net_if_addrs().items():
+            if iface.startswith(_SKIP_IFACE_PREFIXES):
+                continue
+            for addr in addrs:
+                if addr.family != socket.AF_INET:
+                    continue
+                ip = addr.address
+                if ip.startswith("127.") or ip.startswith("169.254."):
+                    continue
+                found.append((iface, ip))
+    except Exception:
+        if primary:
+            found.append(("default", primary))
+    found.sort(key=lambda t: t[1] != primary)  # primary outbound interface first
+    return found
+
+
+def _print_xr_connect_help() -> None:
+    """Print how to connect the headset to this workstation over CloudXR."""
+    ips = _candidate_ipv4s()
+    print("\n" + "=" * 76)
+    print("Connect your XR headset to this workstation over NVIDIA CloudXR:")
+    print(f"  1. In the headset, open the CloudXR web client:  {_CLOUDXR_WEB_CLIENT_URL}")
+    print("  2. Enter this workstation's IP address:")
+    if ips:
+        for iface, ip in ips:
+            print(f"        {ip:<15}  ({iface})")
+        if len(ips) > 1:
+            print("     (use the address on the same network as your headset)")
+    else:
+        print("        <could not determine — check `hostname -I` / `ip addr`>")
+    print(f"  3. Accept the self-signed cert at https://<that-ip>:{_CLOUDXR_WSS_PORT}/ , then Connect.")
+    print("=" * 76 + "\n")
+
+
+def _wait_for_xr_controller(teleop_device: XRController) -> None:
+    """Block until the XR controller is tracked, polling ``get_action()`` and re-printing a
+    reminder every ``_XR_CONNECT_REMINDER_S``. User-paced; ``Ctrl-C`` aborts (no hard timeout).
+    """
+    _print_xr_connect_help()
+    print("Waiting for the headset controllers to start streaming…  (Ctrl-C to abort)")
+    last_reminder = time.time()
+    while True:
+        teleop_device.get_action()  # steps the session; updates is_tracking
+        if teleop_device.is_tracking:
+            print("Headset connected — controllers are streaming.")
+            return
+        if time.time() - last_reminder >= _XR_CONNECT_REMINDER_S:
+            print("…still waiting for the headset to connect (Ctrl-C to abort).")
+            last_reminder = time.time()
+        time.sleep(1.0 / FPS)
+
+
+def setup_xr(cfg: LoopConfig, robot, motor_names: list[str]) -> Device:
+    """Build the XR controller device bundle (clutch + soft-orientation IK pipeline)."""
+    kinematics_solver = RobotKinematics(
+        urdf_path=_ensure_so101_urdf(),
+        target_frame_name="gripper_frame_link",
+        joint_names=motor_names,
+    )
+
+    teleop_config = cfg.teleop  # XRControllerConfig (selected via --teleop.type=xr_controller)
+    teleop_device = XRController(teleop_config)
+
+    # The clutch (below) turns the raw grip pose into an absolute base-frame ee_pose; this
+    # pipeline maps it to joint targets: rename -> bounds/rate-limit -> IK.
+    xr_to_robot_joints_processor = RobotProcessorPipeline[tuple[RobotAction, RobotObservation], RobotAction](
+        steps=[
+            MapXRControllerActionToRobotAction(),
+            # raise_on_jump=False: an over-limit step (e.g. a tracking glitch) is clamped +
+            # warned instead of raised, since a crash mid-loop would leave the arm uncontrolled.
+            # z floor 0.0 keeps a stray target above the table; x/y stay at a loose [-1,1]m box.
+            EEBoundsAndSafety(
+                end_effector_bounds={"min": [-1.0, -1.0, 0.0], "max": [1.0, 1.0, 1.0]},
+                max_ee_step_m=MAX_EE_STEP_M,
+                raise_on_jump=False,
+            ),
+            # initial_guess_current_joints=False: warm-start from the previous IK solution so
+            # the joint trajectory stays continuous frame-to-frame.
+            InverseKinematicsEEToJoints(
+                kinematics=kinematics_solver,
+                motor_names=motor_names,
+                initial_guess_current_joints=False,
+                orientation_weight=IK_ORIENTATION_WEIGHT,
+            ),
+        ],
+        to_transition=robot_action_observation_to_transition,
+        to_output=transition_to_robot_action,
+    )
+
+    # The clutch is built in startup() (after the optional reset slew, seeded from the
+    # post-slew MEASURED pose) and shared with compute() via nonlocal.
+    clutch: Clutch | None = None
+    prev_enabled = False
+
+    def startup() -> None:
+        nonlocal clutch
+        # Connect and wait for the operator to don the headset BEFORE moving the arm, so the
+        # reset slew happens while they are watching in VR.
+        teleop_device.connect()
+        if not teleop_device.is_connected:
+            raise ValueError("Teleop is not connected!")
+        _wait_for_xr_controller(teleop_device)
+
+        if cfg.reset_to_origin:
+            reset_pose_file = Path(RESET_POSE_FILE.format(robot_name=robot.name, robot_id=robot.id))
+            target = _load_reset_target(reset_pose_file, motor_names)
+            source = str(reset_pose_file) if reset_pose_file.exists() else "hardcoded defaults"
+            print(f"Reset target source: {source}")
+            print(f"Resetting to origin over {cfg.reset_duration:.1f} s…")
+            slew(robot, motor_names, lambda: target, cfg.reset_duration)
+            print("Reset complete.")
+
+        # Seed the clutch home from the arm's measured pose (FK of the current joints) so the
+        # first engage is jump-free, whether or not a reset slew ran.
+        obs0 = robot.get_observation()
+        q_measured_deg = np.array([float(obs0[f"{name}.pos"]) for name in motor_names], dtype=float)
+        home_base_T_ee = kinematics_solver.forward_kinematics(q_measured_deg)  # noqa: N806
+        clutch = Clutch(home_base_T_ee)
+
+        print("Starting teleop loop. Squeeze and move the controller to teleoperate the robot...")
+
+    def compute(robot_obs: RobotObservation | None) -> RobotAction | None:
+        nonlocal prev_enabled
+        if clutch is None:  # set in startup(), which runs before compute()
+            raise RuntimeError("compute() called before startup(); the clutch is not initialized")
+        xr_action = teleop_device.get_action()
+        grip_pos = np.asarray(xr_action["grip_pos"], dtype=float)
+        grip_quat = np.asarray(xr_action["grip_quat"], dtype=float)
+        squeeze = float(xr_action["squeeze"])
+        trigger = float(xr_action["trigger"])
+        enabled = squeeze > teleop_config.clutch_threshold
+
+        # On the engage edge, latch the clutch home at the arm's MEASURED EE pose (FK of
+        # the live joints) and the controller origin so the per-frame delta starts at zero.
+        # Latching the last commanded pose instead would snap the arm back to it at full
+        # servo speed if the arm moved while disengaged (gravity sag, external contact).
+        is_engage_frame = enabled and not prev_enabled
+        if is_engage_frame:
+            q_measured = np.array([float(robot_obs[f"{name}.pos"]) for name in motor_names], dtype=float)
+            measured_base_T_ee = kinematics_solver.forward_kinematics(q_measured)  # noqa: N806
+            clutch.engage(grip_pos, grip_quat, measured_base_T_ee=measured_base_T_ee)
+            # Re-anchor the pipeline state at the measured pose as well: EEBoundsAndSafety's
+            # rate limiter and the IK warm start otherwise still reference the stale
+            # pre-disengage command and would fight the fresh home for several frames.
+            xr_to_robot_joints_processor.reset()
+        prev_enabled = enabled
+
+        # SAFETY GATE: command the robot ONLY while the clutch is engaged; otherwise return
+        # None so the loop holds the measured joints (releasing the clutch freezes the arm).
+        if not enabled:
+            return None
+
+        # Rebase the raw grip pose onto the EE, then run the pipeline. closedness = trigger.
+        ee_pos, ee_quat = clutch.rebase(grip_pos, grip_quat)
+        ee_action = {
+            "ee_pose": np.concatenate([ee_pos, ee_quat]).astype(np.float32),
+            "closedness": trigger,
+        }
+        return xr_to_robot_joints_processor((ee_action, robot_obs))
+
+    return Device(compute=compute, startup=startup, cleanup=teleop_device.disconnect)
+
+
+# ============================================================================
+# SO-101 leader arm device
+# ============================================================================
+
+# Default duration [s] for the startup alignment slew (follower current -> leader first pose).
+ALIGN_DURATION_S = 3.0
+
+# How long to wait for the leader plugin to start streaming before aligning / looping.
+LEADER_WARMUP_TIMEOUT_S = 20.0
+
+# The plugin converts the leader's servo ticks to radians, so it reuses the serial SO-101
+# leader's calibration, stored by lerobot-calibrate under SO101Leader.name == "so_leader".
+SO_LEADER_CALIBRATION_NAME = "so_leader"
+
+
+def _leader_calibration_path(cfg: LoopConfig) -> Path | None:
+    """Infer the calibration JSON the launched plugin should read, or None.
+
+    Path convention: ``HF_LEROBOT_CALIBRATION / teleoperators / so_leader / {--teleop.id}.json``
+    (or ``--teleop.calibration_dir`` if set). Returns None (plugin falls back to defaults) when
+    it does not exist, warning if an id was given, or when no ``--teleop.id`` is set.
+    """
+    if not cfg.teleop.id:
+        return None
+    calib_dir = cfg.teleop.calibration_dir or (
+        HF_LEROBOT_CALIBRATION / TELEOPERATORS / SO_LEADER_CALIBRATION_NAME
+    )
+    calib_path = Path(calib_dir) / f"{cfg.teleop.id}.json"
+    if calib_path.is_file():
+        return calib_path
+    print(
+        f"WARNING: no leader calibration at {calib_path}; the plugin will use built-in defaults. "
+        f"Calibrate with the serial leader (`lerobot-calibrate --teleop.type=so101_leader "
+        f"--teleop.id={cfg.teleop.id}`) or the plugin's `calibrate` subcommand."
+    )
+    return None
+
+
+def _wait_for_leader(teleop: SO101LeaderArm, timeout_s: float) -> dict[str, float]:
+    """Poll the leader until it streams a live frame; return that frame's ``{joint}.pos``.
+
+    Raises ``SystemExit`` if no live frame arrives within ``timeout_s`` (plugin not pushing,
+    wrong ``--teleop.collection_id``, or CloudXR not up).
+    """
+    print(f"Waiting up to {timeout_s:.0f}s for the so101_leader plugin to stream…")
+    deadline = time.time() + timeout_s
+    while time.time() < deadline:
+        action = teleop.get_action()
+        if teleop.is_tracking:
+            print("Leader is streaming.")
+            return action
+        time.sleep(1.0 / FPS)
+    raise SystemExit(
+        f"FAILED: leader did not stream within {timeout_s:.0f}s. Is the so101_leader plugin "
+        "running and pushing (check --teleop.collection_id)? Is CloudXR up?"
+    )
+
+
+def _maybe_launch_plugin(cfg: LoopConfig) -> subprocess.Popen | None:
+    """Spawn the so101_leader plugin if ``--launch_plugin <path>`` was given (after connect())."""
+    if cfg.launch_plugin is None:
+        return None
+    if not Path(cfg.launch_plugin).exists():
+        raise SystemExit(
+            f"plugin binary not found: {cfg.launch_plugin} (build it in the IsaacTeleop repo first)"
+        )
+    leader_port = cfg.teleop.port  # SO101LeaderArmConfig.port, forwarded to the plugin
+    backend = f"leader on {leader_port}" if leader_port else "synthetic trajectory"
+    print(f"launching plugin: {cfg.launch_plugin} ({backend})")
+    # Positional args: [device_path] [collection_id] [calibration_file]. Empty device_path ->
+    # synthetic backend. Calibration (only real hardware needs it) is appended when a port is set.
+    argv = [cfg.launch_plugin, leader_port, cfg.teleop.collection_id]
+    if leader_port:
+        calib_path = _leader_calibration_path(cfg)
+        if calib_path is not None:
+            argv.append(str(calib_path))
+            print(f"  leader calibration: {calib_path}")
+    # Spawned after connect() so it inherits the CloudXR runtime env (XR_RUNTIME_JSON, ...).
+    proc = subprocess.Popen(argv)
+    time.sleep(1.5)  # let it create its OpenXR session and start pushing
+    return proc
+
+
+def setup_leader(cfg: LoopConfig, robot, motor_names: list[str]) -> Device:
+    """Build the SO-101 leader arm device bundle (1:1 joint mirror)."""
+    teleop_config = cfg.teleop  # SO101LeaderArmConfig (selected via --teleop.type=so101_leader)
+    teleop = SO101LeaderArm(teleop_config)
+
+    plugin_proc: subprocess.Popen | None = None
+
+    def startup() -> None:
+        nonlocal plugin_proc
+        # connect() auto-launches CloudXR (unless opted out); spawn the plugin AFTER so it
+        # inherits the runtime env. The plugin is reaped in cleanup().
+        teleop.connect()
+        plugin_proc = _maybe_launch_plugin(cfg)
+
+        if not teleop.is_connected:
+            raise ValueError("Teleop is not connected!")
+
+        # Block until the leader streams a live frame (clear error if it never does).
+        _wait_for_leader(teleop, LEADER_WARMUP_TIMEOUT_S)
+
+        if cfg.align:
+            print(f"Aligning follower to leader over {cfg.align_duration:.1f}s…")
+
+            # Re-read the live leader pose once per step so alpha=1 lands on its current pose
+            # from a single coherent frame.
+            def _leader_target() -> dict[str, float]:
+                leader_now = teleop.get_action()
+                return {name: float(leader_now[f"{name}.pos"]) for name in motor_names}
+
+            slew(robot, motor_names, _leader_target, cfg.align_duration)
+            print("Alignment complete.")
+
+        print(
+            "Starting joint-mirror loop. Back-drive the leader to teleoperate the follower… (Ctrl-C to stop)"
+        )
+
+    def compute(robot_obs: RobotObservation | None) -> RobotAction | None:
+        leader_action = teleop.get_action()
+        # Hold the follower at its measured pose when the leader drops out (stale stream)
+        # rather than commanding a possibly-old target.
+        if not teleop.is_tracking:
+            return None
+        return leader_action
+
+    def cleanup() -> None:
+        # A plugin-reaping failure must not skip the session disconnect (and vice versa
+        # the disconnect runs after the plugin stops pushing on it).
+        try:
+            if plugin_proc is not None:
+                plugin_proc.terminate()
+                try:
+                    plugin_proc.wait(timeout=5)
+                except subprocess.TimeoutExpired:
+                    plugin_proc.kill()
+        finally:
+            teleop.disconnect()
+
+    return Device(compute=compute, startup=startup, cleanup=cleanup)
+
+
+# ============================================================================
+# Shared setup
+# ============================================================================
+
+
+def build_device(cfg: LoopConfig) -> tuple:
+    """Connect the follower, build the selected Isaac device, and run its pre-loop startup.
+
+    Connects the follower FIRST (so the startup slew / clutch-home seed can read live joints),
+    dispatches on ``--teleop.type``, then runs ``device.startup()`` before returning. On any
+    failure after ``connect()`` the follower is disconnected so the connection never leaks.
+
+    Returns ``(robot, device, motor_names)``.
+    """
+    # Default the CloudXR input profile to this example's default.env unless the user overrode
+    # it via --teleop.cloudxr_env_file.
+    if cfg.teleop.cloudxr_env_file is None:
+        cfg.teleop.cloudxr_env_file = CLOUDXR_ENV_FILE
+
+    # SO-101/SO-100 only (both share the SO-101 URDF), reject other followers.
+    supported_robots = {"so101_follower", "so100_follower"}
+    if cfg.robot.type not in supported_robots:
+        raise ValueError(
+            f"This example only supports SO-101/SO-100 followers ({sorted(supported_robots)}), "
+            f"but got --robot.type={cfg.robot.type}."
+        )
+
+    # The degree-based pipeline relies on --robot.use_degrees (default True).
+    robot = make_robot_from_config(cfg.robot)
+    # Connect FIRST so the startup slew and clutch-home seed can read live joints.
+    robot.connect()
+    # Everything after connect() can fail; this runs outside the callers' try/finally, so
+    # disconnect the follower on any failure to avoid leaking the connection.
+    device: Device | None = None
+    try:
+        # Joint names in action order, read from {name}.pos action features (robot-agnostic).
+        motor_names = [key.removesuffix(".pos") for key in robot.action_features if key.endswith(".pos")]
+
+        if isinstance(cfg.teleop, SO101LeaderArmConfig):
+            device = setup_leader(cfg, robot, motor_names)
+        else:
+            device = setup_xr(cfg, robot, motor_names)
+
+        device.startup()
+    except BaseException:
+        # Reap a partially-started device, then always disconnect the follower.
+        if device is not None:
+            with suppress(Exception):
+                device.cleanup()
+        robot.disconnect()
+        raise
+
+    return robot, device, motor_names
+
+
+# ============================================================================
+# Keyboard control
+# ============================================================================
+
+
+def init_keyboard_listener():
+    """Recording shortcuts, terminal-first so they work over SSH.
+
+    Whenever stdin is a TTY we use the stdlib :class:`TerminalKeyListener` directly rather
+    than upstream's pynput-first :func:`init_keyboard_listener`, whose global listener would
+    capture the workstation console instead of this (often SSH) terminal. With no TTY we defer
+    to upstream (pynput on a GUI, else headless no-op).
+    """
+    if not (sys.stdin is not None and sys.stdin.isatty()):
+        from lerobot.utils.keyboard_input import init_keyboard_listener as _upstream
+
+        return _upstream()
+
+    from lerobot.utils.keyboard_input import TerminalKeyListener, apply_recording_control
+
+    events = {"exit_early": False, "rerecord_episode": False, "stop_recording": False}
+
+    # n/r/q are the arrow/Esc equivalents that survive escape-sequence splitting over laggy
+    # SSH/VNC links. Case-insensitive so Shift+letter still works.
+    def on_key(name: str) -> None:
+        key = name.lower()
+        if key in ("right", "n"):
+            apply_recording_control("right", events)
+        elif key in ("left", "r"):
+            apply_recording_control("left", events)
+        elif key in ("esc", "q"):
+            apply_recording_control("esc", events)
+
+    listener = TerminalKeyListener(on_key)
+    listener.start()
+    logging.info(
+        "Keyboard control via terminal — keep this terminal focused: "
+        "Right/n = end episode early, Left/r = re-record, Esc/q = stop."
+    )
+    return listener, events

examples/isaac_teleop_to_so101/default.env

@@ -0,0 +1,21 @@
+# CloudXR device-profile overrides for the Isaac Teleop XR -> SO-101 example.
+#
+# Passed to isaacteleop's CloudXRLauncher as `env_config` (via
+# XRControllerConfig.cloudxr_env_file). Format: KEY=value, one per line; `#`
+# comments and blank lines ignored; $VARS / ~ expanded. See
+# isaacteleop/cloudxr/env_config.py::_load_env_file.
+#
+# Runtime-resolved keys (XR_RUNTIME_JSON, XRT_NO_STDIN, NV_CXR_RUNTIME_DIR,
+# NV_CXR_OUTPUT_DIR) are reserved and ignored if set here.
+
+# Transport profile the runtime advertises (CloudXR default: auto-webrtc).
+# "Quest3" also covers the Pico 4. Other values: auto-native, AppleVisionPro.
+NV_DEVICE_PROFILE=Quest3
+
+# Input device discovery channels (both default to true; pinned for clarity).
+NV_CXR_ENABLE_PUSH_DEVICES=true
+NV_CXR_ENABLE_TENSOR_DATA=true
+
+# Runtime logs to ~/.cloudxr/logs — helps debug connection issues
+# (e.g. "Failed to get OpenXR system: -35").
+NV_CXR_FILE_LOGGING=true

examples/isaac_teleop_to_so101/isaac_teleop/__init__.py

@@ -0,0 +1,40 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""NVIDIA Isaac Teleop teleoperators for LeRobot.
+
+Each input device is an :class:`IsaacTeleopTeleoperator` subclass: :class:`XRController`
+(XR/VR controller) and :class:`SO101LeaderArm` (back-drivable SO-101 leader arm) ship today.
+"""
+
+from .base import IsaacTeleopTeleoperator
+from .clutch import Clutch
+from .config_isaac_teleop import IsaacTeleopConfig, SO101LeaderArmConfig, XRControllerConfig
+from .teleop_so101_leader_arm import SO101LeaderArm, leader_joints_to_robot_action
+from .teleop_xr_controller import XRController
+from .xr_controller_processor import MapXRControllerActionToRobotAction
+
+__all__ = [
+    "Clutch",
+    "IsaacTeleopConfig",
+    "IsaacTeleopTeleoperator",
+    "MapXRControllerActionToRobotAction",
+    "SO101LeaderArm",
+    "SO101LeaderArmConfig",
+    "XRController",
+    "XRControllerConfig",
+    "leader_joints_to_robot_action",
+]

examples/isaac_teleop_to_so101/isaac_teleop/base.py

@@ -0,0 +1,282 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""Shared base for NVIDIA Isaac Teleop-backed LeRobot teleoperators.
+
+Isaac Teleop is a multi-modal framework: a single ``TeleopSession`` can be driven by
+XR controllers, hand tracking, Manus gloves, etc. Each modality is a
+:class:`Teleoperator` subclass in its own ``teleop_<device>.py``.
+
+:class:`IsaacTeleopTeleoperator` owns what those devices share — the session
+lifecycle, the per-step staleness/worker-health guard, and the no-op calibration
+tracking devices need. A concrete device implements :meth:`_build_pipeline` (its
+retargeting graph) and :meth:`get_action` (usually via :meth:`_step`).
+
+``isaacteleop`` is an optional NVIDIA dependency (install instructions in the example's
+``README.md``); its imports are guarded behind an availability check at module top, so this
+module imports without it and constructing a device fails fast with install instructions.
+"""
+
+from __future__ import annotations
+
+import abc
+import logging
+import os
+from collections.abc import Mapping
+from pathlib import Path
+from typing import TYPE_CHECKING, Any
+
+from lerobot.teleoperators.teleoperator import Teleoperator
+from lerobot.utils.import_utils import is_package_available
+
+from .config_isaac_teleop import IsaacTeleopConfig
+
+_isaacteleop_available = is_package_available("isaacteleop")
+
+if TYPE_CHECKING or _isaacteleop_available:
+    from isaacteleop.cloudxr import CloudXRLauncher
+    from isaacteleop.retargeting_engine.interface import (
+        ExecutionEvents,
+        ExecutionState,
+        GraphExecutable,
+        RetargeterIO,
+    )
+    from isaacteleop.teleop_session_manager import TeleopSession, TeleopSessionConfig
+else:
+    CloudXRLauncher = None
+    ExecutionEvents = None
+    ExecutionState = None
+    GraphExecutable = None
+    RetargeterIO = None
+    TeleopSession = None
+    TeleopSessionConfig = None
+
+logger = logging.getLogger(__name__)
+
+# Gripper closedness [0, 1] -> SO-101 follower motor units [0, 100] (RANGE_0_100, 100 = OPEN).
+# Shared by the XR processor and leader device, which invert via ``pos = (1 - c) * SCALE``.
+_GRIPPER_MOTOR_SCALE = 100.0
+
+
+def _require_isaacteleop() -> None:
+    """Fail fast with install pointers when the optional ``isaacteleop`` package is missing."""
+    if not _isaacteleop_available:
+        raise ImportError(
+            "The 'isaacteleop' package is required for Isaac Teleop devices but is not "
+            "installed. See examples/isaac_teleop_to_so101/README.md for install instructions."
+        )
+
+
+class IsaacTeleopTeleoperator(Teleoperator):
+    """Abstract base for teleoperators backed by an Isaac Teleop ``TeleopSession``.
+
+    Owns the session lifecycle and the per-step health guard; subclasses supply
+    :meth:`_build_pipeline` and :meth:`get_action`.
+    """
+
+    config_class = IsaacTeleopConfig
+
+    def __init__(self, config: IsaacTeleopConfig):
+        _require_isaacteleop()
+        super().__init__(config)
+        self.config: IsaacTeleopConfig = config
+        self._session: TeleopSession | None = None
+        self._cloudxr_launcher: CloudXRLauncher | None = None
+
+    # ------------------------------------------------------------------
+    # Pipeline construction (device override point)
+    # ------------------------------------------------------------------
+
+    @abc.abstractmethod
+    def _build_pipeline(self) -> GraphExecutable:
+        """Build this device's retargeting pipeline (the ``GraphExecutable`` for
+        ``TeleopSessionConfig.pipeline``). Called once in :meth:`connect`; its output
+        keys must match what :meth:`get_action` unpacks.
+        """
+        raise NotImplementedError
+
+    # ------------------------------------------------------------------
+    # Lifecycle (shared)
+    # ------------------------------------------------------------------
+
+    @property
+    def is_connected(self) -> bool:
+        return self._session is not None
+
+    @property
+    def is_calibrated(self) -> bool:
+        return True  # Tracking devices are self-calibrating.
+
+    def calibrate(self) -> None:
+        pass
+
+    def configure(self) -> None:
+        pass
+
+    def connect(self, calibrate: bool = True) -> None:
+        """Auto-launch the CloudXR runtime (unless opted out) and open the session.
+
+        The CloudXR launch blocks ~30s and, on the first run, prompts on stdin for the
+        EULA (accept once via ``python -m isaacteleop.cloudxr --accept-eula``). Opt out
+        when CloudXR runs externally via ``config.auto_launch_cloudxr=False`` or
+        ``LEROBOT_CLOUDXR_SKIP_AUTOLAUNCH=1`` (env var wins).
+        """
+        if self._session is not None:
+            raise RuntimeError("Already connected. Call disconnect() first.")
+
+        self._ensure_cloudxr_runtime()
+
+        try:
+            pipeline = self._build_pipeline()
+            session_config = TeleopSessionConfig(app_name=self.config.app_name, pipeline=pipeline)
+            self._session = TeleopSession(session_config)
+            self._session.__enter__()
+        except Exception:
+            self._session = None
+            try:
+                self._stop_cloudxr_runtime()
+            except Exception:
+                logger.exception("Failed to stop CloudXR runtime during connect() rollback")
+            raise
+        logger.info("Isaac Teleop session started: %s", self.config.app_name)
+
+    def disconnect(self) -> None:
+        try:
+            if self._session is not None:
+                # Null the handle BEFORE __exit__: even a failed session teardown must not
+                # wedge the device as is_connected (blocking every later connect/disconnect).
+                session = self._session
+                self._session = None
+                session.__exit__(None, None, None)
+                logger.info("Isaac Teleop session ended")
+        finally:
+            # Reap the CloudXR runtime even if session teardown raised, and even if no
+            # session was ever established (e.g. the launcher came up but session creation
+            # failed before this point); a no-op when we never launched CloudXR (opt-out /
+            # externally-owned runtime), so we never stop a runtime we don't own.
+            self._stop_cloudxr_runtime()
+
+    # ------------------------------------------------------------------
+    # CloudXR runtime (shared)
+    # ------------------------------------------------------------------
+
+    def _ensure_cloudxr_runtime(self) -> None:
+        """Auto-launch the CloudXR runtime once, unless opted out.
+
+        Idempotent (no-op once the launcher is up). ``LEROBOT_CLOUDXR_SKIP_AUTOLAUNCH``
+        is checked first and wins over ``config.auto_launch_cloudxr``. Constructing
+        :class:`CloudXRLauncher` mutates the process env (``XR_RUNTIME_JSON`` etc.) and
+        blocks until the runtime is ready or raises :class:`RuntimeError`.
+        """
+        if self._cloudxr_launcher is not None:
+            return
+
+        if os.environ.get("LEROBOT_CLOUDXR_SKIP_AUTOLAUNCH", "").strip() == "1":
+            logger.info(
+                "LEROBOT_CLOUDXR_SKIP_AUTOLAUNCH=1 set; skipping CloudXR auto-launch "
+                "(assuming CloudXR is already running externally)"
+            )
+            return
+
+        if not self.config.auto_launch_cloudxr:
+            logger.info(
+                "config.auto_launch_cloudxr is False; skipping CloudXR auto-launch "
+                "(assuming CloudXR is already running externally)"
+            )
+            return
+
+        logger.info("Launching CloudXR runtime (first run may prompt for EULA and take ~30s)...")
+
+        self._cloudxr_launcher = CloudXRLauncher(
+            install_dir=str(Path.home() / ".cloudxr"),
+            env_config=self.config.cloudxr_env_file,
+            accept_eula=False,
+        )
+
+    def _stop_cloudxr_runtime(self) -> None:
+        """Stop the auto-launched CloudXR runtime, if any.
+
+        Clean stop nulls the handle. On :class:`RuntimeError` the handle is RETAINED so
+        the launcher's ``atexit`` hook owns the retry — a later :meth:`connect` then
+        treats the retained runtime as still up and will not relaunch.
+        """
+        if self._cloudxr_launcher is None:
+            return
+        try:
+            self._cloudxr_launcher.stop()
+        except RuntimeError:
+            logger.warning("CloudXR runtime could not be terminated; handle retained for atexit cleanup")
+        else:
+            self._cloudxr_launcher = None
+            logger.info("CloudXR runtime stopped")
+
+    def send_feedback(self, feedback: dict[str, Any]) -> None:
+        pass  # Haptic feedback not yet implemented.
+
+    # ------------------------------------------------------------------
+    # Stepping (shared)
+    # ------------------------------------------------------------------
+
+    def _running_events(self) -> ExecutionEvents:
+        """Constant ``RUNNING`` ``ExecutionEvents`` for a device with no clutch lifecycle.
+
+        Keeps the stream flowing; ``reset`` stays ``False``. A clutched device that needs
+        a real lifecycle should build its own ``ExecutionEvents`` instead.
+        """
+        return ExecutionEvents(execution_state=ExecutionState.RUNNING, reset=False)
+
+    def _step(
+        self,
+        *,
+        execution_events: ExecutionEvents | None = None,
+        external_inputs: Mapping[str, Any] | None = None,
+    ) -> RetargeterIO:
+        """Step the session once and return the raw pipeline outputs.
+
+        Applies the shared guard: re-raises a retargeting-worker exception and warns on a
+        stale frame. Subclasses call this from :meth:`get_action`.
+
+        Args:
+            execution_events: The ``ExecutionEvents`` driving the session this frame.
+                Devices with a lifecycle (clutch) MUST pass this every frame — when
+                ``None``, ``TeleopSession.step`` auto-fires ``RUNNING`` (the clutch would
+                latch immediately and never stop).
+            external_inputs: Per-step inputs (e.g. a static ``base_T_anchor``) in the
+                ``{leaf_node_name: {output_port_name: TensorGroup}}`` shape ``step`` expects.
+
+        Raises:
+            RuntimeError: If not connected, or if the retargeting worker raised.
+        """
+        if self._session is None:
+            raise RuntimeError("Not connected. Call connect() first.")
+
+        result = self._session.step(
+            execution_events=execution_events,
+            external_inputs=external_inputs,
+        )
+
+        info = self._session.last_step_info
+        if info is not None:
+            if info.worker_exception is not None:
+                raise RuntimeError(
+                    "Isaac Teleop retargeting worker raised an exception"
+                ) from info.worker_exception
+            if info.frame_deadline_miss:
+                logger.warning(
+                    "Isaac Teleop frame deadline miss (returned_age_frames=%s)",
+                    info.returned_age_frames,
+                )
+        return result

examples/isaac_teleop_to_so101/isaac_teleop/clutch.py

@@ -0,0 +1,102 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""Engage-relative clutch for the XR -> SO-101 teleop loop.
+
+Turns the raw controller grip pose into an absolute base-frame EE target, so the XR
+device can stay a thin raw-pose reader. Pure numpy + the local ``Rotation`` helper (no
+``isaacteleop``), so it is unit-testable without the XR runtime.
+"""
+
+from __future__ import annotations
+
+import numpy as np
+
+from lerobot.utils.rotation import Rotation
+
+
+class Clutch:
+    """Engage-relative clutch for both position AND orientation.
+
+    Latch an origin on engage, then track the base-frame delta from it, applied
+    independently to position and orientation. State:
+
+    - ``_last_commanded_pos`` / ``_last_commanded_rot``: last commanded EE pose; held
+      while disengaged so the arm freezes where it was left.
+    - ``_home_pos`` / ``_home_rot``: latched on engage — the EE pose the delta applies to.
+      The position comes from the arm's MEASURED pose when the caller provides it (so an
+      arm that moved while disengaged is not snapped back to a stale command); the
+      orientation always comes from the last commanded rotation (see NOTE below).
+    - ``_origin_pos`` / ``_origin_rot``: latched on engage — the controller pose the delta
+      is measured against.
+
+    Each engaged frame :meth:`rebase` returns::
+
+        pos = home_pos + (grip_pos - origin_pos)  # 1:1 controller -> EE translation
+        rot = (R_ctrl @ R_origin ^ -1) @ R_home  # base-frame delta, left-composed
+
+    On the engage edge the output is exactly the home pose (no teleport). The orientation
+    delta is left-composed (base frame), so hand rotation about base Z maps to EE rotation
+    about base Z. A re-clutch latches a fresh home/origin.
+
+    NOTE: ``_home_rot`` is the last *commanded* orientation even when the measured pose is
+    supplied: the 5-DOF SO-101 tracks orientation only softly, so its measured wrist
+    orientation persistently differs from the command, and latching the measurement would
+    inject that offset into the commanded signal on every re-clutch. Position has no such
+    tracking gap, and there latching the measurement is what prevents the snap-back.
+    """
+
+    def __init__(self, home_base_T_ee: np.ndarray):  # noqa: N803
+        # Seed the held pose from the arm's measured startup EE pose so the first
+        # engage latches home there (no jump on the first squeeze).
+        home = np.asarray(home_base_T_ee, dtype=float)
+        self._last_commanded_pos = home[:3, 3].copy()
+        self._last_commanded_rot = Rotation.from_matrix(home[:3, :3])
+        self._home_pos = self._last_commanded_pos.copy()
+        self._home_rot = self._last_commanded_rot
+        self._origin_pos = np.zeros(3, dtype=float)
+        self._origin_rot = Rotation.from_quat(np.array([0.0, 0.0, 0.0, 1.0]))
+
+    def engage(
+        self,
+        grip_pos: np.ndarray,
+        grip_quat: np.ndarray,
+        measured_base_T_ee: np.ndarray | None = None,  # noqa: N803
+    ) -> None:
+        """Latch the engage home (where the arm is now) and controller origin.
+
+        Pass ``measured_base_T_ee`` (FK of the measured joints) so the home POSITION is
+        where the arm physically is — if the arm moved while disengaged (gravity sag,
+        external contact), latching the stale last-commanded position would make the
+        first engaged frame command a full-speed jump back to it. The home ORIENTATION
+        always stays the last commanded one (see the class NOTE).
+        """
+        if measured_base_T_ee is not None:
+            self._home_pos = np.asarray(measured_base_T_ee, dtype=float)[:3, 3].copy()
+        else:
+            self._home_pos = self._last_commanded_pos.copy()
+        self._home_rot = self._last_commanded_rot
+        self._origin_pos = np.asarray(grip_pos, dtype=float).copy()
+        self._origin_rot = Rotation.from_quat(np.asarray(grip_quat, dtype=float))
+
+    def rebase(self, grip_pos: np.ndarray, grip_quat: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
+        """Return the absolute base-frame EE target ``(pos [m], quat [xyzw])`` for this frame."""
+        pos = self._home_pos + (np.asarray(grip_pos, dtype=float) - self._origin_pos)
+        rot_ctrl = Rotation.from_quat(np.asarray(grip_quat, dtype=float))
+        rot = (rot_ctrl * self._origin_rot.inv()) * self._home_rot
+        self._last_commanded_pos = pos.copy()
+        self._last_commanded_rot = rot
+        return pos, rot.as_quat()

examples/isaac_teleop_to_so101/isaac_teleop/config_isaac_teleop.py

@@ -0,0 +1,135 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""Configuration dataclasses for NVIDIA Isaac Teleop-backed teleoperators.
+
+:class:`IsaacTeleopConfig` holds the shared fields; each device adds its own subclass
+(e.g. :class:`XRControllerConfig`, :class:`SO101LeaderArmConfig`).
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import ClassVar
+
+from lerobot.teleoperators.config import TeleoperatorConfig
+
+
+@dataclass(kw_only=True)
+class IsaacTeleopConfig(TeleoperatorConfig):
+    """Shared config for all Isaac Teleop-backed teleoperators.
+
+    Uses its own draccus ``_choice_registry`` (decoupled from the global
+    :class:`TeleoperatorConfig` one) so ``--teleop.type`` on a field typed
+    ``IsaacTeleopConfig`` resolves against ONLY the Isaac devices — letting them claim
+    short names (``xr_controller``, ``so101_leader``) without colliding with the global
+    registry. These devices are selected by the example scripts, not routed through
+    ``make_teleoperator_from_config``.
+    """
+
+    _choice_registry: ClassVar[dict] = {}
+
+    app_name: str = "LeTeleop"
+    """Application name for the OpenXR / Isaac Teleop session."""
+
+    auto_launch_cloudxr: bool = True
+    """Auto-launch the CloudXR runtime on :meth:`connect`. Set ``False`` (or export
+    ``LEROBOT_CLOUDXR_SKIP_AUTOLAUNCH=1``, which wins) when CloudXR runs externally.
+    """
+
+    cloudxr_env_file: str | None = None
+    """Optional CloudXR device-profile ``.env`` (an INPUT profile selecting the headset
+    transport) passed to ``CloudXRLauncher``. ``None`` keeps the default auto-WebRTC profile.
+    """
+
+
+# Static rebase from the OpenXR controller anchor frame (X=Right, Y=Up, Z=Backward) into the
+# robot base frame (X=Forward, Y=Left, Z=Up). A proper rotation (det=+1): controller motion
+# forward -> robot +X, right -> robot -Y (i.e. rightward), up -> robot +Z.
+_DEFAULT_BASE_T_ANCHOR: list[list[float]] = [
+    [0.0, 0.0, -1.0, 0.0],
+    [-1.0, 0.0, 0.0, 0.0],
+    [0.0, 1.0, 0.0, 0.0],
+    [0.0, 0.0, 0.0, 1.0],
+]
+
+
+@IsaacTeleopConfig.register_subclass("xr_controller")
+@dataclass(kw_only=True)
+class XRControllerConfig(IsaacTeleopConfig):
+    """Config for Isaac Teleop XR (VR) controller teleoperation.
+
+    Exposes the raw base-frame grip pose, squeeze, and trigger via ``ControllersSource``.
+    No retargeters: the clutch and gripper mapping live in the owning loop.
+    """
+
+    hand_side: str = "right"
+    """Which controller hand to use: ``"left"`` or ``"right"``. A plain ``str`` (validated in
+    ``__post_init__``) because draccus cannot decode ``Literal``-typed fields from the CLI."""
+
+    clutch_threshold: float = 0.5
+    """Squeeze value above which the owning loop's clutch engages (held-to-enable). The
+    device reports only the raw squeeze; the threshold is applied by the loop."""
+
+    base_T_anchor: list[list[float]] = field(  # noqa: N815  (frameA_T_frameB transform-matrix convention)
+        # Fresh copy per instance: returning the module-level list itself would alias one
+        # mutable matrix across every config.
+        default_factory=lambda: [row.copy() for row in _DEFAULT_BASE_T_ANCHOR]
+    )
+    """Static 4x4 [row-major] transform rebasing the OpenXR controller anchor frame into
+    the robot base frame. Defaults to OpenXR (X=Right, Y=Up, Z=Backward) -> robot
+    (X=Forward, Y=Left, Z=Up). Plain nested lists so the config stays serializable.
+    """
+
+    def __post_init__(self):
+        if self.hand_side not in ("left", "right"):
+            raise ValueError(f"hand_side must be 'left' or 'right', got {self.hand_side!r}")
+
+
+# Provisional gripper open/close endpoints [rad], normalizing the streamed gripper angle
+# into the follower's RANGE_0_100 jaw target. Derived from the so101_leader plugin README's
+# example calibration (home_ticks=2048, range 2000..3000; angle = (ticks-home)*2*pi/4096).
+_DEFAULT_GRIPPER_OPEN_RAD = -0.074
+_DEFAULT_GRIPPER_CLOSE_RAD = 1.460
+
+
+@IsaacTeleopConfig.register_subclass("so101_leader")
+@dataclass(kw_only=True)
+class SO101LeaderArmConfig(IsaacTeleopConfig):
+    """Config for an Isaac Teleop SO-101 *leader arm* (generic joint-space device).
+
+    Mirrors the leader's joint angles 1:1 onto a follower SO-101. The leader state is
+    streamed in radians by the native ``so101_leader`` plugin and read via a
+    ``JointStateSource``; the device converts arm joints to degrees and the gripper to the
+    follower's RANGE_0_100 jaw target (no IK/clutch/retargeter on the LeRobot side).
+    """
+
+    port: str = ""
+    """Serial port of the physical LEADER arm (e.g. ``/dev/ttyACM1``), forwarded to the
+    plugin (which reads the servos) when the example launches it. Empty -> the plugin runs
+    its synthetic trajectory."""
+
+    collection_id: str = "so101_leader"
+    """Tensor collection id the leader plugin pushes on; must match the running
+    ``so101_leader`` plugin (its second positional arg, default ``"so101_leader"``)."""
+
+    gripper_open_rad: float = _DEFAULT_GRIPPER_OPEN_RAD
+    """Leader gripper angle [rad] at fully OPEN -> follower jaw 100. Provisional default;
+    set from the plugin's ``calibrate`` subcommand. See ``_DEFAULT_GRIPPER_OPEN_RAD``."""
+
+    gripper_close_rad: float = _DEFAULT_GRIPPER_CLOSE_RAD
+    """Leader gripper angle [rad] at fully CLOSED -> follower jaw 0. Provisional default;
+    set from the plugin's ``calibrate`` subcommand. See ``_DEFAULT_GRIPPER_CLOSE_RAD``."""

examples/isaac_teleop_to_so101/isaac_teleop/teleop_so101_leader_arm.py

@@ -0,0 +1,186 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""SO-101 leader-arm device for NVIDIA Isaac Teleop, exposed to LeRobot.
+
+The leader is a back-drivable SO-101 whose six joint angles are streamed (in radians) by
+the native ``so101_leader`` plugin; this device reads them via a ``JointStateSource`` and
+converts them into follower-ready ``{joint}.pos``. Same kinematics as the follower, so it
+needs no retargeting — a 1:1 joint mirror, direct joint drive.
+
+Units (converted in the device so the output is always follower-valid):
+
+* arm joints: ``rad2deg`` — correct only if the leader's calibrated zero and the follower's
+  homing map to the same physical zero (the standard same-hardware assumption).
+* gripper: normalized from ``[gripper_open_rad, gripper_close_rad]`` to RANGE_0_100.
+
+``isaacteleop`` imports are guarded behind the availability flag so this module — and the
+pure :func:`leader_joints_to_robot_action` converter — import without it (construction
+fails fast via the base class).
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import numpy as np
+
+from lerobot.types import RobotAction
+
+from .base import _GRIPPER_MOTOR_SCALE, IsaacTeleopTeleoperator, _isaacteleop_available
+from .config_isaac_teleop import SO101LeaderArmConfig
+
+if TYPE_CHECKING or _isaacteleop_available:
+    from isaacteleop.retargeting_engine.deviceio_source_nodes import JointStateSource
+    from isaacteleop.retargeting_engine.interface import OutputCombiner
+else:
+    JointStateSource = None
+    OutputCombiner = None
+
+# Canonical SO-101 DOF names and order — matches the plugin stream and the follower's motor
+# order. Passed to the ``JointStateSource`` as its output layout; the source maps by name and
+# :func:`_joints_group_to_rad` reads back by name, so a layout mismatch can't mislabel a DOF.
+SO101_LEADER_JOINTS = [
+    "shoulder_pan",
+    "shoulder_lift",
+    "elbow_flex",
+    "wrist_flex",
+    "wrist_roll",
+    "gripper",
+]
+
+
+def leader_joints_to_robot_action(
+    joints_rad: dict[str, float],
+    *,
+    gripper_joint: str,
+    gripper_open_rad: float,
+    gripper_close_rad: float,
+) -> RobotAction:
+    """Convert streamed leader joint angles [rad] to follower-ready ``{joint}.pos``.
+
+    Pure (no ``isaacteleop``, no I/O). Iteration follows ``joints_rad`` insertion order, so
+    pass it in :data:`SO101_LEADER_JOINTS` order for a stable layout. Arm joints are
+    converted ``rad2deg``; ``gripper_joint`` is normalized from
+    ``[gripper_open_rad, gripper_close_rad]`` to RANGE_0_100 (clipped).
+    """
+    action: RobotAction = {}
+    span = gripper_close_rad - gripper_open_rad
+    for name, rad in joints_rad.items():
+        if name == gripper_joint:
+            # Closedness c=0 at open, c=1 at closed; invert to the follower's 100=open jaw.
+            closedness = 0.0 if span == 0.0 else (rad - gripper_open_rad) / span
+            closedness = min(1.0, max(0.0, closedness))
+            action[f"{name}.pos"] = (1.0 - closedness) * _GRIPPER_MOTOR_SCALE
+        else:
+            action[f"{name}.pos"] = float(np.rad2deg(rad))
+    return action
+
+
+def _joints_group_to_rad(joints) -> dict[str, float]:
+    """Read a ``JointStateSource`` output group into ``{joint_name: angle [rad]}``.
+
+    Pure (duck-typed on the group). The group is positional but each slot carries its joint
+    name in ``group.group_type.types``; we key off those names (not a positional index) so a
+    layout mismatch surfaces as a wrong/missing key here rather than a mislabeled DOF.
+    """
+    names = [t.name for t in joints.group_type.types]
+    return {name: float(joints[i]) for i, name in enumerate(names)}
+
+
+class SO101LeaderArm(IsaacTeleopTeleoperator):
+    """SO-101 leader-arm teleoperator (joint-space), direct joint mirror to the follower.
+
+    Reads the six joint angles off a single ``JointStateSource`` each frame; no retargeter,
+    no clutch. When the leader is not streaming, :meth:`get_action` returns the held-last
+    joints and :attr:`is_tracking` is ``False`` so the owning loop can hold the follower.
+    """
+
+    config_class = SO101LeaderArmConfig
+    name = "isaac_teleop_so101_leader"
+
+    def __init__(self, config: SO101LeaderArmConfig):
+        super().__init__(config)
+        self.config: SO101LeaderArmConfig = config
+        # Held-last joint angles [rad], seeded at zero (URDF/home pose) so the first frames
+        # before the plugin starts pushing read as the home pose, not garbage.
+        self._last_joints_rad: dict[str, float] = dict.fromkeys(SO101_LEADER_JOINTS, 0.0)
+        # Whether the most recent get_action() read live leader data (vs held-last).
+        self._is_tracking = False
+
+    # ------------------------------------------------------------------
+    # Pipeline construction
+    # ------------------------------------------------------------------
+
+    def _build_pipeline(self) -> OutputCombiner:
+        """Build the joint-mirror pipeline: a single ``JointStateSource`` leaf that converts
+        the raw stream into a name-keyed joint group. No retargeter (shared kinematics)."""
+        source = JointStateSource(
+            name="so101_leader",
+            collection_id=self.config.collection_id,
+            joint_names=SO101_LEADER_JOINTS,
+        )
+        return OutputCombiner({"joints": source.output(JointStateSource.JOINTS)})
+
+    # ------------------------------------------------------------------
+    # Action features
+    # ------------------------------------------------------------------
+
+    @property
+    def action_features(self) -> dict[str, type]:
+        # Matches the serial SOLeader's action features so this is a drop-in joint-space
+        # leader: one float `{joint}.pos` per DOF, sendable straight to an SO-101 follower.
+        return {f"{name}.pos": float for name in SO101_LEADER_JOINTS}
+
+    @property
+    def feedback_features(self) -> dict[str, type]:
+        return {}
+
+    @property
+    def is_tracking(self) -> bool:
+        """Whether the last :meth:`get_action` read live leader data (vs held-last)."""
+        return self._is_tracking
+
+    # ------------------------------------------------------------------
+    # Action extraction
+    # ------------------------------------------------------------------
+
+    def get_action(self) -> RobotAction:
+        """Step the session and return the leader joints as follower-ready ``{joint}.pos``.
+
+        When the leader is streaming, the live angles are cached and converted; otherwise the
+        held-last angles are reused and :attr:`is_tracking` is set ``False``.
+        """
+        result = self._step(execution_events=self._running_events())
+
+        joints = result["joints"]
+        # The JointStateSource output is Optional: absent (is_none) when the device is
+        # inactive. Treat that as "not tracking" and reuse the held-last angles.
+        self._is_tracking = not getattr(joints, "is_none", False)
+        if self._is_tracking:
+            try:
+                self._last_joints_rad = _joints_group_to_rad(joints)
+            except (AttributeError, IndexError, KeyError, TypeError, ValueError):
+                # A partially-populated / malformed group on an odd frame: keep held-last, but
+                # report it as not-tracking so the loop holds the follower rather than trusting it.
+                self._is_tracking = False
+
+        return leader_joints_to_robot_action(
+            self._last_joints_rad,
+            gripper_joint="gripper",
+            gripper_open_rad=self.config.gripper_open_rad,
+            gripper_close_rad=self.config.gripper_close_rad,
+        )

examples/isaac_teleop_to_so101/isaac_teleop/teleop_xr_controller.py

@@ -0,0 +1,204 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""XR (VR) controller device for NVIDIA Isaac Teleop, exposed to LeRobot.
+
+A deliberately thin reader: exposes the raw controller grip pose off
+``ControllersSource`` (statically rebased into the robot base frame by
+``ControllerTransform``), plus squeeze and trigger. No retargeters and no clutch —
+the clutch rebasing and gripper mapping live downstream in the owning loop, so this
+device is stateless across frames.
+
+``isaacteleop`` imports are guarded behind the availability flag so this module imports
+without it (construction fails fast via the base class).
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+import numpy as np
+
+from lerobot.types import RobotAction
+
+from .base import IsaacTeleopTeleoperator, _isaacteleop_available
+from .config_isaac_teleop import XRControllerConfig
+
+if TYPE_CHECKING or _isaacteleop_available:
+    from isaacteleop.retargeting_engine.deviceio_source_nodes import ControllersSource
+    from isaacteleop.retargeting_engine.interface import OutputCombiner, TensorGroup, ValueInput
+    from isaacteleop.retargeting_engine.tensor_types import TransformMatrix
+    from isaacteleop.retargeting_engine.tensor_types.indices import ControllerInputIndex
+else:
+    ControllersSource = None
+    OutputCombiner = None
+    TensorGroup = None
+    ValueInput = None
+    TransformMatrix = None
+    ControllerInputIndex = None
+
+# Source-node name for the static base_T_anchor rebase input fed via
+# ``TeleopSession.step(external_inputs=...)`` each frame.
+_BASE_T_ANCHOR_INPUT = "base_T_anchor"
+
+
+class XRController(IsaacTeleopTeleoperator):
+    """Raw XR controller grip-pose teleoperator (base-frame), no retargeters.
+
+    Reads the raw grip pose + squeeze + trigger off a ``ControllersSource`` rebased into
+    the robot base frame. :meth:`get_action` returns the absolute base-frame grip pose
+    untouched; the owning loop owns the clutch and gripper mapping.
+    """
+
+    config_class = XRControllerConfig
+    name = "isaac_teleop_controller"
+
+    def __init__(self, config: XRControllerConfig):
+        super().__init__(config)
+        self.config: XRControllerConfig = config
+
+        # Constant base_T_anchor input, built once in connect() (a TensorGroup is heavy and
+        # isaacteleop-backed) and reused every step.
+        self._external_inputs: dict[str, Any] | None = None
+        # Whether the last get_action() read a tracked controller; the owning loop polls this
+        # to wait for the operator to connect before driving the arm.
+        self._is_tracking = False
+
+    # ------------------------------------------------------------------
+    # Pipeline construction
+    # ------------------------------------------------------------------
+
+    def _build_pipeline(self) -> OutputCombiner:
+        """Build the raw-grip-pose pipeline: a ``ControllersSource`` rebased into the base
+        frame by ``ControllerTransform``, exposed verbatim as ``"controller"``. No retargeters.
+        """
+        side = self.config.hand_side
+        controller_key = f"controller_{side}"
+
+        controllers = ControllersSource(name="controllers")
+        # Static base_T_anchor rebase fed via external_inputs each step.
+        xform = ValueInput(_BASE_T_ANCHOR_INPUT, TransformMatrix())
+        transformed = controllers.transformed(xform.output("value"))
+        ctrl = transformed.output(controller_key)
+
+        return OutputCombiner({"controller": ctrl})
+
+    def _build_external_inputs(self) -> dict[str, Any]:
+        """Materialize the constant ``base_T_anchor`` external input (once, in connect)."""
+        tg = TensorGroup(TransformMatrix())
+        tg[0] = np.asarray(self.config.base_T_anchor, dtype=np.float32)
+        return {_BASE_T_ANCHOR_INPUT: {"value": tg}}
+
+    def connect(self, calibrate: bool = True) -> None:
+        super().connect(calibrate=calibrate)
+        try:
+            self._external_inputs = self._build_external_inputs()
+        except Exception:
+            # Roll the session/runtime back so a failed connect() leaves no half-state
+            # (a live session behind a raised connect would leak the CloudXR runtime).
+            self.disconnect()
+            raise
+
+    # ------------------------------------------------------------------
+    # Action features
+    # ------------------------------------------------------------------
+
+    @property
+    def action_features(self) -> dict:
+        return {
+            "grip_pos": {
+                "dtype": "float32",
+                "shape": (3,),
+                "names": {"x": 0, "y": 1, "z": 2},
+            },
+            "grip_quat": {
+                "dtype": "float32",
+                "shape": (4,),
+                "names": {"qx": 0, "qy": 1, "qz": 2, "qw": 3},
+            },
+            # ``get_action`` returns scalars for these two, so the advertised
+            # shape is () (0-d) to stay consistent with the returned values.
+            "squeeze": {
+                "dtype": "float32",
+                "shape": (),
+                "names": None,
+            },
+            "trigger": {
+                "dtype": "float32",
+                "shape": (),
+                "names": None,
+            },
+        }
+
+    @property
+    def feedback_features(self) -> dict:
+        return {}
+
+    @property
+    def is_tracking(self) -> bool:
+        """Whether the last :meth:`get_action` read a tracked controller. ``False`` until the
+        headset is connected over CloudXR and its controllers are live; the owning loop polls
+        it to wait for the operator before commanding the arm."""
+        return self._is_tracking
+
+    # ------------------------------------------------------------------
+    # Action extraction
+    # ------------------------------------------------------------------
+
+    def get_action(self) -> RobotAction:
+        """Step the session and return the raw base-frame grip pose.
+
+        Reads the grip pose + squeeze + trigger off the transformed controller stream (with
+        the constant ``base_T_anchor`` rebase). When the controller is not tracked, returns
+        identity pose and squeeze/trigger = 0.0 so the owning loop freezes the arm.
+
+        Returns:
+            ``{"grip_pos": (3,) [m], "grip_quat": (4,) [qx,qy,qz,qw], "squeeze": float,
+            "trigger": float}`` — pose in the robot base frame; squeeze/trigger in ``[0, 1]``.
+        """
+        result = self._step(execution_events=self._running_events(), external_inputs=self._external_inputs)
+
+        # Optional controller group is None until the headset is connected and its controllers
+        # are live; expose that as is_tracking so the loop can wait before driving the arm.
+        controller = result["controller"]
+        grip_pos = np.zeros(3, dtype=np.float32)
+        grip_quat = np.array([0.0, 0.0, 0.0, 1.0], dtype=np.float32)
+        squeeze = 0.0
+        trigger = 0.0
+        self._is_tracking = not getattr(controller, "is_none", False)
+        if self._is_tracking:
+            # Read ALL four fields into locals before committing any of them: a failure on a
+            # partially-populated frame must not mix live values with the safe defaults (a
+            # live squeeze paired with a defaulted trigger=0.0 would keep the clutch engaged
+            # while commanding the gripper fully open, dropping whatever is grasped). On
+            # failure the defaults stand untouched and the frame reports not-tracked.
+            try:
+                pos = np.asarray(controller[ControllerInputIndex.GRIP_POSITION], dtype=np.float32)
+                quat = np.asarray(controller[ControllerInputIndex.GRIP_ORIENTATION], dtype=np.float32)
+                squeeze_val = float(controller[ControllerInputIndex.SQUEEZE_VALUE])
+                trigger_val = float(controller[ControllerInputIndex.TRIGGER_VALUE])
+            except (IndexError, KeyError, TypeError, ValueError):
+                self._is_tracking = False
+            else:
+                grip_pos, grip_quat = pos, quat
+                squeeze, trigger = squeeze_val, trigger_val
+
+        return {
+            "grip_pos": grip_pos,
+            "grip_quat": grip_quat,
+            "squeeze": squeeze,
+            "trigger": trigger,
+        }

examples/isaac_teleop_to_so101/isaac_teleop/xr_controller_processor.py

@@ -0,0 +1,87 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""Processor step that maps XR controller actions to robot EE targets.
+
+Analogous to ``MapPhoneActionToRobotAction``, this bridges the clutch-rebased EE pose to
+the IK pipeline's input contract (``EEBoundsAndSafety`` -> ``InverseKinematicsEEToJoints``).
+Pure (no ``isaacteleop``), so it is unit-testable without the XR runtime.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from lerobot.configs.types import FeatureType, PipelineFeatureType, PolicyFeature
+from lerobot.processor import ProcessorStepRegistry, RobotActionProcessorStep
+from lerobot.types import RobotAction
+from lerobot.utils.rotation import Rotation
+
+from .base import _GRIPPER_MOTOR_SCALE
+
+
+@ProcessorStepRegistry.register("map_xr_controller_action_to_robot_action")
+@dataclass
+class MapXRControllerActionToRobotAction(RobotActionProcessorStep):
+    """Maps an absolute base-frame EE pose + gripper closedness to the IK input contract.
+
+    Pure, stateless rename (the owning loop's clutch already produced the absolute base-frame
+    target). Each frame it writes:
+
+    - ``ee.x/y/z`` = ``ee_pose[:3]`` (position [m]);
+    - ``ee.wx/wy/wz`` = rotvec of ``ee_pose[3:7]`` (orientation; the IK tracks it softly at a
+      small ``orientation_weight`` on the 5-DOF SO-101);
+    - ``ee.gripper_pos`` = ``(1 - closedness) * _GRIPPER_MOTOR_SCALE`` (jaw target [0, 100],
+      RANGE_0_100 where 100 = open, so closedness is inverted).
+
+    Input keys: ``ee_pose`` ``(7,)`` ``[x,y,z,qx,qy,qz,qw]``, ``closedness`` float in [0, 1].
+    """
+
+    def action(self, action: RobotAction) -> RobotAction:
+        ee_pose = action.pop("ee_pose")
+        closedness = float(action.pop("closedness"))
+
+        action["ee.x"] = float(ee_pose[0])
+        action["ee.y"] = float(ee_pose[1])
+        action["ee.z"] = float(ee_pose[2])
+        # Orientation target as a rotvec (quat [qx,qy,qz,qw] -> axis-angle); the IK
+        # consumes ee.w* as a rotvec and tracks it with orientation_weight.
+        rotvec = Rotation.from_quat(ee_pose[3:7]).as_rotvec()
+        action["ee.wx"] = float(rotvec[0])
+        action["ee.wy"] = float(rotvec[1])
+        action["ee.wz"] = float(rotvec[2])
+        # Inverted: closedness c=1 (closed) -> 0, c=0 (open) -> 100 (SO-101 calibration).
+        action["ee.gripper_pos"] = (1.0 - closedness) * _GRIPPER_MOTOR_SCALE
+        return action
+
+    def transform_features(
+        self, features: dict[PipelineFeatureType, dict[str, PolicyFeature]]
+    ) -> dict[PipelineFeatureType, dict[str, PolicyFeature]]:
+        for feat in ["ee_pose", "closedness"]:
+            features[PipelineFeatureType.ACTION].pop(feat, None)
+
+        for feat in [
+            "ee.x",
+            "ee.y",
+            "ee.z",
+            "ee.wx",
+            "ee.wy",
+            "ee.wz",
+            "ee.gripper_pos",
+        ]:
+            features[PipelineFeatureType.ACTION][feat] = PolicyFeature(type=FeatureType.ACTION, shape=(1,))
+
+        return features

examples/isaac_teleop_to_so101/override_reset_pose.py

@@ -0,0 +1,73 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""Save the current SO-101 joint positions as the reset-origin pose (override).
+
+Move the arm to the desired reset pose by hand (torque off), then run this script to write
+those joints to a per-arm file in the LeRobot cache. ``teleoperate.py`` / ``record.py`` load
+it on startup (matched by ``--robot.id``) as the reset target instead of the defaults.
+
+Usage::
+
+    # 1. Move arm to desired reset pose by hand
+    python -m examples.isaac_teleop_to_so101.override_reset_pose [--port /dev/ttyACM0] [--id so101_follower_arm]
+
+    # 2. Launch teleop with the SAME --robot.id — it will now reset to this pose on startup
+    python -m examples.isaac_teleop_to_so101.teleoperate --robot.type=so101_follower --robot.port=/dev/ttyACM0 --robot.id=so101_follower_arm --teleop.type=xr_controller
+"""
+
+import argparse
+import json
+from pathlib import Path
+
+from lerobot.robots.so_follower import SO100Follower, SO100FollowerConfig
+
+from .common import RESET_POSE_FILE
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter
+    )
+    parser.add_argument("--port", type=str, default="/dev/ttyACM0")
+    parser.add_argument("--id", type=str, default="so101_follower_arm")
+    return parser.parse_args()
+
+
+def main():
+    args = parse_args()
+    robot = SO100Follower(SO100FollowerConfig(port=args.port, id=args.id, use_degrees=True))
+    robot.connect()
+    # Always disconnect the follower so a failure never leaks the serial connection.
+    try:
+        obs = robot.get_observation()
+        motor_names = list(robot.bus.motors.keys())
+        pose = {name: float(obs[f"{name}.pos"]) for name in motor_names}
+    finally:
+        robot.disconnect()
+
+    print("Current joint positions:")
+    for name, val in pose.items():
+        print(f"  {name:20s}: {val:.2f}")
+
+    reset_pose_file = Path(RESET_POSE_FILE.format(robot_name=robot.name, robot_id=robot.id))
+    reset_pose_file.parent.mkdir(parents=True, exist_ok=True)
+    reset_pose_file.write_text(json.dumps(pose, indent=2))
+    print(f"\nSaved to {reset_pose_file}")
+
+
+if __name__ == "__main__":
+    main()

examples/isaac_teleop_to_so101/record.py

@@ -0,0 +1,321 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""Record a LeRobot dataset via NVIDIA Isaac Teleop -> SO-101.
+
+Runs ``teleoperate.py``'s control loop while also saving each frame to a LeRobot dataset.
+``--teleop.type`` selects the device (``xr_controller`` | ``so101_leader``) as in
+``teleoperate.py``.
+
+Usage::
+
+    # XR (VR) controller: clutch + soft-orientation IK
+    python -m examples.isaac_teleop_to_so101.record \\
+        --robot.type=so101_follower \\
+        --robot.port=/dev/ttyACM0 \\
+        --robot.id=so101_follower_arm \\
+        --teleop.type=xr_controller \\
+        --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30}}" \\
+        --dataset.repo_id=<hf_user>/<dataset_name> \\
+        --dataset.single_task="Pick up vial from rack on the left side" \\
+        --dataset.num_episodes=3 \\
+        --dataset.episode_time_s=20 \\
+        --dataset.reset_time_s=5
+
+    # SO-101 leader arm: 1:1 joint mirror (real leader on /dev/ttyACM1)
+    python -m examples.isaac_teleop_to_so101.record \\
+        --robot.type=so101_follower --robot.port=/dev/ttyACM0 --robot.id=so101_follower_arm \\
+        --teleop.type=so101_leader --teleop.port=/dev/ttyACM1 --teleop.id=so101_leader_arm \\
+        --launch_plugin=/path/to/IsaacTeleop/install/plugins/so101_leader/so101_leader_plugin \\
+        --dataset.repo_id=<hf_user>/<dataset_name> --dataset.single_task="Pick up the cube" \\
+        --dataset.num_episodes=3 --dataset.episode_time_s=20 --dataset.reset_time_s=5
+
+The loop/launch knobs mirror ``teleoperate.py`` (tagged ``[xr]`` / ``[leader]`` below).
+
+Keyboard shortcuts: Right/n = end episode early and save, Left/r = discard + re-record,
+Esc/q = stop after the current episode. All frames are recorded (including hold frames).
+"""
+
+import logging
+import time
+from dataclasses import asdict, dataclass
+from pprint import pformat
+
+from lerobot.cameras import CameraConfig  # noqa: F401
+from lerobot.cameras.opencv import OpenCVCameraConfig  # noqa: F401
+from lerobot.common.control_utils import sanity_check_dataset_robot_compatibility
+from lerobot.configs import parser
+from lerobot.configs.dataset import DatasetRecordConfig
+from lerobot.datasets import (
+    LeRobotDataset,
+    VideoEncodingManager,
+    aggregate_pipeline_dataset_features,
+    create_initial_features,
+    safe_stop_image_writer,
+)
+from lerobot.processor import make_default_processors
+from lerobot.robots import RobotConfig
+from lerobot.robots.so_follower import SOFollowerConfig  # noqa: F401  (registers so101_follower)
+from lerobot.utils.constants import ACTION, OBS_STR
+from lerobot.utils.feature_utils import build_dataset_frame, combine_feature_dicts
+from lerobot.utils.robot_utils import precise_sleep
+from lerobot.utils.utils import init_logging
+
+from .common import (
+    ALIGN_DURATION_S,
+    RESET_DURATION_S,
+    Device,
+    HoldLatch,
+    build_device,
+    init_keyboard_listener,
+)
+from .isaac_teleop import IsaacTeleopConfig
+
+
+@dataclass
+class RecordConfig:
+    """CLI config for Isaac Teleop -> SO-101 dataset recording.
+
+    ``--robot.*`` / ``--teleop.*`` / ``--dataset.*`` configure the follower, device, and
+    recording; the loop/launch knobs below carry the same ``[xr]`` / ``[leader]`` tags as
+    ``teleoperate.py``. Use ``--flag=false`` for booleans (draccus style).
+    """
+
+    robot: RobotConfig
+    # --teleop.type=xr_controller|so101_leader, resolved against IsaacTeleopConfig's registry.
+    teleop: IsaacTeleopConfig
+    dataset: DatasetRecordConfig
+
+    # [leader] Path to the so101_leader plugin binary to spawn after CloudXR is up (it then
+    # inherits the runtime env). None (default) -> assume the plugin already runs externally.
+    launch_plugin: str | None = None
+
+    # [xr] Slew all joints to the reset pose before the first episode (--reset_to_origin=false to
+    # keep the arm where it is). After the slew the clutch seeds its home from the measured pose.
+    reset_to_origin: bool = True
+    # [xr] Duration [s] of the reset-to-origin slew (passed through to setup_xr).
+    reset_duration: float = RESET_DURATION_S
+
+    # [leader] Slew the follower to the leader's first pose before mirroring (--align=false to
+    # begin the 1:1 mirror immediately; the follower may snap).
+    align: bool = True
+    # [leader] Duration [s] of the startup alignment slew.
+    align_duration: float = ALIGN_DURATION_S
+
+    # Resume recording on an existing (previously interrupted) dataset.
+    resume: bool = False
+
+
+@safe_stop_image_writer
+def _record_loop(
+    robot,
+    device: Device,
+    motor_names: list[str],
+    events: dict,
+    fps: int,
+    dataset: LeRobotDataset | None = None,
+    control_time_s: float = 0.0,
+    single_task: str | None = None,
+) -> None:
+    """Run one episode (or reset phase) of the control loop.
+
+    When ``dataset`` is None the loop still controls the robot (so the operator
+    can reposition the arm during the reset window) but does not record frames.
+    """
+    control_interval = 1.0 / fps
+    timestamp = 0.0
+    start_t = time.perf_counter()
+    record_frames = dataset is not None
+    hold = HoldLatch(motor_names)
+
+    while timestamp < control_time_s:
+        loop_start = time.perf_counter()
+
+        if events["exit_early"]:
+            events["exit_early"] = False
+            break
+
+        obs = robot.get_observation()
+
+        if record_frames:
+            observation_frame = build_dataset_frame(dataset.features, obs, prefix=OBS_STR)
+
+        # Device idle (XR clutch disengaged, or leader stream stale) -> hold the pose
+        # latched on the active->idle edge.
+        action = hold.resolve(device.compute(obs), obs)
+
+        robot.send_action(action)
+
+        if record_frames:
+            action_frame = build_dataset_frame(dataset.features, action, prefix=ACTION)
+            dataset.add_frame({**observation_frame, **action_frame, "task": single_task})
+
+        dt_s = time.perf_counter() - loop_start
+        precise_sleep(max(control_interval - dt_s, 0.0))
+        timestamp = time.perf_counter() - start_t
+
+
+@parser.wrap()
+def record(cfg: RecordConfig) -> LeRobotDataset:
+    init_logging()
+    logging.info(pformat(asdict(cfg)))
+
+    # Connect the follower, build the selected Isaac device, and run its pre-loop startup
+    # (reset slew / leader align) — shared with teleoperate.py.
+    robot, device, motor_names = build_device(cfg)
+
+    # Build dataset feature spec.  The IK pipeline lives inside device.compute(), so the
+    # action features are exactly robot.action_features (joint positions in degrees).
+    teleop_proc, _, obs_proc = make_default_processors()
+    dataset_features = combine_feature_dicts(
+        aggregate_pipeline_dataset_features(
+            pipeline=teleop_proc,
+            initial_features=create_initial_features(action=robot.action_features),
+            use_videos=cfg.dataset.video,
+        ),
+        aggregate_pipeline_dataset_features(
+            pipeline=obs_proc,
+            initial_features=create_initial_features(observation=robot.observation_features),
+            use_videos=cfg.dataset.video,
+        ),
+    )
+
+    num_cameras = len(robot.cameras) if hasattr(robot, "cameras") else 0
+    image_writer_threads = cfg.dataset.num_image_writer_threads_per_camera * num_cameras
+
+    dataset: LeRobotDataset | None = None
+    listener = None
+    try:
+        if cfg.resume:
+            dataset = LeRobotDataset.resume(
+                cfg.dataset.repo_id,
+                root=cfg.dataset.root,
+                batch_encoding_size=cfg.dataset.video_encoding_batch_size,
+                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,
+                image_writer_processes=cfg.dataset.num_image_writer_processes if num_cameras > 0 else 0,
+                image_writer_threads=image_writer_threads if num_cameras > 0 else 0,
+            )
+            sanity_check_dataset_robot_compatibility(dataset, robot, cfg.dataset.fps, dataset_features)
+        else:
+            cfg.dataset.stamp_repo_id()
+            dataset = LeRobotDataset.create(
+                cfg.dataset.repo_id,
+                cfg.dataset.fps,
+                root=cfg.dataset.root,
+                robot_type=robot.name,
+                features=dataset_features,
+                use_videos=cfg.dataset.video,
+                image_writer_processes=cfg.dataset.num_image_writer_processes,
+                image_writer_threads=image_writer_threads,
+                batch_encoding_size=cfg.dataset.video_encoding_batch_size,
+                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,
+            )
+
+        listener, events = init_keyboard_listener()
+
+        loop_kwargs = {
+            "robot": robot,
+            "device": device,
+            "motor_names": motor_names,
+            "events": events,
+            "fps": cfg.dataset.fps,
+            "single_task": cfg.dataset.single_task,
+        }
+
+        with VideoEncodingManager(dataset):
+            recorded_episodes = 0
+            while recorded_episodes < cfg.dataset.num_episodes and not events["stop_recording"]:
+                logging.info(f"Recording episode {dataset.num_episodes}")
+                _record_loop(
+                    **loop_kwargs,
+                    dataset=dataset,
+                    control_time_s=cfg.dataset.episode_time_s,
+                )
+
+                # Reset window: give the operator time to reposition the scene.
+                # Skipped for the last episode (or if stop_recording was set).
+                if not events["stop_recording"] and (
+                    recorded_episodes < cfg.dataset.num_episodes - 1 or events["rerecord_episode"]
+                ):
+                    logging.info("Reset the environment")
+                    _record_loop(
+                        **loop_kwargs,
+                        dataset=None,
+                        control_time_s=cfg.dataset.reset_time_s,
+                    )
+
+                if events["rerecord_episode"]:
+                    logging.info("Re-record episode")
+                    events["rerecord_episode"] = False
+                    events["exit_early"] = False
+                    dataset.clear_episode_buffer()
+                    continue
+
+                dataset.save_episode()
+                recorded_episodes += 1
+
+    finally:
+        logging.info("Stop recording")
+
+        # Hardware teardown FIRST, each step guarded: the arm must be freed promptly (not
+        # after a potentially long finalize/encode), a cleanup failure must not skip the
+        # follower disconnect (which is what disables torque), and neither must prevent
+        # the dataset from being finalized below.
+        try:
+            device.cleanup()
+        except Exception:
+            logging.exception("Device cleanup failed")
+        try:
+            if robot.is_connected:
+                robot.disconnect()
+        except Exception:
+            logging.exception("Robot disconnect failed")
+
+        # Restore the terminal before the (potentially long) finalize/encode.
+        if listener is not None:
+            try:
+                listener.stop()
+            except Exception:
+                logging.exception("Keyboard listener stop failed")
+
+        if dataset is not None:
+            dataset.finalize()
+
+        if cfg.dataset.push_to_hub:
+            if dataset is not None and dataset.num_episodes > 0:
+                dataset.push_to_hub(tags=cfg.dataset.tags, private=cfg.dataset.private)
+            else:
+                logging.warning("No episodes saved — skipping push to hub")
+
+        logging.info("Exiting")
+
+    return dataset
+
+
+def main():
+    record()
+
+
+if __name__ == "__main__":
+    main()

examples/isaac_teleop_to_so101/teleoperate.py

@@ -0,0 +1,117 @@
+#!/usr/bin/env python
+
+# Copyright 2026 NVIDIA Corporation and 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.
+
+"""Teleoperate an SO-101 follower arm via NVIDIA Isaac Teleop.
+
+``lerobot-teleoperate``-style CLI (draccus): ``--teleop.type`` selects the Isaac device
+(``xr_controller`` | ``so101_leader``), ``--robot.*`` the follower::
+
+    # XR (VR) controller: clutch + soft-orientation IK
+    python -m examples.isaac_teleop_to_so101.teleoperate --robot.type=so101_follower \
+        --robot.port=/dev/ttyACM0 --robot.id=so101_follower_arm --teleop.type=xr_controller
+
+    # SO-101 leader arm: 1:1 joint mirror (real leader on /dev/ttyACM1)
+    python -m examples.isaac_teleop_to_so101.teleoperate --robot.type=so101_follower \
+        --robot.port=/dev/ttyACM0 --robot.id=so101_follower_arm --teleop.type=so101_leader \
+        --teleop.port=/dev/ttyACM1 --teleop.id=so101_leader_arm \
+        --launch_plugin=/code/Teleop/install/plugins/so101_leader/so101_leader_plugin
+
+``--teleop.type`` resolves against the Isaac device registry (see :class:`IsaacTeleopConfig`),
+distinct from the serial ``so101_leader``. The pipelines, clutch/IK/align internals, and
+reset-pose behavior live in ``common.py``. Requires the ``isaacteleop`` package and an OpenXR
+runtime (install instructions in this folder's ``README.md``).
+"""
+
+import time
+from dataclasses import dataclass
+
+from lerobot.configs import parser
+from lerobot.robots import RobotConfig
+from lerobot.robots.so_follower import SOFollowerConfig  # noqa: F401  (registers so101_follower)
+from lerobot.utils.robot_utils import precise_sleep
+
+from .common import (
+    ALIGN_DURATION_S,
+    FPS,
+    RESET_DURATION_S,
+    HoldLatch,
+    build_device,
+)
+from .isaac_teleop import IsaacTeleopConfig
+
+
+@dataclass
+class TeleoperateConfig:
+    """``lerobot-teleoperate``-style CLI for the Isaac Teleop -> SO-101 example.
+
+    The fields below are the loop/launch knobs (not part of either device's config); the
+    ``[xr]`` / ``[leader]`` tags mark which device a knob applies to. Use ``--flag=false``
+    for booleans (draccus style).
+    """
+
+    # Isaac Teleop input device + its knobs (--teleop.type=xr_controller|so101_leader,
+    # then --teleop.<field>=...). Resolved against IsaacTeleopConfig's own choice registry.
+    teleop: IsaacTeleopConfig
+    # SO-101 FOLLOWER arm (--robot.type=so101_follower --robot.port=/dev/ttyACM0 --robot.id=...).
+    robot: RobotConfig
+
+    # [leader] Path to the so101_leader plugin binary to spawn AFTER CloudXR is up (it then
+    # inherits the runtime env). None (default) -> assume the plugin already runs externally.
+    # The leader's serial port is --teleop.port (forwarded to the plugin; empty -> synthetic).
+    launch_plugin: str | None = None
+
+    # [xr] Slew all joints to a default reset pose before the loop (--reset_to_origin=false to
+    # keep the arm where it is). After the slew the clutch seeds its home from the measured pose.
+    reset_to_origin: bool = True
+    # [xr] Duration [s] of the reset-to-origin slew.
+    reset_duration: float = RESET_DURATION_S
+
+    # [leader] Slew the follower to the leader's first pose before mirroring (--align=false to
+    # begin the 1:1 mirror immediately; the follower may snap).
+    align: bool = True
+    # [leader] Duration [s] of the startup alignment slew.
+    align_duration: float = ALIGN_DURATION_S
+
+
+@parser.wrap()
+def teleoperate(cfg: TeleoperateConfig):
+    robot, device, motor_names = build_device(cfg)
+    hold = HoldLatch(motor_names)
+    try:
+        while True:
+            t0 = time.perf_counter()
+            obs = robot.get_observation()
+            # Idle (compute() -> None) holds the pose latched on the active->idle edge.
+            action = hold.resolve(device.compute(obs), obs)
+            robot.send_action(action)
+            precise_sleep(max(1.0 / FPS - (time.perf_counter() - t0), 0.0))
+    except KeyboardInterrupt:
+        pass
+    finally:
+        # A failing device cleanup must not skip the follower disconnect (which is what
+        # disables torque on the arm).
+        try:
+            device.cleanup()
+        finally:
+            robot.disconnect()
+
+
+def main():
+    teleoperate()
+
+
+if __name__ == "__main__":
+    main()

examples/lekiwi/evaluate.py

@@ -17,7 +17,7 @@
 import logging
 import time
 
-from lerobot.common.control_utils import init_keyboard_listener, predict_action
+from lerobot.common.control_utils import predict_action
 from lerobot.datasets import LeRobotDataset
 from lerobot.policies import make_pre_post_processors
 from lerobot.policies.act import ACTPolicy
@@ -26,6 +26,7 @@ from lerobot.processor import make_default_processors
 from lerobot.robots.lekiwi import LeKiwiClient, LeKiwiClientConfig
 from lerobot.utils.constants import ACTION, OBS_STR
 from lerobot.utils.feature_utils import build_dataset_frame, hw_to_dataset_features
+from lerobot.utils.keyboard_input import init_keyboard_listener
 from lerobot.utils.robot_utils import precise_sleep
 from lerobot.utils.utils import log_say
 from lerobot.utils.visualization_utils import init_rerun, log_rerun_data

examples/lekiwi/record.py

@@ -14,7 +14,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from lerobot.common.control_utils import init_keyboard_listener
 from lerobot.datasets import LeRobotDataset
 from lerobot.processor import make_default_processors
 from lerobot.robots.lekiwi import LeKiwiClient, LeKiwiClientConfig
@@ -23,6 +22,7 @@ from lerobot.teleoperators.keyboard import KeyboardTeleop, KeyboardTeleopConfig
 from lerobot.teleoperators.so_leader import SO100Leader, SO100LeaderConfig
 from lerobot.utils.constants import ACTION, OBS_STR
 from lerobot.utils.feature_utils import hw_to_dataset_features
+from lerobot.utils.keyboard_input import init_keyboard_listener
 from lerobot.utils.utils import log_say
 from lerobot.utils.visualization_utils import init_rerun
 

examples/phone_to_so100/evaluate.py

@@ -18,7 +18,7 @@ import logging
 import time
 
 from lerobot.cameras.opencv import OpenCVCameraConfig
-from lerobot.common.control_utils import init_keyboard_listener, predict_action
+from lerobot.common.control_utils import predict_action
 from lerobot.configs import FeatureType, PolicyFeature
 from lerobot.datasets import LeRobotDataset, aggregate_pipeline_dataset_features, create_initial_features
 from lerobot.model.kinematics import RobotKinematics
@@ -41,6 +41,7 @@ from lerobot.robots.so_follower.robot_kinematic_processor import (
 from lerobot.types import RobotAction, RobotObservation
 from lerobot.utils.constants import ACTION, OBS_STR
 from lerobot.utils.feature_utils import build_dataset_frame, combine_feature_dicts
+from lerobot.utils.keyboard_input import init_keyboard_listener
 from lerobot.utils.robot_utils import precise_sleep
 from lerobot.utils.utils import log_say
 from lerobot.utils.visualization_utils import init_rerun, log_rerun_data

examples/phone_to_so100/record.py

@@ -15,7 +15,6 @@
 # limitations under the License.
 
 from lerobot.cameras.opencv import OpenCVCameraConfig
-from lerobot.common.control_utils import init_keyboard_listener
 from lerobot.datasets import LeRobotDataset, aggregate_pipeline_dataset_features, create_initial_features
 from lerobot.model.kinematics import RobotKinematics
 from lerobot.processor import (
@@ -39,6 +38,7 @@ from lerobot.teleoperators.phone.config_phone import PhoneOS
 from lerobot.teleoperators.phone.phone_processor import MapPhoneActionToRobotAction
 from lerobot.types import RobotAction, RobotObservation
 from lerobot.utils.feature_utils import combine_feature_dicts
+from lerobot.utils.keyboard_input import init_keyboard_listener
 from lerobot.utils.utils import log_say
 from lerobot.utils.visualization_utils import init_rerun
 

examples/so100_to_so100_EE/evaluate.py

@@ -18,7 +18,7 @@ import logging
 import time
 
 from lerobot.cameras.opencv import OpenCVCameraConfig
-from lerobot.common.control_utils import init_keyboard_listener, predict_action
+from lerobot.common.control_utils import predict_action
 from lerobot.configs import FeatureType, PolicyFeature
 from lerobot.datasets import LeRobotDataset, aggregate_pipeline_dataset_features, create_initial_features
 from lerobot.model.kinematics import RobotKinematics
@@ -41,6 +41,7 @@ from lerobot.robots.so_follower.robot_kinematic_processor import (
 from lerobot.types import RobotAction, RobotObservation
 from lerobot.utils.constants import ACTION, OBS_STR
 from lerobot.utils.feature_utils import build_dataset_frame, combine_feature_dicts
+from lerobot.utils.keyboard_input import init_keyboard_listener
 from lerobot.utils.robot_utils import precise_sleep
 from lerobot.utils.utils import log_say
 from lerobot.utils.visualization_utils import init_rerun, log_rerun_data

examples/so100_to_so100_EE/record.py

@@ -16,7 +16,6 @@
 
 
 from lerobot.cameras.opencv import OpenCVCameraConfig
-from lerobot.common.control_utils import init_keyboard_listener
 from lerobot.datasets import LeRobotDataset, aggregate_pipeline_dataset_features, create_initial_features
 from lerobot.model.kinematics import RobotKinematics
 from lerobot.processor import (
@@ -36,6 +35,7 @@ from lerobot.scripts.lerobot_record import record_loop
 from lerobot.teleoperators.so_leader import SO100Leader, SO100LeaderConfig
 from lerobot.types import RobotAction, RobotObservation
 from lerobot.utils.feature_utils import combine_feature_dicts
+from lerobot.utils.keyboard_input import init_keyboard_listener
 from lerobot.utils.utils import log_say
 from lerobot.utils.visualization_utils import init_rerun
 

pyproject.toml

@@ -25,7 +25,7 @@ discord = "https://discord.gg/s3KuuzsPFb"
 
 [project]
 name = "lerobot"
-version = "0.5.2"
+version = "0.6.0"
 description = "🤗 LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch"
 dynamic = ["readme"]
 license = { text = "Apache-2.0" }
@@ -124,7 +124,8 @@ hardware = [
     "lerobot[deepdiff-dep]",
 ]
 viz = [
-    "rerun-sdk>=0.24.0,<0.27.0",
+    "rerun-sdk>=0.24.0,<0.34.0",
+    "foxglove-sdk>=0.25.1,<0.26.0",
 ]
 # ── User-facing composite extras (map to CLI scripts) ─────
 # lerobot-record, lerobot-replay, lerobot-calibrate, lerobot-teleoperate, etc.
@@ -140,7 +141,14 @@ av-dep = ["av>=15.0.0,<16.0.0"]
 pygame-dep = ["pygame>=2.5.1,<2.7.0"]
 # NOTE: 0.9.16 links against liburdfdom_sensor.so.4, which is unavailable on Ubuntu 24.04
 # (noble ships urdfdom 3.x). Cap below 0.9.16 until system urdfdom 4.x is broadly available.
-placo-dep = ["placo>=0.9.6,<0.9.16"]
+#
+# NOTE: placo pulls in pin (Pinocchio), whose binary wheels dlopen specific cmeel sonames
+# (liburdfdom_sensor.so.4.0, libtinyxml2.so.10) but declare only `>=` floors on their cmeel
+# packages. The 2026-05-21 major bumps (cmeel-urdfdom 6.0.0 -> .so.6, cmeel-tinyxml2 11.0.0
+# -> .so.11) ship newer sonames, so left unpinned the resolver grabs them and `import placo`
+# fails at load with "liburdfdom_sensor.so.4.0: cannot open shared object file" (see #3755).
+# There is no cmeel-urdfdom 5.x; <5 selects the 4.x ABI the placo/pin wheels are built against.
+placo-dep = ["placo>=0.9.6,<0.9.16", "cmeel-urdfdom>=4,<5", "cmeel-tinyxml2<11"]
 transformers-dep = ["transformers>=5.4.0,<5.6.0"]
 grpcio-dep = ["grpcio>=1.73.1,<2.0.0", "protobuf>=6.31.1,<8.0.0"]
 accelerate-dep = ["accelerate>=1.14.0,<2.0.0"]
@@ -156,6 +164,7 @@ pynput-dep = ["pynput>=1.7.8,<1.9.0"]
 pyzmq-dep = ["pyzmq>=26.2.1,<28.0.0"]
 motorbridge-dep = ["motorbridge>=0.3.2,<0.4.0"]
 motorbridge-smart-servo-dep = ["motorbridge-smart-servo>=0.0.4,<0.1.0"]
+timm-dep = ["timm>=1.0.0,<1.1.0"]
 
 # Motors
 feetech = ["feetech-servo-sdk>=1.0.0,<2.0.0", "lerobot[pyserial-dep]", "lerobot[deepdiff-dep]"]
@@ -211,24 +220,44 @@ 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",
+    "lerobot[timm-dep]",
     "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]"]
 topreward = ["lerobot[transformers-dep]"]
 xvla = ["lerobot[transformers-dep]"]
 eo1 = ["lerobot[transformers-dep]", "lerobot[qwen-vl-utils-dep]"]
+fastwam = [
+    "lerobot[transformers-dep]",
+    "lerobot[diffusers-dep]",
+]
+evo1 = ["lerobot[transformers-dep]"]
 hilserl = ["lerobot[transformers-dep]", "lerobot[dataset]", "gym-hil>=0.1.14,<0.2.0", "lerobot[grpcio-dep]", "lerobot[placo-dep]"]
 vla_jepa = ["lerobot[transformers-dep]", "lerobot[diffusers-dep]", "lerobot[qwen-vl-utils-dep]"]
+lingbot_va = ["lerobot[transformers-dep]", "lerobot[diffusers-dep]", "lerobot[accelerate-dep]"]
 
 # Features
 async = ["lerobot[grpcio-dep]", "lerobot[matplotlib-dep]"]
 peft = ["lerobot[transformers-dep]", "lerobot[peft-dep]"]
 
+# Annotation pipeline (lerobot-annotate). The only backend is ``openai``,
+# which talks to any OpenAI-compatible server (``vllm serve`` /
+# ``transformers serve`` / hosted). Distributed runs use Hugging Face Jobs
+# (see examples/annotations/run_hf_job.py).
+annotations = [
+    "lerobot[dataset]",
+    "lerobot[transformers-dep]",
+    "openai>=1.40,<2.0",
+    # ``vllm`` is intentionally NOT a hard dep: it pins an older torch, and
+    # uv's single unified lock would then cap ``torch`` for every extra
+    # (e.g. forcing 2.8 while ``torchcodec`` in [dataset] needs 2.11 -> ABI
+    # break in CI). The HF Jobs image (``vllm/vllm-openai``) provides vLLM;
+    # install it locally only if you run your own ``vllm serve``.
+]
+
 # Development
 dev = ["pre-commit>=3.7.0,<5.0.0", "debugpy>=1.8.1,<1.9.0", "lerobot[grpcio-dep]", "grpcio-tools>=1.73.1,<2.0.0", "mypy>=1.19.1", "ruff>=0.14.1", "lerobot[notebook]"]
 notebook = ["jupyter>=1.0.0,<2.0.0", "ipykernel>=6.0.0,<7.0.0"]
@@ -286,10 +315,13 @@ all = [
     "lerobot[pi]",
     "lerobot[molmoact2]",
     "lerobot[smolvla]",
-    # "lerobot[groot]", TODO(Steven): Gr00t requires specific installation instructions for flash-attn
+    "lerobot[fastwam]",
+    "lerobot[groot]",
     "lerobot[xvla]",
+    "lerobot[evo1]",
     "lerobot[hilserl]",
     "lerobot[vla_jepa]",
+    "lerobot[lingbot_va]",
     "lerobot[async]",
     "lerobot[dev]",
     "lerobot[test]",
@@ -323,6 +355,7 @@ lerobot-find-joint-limits="lerobot.scripts.lerobot_find_joint_limits:main"
 lerobot-imgtransform-viz="lerobot.scripts.lerobot_imgtransform_viz:main"
 lerobot-edit-dataset="lerobot.scripts.lerobot_edit_dataset:main"
 lerobot-setup-can="lerobot.scripts.lerobot_setup_can:main"
+lerobot-annotate="lerobot.scripts.lerobot_annotate:main"
 lerobot-rollout="lerobot.scripts.lerobot_rollout:main"
 
 # ---------------- Tool Configurations ----------------
@@ -341,7 +374,7 @@ torch = [{ index = "pytorch-cu128", marker = "sys_platform == 'linux'" }]
 torchvision = [{ index = "pytorch-cu128", marker = "sys_platform == 'linux'" }]
 
 [tool.setuptools.package-data]
-lerobot = ["envs/*.json"]
+lerobot = ["envs/*.json", "annotations/steerable_pipeline/prompts/*.txt"]
 
 [tool.setuptools.packages.find]
 where = ["src"]
@@ -421,7 +454,8 @@ default.extend-ignore-identifiers-re = [
     "is_compileable",
     "ROBOTIS",
     "OT_VALUE",
-    "VanderBilt"
+    "VanderBilt",
+    "seperated_timestep",
 ]
 
 # TODO: Uncomment when ready to use

requirements-macos.txt

@@ -1,729 +0,0 @@
-#
-# This file is autogenerated by pip-compile with Python 3.12
-# by the following command:
-#
-#    pip-compile --output-file=requirements-macos.txt requirements.in
-#
--e .[all]
-    # via -[all]
-absl-py==2.4.0
-    # via
-    #   dm-control
-    #   dm-env
-    #   dm-tree
-    #   labmaze
-    #   mujoco
-accelerate==1.13.0
-    # via
-    #   lerobot
-    #   peft
-aiohappyeyeballs==2.6.1
-    # via aiohttp
-aiohttp==3.13.3
-    # via fsspec
-aiosignal==1.4.0
-    # via aiohttp
-annotated-doc==0.0.4
-    # via
-    #   fastapi
-    #   typer
-annotated-types==0.7.0
-    # via pydantic
-anyio==4.12.1
-    # via
-    #   httpx
-    #   starlette
-    #   watchfiles
-asttokens==3.0.1
-    # via stack-data
-attrs==25.4.0
-    # via
-    #   aiohttp
-    #   dm-tree
-    #   jsonlines
-    #   rerun-sdk
-av==15.1.0
-    # via
-    #   lerobot
-    #   qwen-vl-utils
-certifi==2026.2.25
-    # via
-    #   httpcore
-    #   httpx
-    #   requests
-    #   sentry-sdk
-cffi==2.0.0
-    # via pymunk
-cfgv==3.5.0
-    # via pre-commit
-charset-normalizer==3.4.5
-    # via requests
-click==8.3.1
-    # via
-    #   typer
-    #   uvicorn
-    #   wandb
-cloudpickle==3.1.2
-    # via gymnasium
-cmake==4.1.3
-    # via lerobot
-cmeel==0.59.0
-    # via
-    #   cmeel-assimp
-    #   cmeel-boost
-    #   cmeel-console-bridge
-    #   cmeel-octomap
-    #   cmeel-qhull
-    #   cmeel-tinyxml2
-    #   cmeel-urdfdom
-    #   cmeel-zlib
-    #   coal-library
-    #   eigenpy
-    #   eiquadprog
-    #   pin
-    #   placo
-    #   rhoban-cmeel-jsoncpp
-cmeel-assimp==5.4.3.1
-    # via coal-library
-cmeel-boost==1.87.0.1
-    # via
-    #   coal-library
-    #   eigenpy
-    #   eiquadprog
-    #   pin
-cmeel-console-bridge==1.0.2.3
-    # via cmeel-urdfdom
-cmeel-octomap==1.10.0
-    # via coal-library
-cmeel-qhull==8.0.2.1
-    # via coal-library
-cmeel-tinyxml2==10.0.0
-    # via cmeel-urdfdom
-cmeel-urdfdom==4.0.1
-    # via pin
-cmeel-zlib==1.3.1
-    # via cmeel-assimp
-coal-library==3.0.1
-    # via pin
-contourpy==1.3.3
-    # via
-    #   lerobot
-    #   matplotlib
-coverage[toml]==7.13.4
-    # via pytest-cov
-cycler==0.12.1
-    # via matplotlib
-datasets==4.6.1
-    # via lerobot
-debugpy==1.8.20
-    # via lerobot
-decorator==5.2.1
-    # via ipython
-deepdiff==8.6.1
-    # via lerobot
-diffusers==0.35.2
-    # via lerobot
-dill==0.4.0
-    # via
-    #   datasets
-    #   multiprocess
-distlib==0.4.0
-    # via virtualenv
-dm-control==1.0.37
-    # via gym-aloha
-dm-env==1.6
-    # via dm-control
-dm-tree==0.1.9
-    # via
-    #   dm-control
-    #   dm-env
-docopt==0.6.2
-    # via num2words
-draccus==0.10.0
-    # via lerobot
-dynamixel-sdk==3.8.4
-    # via lerobot
-eigenpy==3.10.3
-    # via coal-library
-einops==0.8.2
-    # via lerobot
-eiquadprog==1.2.9
-    # via placo
-etils[epath,epy]==1.14.0
-    # via mujoco
-executing==2.2.1
-    # via stack-data
-faker==34.0.2
-    # via lerobot
-farama-notifications==0.0.4
-    # via gymnasium
-fastapi==0.135.1
-    # via
-    #   lerobot
-    #   teleop
-feetech-servo-sdk==1.0.0
-    # via lerobot
-filelock==3.25.0
-    # via
-    #   datasets
-    #   diffusers
-    #   huggingface-hub
-    #   python-discovery
-    #   torch
-    #   virtualenv
-fonttools==4.61.1
-    # via matplotlib
-frozenlist==1.8.0
-    # via
-    #   aiohttp
-    #   aiosignal
-fsspec[http]==2026.2.0
-    # via
-    #   datasets
-    #   etils
-    #   huggingface-hub
-    #   torch
-gitdb==4.0.12
-    # via gitpython
-gitpython==3.1.46
-    # via wandb
-glfw==2.10.0
-    # via
-    #   dm-control
-    #   mujoco
-grpcio==1.73.1
-    # via
-    #   grpcio-tools
-    #   lerobot
-    #   reachy2-sdk
-    #   reachy2-sdk-api
-grpcio-tools==1.73.1
-    # via
-    #   lerobot
-    #   reachy2-sdk-api
-gym-aloha==0.1.3
-    # via lerobot
-gym-hil==0.1.13
-    # via lerobot
-gym-pusht==0.1.6
-    # via lerobot
-gymnasium==1.2.3
-    # via
-    #   gym-aloha
-    #   gym-hil
-    #   gym-pusht
-    #   lerobot
-    #   metaworld
-h11==0.16.0
-    # via
-    #   httpcore
-    #   uvicorn
-hebi-py==2.11.0
-    # via lerobot
-hf-xet==1.3.2
-    # via huggingface-hub
-hidapi==0.14.0.post4
-    # via
-    #   gym-hil
-    #   lerobot
-httpcore==1.0.9
-    # via httpx
-httptools==0.7.1
-    # via uvicorn
-httpx==0.28.1
-    # via
-    #   datasets
-    #   huggingface-hub
-huggingface-hub==1.6.0
-    # via
-    #   accelerate
-    #   datasets
-    #   diffusers
-    #   lerobot
-    #   peft
-    #   tokenizers
-    #   transformers
-identify==2.6.17
-    # via pre-commit
-idna==3.11
-    # via
-    #   anyio
-    #   httpx
-    #   requests
-    #   yarl
-imageio[ffmpeg]==2.37.2
-    # via
-    #   gym-aloha
-    #   gym-hil
-    #   lerobot
-    #   metaworld
-    #   scikit-image
-imageio-ffmpeg==0.6.0
-    # via imageio
-importlib-metadata==8.7.1
-    # via diffusers
-iniconfig==2.3.0
-    # via pytest
-ipython==9.11.0
-    # via meshcat
-ipython-pygments-lexers==1.1.1
-    # via ipython
-ischedule==1.2.7
-    # via placo
-jedi==0.19.2
-    # via ipython
-jinja2==3.1.6
-    # via torch
-jsonlines==4.0.0
-    # via lerobot
-kiwisolver==1.4.9
-    # via matplotlib
-labmaze==1.0.6
-    # via dm-control
-lazy-loader==0.5
-    # via scikit-image
-librt==0.8.1
-    # via mypy
-lxml==6.0.2
-    # via dm-control
-markdown-it-py==4.0.0
-    # via rich
-markupsafe==3.0.3
-    # via jinja2
-matplotlib==3.10.8
-    # via lerobot
-matplotlib-inline==0.2.1
-    # via ipython
-mdurl==0.1.2
-    # via markdown-it-py
-mergedeep==1.3.4
-    # via draccus
-meshcat==0.3.2
-    # via placo
-metaworld==3.0.0
-    # via lerobot
-mock-serial==0.0.1
-    # via lerobot
-mpmath==1.3.0
-    # via sympy
-mujoco==3.5.0
-    # via
-    #   dm-control
-    #   gym-aloha
-    #   gym-hil
-    #   metaworld
-multidict==6.7.1
-    # via
-    #   aiohttp
-    #   yarl
-multiprocess==0.70.18
-    # via datasets
-mypy==1.19.1
-    # via lerobot
-mypy-extensions==1.1.0
-    # via
-    #   mypy
-    #   typing-inspect
-networkx==3.6.1
-    # via
-    #   scikit-image
-    #   torch
-nodeenv==1.10.0
-    # via pre-commit
-num2words==0.5.14
-    # via lerobot
-numpy==2.2.6
-    # via
-    #   accelerate
-    #   cmeel-boost
-    #   contourpy
-    #   datasets
-    #   diffusers
-    #   dm-control
-    #   dm-env
-    #   dm-tree
-    #   gymnasium
-    #   hebi-py
-    #   imageio
-    #   labmaze
-    #   lerobot
-    #   matplotlib
-    #   meshcat
-    #   metaworld
-    #   mujoco
-    #   opencv-python
-    #   opencv-python-headless
-    #   pandas
-    #   peft
-    #   pyquaternion
-    #   reachy2-sdk
-    #   rerun-sdk
-    #   scikit-image
-    #   scipy
-    #   shapely
-    #   teleop
-    #   tifffile
-    #   torchvision
-    #   transformers
-    #   transforms3d
-opencv-python==4.13.0.92
-    # via
-    #   gym-pusht
-    #   reachy2-sdk
-opencv-python-headless==4.12.0.88
-    # via lerobot
-orderly-set==5.5.0
-    # via deepdiff
-packaging==25.0
-    # via
-    #   accelerate
-    #   datasets
-    #   huggingface-hub
-    #   lazy-loader
-    #   lerobot
-    #   matplotlib
-    #   peft
-    #   pytest
-    #   qwen-vl-utils
-    #   reachy2-sdk
-    #   scikit-image
-    #   transformers
-    #   wandb
-pandas==2.3.3
-    # via
-    #   datasets
-    #   lerobot
-parso==0.8.6
-    # via jedi
-pathspec==1.0.4
-    # via mypy
-peft==0.18.1
-    # via lerobot
-pexpect==4.9.0
-    # via ipython
-pillow==12.1.1
-    # via
-    #   diffusers
-    #   imageio
-    #   matplotlib
-    #   meshcat
-    #   qwen-vl-utils
-    #   rerun-sdk
-    #   scikit-image
-    #   torchvision
-pin==3.4.0
-    # via placo
-placo==0.9.16
-    # via lerobot
-platformdirs==4.9.4
-    # via
-    #   python-discovery
-    #   virtualenv
-    #   wandb
-pluggy==1.6.0
-    # via
-    #   pytest
-    #   pytest-cov
-pre-commit==4.5.1
-    # via lerobot
-prompt-toolkit==3.0.52
-    # via ipython
-propcache==0.4.1
-    # via
-    #   aiohttp
-    #   yarl
-protobuf==6.31.1
-    # via
-    #   dm-control
-    #   grpcio-tools
-    #   lerobot
-    #   reachy2-sdk
-    #   reachy2-sdk-api
-    #   wandb
-psutil==7.2.2
-    # via
-    #   accelerate
-    #   imageio
-    #   peft
-ptyprocess==0.7.0
-    # via pexpect
-pure-eval==0.2.3
-    # via stack-data
-pyarrow==23.0.1
-    # via
-    #   datasets
-    #   rerun-sdk
-pycparser==3.0
-    # via cffi
-pydantic==2.12.5
-    # via
-    #   fastapi
-    #   wandb
-pydantic-core==2.41.5
-    # via pydantic
-pygame==2.6.1
-    # via
-    #   gym-hil
-    #   gym-pusht
-    #   lerobot
-pygments==2.19.2
-    # via
-    #   ipython
-    #   ipython-pygments-lexers
-    #   pytest
-    #   rich
-pymunk==6.11.1
-    # via
-    #   gym-pusht
-    #   lerobot
-pyngrok==7.5.1
-    # via meshcat
-pynput==1.8.1
-    # via
-    #   gym-hil
-    #   lerobot
-pyobjc-core==12.1
-    # via
-    #   pyobjc-framework-applicationservices
-    #   pyobjc-framework-cocoa
-    #   pyobjc-framework-coretext
-    #   pyobjc-framework-quartz
-pyobjc-framework-applicationservices==12.1
-    # via pynput
-pyobjc-framework-cocoa==12.1
-    # via
-    #   pyobjc-framework-applicationservices
-    #   pyobjc-framework-coretext
-    #   pyobjc-framework-quartz
-pyobjc-framework-coretext==12.1
-    # via pyobjc-framework-applicationservices
-pyobjc-framework-quartz==12.1
-    # via
-    #   pynput
-    #   pyobjc-framework-applicationservices
-    #   pyobjc-framework-coretext
-pyopengl==3.1.10
-    # via
-    #   dm-control
-    #   mujoco
-pyparsing==3.3.2
-    # via
-    #   dm-control
-    #   matplotlib
-pyquaternion==0.9.9
-    # via reachy2-sdk
-pyrealsense2-macosx==2.56.5
-    # via lerobot
-pyserial==3.5
-    # via
-    #   dynamixel-sdk
-    #   feetech-servo-sdk
-    #   lerobot
-pytest==8.4.2
-    # via
-    #   lerobot
-    #   pytest-cov
-    #   pytest-timeout
-    #   teleop
-pytest-cov==7.0.0
-    # via lerobot
-pytest-timeout==2.4.0
-    # via lerobot
-python-dateutil==2.9.0.post0
-    # via
-    #   faker
-    #   matplotlib
-    #   pandas
-python-discovery==1.1.1
-    # via virtualenv
-python-dotenv==1.2.2
-    # via uvicorn
-pytz==2026.1.post1
-    # via pandas
-pyyaml==6.0.3
-    # via
-    #   accelerate
-    #   datasets
-    #   draccus
-    #   hebi-py
-    #   huggingface-hub
-    #   peft
-    #   pre-commit
-    #   pyngrok
-    #   pyyaml-include
-    #   transformers
-    #   uvicorn
-    #   wandb
-pyyaml-include==1.4.1
-    # via draccus
-pyzmq==27.1.0
-    # via
-    #   lerobot
-    #   meshcat
-qwen-vl-utils==0.0.14
-    # via lerobot
-reachy2-sdk==1.0.15
-    # via lerobot
-reachy2-sdk-api==1.0.21
-    # via reachy2-sdk
-regex==2026.2.28
-    # via
-    #   diffusers
-    #   transformers
-requests==2.32.5
-    # via
-    #   datasets
-    #   diffusers
-    #   dm-control
-    #   qwen-vl-utils
-    #   teleop
-    #   wandb
-rerun-sdk==0.26.2
-    # via lerobot
-rhoban-cmeel-jsoncpp==1.9.4.9
-    # via placo
-rich==14.3.3
-    # via typer
-safetensors==0.7.0
-    # via
-    #   accelerate
-    #   diffusers
-    #   lerobot
-    #   peft
-    #   transformers
-scikit-image==0.25.2
-    # via
-    #   gym-pusht
-    #   lerobot
-scipy==1.17.1
-    # via
-    #   dm-control
-    #   lerobot
-    #   metaworld
-    #   scikit-image
-    #   torchdiffeq
-sentry-sdk==2.54.0
-    # via wandb
-shapely==2.1.2
-    # via gym-pusht
-shellingham==1.5.4
-    # via typer
-six==1.17.0
-    # via
-    #   pynput
-    #   python-dateutil
-smmap==5.0.3
-    # via gitdb
-stack-data==0.6.3
-    # via ipython
-starlette==0.52.1
-    # via fastapi
-sympy==1.14.0
-    # via torch
-teleop==0.1.4
-    # via lerobot
-termcolor==3.3.0
-    # via lerobot
-tifffile==2026.3.3
-    # via scikit-image
-tokenizers==0.22.2
-    # via transformers
-toml==0.10.2
-    # via draccus
-torch==2.10.0
-    # via
-    #   accelerate
-    #   lerobot
-    #   peft
-    #   torchdiffeq
-    #   torchvision
-torchcodec==0.10.0
-    # via lerobot
-torchdiffeq==0.2.5
-    # via lerobot
-torchvision==0.25.0
-    # via lerobot
-tornado==6.5.4
-    # via meshcat
-tqdm==4.67.3
-    # via
-    #   datasets
-    #   dm-control
-    #   huggingface-hub
-    #   peft
-    #   transformers
-traitlets==5.14.3
-    # via
-    #   ipython
-    #   matplotlib-inline
-transformers==5.3.0
-    # via
-    #   lerobot
-    #   peft
-transforms3d==0.4.2
-    # via teleop
-typer==0.24.1
-    # via
-    #   huggingface-hub
-    #   transformers
-typing-extensions==4.15.0
-    # via
-    #   aiosignal
-    #   anyio
-    #   etils
-    #   faker
-    #   fastapi
-    #   gymnasium
-    #   huggingface-hub
-    #   mypy
-    #   pydantic
-    #   pydantic-core
-    #   rerun-sdk
-    #   starlette
-    #   torch
-    #   typing-inspect
-    #   typing-inspection
-    #   wandb
-typing-inspect==0.9.0
-    # via draccus
-typing-inspection==0.4.2
-    # via
-    #   fastapi
-    #   pydantic
-tzdata==2025.3
-    # via pandas
-u-msgpack-python==2.8.0
-    # via meshcat
-urllib3==2.6.3
-    # via
-    #   requests
-    #   sentry-sdk
-uvicorn[standard]==0.41.0
-    # via teleop
-uvloop==0.22.1
-    # via uvicorn
-virtualenv==21.1.0
-    # via pre-commit
-wandb==0.24.2
-    # via lerobot
-watchfiles==1.1.1
-    # via uvicorn
-wcwidth==0.6.0
-    # via prompt-toolkit
-websocket-client==1.9.0
-    # via teleop
-websockets==16.0
-    # via uvicorn
-wrapt==2.1.2
-    # via dm-tree
-xxhash==3.6.0
-    # via datasets
-yarl==1.23.0
-    # via aiohttp
-zipp==3.23.0
-    # via
-    #   etils
-    #   importlib-metadata
-
-# The following packages are considered to be unsafe in a requirements file:
-# setuptools

requirements-ubuntu.txt

@@ -1,882 +0,0 @@
-#
-# This file is autogenerated by pip-compile with Python 3.12
-# by the following command:
-#
-#    pip-compile --output-file=requirements-ubuntu.txt requirements.in
-#
--e .[all]
-    # via -[all]
-absl-py==2.4.0
-    # via
-    #   dm-control
-    #   dm-env
-    #   dm-tree
-    #   labmaze
-    #   mujoco
-    #   tensorboard
-accelerate==1.13.0
-    # via
-    #   lerobot
-    #   peft
-aiohappyeyeballs==2.6.1
-    # via aiohttp
-aiohttp==3.13.3
-    # via fsspec
-aiosignal==1.4.0
-    # via aiohttp
-annotated-doc==0.0.4
-    # via
-    #   fastapi
-    #   typer
-annotated-types==0.7.0
-    # via pydantic
-antlr4-python3-runtime==4.9.3
-    # via
-    #   hydra-core
-    #   omegaconf
-anyio==4.12.1
-    # via
-    #   httpx
-    #   starlette
-    #   watchfiles
-asttokens==3.0.1
-    # via stack-data
-attrs==25.4.0
-    # via
-    #   aiohttp
-    #   dm-tree
-    #   jsonlines
-    #   jsonschema
-    #   referencing
-    #   rerun-sdk
-av==15.1.0
-    # via
-    #   lerobot
-    #   qwen-vl-utils
-bddl==1.0.1
-    # via hf-libero
-certifi==2026.2.25
-    # via
-    #   httpcore
-    #   httpx
-    #   requests
-    #   sentry-sdk
-cffi==2.0.0
-    # via pymunk
-cfgv==3.5.0
-    # via pre-commit
-charset-normalizer==3.4.5
-    # via requests
-click==8.3.1
-    # via
-    #   typer
-    #   uvicorn
-    #   wandb
-cloudpickle==3.1.2
-    # via
-    #   gymnasium
-    #   hf-libero
-cmake==4.1.3
-    # via lerobot
-cmeel==0.59.0
-    # via
-    #   cmeel-assimp
-    #   cmeel-boost
-    #   cmeel-console-bridge
-    #   cmeel-octomap
-    #   cmeel-qhull
-    #   cmeel-tinyxml2
-    #   cmeel-urdfdom
-    #   cmeel-zlib
-    #   coal-library
-    #   eigenpy
-    #   eiquadprog
-    #   pin
-    #   placo
-    #   rhoban-cmeel-jsoncpp
-cmeel-assimp==5.4.3.1
-    # via coal-library
-cmeel-boost==1.87.0.1
-    # via
-    #   coal-library
-    #   eigenpy
-    #   eiquadprog
-    #   pin
-cmeel-console-bridge==1.0.2.3
-    # via cmeel-urdfdom
-cmeel-octomap==1.10.0
-    # via coal-library
-cmeel-qhull==8.0.2.1
-    # via coal-library
-cmeel-tinyxml2==10.0.0
-    # via cmeel-urdfdom
-cmeel-urdfdom==4.0.1
-    # via pin
-cmeel-zlib==1.3.1
-    # via cmeel-assimp
-coal-library==3.0.1
-    # via pin
-contourpy==1.3.3
-    # via
-    #   lerobot
-    #   matplotlib
-coverage[toml]==7.13.4
-    # via pytest-cov
-cuda-bindings==12.9.4
-    # via torch
-cuda-pathfinder==1.4.1
-    # via cuda-bindings
-cycler==0.12.1
-    # via matplotlib
-datasets==4.6.1
-    # via lerobot
-debugpy==1.8.20
-    # via lerobot
-decorator==5.2.1
-    # via ipython
-deepdiff==8.6.1
-    # via lerobot
-diffusers==0.35.2
-    # via lerobot
-dill==0.4.0
-    # via
-    #   datasets
-    #   multiprocess
-distlib==0.4.0
-    # via virtualenv
-dm-control==1.0.37
-    # via gym-aloha
-dm-env==1.6
-    # via dm-control
-dm-tree==0.1.9
-    # via
-    #   dm-control
-    #   dm-env
-docopt==0.6.2
-    # via num2words
-draccus==0.10.0
-    # via lerobot
-dynamixel-sdk==3.8.4
-    # via lerobot
-easydict==1.13
-    # via hf-libero
-egl-probe==1.0.2
-    # via robomimic
-eigenpy==3.10.3
-    # via coal-library
-einops==0.8.2
-    # via
-    #   hf-libero
-    #   lerobot
-eiquadprog==1.2.9
-    # via placo
-etils[epath,epy]==1.14.0
-    # via mujoco
-evdev==1.9.3
-    # via pynput
-executing==2.2.1
-    # via stack-data
-faker==34.0.2
-    # via lerobot
-farama-notifications==0.0.4
-    # via gymnasium
-fastapi==0.135.1
-    # via
-    #   lerobot
-    #   teleop
-fastjsonschema==2.21.2
-    # via nbformat
-feetech-servo-sdk==1.0.0
-    # via lerobot
-filelock==3.25.0
-    # via
-    #   datasets
-    #   diffusers
-    #   huggingface-hub
-    #   python-discovery
-    #   torch
-    #   virtualenv
-fonttools==4.61.1
-    # via matplotlib
-frozenlist==1.8.0
-    # via
-    #   aiohttp
-    #   aiosignal
-fsspec[http]==2026.2.0
-    # via
-    #   datasets
-    #   etils
-    #   huggingface-hub
-    #   torch
-future==1.0.0
-    # via hf-libero
-gitdb==4.0.12
-    # via gitpython
-gitpython==3.1.46
-    # via wandb
-glfw==2.10.0
-    # via
-    #   dm-control
-    #   mujoco
-grpcio==1.73.1
-    # via
-    #   grpcio-tools
-    #   lerobot
-    #   reachy2-sdk
-    #   reachy2-sdk-api
-    #   tensorboard
-grpcio-tools==1.73.1
-    # via
-    #   lerobot
-    #   reachy2-sdk-api
-gym-aloha==0.1.3
-    # via lerobot
-gym-hil==0.1.13
-    # via lerobot
-gym-pusht==0.1.6
-    # via lerobot
-gymnasium==1.2.3
-    # via
-    #   gym-aloha
-    #   gym-hil
-    #   gym-pusht
-    #   hf-libero
-    #   lerobot
-    #   metaworld
-h11==0.16.0
-    # via
-    #   httpcore
-    #   uvicorn
-h5py==3.16.0
-    # via robomimic
-hebi-py==2.11.0
-    # via lerobot
-hf-egl-probe==1.0.2
-    # via hf-libero
-hf-libero==0.1.3
-    # via lerobot
-hf-xet==1.3.2
-    # via huggingface-hub
-hidapi==0.14.0.post4
-    # via
-    #   gym-hil
-    #   lerobot
-httpcore==1.0.9
-    # via httpx
-httptools==0.7.1
-    # via uvicorn
-httpx==0.28.1
-    # via
-    #   datasets
-    #   huggingface-hub
-huggingface-hub==1.6.0
-    # via
-    #   accelerate
-    #   datasets
-    #   diffusers
-    #   lerobot
-    #   peft
-    #   tokenizers
-    #   transformers
-hydra-core==1.3.2
-    # via hf-libero
-identify==2.6.17
-    # via pre-commit
-idna==3.11
-    # via
-    #   anyio
-    #   httpx
-    #   requests
-    #   yarl
-imageio[ffmpeg]==2.37.2
-    # via
-    #   gym-aloha
-    #   gym-hil
-    #   lerobot
-    #   metaworld
-    #   robomimic
-    #   scikit-image
-imageio-ffmpeg==0.6.0
-    # via
-    #   imageio
-    #   robomimic
-importlib-metadata==8.7.1
-    # via diffusers
-iniconfig==2.3.0
-    # via pytest
-ipython==9.11.0
-    # via meshcat
-ipython-pygments-lexers==1.1.1
-    # via ipython
-ischedule==1.2.7
-    # via placo
-jedi==0.19.2
-    # via ipython
-jinja2==3.1.6
-    # via torch
-jsonlines==4.0.0
-    # via lerobot
-jsonschema==4.26.0
-    # via nbformat
-jsonschema-specifications==2025.9.1
-    # via jsonschema
-jupyter-core==5.9.1
-    # via nbformat
-jupytext==1.19.1
-    # via bddl
-kiwisolver==1.4.9
-    # via matplotlib
-labmaze==1.0.6
-    # via dm-control
-lazy-loader==0.5
-    # via scikit-image
-librt==0.8.1
-    # via mypy
-llvmlite==0.46.0
-    # via numba
-lxml==6.0.2
-    # via dm-control
-markdown==3.10.2
-    # via tensorboard
-markdown-it-py==4.0.0
-    # via
-    #   jupytext
-    #   mdit-py-plugins
-    #   rich
-markupsafe==3.0.3
-    # via
-    #   jinja2
-    #   werkzeug
-matplotlib==3.10.8
-    # via
-    #   hf-libero
-    #   lerobot
-matplotlib-inline==0.2.1
-    # via ipython
-mdit-py-plugins==0.5.0
-    # via jupytext
-mdurl==0.1.2
-    # via markdown-it-py
-mergedeep==1.3.4
-    # via draccus
-meshcat==0.3.2
-    # via placo
-metaworld==3.0.0
-    # via lerobot
-mock-serial==0.0.1
-    # via lerobot
-mpmath==1.3.0
-    # via sympy
-mujoco==3.5.0
-    # via
-    #   dm-control
-    #   gym-aloha
-    #   gym-hil
-    #   hf-libero
-    #   metaworld
-    #   robosuite
-multidict==6.7.1
-    # via
-    #   aiohttp
-    #   yarl
-multiprocess==0.70.18
-    # via datasets
-mypy==1.19.1
-    # via lerobot
-mypy-extensions==1.1.0
-    # via
-    #   mypy
-    #   typing-inspect
-nbformat==5.10.4
-    # via jupytext
-networkx==3.6.1
-    # via
-    #   bddl
-    #   scikit-image
-    #   torch
-nodeenv==1.10.0
-    # via pre-commit
-num2words==0.5.14
-    # via lerobot
-numba==0.64.0
-    # via robosuite
-numpy==2.2.6
-    # via
-    #   accelerate
-    #   bddl
-    #   cmeel-boost
-    #   contourpy
-    #   datasets
-    #   diffusers
-    #   dm-control
-    #   dm-env
-    #   dm-tree
-    #   gymnasium
-    #   h5py
-    #   hebi-py
-    #   hf-libero
-    #   imageio
-    #   labmaze
-    #   lerobot
-    #   matplotlib
-    #   meshcat
-    #   metaworld
-    #   mujoco
-    #   numba
-    #   opencv-python
-    #   opencv-python-headless
-    #   pandas
-    #   peft
-    #   pyquaternion
-    #   reachy2-sdk
-    #   rerun-sdk
-    #   robomimic
-    #   robosuite
-    #   scikit-image
-    #   scipy
-    #   shapely
-    #   teleop
-    #   tensorboard
-    #   tensorboardx
-    #   tifffile
-    #   torchvision
-    #   transformers
-    #   transforms3d
-nvidia-cublas-cu12==12.8.4.1
-    # via
-    #   nvidia-cudnn-cu12
-    #   nvidia-cusolver-cu12
-    #   torch
-nvidia-cuda-cupti-cu12==12.8.90
-    # via torch
-nvidia-cuda-nvrtc-cu12==12.8.93
-    # via torch
-nvidia-cuda-runtime-cu12==12.8.90
-    # via torch
-nvidia-cudnn-cu12==9.10.2.21
-    # via torch
-nvidia-cufft-cu12==11.3.3.83
-    # via torch
-nvidia-cufile-cu12==1.13.1.3
-    # via torch
-nvidia-curand-cu12==10.3.9.90
-    # via torch
-nvidia-cusolver-cu12==11.7.3.90
-    # via torch
-nvidia-cusparse-cu12==12.5.8.93
-    # via
-    #   nvidia-cusolver-cu12
-    #   torch
-nvidia-cusparselt-cu12==0.7.1
-    # via torch
-nvidia-nccl-cu12==2.27.5
-    # via torch
-nvidia-nvjitlink-cu12==12.8.93
-    # via
-    #   nvidia-cufft-cu12
-    #   nvidia-cusolver-cu12
-    #   nvidia-cusparse-cu12
-    #   torch
-nvidia-nvshmem-cu12==3.4.5
-    # via torch
-nvidia-nvtx-cu12==12.8.90
-    # via torch
-omegaconf==2.3.0
-    # via hydra-core
-opencv-python==4.13.0.92
-    # via
-    #   gym-pusht
-    #   hf-libero
-    #   reachy2-sdk
-    #   robosuite
-opencv-python-headless==4.12.0.88
-    # via lerobot
-orderly-set==5.5.0
-    # via deepdiff
-packaging==25.0
-    # via
-    #   accelerate
-    #   datasets
-    #   huggingface-hub
-    #   hydra-core
-    #   jupytext
-    #   lazy-loader
-    #   lerobot
-    #   matplotlib
-    #   peft
-    #   pytest
-    #   qwen-vl-utils
-    #   reachy2-sdk
-    #   scikit-image
-    #   tensorboard
-    #   tensorboardx
-    #   transformers
-    #   wandb
-pandas==2.3.3
-    # via
-    #   datasets
-    #   lerobot
-parso==0.8.6
-    # via jedi
-pathspec==1.0.4
-    # via mypy
-peft==0.18.1
-    # via lerobot
-pexpect==4.9.0
-    # via ipython
-pillow==12.1.1
-    # via
-    #   diffusers
-    #   imageio
-    #   matplotlib
-    #   meshcat
-    #   qwen-vl-utils
-    #   rerun-sdk
-    #   robosuite
-    #   scikit-image
-    #   tensorboard
-    #   torchvision
-pin==3.4.0
-    # via placo
-placo==0.9.16
-    # via lerobot
-platformdirs==4.9.4
-    # via
-    #   jupyter-core
-    #   python-discovery
-    #   virtualenv
-    #   wandb
-pluggy==1.6.0
-    # via
-    #   pytest
-    #   pytest-cov
-pre-commit==4.5.1
-    # via lerobot
-prompt-toolkit==3.0.52
-    # via ipython
-propcache==0.4.1
-    # via
-    #   aiohttp
-    #   yarl
-protobuf==6.31.1
-    # via
-    #   dm-control
-    #   grpcio-tools
-    #   lerobot
-    #   reachy2-sdk
-    #   reachy2-sdk-api
-    #   tensorboard
-    #   tensorboardx
-    #   wandb
-psutil==7.2.2
-    # via
-    #   accelerate
-    #   imageio
-    #   peft
-    #   robomimic
-ptyprocess==0.7.0
-    # via pexpect
-pure-eval==0.2.3
-    # via stack-data
-pyarrow==23.0.1
-    # via
-    #   datasets
-    #   rerun-sdk
-pycparser==3.0
-    # via cffi
-pydantic==2.12.5
-    # via
-    #   fastapi
-    #   wandb
-pydantic-core==2.41.5
-    # via pydantic
-pygame==2.6.1
-    # via
-    #   gym-hil
-    #   gym-pusht
-    #   lerobot
-pygments==2.19.2
-    # via
-    #   ipython
-    #   ipython-pygments-lexers
-    #   pytest
-    #   rich
-pymunk==6.11.1
-    # via
-    #   gym-pusht
-    #   lerobot
-pyngrok==7.5.1
-    # via meshcat
-pynput==1.8.1
-    # via
-    #   gym-hil
-    #   lerobot
-pyopengl==3.1.10
-    # via
-    #   dm-control
-    #   mujoco
-pyparsing==3.3.2
-    # via
-    #   dm-control
-    #   matplotlib
-pyquaternion==0.9.9
-    # via reachy2-sdk
-pyrealsense2==2.56.5.9235
-    # via lerobot
-pyserial==3.5
-    # via
-    #   dynamixel-sdk
-    #   feetech-servo-sdk
-    #   lerobot
-pytest==8.4.2
-    # via
-    #   bddl
-    #   lerobot
-    #   pytest-cov
-    #   pytest-timeout
-    #   teleop
-pytest-cov==7.0.0
-    # via lerobot
-pytest-timeout==2.4.0
-    # via lerobot
-python-dateutil==2.9.0.post0
-    # via
-    #   faker
-    #   matplotlib
-    #   pandas
-python-discovery==1.1.1
-    # via virtualenv
-python-dotenv==1.2.2
-    # via uvicorn
-python-xlib==0.33
-    # via pynput
-pytz==2026.1.post1
-    # via pandas
-pyyaml==6.0.3
-    # via
-    #   accelerate
-    #   datasets
-    #   draccus
-    #   hebi-py
-    #   huggingface-hub
-    #   jupytext
-    #   omegaconf
-    #   peft
-    #   pre-commit
-    #   pyngrok
-    #   pyyaml-include
-    #   transformers
-    #   uvicorn
-    #   wandb
-pyyaml-include==1.4.1
-    # via draccus
-pyzmq==27.1.0
-    # via
-    #   lerobot
-    #   meshcat
-qwen-vl-utils==0.0.14
-    # via lerobot
-reachy2-sdk==1.0.15
-    # via lerobot
-reachy2-sdk-api==1.0.21
-    # via reachy2-sdk
-referencing==0.37.0
-    # via
-    #   jsonschema
-    #   jsonschema-specifications
-regex==2026.2.28
-    # via
-    #   diffusers
-    #   transformers
-requests==2.32.5
-    # via
-    #   datasets
-    #   diffusers
-    #   dm-control
-    #   qwen-vl-utils
-    #   teleop
-    #   wandb
-rerun-sdk==0.26.2
-    # via lerobot
-rhoban-cmeel-jsoncpp==1.9.4.9
-    # via placo
-rich==14.3.3
-    # via typer
-robomimic==0.2.0
-    # via hf-libero
-robosuite==1.4.0
-    # via hf-libero
-rpds-py==0.30.0
-    # via
-    #   jsonschema
-    #   referencing
-safetensors==0.7.0
-    # via
-    #   accelerate
-    #   diffusers
-    #   lerobot
-    #   peft
-    #   transformers
-scikit-image==0.25.2
-    # via
-    #   gym-pusht
-    #   lerobot
-scipy==1.17.1
-    # via
-    #   dm-control
-    #   lerobot
-    #   metaworld
-    #   robosuite
-    #   scikit-image
-    #   torchdiffeq
-sentry-sdk==2.54.0
-    # via wandb
-shapely==2.1.2
-    # via gym-pusht
-shellingham==1.5.4
-    # via typer
-six==1.17.0
-    # via
-    #   pynput
-    #   python-dateutil
-    #   python-xlib
-smmap==5.0.3
-    # via gitdb
-stack-data==0.6.3
-    # via ipython
-starlette==0.52.1
-    # via fastapi
-sympy==1.14.0
-    # via torch
-teleop==0.1.4
-    # via lerobot
-tensorboard==2.20.0
-    # via robomimic
-tensorboard-data-server==0.7.2
-    # via tensorboard
-tensorboardx==2.6.4
-    # via robomimic
-termcolor==3.3.0
-    # via
-    #   lerobot
-    #   robomimic
-thop==0.1.1.post2209072238
-    # via hf-libero
-tifffile==2026.3.3
-    # via scikit-image
-tokenizers==0.22.2
-    # via transformers
-toml==0.10.2
-    # via draccus
-torch==2.10.0
-    # via
-    #   accelerate
-    #   lerobot
-    #   peft
-    #   robomimic
-    #   thop
-    #   torchdiffeq
-    #   torchvision
-torchcodec==0.10.0
-    # via lerobot
-torchdiffeq==0.2.5
-    # via lerobot
-torchvision==0.25.0
-    # via
-    #   lerobot
-    #   robomimic
-tornado==6.5.4
-    # via meshcat
-tqdm==4.67.3
-    # via
-    #   datasets
-    #   dm-control
-    #   huggingface-hub
-    #   peft
-    #   robomimic
-    #   transformers
-traitlets==5.14.3
-    # via
-    #   ipython
-    #   jupyter-core
-    #   matplotlib-inline
-    #   nbformat
-transformers==5.3.0
-    # via
-    #   hf-libero
-    #   lerobot
-    #   peft
-transforms3d==0.4.2
-    # via teleop
-triton==3.6.0
-    # via torch
-typer==0.24.1
-    # via
-    #   huggingface-hub
-    #   transformers
-typing-extensions==4.15.0
-    # via
-    #   aiosignal
-    #   anyio
-    #   etils
-    #   faker
-    #   fastapi
-    #   gymnasium
-    #   huggingface-hub
-    #   mypy
-    #   pydantic
-    #   pydantic-core
-    #   referencing
-    #   rerun-sdk
-    #   starlette
-    #   torch
-    #   typing-inspect
-    #   typing-inspection
-    #   wandb
-typing-inspect==0.9.0
-    # via draccus
-typing-inspection==0.4.2
-    # via
-    #   fastapi
-    #   pydantic
-tzdata==2025.3
-    # via pandas
-u-msgpack-python==2.8.0
-    # via meshcat
-urllib3==2.6.3
-    # via
-    #   requests
-    #   sentry-sdk
-uvicorn[standard]==0.41.0
-    # via teleop
-uvloop==0.22.1
-    # via uvicorn
-virtualenv==21.1.0
-    # via pre-commit
-wandb==0.24.2
-    # via
-    #   hf-libero
-    #   lerobot
-watchfiles==1.1.1
-    # via uvicorn
-wcwidth==0.6.0
-    # via prompt-toolkit
-websocket-client==1.9.0
-    # via teleop
-websockets==16.0
-    # via uvicorn
-werkzeug==3.1.6
-    # via tensorboard
-wrapt==2.1.2
-    # via dm-tree
-xxhash==3.6.0
-    # via datasets
-yarl==1.23.0
-    # via aiohttp
-zipp==3.23.0
-    # via
-    #   etils
-    #   importlib-metadata
-
-# The following packages are considered to be unsafe in a requirements file:
-# setuptools

requirements.in

@@ -1,9 +0,0 @@
-# requirements.in
-
-# requirements-macos.txt was generated on macOS and is platform-specific (macOS 26.3.1 25D2128 arm64).
-# Darwin MacBook-Pro.local 25.3.0 Darwin Kernel Version 25.3.0: Wed Jan 28 20:54:55 PST 2026; root:xnu-12377.91.3~2/RELEASE_ARM64_T8132 arm64
-
-# requirements-ubuntu.txt was generated on Linux and is platform-specific (Ubuntu 24.04.4 LTS x86_64).
-# Linux lerobot-linux 6.17.0-14-generic #14~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Thu Jan 15 15:52:10 UTC 2 x86_64 x86_64 x86_64 GNU/Linux
-
--e .[all]

src/lerobot/annotations/__init__.py

@@ -0,0 +1,15 @@
+#!/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.

src/lerobot/annotations/steerable_pipeline/__init__.py

@@ -0,0 +1,36 @@
+#!/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.
+"""Steerable annotation pipeline producing ``language_persistent`` and
+``language_events`` columns for LeRobot datasets.
+
+The pipeline is decomposed into three independently runnable modules whose
+outputs are staged per-episode before a final parquet rewrite:
+
+- :mod:`.modules.plan_subtasks_memory` (the ``plan`` module) — persistent styles
+- :mod:`.modules.interjections_and_speech` (the ``interjections`` module) — event styles + speech
+- :mod:`.modules.general_vqa` (the ``vqa`` module) — event-style VQA pairs
+"""
+
+from .config import AnnotationPipelineConfig
+from .validator import StagingValidator, ValidationReport
+from .writer import LanguageColumnsWriter
+
+__all__ = [
+    "AnnotationPipelineConfig",
+    "LanguageColumnsWriter",
+    "StagingValidator",
+    "ValidationReport",
+]

src/lerobot/annotations/steerable_pipeline/config.py

@@ -0,0 +1,211 @@
+#!/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.
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+
+@dataclass
+class PlanConfig:
+    """``plan`` module: subtasks + plan + memory + task augmentation."""
+
+    enabled: bool = True
+
+    # ``task_aug`` rephrasings at t=0 (renderer rotates ${task} among them); 0 disables.
+    n_task_rephrasings: int = 10
+
+    # Derive the task from video instead of episode_task: off / if_short / always.
+    # Affects prompts only; ``meta/tasks.parquet`` is untouched.
+    derive_task_from_video: str = "if_short"
+    derive_task_min_words: int = 3
+
+    # --- Frame input: timestamped contact sheets (always on) ---------------
+    # The subtask describe/segment passes ALWAYS render the episode as
+    # macrodata/refiner-style contact sheets: sampled frames packed into JPEG
+    # grids with each frame's timestamp burned into its corner, so the VLM
+    # cites the exact source time of a boundary directly. This is far cheaper
+    # in vision tokens than one image per frame (≈2× faster subtask generation
+    # in practice), which is why the sampling is dense by default.
+    #
+    # ``frames_per_second`` is the sampling rate: 2.0 = one frame every 0.5s.
+    frames_per_second: float = 2.0
+    # Frame budget per VLM call (= columns × rows × sheets). When a whole
+    # episode sampled at ``frames_per_second`` exceeds this, the episode is
+    # AUTOMATICALLY split into consecutive windows of
+    # ``max_frames_per_prompt`` frames each (one describe→segment call per
+    # window, still at the full ``frames_per_second`` density), and the
+    # per-window spans are merged + stitched into one contiguous cover. So an
+    # episode of any length is always covered at the full sampling density.
+    max_frames_per_prompt: int = 60
+    contact_sheet_columns: int = 5
+    contact_sheet_frames_per_sheet: int = 20
+    contact_sheet_frame_width: int = 224
+    contact_sheet_quality: int = 84
+
+    min_subtask_seconds: float = 1.5
+    plan_max_steps: int = 8
+
+    # Narrate-only grounding pass before segmenting — best defense against subtasks
+    # invented from the task text (+1 VLM call/episode).
+    subtask_describe_first: bool = True
+
+    # Emit ``style="plan"`` rows at each boundary; False = subtasks + memory only.
+    emit_plan: bool = True
+
+    # Emit ``style="memory"`` rows at each boundary; False = subtasks (+ plan) only.
+    # Symmetric counterpart of ``emit_plan``.
+    emit_memory: bool = True
+
+    # (subtask spans are always stitched to a contiguous full-episode cover; not configurable.)
+
+    # Optional EgoMimic-style 5-axis task augmentation; replaces n_task_rephrasings.
+    task_aug_axes: TaskAugAxesConfig = field(default_factory=lambda: TaskAugAxesConfig())
+
+
+@dataclass
+class TaskAugAxesConfig:
+    """5-axis t=0 task augmentation (EgoMimic-style): synonym / omit_arm /
+    omit_orientation / omit_grasp_method / combined. Replaces n_task_rephrasings
+    when enabled; each variant becomes a ``task_aug`` row. Axes with nothing to
+    omit emit fewer entries. Defaults (3+3+2+2+2) match EgoMimic."""
+
+    enabled: bool = False
+
+    synonym_paraphrase: int = 3
+    omit_arm: int = 3
+    omit_orientation: int = 2
+    omit_grasp_method: int = 2
+    combined_omissions: int = 2
+
+
+@dataclass
+class InterjectionsConfig:
+    """``interjections`` module: interjections + paired speech."""
+
+    enabled: bool = True
+
+    # Each emits a paired (interjection, speech) row + a plan refresh at that ts.
+    max_interjections_per_episode: int = 3
+    interjection_min_t: float = 2.0
+
+    # Frame window centered on the timestamp so the VLM sees motion, not one frame.
+    interjection_window_seconds: float = 2.0
+    interjection_window_frames: int = 4
+
+
+@dataclass
+class VqaConfig:
+    """``vqa`` module: general VQA."""
+
+    enabled: bool = True
+    vqa_emission_hz: float = 1.0
+    K: int = 1
+    """Consecutive frames per emission tick. The VLM grounds on the FIRST frame,
+    so K>1 smears stale labels onto moved frames. Default 1 (no smear)."""
+    question_types: tuple[str, ...] = ("bbox", "keypoint", "count", "attribute", "spatial")
+
+    # True: ground VQA only on --vlm.camera_key (default: every camera).
+    restrict_to_default_camera: bool = False
+
+
+@dataclass
+class VlmConfig:
+    """Shared Qwen-VL client configuration."""
+
+    # Only ``openai`` (OpenAI-compatible vLLM server, auto-spawned when
+    # auto_serve=True); ``stub`` is for tests.
+    backend: str = "openai"
+    model_id: str = "Qwen/Qwen3.6-27B"
+
+    # OpenAI-compatible endpoint; ``EMPTY`` key works for local servers.
+    api_base: str = "http://localhost:8000/v1"
+    api_key: str = "EMPTY"
+
+    # Spawn a server if none answers api_base; False = fail fast on a remote.
+    auto_serve: bool = True
+    serve_port: int = 8000
+    # Override the auto-serve command; ``{port}`` substituted per replica.
+    serve_command: str | None = None
+
+    # Independent servers for round-robin routing (one per GPU). num_gpus=0 = one each.
+    parallel_servers: int = 1
+    num_gpus: int = 0
+    client_concurrency: int = 16
+    serve_ready_timeout_s: float = 600.0
+
+    max_new_tokens: int = 512
+    temperature: float = 0.2
+
+    # Auto-serve context length (None → 32768); other vLLM flags go in serve_command.
+    max_model_len: int | None = None
+
+    # Camera for keyframes; None → first ``observation.images.*`` key.
+    camera_key: str | None = None
+    # Forwarded as extra_body.chat_template_kwargs (e.g. {"enable_thinking": false}).
+    chat_template_kwargs: dict[str, Any] | None = None
+
+
+@dataclass
+class ExecutorConfig:
+    """Executor settings (intra-process episode concurrency; distribution via HF Jobs)."""
+
+    # Episodes processed concurrently per phase; main knob for saturating the servers.
+    episode_parallelism: int = 16
+
+
+@dataclass
+class AnnotationPipelineConfig:
+    """Top-level config for ``lerobot-annotate`` (rewrites data shards in place)."""
+
+    # Hub dataset: download source when ``root`` unset; push target when push_to_hub
+    # is on and ``new_repo_id`` unset.
+    repo_id: str | None = None
+
+    # Separate push target (matches the LeRobot edit tools). Unset → push in place.
+    new_repo_id: str | None = None
+
+    root: Path | None = None
+
+    # Defaults to ``<root>/.annotate_staging/``.
+    staging_dir: Path | None = None
+
+    seed: int = 1729
+
+    plan: PlanConfig = field(default_factory=PlanConfig)
+    interjections: InterjectionsConfig = field(default_factory=InterjectionsConfig)
+    vqa: VqaConfig = field(default_factory=VqaConfig)
+
+    vlm: VlmConfig = field(default_factory=VlmConfig)
+    executor: ExecutorConfig = field(default_factory=ExecutorConfig)
+
+    skip_validation: bool = False
+    only_episodes: tuple[int, ...] | None = None
+
+    # Keyframe decode backend forwarded to ``decode_video_frames``. None →
+    # library default (torchcodec when available, else PyAV). Or pin
+    # ``"torchcodec"`` / ``"pyav"`` explicitly.
+    video_backend: str | None = None
+
+    # Upload to the Hub (new_repo_id if set, else repo_id; one must be set).
+    push_to_hub: bool = False
+    push_private: bool = False
+    push_commit_message: str | None = None
+
+    def resolved_staging_dir(self, root: Path) -> Path:
+        return self.staging_dir if self.staging_dir is not None else root / ".annotate_staging"

src/lerobot/annotations/steerable_pipeline/executor.py

@@ -0,0 +1,253 @@
+#!/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.
+"""In-process executor that runs the annotation phases.
+
+The executor runs **six phases** in dependency order:
+
+    phase 1: ``plan`` module (plan + subtasks + memory)
+    phase 2: ``interjections`` module (interjections + speech)
+    phase 3: ``plan`` plan-update pass — re-runs plan emission at every
+             interjection timestamp produced by phase 2
+    phase 4: ``vqa`` module (VQA)
+    phase 5: validator
+    phase 6: writer
+
+Phase 3 is why the ``plan`` module must be re-entered after the
+``interjections`` module — to refresh ``plan`` rows at interjection
+timestamps.
+
+Distributed execution is provided by Hugging Face Jobs (see
+``examples/annotations/run_hf_job.py``); the runner inside the job
+invokes ``lerobot-annotate`` which uses this in-process executor.
+Episode-level concurrency is controlled by
+``ExecutorConfig.episode_parallelism``.
+"""
+
+from __future__ import annotations
+
+import logging
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+from .config import AnnotationPipelineConfig
+from .reader import EpisodeRecord, iter_episodes
+from .staging import EpisodeStaging
+from .validator import StagingValidator
+from .writer import LanguageColumnsWriter
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class PhaseResult:
+    """Summary of one pipeline phase across all episodes."""
+
+    name: str
+    episodes_processed: int
+    episodes_skipped: int
+
+
+@dataclass
+class PipelineRunSummary:
+    """Aggregated result returned by :meth:`Executor.run`."""
+
+    phases: list[PhaseResult]
+    written_paths: list[Path]
+    validation_report: Any  # ValidationReport, kept Any to avoid import cycle
+
+
+@dataclass
+class Executor:
+    """Run all six phases over a dataset root in-process.
+
+    Episode-level concurrency comes from ``ExecutorConfig.episode_parallelism``
+    (a thread pool); cluster-level concurrency comes from running this
+    executor inside a Hugging Face Job. Tests construct the executor
+    directly with stub modules.
+    """
+
+    config: AnnotationPipelineConfig
+    plan: Any  # PlanSubtasksMemoryModule
+    interjections: Any  # InterjectionsAndSpeechModule
+    vqa: Any  # GeneralVqaModule
+    writer: LanguageColumnsWriter
+    validator: StagingValidator
+
+    def run(self, root: Path) -> PipelineRunSummary:
+        records = list(iter_episodes(root, only_episodes=self.config.only_episodes))
+        n = len(records)
+        if n == 0:
+            raise ValueError(f"No episodes found under {root}/data/")
+
+        print(f"[annotate] {n} episodes total", flush=True)
+
+        staging_dir = self.config.resolved_staging_dir(root)
+        staging_dir.mkdir(parents=True, exist_ok=True)
+
+        phases: list[PhaseResult] = []
+
+        # Phase 1: ``plan`` module (plan + subtasks + memory)
+        phases.append(self._run_module_phase("plan", records, staging_dir, self.plan))
+        # Phase 2: ``interjections`` module (interjections + speech). It
+        # reads the ``plan`` module's subtask rows from the same staging
+        # tree to ground the interjection prompt in the correct local subtask.
+        phases.append(self._run_module_phase("interjections", records, staging_dir, self.interjections))
+        # Phase 3: ``plan`` plan-update pass at interjection timestamps.
+        phases.append(self._run_plan_update_phase(records, staging_dir))
+        # Phase 4: ``vqa`` module (VQA)
+        phases.append(self._run_module_phase("vqa", records, staging_dir, self.vqa))
+
+        print("[annotate] running validator...", flush=True)
+        report = self.validator.validate(records, staging_dir)
+        if not report.ok and not self.config.skip_validation:
+            raise RuntimeError(f"Staging validation failed: {report.summary()}")
+        print(f"[annotate] validator: {report.summary()}", flush=True)
+
+        print(f"[annotate] writing parquet shards into {root}/data/...", flush=True)
+        written = self.writer.write_all(records, staging_dir, root)
+        print(f"[annotate] wrote {len(written)} shard(s); pipeline complete", flush=True)
+
+        # Keep meta/info.json aligned with the parquet schema we just wrote.
+        # Idempotent and additive: existing user metadata is preserved.
+        self._ensure_annotation_metadata_in_info(root)
+
+        return PipelineRunSummary(phases=phases, written_paths=written, validation_report=report)
+
+    @staticmethod
+    def _ensure_annotation_metadata_in_info(root: Path) -> None:
+        """Write language features and canonical tools to ``meta/info.json``.
+
+        ``LanguageColumnsWriter`` adds ``language_persistent`` and
+        ``language_events`` to parquet shards. The metadata must advertise
+        those columns too, otherwise non-streaming ``LeRobotDataset`` loads
+        cast against the old schema and fail on the extra parquet columns.
+        """
+        from lerobot.datasets.io_utils import load_info, write_info  # noqa: PLC0415
+        from lerobot.datasets.language import SAY_TOOL_SCHEMA, language_feature_info  # noqa: PLC0415
+
+        info_path = root / "meta" / "info.json"
+        if not info_path.exists():
+            return
+        try:
+            info = load_info(root)
+        except Exception as exc:  # noqa: BLE001
+            print(f"[annotate] could not read {info_path}: {exc}", flush=True)
+            return
+
+        changed = False
+
+        merged_features = {**info.features, **language_feature_info()}
+        if merged_features != info.features:
+            info.features = merged_features
+            changed = True
+
+        existing = info.tools or []
+        names = {(t.get("function") or {}).get("name") for t in existing if isinstance(t, dict)}
+        if SAY_TOOL_SCHEMA["function"]["name"] not in names:
+            info.tools = [*existing, SAY_TOOL_SCHEMA]
+            changed = True
+
+        if changed:
+            write_info(info, root)
+            print(
+                "[annotate] meta/info.json: "
+                f"language_features={list(language_feature_info())}, "
+                f"tools={[t['function']['name'] for t in (info.tools or [])]}",
+                flush=True,
+            )
+
+    def _run_module_phase(
+        self,
+        name: str,
+        records: list[EpisodeRecord],
+        staging_dir: Path,
+        module: Any,
+    ) -> PhaseResult:
+        if not module.enabled:
+            print(f"[annotate] phase={name} skipped (module disabled)", flush=True)
+            return PhaseResult(name=name, episodes_processed=0, episodes_skipped=len(records))
+        n = len(records)
+        parallelism = max(1, min(self.config.executor.episode_parallelism, n))
+        print(
+            f"[annotate] phase={name} starting on {n} episode(s) (parallelism={parallelism})",
+            flush=True,
+        )
+        t0 = time.time()
+
+        def _do(idx_record: tuple[int, EpisodeRecord]) -> tuple[int, int, float]:
+            i, record = idx_record
+            ep_start = time.time()
+            staging = EpisodeStaging(staging_dir, record.episode_index)
+            module.run_episode(record, staging)
+            return i, record.episode_index, time.time() - ep_start
+
+        processed = 0
+        if parallelism == 1:
+            for i, record in enumerate(records, 1):
+                _, ep_idx, elapsed = _do((i, record))
+                processed += 1
+                print(
+                    f"[annotate]   {name} episode {i}/{n} (idx={ep_idx}) done in {elapsed:.1f}s",
+                    flush=True,
+                )
+        else:
+            with ThreadPoolExecutor(max_workers=parallelism) as pool:
+                futures = [pool.submit(_do, (i, r)) for i, r in enumerate(records, 1)]
+                for fut in as_completed(futures):
+                    i, ep_idx, elapsed = fut.result()
+                    processed += 1
+                    print(
+                        f"[annotate]   {name} episode {processed}/{n} "
+                        f"(idx={ep_idx}, submit_order={i}) done in {elapsed:.1f}s",
+                        flush=True,
+                    )
+        total = time.time() - t0
+        print(f"[annotate] phase={name} complete: {processed}/{n} in {total:.1f}s", flush=True)
+        return PhaseResult(name=name, episodes_processed=processed, episodes_skipped=0)
+
+    def _run_plan_update_phase(  # noqa: PLR0915
+        self, records: list[EpisodeRecord], staging_dir: Path
+    ) -> PhaseResult:
+        """Re-emit ``plan`` rows at each timestamp the ``interjections`` module produced.
+
+        The ``plan`` module owns the prompt; the ``interjections`` module
+        produced the timestamps. This phase therefore calls back into the
+        ``plan`` module with the interjection timestamps so its existing
+        prompt path is reused.
+        """
+        if not self.plan.enabled or not self.interjections.enabled:
+            return PhaseResult(name="plan_update", episodes_processed=0, episodes_skipped=len(records))
+        processed = 0
+        for record in records:
+            staging = EpisodeStaging(staging_dir, record.episode_index)
+            interjection_rows = [
+                row for row in staging.read("interjections") if row.get("style") == "interjection"
+            ]
+            interjection_times = [float(row["timestamp"]) for row in interjection_rows]
+            interjection_texts = [str(row.get("content") or "") for row in interjection_rows]
+            if interjection_times:
+                self.plan.run_plan_updates(record, staging, interjection_times, interjection_texts)
+                processed += 1
+        # Episodes without any interjections are skipped (no plan refresh
+        # needed); count them so the summary's processed+skipped == total.
+        return PhaseResult(
+            name="plan_update",
+            episodes_processed=processed,
+            episodes_skipped=len(records) - processed,
+        )

src/lerobot/annotations/steerable_pipeline/frames.py

@@ -0,0 +1,483 @@
+#!/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.
+"""Keyframe extraction for the annotation pipeline.
+
+Modules attach decoded camera frames to their VLM prompts so the model can
+ground subtask decomposition, interjection scenarios, and VQA in actual
+visual content. The pipeline shares one provider across modules and one
+episode at a time, with a small per-episode cache so multiple modules
+querying the same timestamp pay decode cost once.
+"""
+
+from __future__ import annotations
+
+import io
+import logging
+import math
+import threading
+from collections.abc import Sequence
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Protocol
+
+import PIL.Image
+import torch
+
+from lerobot.configs import RGBEncoderConfig
+from lerobot.datasets.video_utils import decode_video_frames, reencode_video
+
+from .reader import EpisodeRecord, snap_to_frame
+
+logger = logging.getLogger(__name__)
+
+
+class FrameProvider(Protocol):
+    """Decodes camera frames at episode-relative timestamps."""
+
+    @property
+    def camera_keys(self) -> list[str]:
+        """All ``observation.images.*`` feature keys this provider can decode."""
+
+    def frames_at(
+        self,
+        record: EpisodeRecord,
+        timestamps: list[float],
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        """Return one decoded frame per timestamp from ``camera_key`` (or default).
+
+        Frames are ``torch.Tensor`` (``C, H, W`` uint8) — the shape
+        :func:`lerobot.datasets.video_utils.decode_video_frames` returns.
+        :func:`to_image_blocks` converts them to PIL only at the VLM-message
+        boundary.
+
+        Empty list if the camera is unavailable. ``camera_key=None`` falls back
+        to the provider's default camera so existing single-camera callers
+        (the ``plan`` and ``interjections`` modules) keep working unchanged.
+        """
+
+    def video_for_episode(
+        self,
+        record: EpisodeRecord,
+        max_frames: int,
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        """Return up to ``max_frames`` decoded frames covering the whole episode.
+
+        Sampling is uniform across the episode duration. Frames are
+        ``torch.Tensor`` (``C, H, W`` uint8); :func:`to_video_block` wraps
+        them into one ``{"type":"video", "video":<list>}`` block for a
+        Qwen-VL-compatible model that pools temporally itself. Empty list if
+        no camera available.
+        """
+
+
+@dataclass
+class _NullProvider:
+    """No-op provider used when the dataset has no video keys or in tests."""
+
+    @property
+    def camera_keys(self) -> list[str]:
+        return []
+
+    def frames_at(
+        self,
+        record: EpisodeRecord,
+        timestamps: list[float],
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        return []
+
+    def video_for_episode(
+        self,
+        record: EpisodeRecord,
+        max_frames: int,
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        return []
+
+
+def null_provider() -> FrameProvider:
+    return _NullProvider()
+
+
+@dataclass
+class VideoFrameProvider:
+    """Decodes frames from the dataset's ``observation.images.*`` streams.
+
+    By default the *first* camera key is used for the ``plan`` module
+    (subtask decomposition) and the ``interjections`` module (interjection
+    scenarios) — those prompts care about *what is happening*, not which
+    angle. The ``vqa`` module instead iterates over every camera in
+    :attr:`camera_keys` so each frame's
+    grounded answer (bbox/keypoint/...) is tagged with the camera it was
+    grounded against.
+
+    ``camera_key`` overrides the default-camera choice but does not restrict
+    :attr:`camera_keys`. Pass ``camera_key`` explicitly to ``frames_at`` /
+    ``video_for_episode`` to read a non-default stream.
+
+    Caches up to ``cache_size`` decoded frames per process to keep
+    co-timestamped ``interjections`` + ``plan`` plan-update calls cheap.
+    """
+
+    root: Path
+    camera_key: str | None = None
+    tolerance_s: float = 1e-2
+    cache_size: int = 256
+    # Keyframe decode backend forwarded to
+    # :func:`lerobot.datasets.video_utils.decode_video_frames`. ``None``
+    # uses the library default (torchcodec when available, else PyAV).
+    video_backend: str | None = None
+    _meta: Any = field(default=None, init=False, repr=False)
+    _cache: dict = field(default_factory=dict, init=False, repr=False)
+    _camera_keys: list[str] = field(default_factory=list, init=False, repr=False)
+    # Pipeline runs the three module phases under a ThreadPoolExecutor (see
+    # ``ExecutorConfig.episode_parallelism``); guard the dict cache and the
+    # one-shot warn flag against concurrent updates from worker threads.
+    _lock: threading.Lock = field(default_factory=threading.Lock, init=False, repr=False)
+    # Serializes decode_video_frames calls: torchcodec hands out one
+    # ``VideoDecoder`` per file from a process-wide cache, and the decoder
+    # is not safe to drive from multiple threads at once.
+    _decode_lock: threading.Lock = field(default_factory=threading.Lock, init=False, repr=False)
+    _warned_decode_fail: bool = field(default=False, init=False, repr=False)
+
+    def __post_init__(self) -> None:
+        from lerobot.datasets.dataset_metadata import LeRobotDatasetMetadata  # noqa: PLC0415
+
+        self._meta = LeRobotDatasetMetadata(repo_id="local", root=self.root)
+        # Only ``video_keys`` are decodable here: the clip/decode paths read
+        # ``videos/<key>/from_timestamp`` from episode metadata, which exists
+        # only for video-stored cameras. Image-stored cameras (also in
+        # ``camera_keys``) would KeyError, so restrict the list — and the
+        # default — to 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.
+        if not keys and self.camera_key:
+            keys = [self.camera_key]
+        self._camera_keys = keys
+        if self.camera_key is None:
+            self.camera_key = keys[0] if keys else None
+
+    @property
+    def camera_keys(self) -> list[str]:
+        """All ``observation.images.*`` keys available on this dataset."""
+        return list(self._camera_keys)
+
+    def frames_at(
+        self,
+        record: EpisodeRecord,
+        timestamps: list[float],
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        target = camera_key if camera_key is not None else self.camera_key
+        if not timestamps or target is None:
+            return []
+        # Snap each request to the nearest real frame timestamp: callers
+        # sample uniform grids whose points land mid-frame, and
+        # ``decode_video_frames`` rejects queries farther than
+        # ``tolerance_s`` from a decodable frame. Snapping also dedupes
+        # repeat queries through the cache.
+        if record.frame_timestamps:
+            timestamps = [snap_to_frame(float(ts), record.frame_timestamps) for ts in timestamps]
+
+        out: list[Any] = []
+        misses: list[float] = []
+        miss_indices: list[int] = []
+        with self._lock:
+            for i, ts in enumerate(timestamps):
+                key = (record.episode_index, target, round(float(ts), 6))
+                cached = self._cache.get(key)
+                if cached is not None:
+                    out.append(cached)
+                else:
+                    out.append(None)
+                    misses.append(float(ts))
+                    miss_indices.append(i)
+
+        if misses:
+            decoded = self._decode(record.episode_index, misses, target)
+            # ``_decode`` returns exactly one frame per requested timestamp,
+            # or an empty list if decoding failed wholesale. A partial list
+            # would mean a frame/timestamp misalignment, so only pair them up
+            # when the counts match (``strict=True`` then guards regressions).
+            if len(decoded) == len(miss_indices):
+                with self._lock:
+                    for i, frame in zip(miss_indices, decoded, strict=True):
+                        out[i] = frame
+                        key = (record.episode_index, target, round(float(timestamps[i]), 6))
+                        if len(self._cache) >= self.cache_size:
+                            self._cache.pop(next(iter(self._cache)))
+                        self._cache[key] = frame
+        # filter out any None left over from decode failures
+        return [frame for frame in out if frame is not None]
+
+    def video_for_episode(
+        self,
+        record: EpisodeRecord,
+        max_frames: int,
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        """Return up to ``max_frames`` frames uniformly sampled across the episode.
+
+        The whole episode duration is covered; the model picks subtask
+        boundaries from the temporal pooling it does internally. Frames are
+        ``torch.Tensor`` (see :meth:`frames_at`).
+        """
+        target = camera_key if camera_key is not None else self.camera_key
+        if max_frames <= 0 or target is None or not record.frame_timestamps:
+            return []
+        n_frames = min(max_frames, len(record.frame_timestamps))
+        if n_frames == len(record.frame_timestamps):
+            timestamps = list(record.frame_timestamps)
+        else:
+            t0 = record.frame_timestamps[0]
+            t_last = record.frame_timestamps[-1]
+            if t_last <= t0:
+                timestamps = [float(t0)] * n_frames
+            else:
+                step = (t_last - t0) / (n_frames - 1) if n_frames > 1 else 0.0
+                timestamps = [float(t0 + i * step) for i in range(n_frames)]
+        return self.frames_at(record, timestamps, camera_key=target)
+
+    def episode_clip_path(self, record: EpisodeRecord, cache_dir: Path) -> Path | None:
+        """Extract the episode's subclip to ``cache_dir/ep_{idx:06d}.mp4``.
+
+        Returns ``None`` if the dataset has no video tracks or extraction
+        failed. Skips re-extract when the cached clip already exists.
+        Re-encodes to H.264 via
+        :func:`lerobot.datasets.video_utils.reencode_video` so the resulting
+        mp4 is decodable by every downstream video processor — stream-copy
+        would inherit the source codec (often AV1 in modern LeRobot
+        datasets), which vllm's libav build cannot decode.
+        """
+        if self.camera_key is None:
+            return None
+        cache_dir.mkdir(parents=True, exist_ok=True)
+        out_path = cache_dir / f"ep_{record.episode_index:06d}.mp4"
+        if out_path.exists() and out_path.stat().st_size > 0:
+            return out_path
+        ep = self._meta.episodes[record.episode_index]
+        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 = RGBEncoderConfig(vcodec="h264", pix_fmt="yuv420p", g=None, crf=23, preset="ultrafast")
+        try:
+            reencode_video(
+                src,
+                out_path,
+                video_encoder=encoder,
+                overwrite=True,
+                start_time_s=from_timestamp,
+                end_time_s=to_timestamp,
+            )
+        except Exception:
+            logger.warning(
+                "clip extraction failed for episode %s (%s)", record.episode_index, src, exc_info=True
+            )
+            return None
+        return out_path if out_path.exists() and out_path.stat().st_size > 0 else None
+
+    def _decode(self, episode_index: int, timestamps: list[float], camera_key: str) -> list[Any]:
+        """Decode ``timestamps`` from the episode's video as ``(C, H, W)`` tensors.
+
+        Delegates to :func:`lerobot.datasets.video_utils.decode_video_frames`
+        (torchcodec when available, PyAV otherwise; ``video_backend`` pins
+        one explicitly). Returns one frame per requested timestamp, or ``[]``
+        if decoding failed — callers treat ``[]`` as "no frames available".
+        """
+        ep = self._meta.episodes[episode_index]
+        from_timestamp = ep[f"videos/{camera_key}/from_timestamp"]
+        shifted = [from_timestamp + ts for ts in timestamps]
+        video_path = self.root / self._meta.get_video_file_path(episode_index, camera_key)
+
+        try:
+            # The module phases decode under a ThreadPoolExecutor (see
+            # ``ExecutorConfig.episode_parallelism``) but torchcodec's cached
+            # per-file decoder is single-threaded, so serialize decodes on a
+            # dedicated lock. Frame extraction is a small fraction of episode
+            # wall time (VLM calls dominate), so the contention is cheap.
+            with self._decode_lock:
+                # Stacked ``(N, C, H, W)`` uint8 tensor; one row per timestamp.
+                decoded = decode_video_frames(
+                    video_path, shifted, self.tolerance_s, backend=self.video_backend, return_uint8=True
+                )
+            return list(decoded)
+        except Exception as exc:
+            # Log loudly the first time so a silent vqa-module no-op (every
+            # prompt skipped because frames_at returned []) is debuggable from
+            # the job log instead of post-hoc parquet inspection. Subsequent
+            # failures stay quiet.
+            with self._lock:
+                already_warned = self._warned_decode_fail
+                if not already_warned:
+                    self._warned_decode_fail = True
+            if not already_warned:
+                logger.warning(
+                    "VideoFrameProvider._decode failed for episode=%s camera=%s video_path=%s backend=%s: %s",
+                    episode_index,
+                    camera_key,
+                    video_path,
+                    self.video_backend,
+                    exc,
+                    exc_info=exc,
+                )
+            return []
+
+
+def make_frame_provider(
+    root: Path, camera_key: str | None = None, video_backend: str | None = None
+) -> FrameProvider:
+    """Build a :class:`VideoFrameProvider` if videos are present, else null."""
+    try:
+        provider = VideoFrameProvider(root=root, camera_key=camera_key, video_backend=video_backend)
+    except Exception:
+        return null_provider()
+    if provider.camera_key is None:
+        return null_provider()
+    return provider
+
+
+def _frame_to_pil(frame: Any) -> Any:
+    """Materialise a decoded frame as a ``PIL.Image`` for the VLM message.
+
+    Frames flow through the provider as ``torch.Tensor`` (``C, H, W`` uint8,
+    straight from :func:`decode_video_frames`); PIL is only created here, at
+    the VLM-message boundary, because the chat backends expect PIL images /
+    data URLs. Non-tensor inputs (e.g. test stubs) pass through untouched.
+    """
+    if not isinstance(frame, torch.Tensor):
+        return frame
+    array = frame.detach().cpu()
+    if array.ndim == 3 and array.shape[0] in (1, 3):
+        array = array.permute(1, 2, 0)  # (C, H, W) -> (H, W, C)
+    if array.shape[-1] == 1:
+        array = array.squeeze(-1)
+    return PIL.Image.fromarray(array.to(torch.uint8).numpy())
+
+
+def to_image_blocks(frames: list[Any]) -> list[dict[str, Any]]:
+    """Convert decoded frames to Qwen-VL-compatible image content blocks."""
+    return [{"type": "image", "image": _frame_to_pil(frame)} for frame in frames]
+
+
+def to_video_block(frames: list[Any]) -> list[dict[str, Any]]:
+    """Wrap a list of decoded frames as one Qwen-VL video block.
+
+    Returns ``[]`` when the list is empty, so the caller can splat the result
+    into a content array without a separate emptiness check.
+    """
+    if not frames:
+        return []
+    return [{"type": "video", "video": [_frame_to_pil(frame) for frame in frames]}]
+
+
+def to_video_url_block(url: str | None, fps: float = 2.0) -> list[dict[str, Any]]:
+    """Wrap a video file URL as one ``video_url`` block.
+
+    Used by the ``openai`` backend (transformers serve / vllm serve /
+    ktransformers serve), where the server handles frame sampling.
+    Returns ``[]`` when ``url`` is ``None`` so the caller can splat.
+    """
+    if not url:
+        return []
+    return [{"type": "video_url", "video_url": {"url": url}, "fps": fps}]
+
+
+def _draw_timestamp_badge(image: PIL.Image.Image, timestamp: float) -> PIL.Image.Image:
+    """Burn ``timestamp`` (seconds) into the top-left corner of ``image``.
+
+    A solid black badge with white text, so a VLM reading a contact sheet can
+    cite the exact source time of each tile (e.g. ``012.50s``) directly,
+    instead of the caller having to map tile position back to time. Mirrors
+    the macrodata/refiner contact-sheet convention.
+    """
+    from PIL import ImageDraw, ImageFont
+
+    result = image.copy()
+    draw = ImageDraw.Draw(result)
+    font = ImageFont.load_default()
+    label = f"{timestamp:06.2f}s"
+    left, top, right, bottom = draw.textbbox((0, 0), label, font=font)
+    text_w, text_h = right - left, bottom - top
+    pad = max(3, round(min(image.width, image.height) * 0.018))
+    draw.rectangle((0, 0, text_w + pad * 2, text_h + pad * 2), fill=(0, 0, 0))
+    draw.text((pad - left, pad - top), label, fill=(255, 255, 255), font=font)
+    return result
+
+
+def to_contact_sheet_blocks(
+    frames: Sequence[Any],
+    timestamps: Sequence[float],
+    *,
+    columns: int = 5,
+    frames_per_sheet: int = 20,
+    frame_width: int = 224,
+    quality: int = 84,
+) -> list[dict[str, Any]]:
+    """Pack decoded frames into timestamped JPEG contact-sheet image blocks.
+
+    Each frame is resized to ``frame_width`` wide, stamped with its
+    episode-relative timestamp, and tiled row-major into grids of
+    ``frames_per_sheet`` (``columns`` wide). One ``{"type":"image", ...}``
+    block is returned per grid; many frames collapse into a few images, so a
+    long episode's temporal coverage stays dense at a fraction of the vision
+    tokens N separate frames would cost. ``frames`` and ``timestamps`` must be
+    aligned and equal length. Returns ``[]`` for empty input.
+    """
+    from PIL import Image
+
+    if not frames:
+        return []
+    columns = max(1, columns)
+    frames_per_sheet = max(1, frames_per_sheet)
+    rows_per_sheet = math.ceil(frames_per_sheet / columns)
+
+    tiles: list[PIL.Image.Image] = []
+    for ts, frame in zip(timestamps, frames, strict=False):
+        img = _frame_to_pil(frame)
+        if not isinstance(img, PIL.Image.Image):
+            continue
+        img = img.convert("RGB")
+        if img.width != frame_width:
+            height = max(1, round(img.height * frame_width / img.width))
+            img = img.resize((frame_width, height), resample=Image.Resampling.BILINEAR)
+        tiles.append(_draw_timestamp_badge(img, float(ts)))
+    if not tiles:
+        return []
+
+    blocks: list[dict[str, Any]] = []
+    for start in range(0, len(tiles), frames_per_sheet):
+        chunk = tiles[start : start + frames_per_sheet]
+        cell_w = max(tile.width for tile in chunk)
+        cell_h = max(tile.height for tile in chunk)
+        sheet = Image.new("RGB", (cell_w * columns, cell_h * rows_per_sheet), color=(0, 0, 0))
+        for i, tile in enumerate(chunk):
+            x = (i % columns) * cell_w
+            y = (i // columns) * cell_h
+            sheet.paste(tile, (x, y))
+        # JPEG round-trip at ``quality`` to match the refiner convention and
+        # shrink the wire payload; vision-token count is set by resolution, so
+        # the real saving is the grid packing, not the codec.
+        buf = io.BytesIO()
+        sheet.save(buf, format="JPEG", quality=quality)
+        buf.seek(0)
+        blocks.append({"type": "image", "image": Image.open(buf).convert("RGB")})
+    return blocks

src/lerobot/annotations/steerable_pipeline/modules/__init__.py

@@ -0,0 +1,25 @@
+#!/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.
+
+from .general_vqa import GeneralVqaModule
+from .interjections_and_speech import InterjectionsAndSpeechModule
+from .plan_subtasks_memory import PlanSubtasksMemoryModule
+
+__all__ = [
+    "GeneralVqaModule",
+    "InterjectionsAndSpeechModule",
+    "PlanSubtasksMemoryModule",
+]

src/lerobot/annotations/steerable_pipeline/modules/general_vqa.py

@@ -0,0 +1,248 @@
+#!/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.
+"""``vqa`` module: general VQA at a timed cadence.
+
+Every ``1/hz`` seconds an emission tick fires; each tick anchors ``K``
+consecutive frames, and every anchored frame gets its own VQA pair. Each
+pair is grounded on that single anchor frame — there is no per-pair frame
+window. For datasets with multiple cameras, every anchored frame produces
+one ``(vqa, user)`` + ``(vqa, assistant)`` pair *per camera*: each pair is
+generated against that camera's frame and stamped with the matching
+``camera`` field on the emitted rows. The resolver disambiguates via
+``camera=...``; recipes that consume VQA do so through one sub-recipe
+per camera (see ``recipes/pi05_hirobot.yaml``).
+
+Within a single (frame, camera) we still emit at most one ``(vqa, user)``
+and one ``(vqa, assistant)`` row, so the resolver contract stays scalar.
+
+Question types covered (per the plan's ``vqa`` table): bbox, keypoint,
+count, attribute, spatial. The assistant's ``content`` is a JSON string
+whose schema depends on the question type. Malformed JSON triggers one
+retry inside :meth:`VlmClient.generate_json`.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import random
+from collections.abc import Sequence
+from dataclasses import dataclass, field
+from typing import Any
+
+from ..config import VqaConfig
+from ..frames import FrameProvider, null_provider, to_image_blocks
+from ..prompts import load as load_prompt
+from ..reader import EpisodeRecord
+from ..staging import EpisodeStaging
+from ..validator import classify_vqa_answer
+from ..vlm_client import VlmClient
+
+
+def _emission_anchor_indices(frame_timestamps: Sequence[float], hz: float, k: int) -> list[int]:
+    """Return the relative frame indices to anchor VQA emissions to.
+
+    For each emission tick (every ``1/hz`` seconds), we anchor ``k``
+    consecutive frames starting at the tick. Ticks fall on the nearest
+    available source frame timestamp.
+    """
+    if hz <= 0 or k <= 0 or not frame_timestamps:
+        return []
+    t0 = frame_timestamps[0]
+    t_last = frame_timestamps[-1]
+    period = 1.0 / hz
+    indices: list[int] = []
+    t = t0
+    while t <= t_last + 1e-9:
+        # find the index of the nearest frame to t
+        nearest_i = min(range(len(frame_timestamps)), key=lambda i: abs(frame_timestamps[i] - t))
+        for offset in range(k):
+            j = nearest_i + offset
+            if j >= len(frame_timestamps):
+                break
+            if not indices or indices[-1] != j:
+                indices.append(j)
+        t += period
+    # dedupe while preserving order
+    seen: set[int] = set()
+    deduped: list[int] = []
+    for i in indices:
+        if i in seen:
+            continue
+        seen.add(i)
+        deduped.append(i)
+    return deduped
+
+
+@dataclass
+class GeneralVqaModule:
+    """Emit grounded VQA pairs at a timed cadence."""
+
+    vlm: VlmClient
+    config: VqaConfig
+    seed: int = 1729
+    frame_provider: FrameProvider = field(default_factory=null_provider)
+    _warned_no_camera: bool = field(default=False, init=False, repr=False)
+
+    @property
+    def enabled(self) -> bool:
+        return self.config.enabled
+
+    def run_episode(self, record: EpisodeRecord, staging: EpisodeStaging) -> None:
+        if not record.frame_timestamps:
+            staging.write("vqa", [])
+            return
+        rng = random.Random(f"{self.seed}:{record.episode_index}:vqa")
+        anchor_idx = _emission_anchor_indices(
+            record.frame_timestamps, self.config.vqa_emission_hz, self.config.K
+        )
+        cameras = self._target_cameras()
+        if not cameras:
+            # No camera available — emit nothing rather than producing
+            # untagged rows that would fail validation. Surface a loud one-
+            # time warning so this is never silently a no-op.
+            if not self._warned_no_camera:
+                logging.getLogger(__name__).warning(
+                    "vqa module found no cameras on the frame provider — "
+                    "every episode will emit zero VQA rows. Check that the "
+                    "dataset declares observation.images.* features in "
+                    "meta/info.json; passing --vlm.camera_key=<key> at the "
+                    "CLI now also seeds the cameras list as a fallback."
+                )
+                self._warned_no_camera = True
+            staging.write("vqa", [])
+            return
+
+        # Build all messages first (one per (frame, camera)), then issue them
+        # as a single batched generate_json call so the client can fan them
+        # out concurrently.
+        per_call: list[tuple[float, str, str, list[dict[str, Any]]]] = []
+        for idx in anchor_idx:
+            ts = float(record.frame_timestamps[idx])
+            qtype = rng.choice(self.config.question_types)
+            for camera in cameras:
+                messages = self._build_messages(record, qtype, ts, camera)
+                # Skip cameras that decoded to zero frames at this ts: no point
+                # asking the VLM to ground a bbox without an image.
+                if not _has_image_block(messages):
+                    continue
+                per_call.append((ts, camera, qtype, messages))
+
+        if not per_call:
+            staging.write("vqa", [])
+            return
+
+        results = self.vlm.generate_json([m for _, _, _, m in per_call])
+
+        rows: list[dict[str, Any]] = []
+        for (ts, camera, _qtype, _messages), result in zip(per_call, results, strict=True):
+            qa = self._postprocess(result)
+            if qa is None:
+                continue
+            question, answer = qa
+            rows.append(
+                {
+                    "role": "user",
+                    "content": question,
+                    "style": "vqa",
+                    "timestamp": ts,
+                    "camera": camera,
+                    "tool_calls": None,
+                }
+            )
+            rows.append(
+                {
+                    "role": "assistant",
+                    "content": json.dumps(answer, sort_keys=True),
+                    "style": "vqa",
+                    "timestamp": ts,
+                    "camera": camera,
+                    "tool_calls": None,
+                }
+            )
+        staging.write("vqa", rows)
+
+    def _target_cameras(self) -> list[str]:
+        """Return the cameras the ``vqa`` module should iterate per anchored frame.
+
+        Defaults to every camera the provider exposes. Datasets with no
+        cameras (or test/null providers) yield an empty list, which makes
+        ``run_episode`` a no-op.
+
+        When ``config.restrict_to_default_camera`` is set, VQA grounds on
+        only the provider's default camera (the single ``--vlm.camera_key``
+        stream), matching the plan / interjection modules so the whole
+        pipeline focuses on one view.
+        """
+        all_cameras = list(getattr(self.frame_provider, "camera_keys", []) or [])
+        if getattr(self.config, "restrict_to_default_camera", False):
+            default = getattr(self.frame_provider, "camera_key", None)
+            if default and default in all_cameras:
+                return [default]
+            # ``restrict_to_default_camera`` is set but the configured default
+            # isn't one the provider exposes. Returning it anyway would make
+            # ``_decode`` raise a KeyError deep in frame extraction, so warn and
+            # fall through to every available camera instead.
+            if default:
+                logging.getLogger(__name__).warning(
+                    "restrict_to_default_camera is set but camera_key=%r is not in the "
+                    "provider's cameras %s; grounding VQA on all available cameras instead.",
+                    default,
+                    all_cameras,
+                )
+        return all_cameras
+
+    def _build_messages(
+        self,
+        record: EpisodeRecord,
+        question_type: str,
+        frame_timestamp: float,
+        camera_key: str,
+    ) -> list[dict[str, Any]]:
+        prompt = load_prompt("vqa").format(
+            episode_task=record.episode_task,
+            question_type=question_type,
+        )
+        images = self.frame_provider.frames_at(record, [frame_timestamp], camera_key=camera_key)
+        content = [*to_image_blocks(images), {"type": "text", "text": prompt}]
+        return [{"role": "user", "content": content}]
+
+    def _postprocess(self, result: Any) -> tuple[str, dict[str, Any]] | None:
+        if not isinstance(result, dict):
+            return None
+        question = result.get("question")
+        answer = result.get("answer")
+        if not isinstance(question, str) or not question.strip():
+            return None
+        if not isinstance(answer, dict):
+            return None
+        # The validator will enforce shape; here we just sanity-check that the
+        # answer matches *some* known shape so we can drop garbage early.
+        if classify_vqa_answer(answer) is None:
+            return None
+        return question.strip(), answer
+
+
+def _has_image_block(messages: list[dict[str, Any]]) -> bool:
+    """Return True if any user content block is a populated image block."""
+    for msg in messages:
+        content = msg.get("content")
+        if not isinstance(content, list):
+            continue
+        for block in content:
+            if isinstance(block, dict) and block.get("type") == "image":
+                return True
+    return False

src/lerobot/annotations/steerable_pipeline/modules/interjections_and_speech.py

@@ -0,0 +1,211 @@
+#!/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.
+"""``interjections`` module: interjections + paired speech (EVENT styles + speech atoms).
+
+Two sub-passes:
+
+1. At ``t=0``, emit ONLY a speech tool-call atom (acknowledgement of the
+   canonical task). No interjection row — the canonical task is already the
+   user utterance from ``meta/tasks.parquet``.
+
+2. For mid-episode interruptions, emit a co-timestamped pair:
+       {role:user, style:interjection, content:<text>}
+       speech atom (role:assistant, style:None, tool_calls=[say(...)])
+   Both rows go in ``language_events`` at the same timestamp.
+
+The ``plan`` module's :meth:`run_plan_updates` reuses this module's
+interjection timestamps to refresh the ``plan`` row at the same instant.
+"""
+
+from __future__ import annotations
+
+import random
+from collections.abc import Sequence
+from dataclasses import dataclass, field
+from typing import Any
+
+from ..config import InterjectionsConfig
+from ..frames import FrameProvider, null_provider, to_image_blocks
+from ..prompts import load as load_prompt
+from ..reader import EpisodeRecord, reconstruct_subtask_spans, snap_to_frame
+from ..staging import EpisodeStaging
+from ..vlm_client import VlmClient
+from ..writer import speech_atom
+
+
+@dataclass
+class InterjectionsAndSpeechModule:
+    """Generate task-start speech and mid-episode interjection/speech pairs."""
+
+    vlm: VlmClient
+    config: InterjectionsConfig
+    seed: int = 1729
+    frame_provider: FrameProvider = field(default_factory=null_provider)
+
+    @property
+    def enabled(self) -> bool:
+        return self.config.enabled
+
+    def run_episode(self, record: EpisodeRecord, staging: EpisodeStaging) -> None:
+        rows: list[dict[str, Any]] = []
+        if record.frame_timestamps:
+            t0 = float(record.frame_timestamps[0])
+            initial = self._initial_speech(record)
+            if initial:
+                rows.append(speech_atom(t0, initial))
+        # Pull the ``plan`` module's subtask spans for this episode so the
+        # interjection prompt can ground itself in the actual current
+        # subtask at each chosen timestamp. The ``plan`` module ran first.
+        episode_end_t = float(record.frame_timestamps[-1]) if record.frame_timestamps else None
+        subtask_spans = reconstruct_subtask_spans(staging.read("plan"), episode_end_t=episode_end_t)
+        rows.extend(self._mid_episode_interjections(record, subtask_spans))
+        staging.write("interjections", rows)
+
+    @staticmethod
+    def _subtask_at(spans: Sequence[dict[str, Any]], t: float) -> str | None:
+        current: str | None = None
+        for span in spans:
+            if float(span["start"]) <= t:
+                current = span.get("text")
+            else:
+                break
+        return current
+
+    def _initial_speech(self, record: EpisodeRecord) -> str | None:
+        prompt = load_prompt("interjections_initial_speech").format(
+            episode_task=record.episode_task,
+        )
+        messages = [{"role": "user", "content": [{"type": "text", "text": prompt}]}]
+        result = self.vlm.generate_json([messages])[0]
+        if isinstance(result, dict) and isinstance(result.get("text"), str):
+            text = result["text"].strip()
+            if text:
+                return text
+        return None
+
+    def _mid_episode_interjections(
+        self,
+        record: EpisodeRecord,
+        subtask_spans: Sequence[dict[str, Any]],
+    ) -> list[dict[str, Any]]:
+        """Generate interjections aligned with the actual demo trajectory.
+
+        Teleop data is frozen — the robot already executed every step in
+        the video. A *counterfactual* interjection like "actually skip
+        the wipe" contradicts what then happens in the video, which is
+        what qwen36moe-10/11 surfaced as low-quality interjections.
+
+        Instead, anchor every interjection at a subtask boundary and
+        write it as a natural user request for the *upcoming* subtask.
+        The robot's visible next behavior IS the interjection's effect,
+        so the training signal stays consistent: interjection text →
+        plan refresh → action stream all line up.
+        """
+        if self.config.max_interjections_per_episode <= 0:
+            return []
+        if len(subtask_spans) < 2:
+            # Need at least one transition (subtask 0 → subtask 1).
+            return []
+        # Deterministic per-episode RNG so reruns are stable across SLURM jobs.
+        rng = random.Random(f"{self.seed}:{record.episode_index}:interjection")
+
+        # Boundaries: the start time of every subtask except the first
+        # (which is just t0 and is covered by the initial-task speech atom).
+        boundaries: list[tuple[float, str, str]] = []
+        for i in range(1, len(subtask_spans)):
+            ts = float(subtask_spans[i]["start"])
+            if ts < self.config.interjection_min_t:
+                continue
+            prev_text = (subtask_spans[i - 1].get("text") or "").strip()
+            next_text = (subtask_spans[i].get("text") or "").strip()
+            if not next_text:
+                continue
+            boundaries.append((ts, prev_text, next_text))
+        if not boundaries:
+            return []
+
+        n = min(self.config.max_interjections_per_episode, len(boundaries))
+        chosen = sorted(rng.sample(boundaries, n), key=lambda b: b[0])
+
+        out: list[dict[str, Any]] = []
+        for t, prev_subtask, next_subtask in chosen:
+            t_snap = snap_to_frame(t, record.frame_timestamps)
+            # Window straddles the boundary so the VLM sees the end of the
+            # previous subtask and the start of the next one — same
+            # conditioning the policy will see at training time.
+            window_ts = self._window_timestamps(t_snap, record.frame_timestamps)
+            prompt = load_prompt("interjections_interjection").format(
+                episode_task=record.episode_task,
+                prev_subtask=prev_subtask or "(starting from initial state)",
+                next_subtask=next_subtask,
+                timestamp=t_snap,
+                window_seconds=self.config.interjection_window_seconds,
+            )
+            images = self.frame_provider.frames_at(record, window_ts)
+            content = [*to_image_blocks(images), {"type": "text", "text": prompt}]
+            messages = [{"role": "user", "content": content}]
+            result = self.vlm.generate_json([messages])[0]
+            if not isinstance(result, dict):
+                continue
+            interjection_text = result.get("interjection")
+            speech_text = result.get("speech")
+            if not isinstance(interjection_text, str) or not interjection_text.strip():
+                continue
+            if not isinstance(speech_text, str) or not speech_text.strip():
+                continue
+            out.append(
+                {
+                    "role": "user",
+                    "content": interjection_text.strip(),
+                    "style": "interjection",
+                    "timestamp": t_snap,
+                    "tool_calls": None,
+                }
+            )
+            out.append(speech_atom(t_snap, speech_text.strip()))
+        return out
+
+    def _window_timestamps(self, t_anchor: float, frame_timestamps: Sequence[float]) -> list[float]:
+        """Return a small set of frame timestamps centered on ``t_anchor``.
+
+        The window straddles the subtask boundary the interjection sits
+        on: roughly half the frames cover the end of the previous
+        subtask, half cover the start of the next one. The VLM therefore
+        sees BOTH what just finished AND what's about to start, which is
+        the conditioning we need to write a natural "now please do X"
+        request that matches the visible upcoming behavior.
+        """
+        if not frame_timestamps:
+            return [t_anchor]
+        n = max(1, int(self.config.interjection_window_frames))
+        if n == 1:
+            return [t_anchor]
+        window = float(self.config.interjection_window_seconds)
+        step = window / max(1, n - 1)
+        # Center the window on the anchor so half lands before, half after.
+        start_offset = -window / 2.0
+        targets = [t_anchor + start_offset + step * i for i in range(n)]
+        first_ts = float(frame_timestamps[0])
+        last_ts = float(frame_timestamps[-1])
+        snapped: list[float] = []
+        seen: set[float] = set()
+        for tgt in targets:
+            clamped = min(last_ts, max(first_ts, tgt))
+            t = snap_to_frame(clamped, frame_timestamps)
+            if t not in seen:
+                seen.add(t)
+                snapped.append(t)
+        return snapped or [t_anchor]

src/lerobot/annotations/steerable_pipeline/modules/plan_subtasks_memory.py

@@ -0,0 +1,780 @@
+#!/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.
+"""``plan`` module: subtask decomposition + plan + memory (PERSISTENT styles)."""
+
+from __future__ import annotations
+
+import logging
+from collections.abc import Sequence
+from dataclasses import dataclass, field
+from typing import Any
+
+from ..config import PlanConfig
+from ..frames import (
+    FrameProvider,
+    null_provider,
+    to_contact_sheet_blocks,
+)
+from ..prompts import load as load_prompt
+from ..reader import EpisodeRecord, reconstruct_subtask_spans, snap_to_frame
+from ..staging import EpisodeStaging
+from ..vlm_client import VlmClient
+
+logger = logging.getLogger(__name__)
+
+
+# Prepended to every describe / segment prompt so the VLM knows the images are
+# timestamped contact-sheet grids, not a single video, and reads the burned-in
+# per-tile timestamp when choosing boundaries.
+def _contact_sheet_preamble(columns: int) -> str:
+    return (
+        "CONTACT SHEETS — how to read the images below:\n"
+        f"- Each image is a grid of sampled video frames, {columns} per row, "
+        "with time running left-to-right then top-to-bottom (row-major).\n"
+        "- Each frame has its timestamp burned into the top-left corner, e.g. "
+        '"012.50s". Use that printed timestamp (not the tile position) when you '
+        "choose start/end times; boundaries should land on or near a printed "
+        "timestamp.\n"
+        "- Frames continue across grids: an action may span the end of one sheet "
+        "and the start of the next, so do not place a boundary just because a new "
+        "image begins.\n\n"
+    )
+
+
+# Appended to every describe (and segment) prompt. A visual, causal definition
+# of where one event ends and the next begins — adapted from macrodata/refiner —
+# to sharpen cut points while the existing prompt keeps owning the imperative
+# phrasing.
+_CAUSAL_BOUNDARY_RULES = (
+    "EVENT BOUNDARIES — where one event ends and the next begins:\n"
+    "- Start a new event whenever the world state changes: an object becomes "
+    "held (the gripper closes on it), an object is released (the gripper opens "
+    "and it stays put), an object reaches a new location, a lid/door/drawer "
+    "changes open/closed state, a tool starts or stops affecting a surface, or "
+    "contents visibly move (e.g. poured).\n"
+    "- If a single action changes the same state gradually and continuously, "
+    "keep it as ONE event — do not split it.\n"
+    "- If the same action repeats on different objects or target locations, "
+    "treat each repetition as a separate event.\n"
+    "- Do NOT create boundaries for idle time, camera motion, hesitation, or "
+    "tiny hand adjustments."
+)
+
+
+@dataclass
+class PlanSubtasksMemoryModule:
+    """Generate subtask spans, plan, and memory rows.
+
+    All output is persistent (lives in ``language_persistent``):
+
+    - ``subtask`` rows: one per span, stamped at the span's *start* timestamp
+      (snapped to an exact frame).
+    - ``plan`` rows: emitted at ``t=0``; refreshed at every interjection
+      timestamp via :meth:`run_plan_updates` (called by the executor after
+      the ``interjections`` module completes).
+    - ``memory`` rows: emitted at each subtask boundary (= subtask start
+      timestamp from the second subtask onward).
+    """
+
+    vlm: VlmClient
+    config: PlanConfig
+    frame_provider: FrameProvider = field(default_factory=null_provider)
+
+    @property
+    def enabled(self) -> bool:
+        return self.config.enabled
+
+    def run_episode(self, record: EpisodeRecord, staging: EpisodeStaging) -> None:
+        rows: list[dict[str, Any]] = []
+        # Task driving every plan-module prompt: canonical episode_task, or a
+        # video-derived one when it's empty/placeholder (see derive_task_*).
+        effective_task = self._resolve_effective_task(record)
+        # task_aug rows at t=0: phrasings the renderer rotates ${task} through.
+        # Either the structured 5-axis taxonomy (task_aug_axes.enabled) or
+        # free-form n_task_rephrasings; the effective task is always emitted
+        # first so the rotation covers the source-of-truth phrasing.
+        t0 = float(record.frame_timestamps[0]) if record.frame_timestamps else 0.0
+        variants: list[str] | None = None
+        if self.config.task_aug_axes.enabled and effective_task:
+            variants = self._generate_task_aug_by_axes(effective_task, self.config.task_aug_axes)
+        elif self.config.n_task_rephrasings > 0 and effective_task:
+            variants = self._generate_task_rephrasings(effective_task, n=self.config.n_task_rephrasings)
+        if variants is not None:
+            rows.extend(self._task_aug_rows([effective_task, *variants], t0))
+
+        subtask_spans = self._generate_subtasks(record, task=effective_task)
+
+        # subtask rows
+        for span in subtask_spans:
+            rows.append(
+                {
+                    "role": "assistant",
+                    "content": span["text"],
+                    "style": "subtask",
+                    "timestamp": snap_to_frame(span["start"], record.frame_timestamps),
+                    "tool_calls": None,
+                }
+            )
+        # Plan rows at every subtask boundary (incl. t=0). The plan is a
+        # numbered list of still-todo subtasks, so re-emitting at each
+        # boundary makes it shrink as work progresses — ${plan} at frame t is
+        # exactly what's left to do.
+        if self.config.emit_plan:
+            for span in subtask_spans:
+                boundary_t = snap_to_frame(span["start"], record.frame_timestamps)
+                plan_text = self._generate_plan(
+                    record, subtask_spans, refresh_t=boundary_t, task=effective_task
+                )
+                if plan_text is not None:
+                    rows.append(
+                        {
+                            "role": "assistant",
+                            "content": plan_text,
+                            "style": "plan",
+                            "timestamp": float(boundary_t),
+                            "tool_calls": None,
+                        }
+                    )
+        # memory rows at every subtask boundary except the very first start;
+        # skipped entirely when ``emit_memory`` is False (subtasks-only / plan-only).
+        prior_memory = ""
+        memory_boundaries = enumerate(subtask_spans[1:], start=1) if self.config.emit_memory else []
+        for i, span in memory_boundaries:
+            completed = subtask_spans[i - 1]["text"]
+            remaining = [s["text"] for s in subtask_spans[i:]]
+            mem_text = self._generate_memory(record, prior_memory, completed, remaining, task=effective_task)
+            if mem_text:
+                ts = snap_to_frame(span["start"], record.frame_timestamps)
+                rows.append(
+                    {
+                        "role": "assistant",
+                        "content": mem_text,
+                        "style": "memory",
+                        "timestamp": ts,
+                        "tool_calls": None,
+                    }
+                )
+                prior_memory = mem_text
+        staging.write("plan", rows)
+
+    # ------------------------------------------------------------------
+    # Task derivation + rephrasings
+    # ------------------------------------------------------------------
+
+    _PLACEHOLDER_TASKS: frozenset[str] = frozenset(
+        {
+            "debug",
+            "test",
+            "tbd",
+            "todo",
+            "n/a",
+            "na",
+            "untitled",
+            "unnamed",
+            "default",
+            "placeholder",
+        }
+    )
+
+    def _resolve_effective_task(self, record: EpisodeRecord) -> str:
+        """Decide which task string drives the ``plan`` module for this episode.
+
+        Returns the user-supplied ``record.episode_task`` unless
+        ``derive_task_from_video`` says otherwise (see config docstring).
+        Falls back gracefully to the canonical task if video derivation
+        fails.
+        """
+        canonical = (record.episode_task or "").strip()
+        mode = (self.config.derive_task_from_video or "off").strip().lower()
+        if mode == "always":
+            derived = self._derive_task_from_video(record)
+            return derived or canonical
+        if mode == "if_short" and self._task_seems_bad(canonical):
+            derived = self._derive_task_from_video(record)
+            if derived:
+                return derived
+        return canonical
+
+    def _task_seems_bad(self, task: str) -> bool:
+        if not task:
+            return True
+        if len(task.split()) < int(self.config.derive_task_min_words):
+            return True
+        return task.lower() in self._PLACEHOLDER_TASKS
+
+    @staticmethod
+    def _task_aug_rows(phrasings: Sequence[str], t0: float) -> list[dict[str, Any]]:
+        """Build deduplicated ``task_aug`` rows (role=user) at ``t0``."""
+        seen: set[str] = set()
+        rows: list[dict[str, Any]] = []
+        for phrasing in phrasings:
+            key = phrasing.strip()
+            if not key or key in seen:
+                continue
+            seen.add(key)
+            rows.append(
+                {"role": "user", "content": key, "style": "task_aug", "timestamp": t0, "tool_calls": None}
+            )
+        return rows
+
+    # ------------------------------------------------------------------
+    # VLM call helpers — every plan-module prompt follows the same shape:
+    # build messages → single VLM call → pull a named field.
+    # ------------------------------------------------------------------
+
+    def _vlm_field(self, messages: list[dict[str, Any]], field: str) -> Any:
+        """Run a single VLM call and return ``result[field]`` or ``None``.
+
+        Centralizes the ``vlm.generate_json([m])[0]`` + ``isinstance(dict)``
+        dance every prompt-call site needs.
+        """
+        result = self.vlm.generate_json([messages])[0]
+        if isinstance(result, dict):
+            return result.get(field)
+        return None
+
+    @staticmethod
+    def _text_message(text: str) -> list[dict[str, Any]]:
+        """One-shot text-only user message wrapped for ``generate_json``."""
+        return [{"role": "user", "content": [{"type": "text", "text": text}]}]
+
+    def _video_message(
+        self,
+        record: EpisodeRecord,
+        prompt: str,
+        window: tuple[float, float] | None = None,
+    ) -> list[dict[str, Any]]:
+        """User message combining the (optionally windowed) contact sheets with ``prompt``.
+
+        The prompt is always prefixed with a short explanation of how to read
+        the timestamped grids, so the model treats them as one ordered
+        sequence of frames rather than unrelated images.
+        """
+        prompt = _contact_sheet_preamble(self.config.contact_sheet_columns) + prompt
+        content = [*self._episode_video_block(record, window=window), {"type": "text", "text": prompt}]
+        return [{"role": "user", "content": content}]
+
+    def _derive_task_from_video(self, record: EpisodeRecord) -> str | None:
+        """Ask the VLM "what is this video about" with no task hint at all."""
+        text = self._vlm_field(self._video_message(record, load_prompt("plan_video_task")), "task")
+        return text.strip() if isinstance(text, str) and text.strip() else None
+
+    def _generate_task_rephrasings(self, base_task: str, *, n: int) -> list[str]:
+        """Generate ``n`` text-only paraphrases of ``base_task``."""
+        if n <= 0 or not base_task:
+            return []
+        prompt = load_prompt("plan_task_rephrasings").format(base_task=base_task, n=n)
+        raw = self._vlm_field(self._text_message(prompt), "rephrasings")
+        if not isinstance(raw, list):
+            return []
+        out = [item.strip().strip('"').strip("'") for item in raw if isinstance(item, str)]
+        return [s for s in out if s][:n]
+
+    # ------------------------------------------------------------------
+    # Structured 5-axis task augmentation (EgoMimic-style taxonomy)
+    # ------------------------------------------------------------------
+
+    def _generate_task_aug_by_axes(self, base_task: str, axes_cfg: Any) -> list[str]:
+        """One VLM call → variants along the 5-axis taxonomy.
+
+        Variants from all axes are flattened into a single list (the
+        downstream pipeline doesn't need to know about the per-axis
+        bucketing — every variant becomes a ``task_aug`` row). Order
+        is preserved for reproducibility: synonym_paraphrase first,
+        then omit_arm, then omit_orientation, then omit_grasp_method,
+        then combined_omissions.
+        """
+        if not base_task:
+            return []
+        prompt = load_prompt("plan_task_aug_axes").format(
+            base_task=base_task,
+            n_synonym=axes_cfg.synonym_paraphrase,
+            n_omit_arm=axes_cfg.omit_arm,
+            n_omit_orientation=axes_cfg.omit_orientation,
+            n_omit_grasp_method=axes_cfg.omit_grasp_method,
+            n_combined=axes_cfg.combined_omissions,
+        )
+        result = self.vlm.generate_json([self._text_message(prompt)])[0]
+        if not isinstance(result, dict):
+            return []
+        ordered_axes = (
+            "synonym_paraphrase",
+            "omit_arm",
+            "omit_orientation",
+            "omit_grasp_method",
+            "combined_omissions",
+        )
+        flat: list[str] = []
+        seen: set[str] = set()
+        for axis in ordered_axes:
+            entries = result.get(axis)
+            if not isinstance(entries, list):
+                continue
+            for item in entries:
+                if not isinstance(item, str):
+                    continue
+                key = item.strip().strip('"').strip("'")
+                if not key or key in seen:
+                    continue
+                seen.add(key)
+                flat.append(key)
+        return flat
+
+    def _episode_video_block(
+        self, record: EpisodeRecord, window: tuple[float, float] | None = None
+    ) -> list[dict[str, Any]]:
+        """Timestamped contact sheets for the describe / segmentation prompts.
+
+        Always renders the (optionally windowed) episode as contact sheets:
+        frames sampled at ``frames_per_second`` and packed into timestamped
+        JPEG grids. ``max_frames_per_prompt`` caps the frame count; whole
+        episodes that exceed it are windowed upstream in
+        :meth:`_generate_subtasks` so each call stays within budget while the
+        full episode keeps its sampling density.
+
+        When ``window=(w0, w1)`` is given the badges are WINDOW-RELATIVE
+        (``ts - w0``) to match the window-relative time frame the
+        segmentation prompt works in (spans are offset back to absolute time
+        afterwards).
+        """
+        if not record.frame_timestamps:
+            return []
+        if window is not None:
+            w0, w1 = float(window[0]), float(window[1])
+            dur = max(0.0, w1 - w0)
+            n = max(1, int(round(dur * self.config.frames_per_second)) + 1)
+            n = min(n, self.config.max_frames_per_prompt)
+            if n <= 1 or dur <= 0.0:
+                timestamps = [0.5 * (w0 + w1)]
+            else:
+                step = dur / (n - 1)
+                timestamps = [w0 + i * step for i in range(n)]
+            frames = self.frame_provider.frames_at(record, timestamps)
+            rel = [ts - w0 for ts in timestamps[: len(frames)]]
+            return self._contact_sheet_blocks(frames, rel)
+        episode_duration = record.frame_timestamps[-1] - record.frame_timestamps[0]
+        n = max(1, int(round(episode_duration * self.config.frames_per_second)) + 1)
+        n = min(n, self.config.max_frames_per_prompt)
+        timestamps = self._uniform_episode_timestamps(record, n)
+        frames = self.frame_provider.frames_at(record, timestamps)
+        return self._contact_sheet_blocks(frames, timestamps[: len(frames)])
+
+    @staticmethod
+    def _uniform_episode_timestamps(record: EpisodeRecord, n: int) -> list[float]:
+        """``n`` episode-relative timestamps spanning ``[t0, t_last]`` uniformly."""
+        ts = record.frame_timestamps
+        if n >= len(ts):
+            return [float(t) for t in ts]
+        t0, t_last = float(ts[0]), float(ts[-1])
+        if t_last <= t0 or n <= 1:
+            return [t0] * max(1, n)
+        step = (t_last - t0) / (n - 1)
+        return [t0 + i * step for i in range(n)]
+
+    def _contact_sheet_blocks(self, frames: list[Any], timestamps: list[float]) -> list[dict[str, Any]]:
+        """Build timestamped contact-sheet image blocks from decoded frames."""
+        return to_contact_sheet_blocks(
+            frames,
+            timestamps,
+            columns=self.config.contact_sheet_columns,
+            frames_per_sheet=self.config.contact_sheet_frames_per_sheet,
+            frame_width=self.config.contact_sheet_frame_width,
+            quality=self.config.contact_sheet_quality,
+        )
+
+    def run_plan_updates(
+        self,
+        record: EpisodeRecord,
+        staging: EpisodeStaging,
+        interjection_times: Sequence[float],
+        interjection_texts: Sequence[str] | None = None,
+    ) -> None:
+        """Append additional ``plan`` rows at every interjection timestamp.
+
+        Plans refresh ONLY on user interjections (event-driven). The
+        interjection text is forwarded into the prompt so the refreshed plan
+        reflects the user's correction.
+        """
+        if not self.config.emit_plan:
+            return
+        existing = staging.read("plan")
+        # Pass the last frame timestamp so the final span is closed (else its
+        # end == start, zero duration, and a refresh inside it is missed).
+        episode_end_t = float(record.frame_timestamps[-1]) if record.frame_timestamps else None
+        spans = reconstruct_subtask_spans(existing, episode_end_t=episode_end_t)
+        already_planned: set[float] = {float(r["timestamp"]) for r in existing if r.get("style") == "plan"}
+        new_rows = list(existing)
+
+        texts: list[str | None] = (
+            [None] * len(interjection_times)
+            if interjection_texts is None
+            else [str(t) if t else None for t in interjection_texts]
+        )
+        for raw_t, inter_text in zip(interjection_times, texts, strict=True):
+            t = snap_to_frame(raw_t, record.frame_timestamps)
+            if t in already_planned:
+                continue
+            already_planned.add(t)
+            plan_text = self._generate_plan(record, spans, refresh_t=t, interjection=inter_text)
+            if plan_text is not None:
+                new_rows.append(
+                    {
+                        "role": "assistant",
+                        "content": plan_text,
+                        "style": "plan",
+                        "timestamp": t,
+                        "tool_calls": None,
+                    }
+                )
+        staging.write("plan", new_rows)
+
+    def _generate_subtasks(self, record: EpisodeRecord, *, task: str | None = None) -> list[dict[str, Any]]:
+        """Generate subtask spans, optionally via a multi-call quality chain.
+
+        Single call (default): watch video → emit subtask JSON.
+
+        Multi-call (opt-in, higher quality, more VLM calls):
+          1. ``subtask_describe_first`` — a grounding pass that narrates
+             ONLY what is visible (no JSON commitment to subtasks yet);
+             its description is injected into the segmentation prompt so
+             the model segments its own grounded observations instead of
+             pattern-matching the task text.
+          2. segmentation — emit subtask JSON (as before).
+        """
+        if record.row_count == 0 or not record.frame_timestamps:
+            return []
+        episode_duration = record.frame_timestamps[-1] - record.frame_timestamps[0]
+        effective_task = task if task is not None else record.episode_task
+
+        # ---- Auto-windowing (keeps the full sampling density) --------
+        # Contact sheets are cheap, but a whole long episode sampled at
+        # ``frames_per_second`` can still exceed ``max_frames_per_prompt``.
+        # When it does, split into consecutive windows of exactly that many
+        # frames (one describe→segment call each, still at the full sampling
+        # density), then merge + stitch — so an episode of any length is
+        # covered at full density rather than subsampled into one sparse call.
+        fps = max(1e-6, float(self.config.frames_per_second))
+        n_whole = int(round(episode_duration * fps)) + 1
+        if n_whole > self.config.max_frames_per_prompt:
+            window_s = self.config.max_frames_per_prompt / fps
+            return self._generate_subtasks_windowed(record, effective_task, window_s)
+
+        # ---- Pass 1 (optional): grounding description ----------------
+        observation_block = ""
+        if getattr(self.config, "subtask_describe_first", False):
+            description = self._describe_episode(record, effective_task)
+            if description:
+                observation_block = (
+                    "You watched this video and described, chronologically, "
+                    "ONLY what the robot actually does:\n"
+                    f'"""{description}"""\n\n'
+                    "Segment THAT grounded description (cross-checked against "
+                    "the video) into atomic subtasks. Do not introduce any "
+                    "action that is not in your description above.\n\n"
+                )
+
+        # ---- Pass 2: segmentation ------------------------------------
+        prompt = self._with_causal_rules(
+            load_prompt("plan_subtasks").format(
+                episode_task=effective_task,
+                min_subtask_seconds=self.config.min_subtask_seconds,
+                max_steps=self.config.plan_max_steps,
+                episode_duration=f"{episode_duration:.3f}",
+                observation_block=observation_block,
+            )
+        )
+        spans = self._vlm_field(self._video_message(record, prompt), "subtasks")
+        cleaned = self._clean_spans(spans, record)
+        if not cleaned:
+            return []
+
+        # ---- Full-episode coverage stitch ----------------------------
+        # The VLM can start after t0 or leave gaps, so frames fall through
+        # with no active subtask. Always stitch into a contiguous
+        # [t0, t_last] cover.
+        cleaned = self._stitch_full_coverage(cleaned, record)
+
+        return cleaned
+
+    def _generate_subtasks_windowed(
+        self, record: EpisodeRecord, task: str, window_s: float
+    ) -> list[dict[str, Any]]:
+        """Subtask generation in fixed-length windows at constant fps.
+
+        Splits ``[t0, t_last]`` into consecutive windows of ``window_s``
+        seconds, runs the describe -> segment chain on each window's own
+        frames (sampled at ``frames_per_second``), offsets
+        each window's spans back to absolute episode time, then merges +
+        stitches into a contiguous whole-episode cover.
+        """
+        t0 = float(record.frame_timestamps[0])
+        t_last = float(record.frame_timestamps[-1])
+        all_spans: list[dict[str, Any]] = []
+        w0 = t0
+        n_windows = 0
+        while w0 < t_last - 1e-6:
+            w1 = min(w0 + window_s, t_last)
+            all_spans.extend(self._subtasks_for_window(record, task, w0, w1))
+            n_windows += 1
+            w0 = w1
+        logger.info(
+            "episode %d: windowed subtask gen over %d window(s) of %.1fs -> %d raw spans",
+            record.episode_index,
+            n_windows,
+            window_s,
+            len(all_spans),
+        )
+        # Merge across windows: clamp to the absolute episode, sort, and
+        # frame-snap to distinct starts (handles any boundary collisions).
+        cleaned = self._clean_spans(all_spans, record)
+        if not cleaned:
+            return []
+        return self._stitch_full_coverage(cleaned, record)
+
+    def _subtasks_for_window(
+        self, record: EpisodeRecord, task: str, w0: float, w1: float
+    ) -> list[dict[str, Any]]:
+        """Run describe -> segment on one ``[w0, w1]`` window.
+
+        The model works in window-RELATIVE time ``[0, L]`` (it perceives
+        the window as a clip starting at 0); spans are offset back to
+        absolute ``[w0, w1]`` before returning.
+        """
+        window = (w0, w1)
+        win_len = max(0.0, w1 - w0)
+
+        observation_block = ""
+        if getattr(self.config, "subtask_describe_first", False):
+            description = self._describe_episode(record, task, window=window)
+            if description:
+                observation_block = (
+                    "You watched this video clip and described, chronologically, "
+                    "ONLY what the robot actually does:\n"
+                    f'"""{description}"""\n\n'
+                    "Segment THAT grounded description (cross-checked against "
+                    "the clip) into atomic subtasks. Do not introduce any "
+                    "action that is not in your description above.\n\n"
+                )
+
+        prompt = self._with_causal_rules(
+            load_prompt("plan_subtasks").format(
+                episode_task=task,
+                min_subtask_seconds=self.config.min_subtask_seconds,
+                max_steps=self.config.plan_max_steps,
+                episode_duration=f"{win_len:.3f}",
+                observation_block=observation_block,
+            )
+        )
+        spans = self._vlm_field(self._video_message(record, prompt, window=window), "subtasks")
+        # Window-relative clamp; no frame-snap dedupe yet (done on the
+        # merged absolute set).
+        cleaned = self._clean_spans(spans, record, bounds=(0.0, win_len), dedupe=False)
+        if not cleaned:
+            return []
+
+        # Offset window-relative spans back to absolute episode time.
+        for s in cleaned:
+            s["start"] = w0 + float(s["start"])
+            s["end"] = w0 + float(s["end"])
+        return cleaned
+
+    def _stitch_full_coverage(
+        self, spans: list[dict[str, Any]], record: EpisodeRecord
+    ) -> list[dict[str, Any]]:
+        """Make subtask spans tile the full episode with no gaps.
+
+        * The first subtask starts at the episode's first frame ``t0``
+          (any idle / approach before the first labelled action is folded
+          into it), so every early frame has an active subtask.
+        * Each subtask's ``end`` is snapped to the next subtask's
+          ``start`` (gaps between spans are closed), and the final
+          subtask's ``end`` extends to the last frame ``t_last``.
+
+        Starts are otherwise left as the (already frame-snapped, distinct)
+        values the VLM produced — only the FIRST start is pulled
+        back to ``t0``, which can't collide with a later span because it
+        was already the earliest. Purely deterministic; runs after the
+        VLM passes.
+        """
+        if not spans or not record.frame_timestamps:
+            return spans
+        t0 = float(record.frame_timestamps[0])
+        t_last = float(record.frame_timestamps[-1])
+        spans = sorted(spans, key=lambda s: float(s["start"]))
+        spans[0]["start"] = t0
+        for i in range(len(spans) - 1):
+            spans[i]["end"] = float(spans[i + 1]["start"])
+        spans[-1]["end"] = t_last
+        for s in spans:
+            if float(s["end"]) < float(s["start"]):
+                s["end"] = float(s["start"])
+        return spans
+
+    @staticmethod
+    def _with_causal_rules(prompt: str) -> str:
+        """Append the causal event-boundary rules to a describe/segment prompt."""
+        return f"{prompt}\n\n{_CAUSAL_BOUNDARY_RULES}"
+
+    def _clean_spans(
+        self,
+        spans: Any,
+        record: EpisodeRecord,
+        bounds: tuple[float, float] | None = None,
+        dedupe: bool = True,
+    ) -> list[dict[str, Any]]:
+        """Clamp / sort / (optionally) dedupe raw VLM subtask spans into valid rows.
+
+        ``bounds`` overrides the clamp range — pass the window's
+        ``(w_lo, w_hi)`` when cleaning window-relative spans, or leave
+        ``None`` to clamp to the whole episode ``[t0, t_last]``.
+        ``dedupe`` runs the frame-snap distinct-start step; skip it for
+        window-relative spans (frame snapping is done once on the merged,
+        absolute-time set).
+        """
+        if not spans:
+            return []
+        if bounds is not None:
+            lo, hi = float(bounds[0]), float(bounds[1])
+        else:
+            lo = record.frame_timestamps[0]
+            hi = record.frame_timestamps[-1]
+        cleaned: list[dict[str, Any]] = []
+        for span in spans:
+            try:
+                start = float(span["start"])
+                end = float(span["end"])
+                text = str(span["text"]).strip()
+            except (KeyError, ValueError, TypeError):
+                continue
+            start = max(lo, min(start, hi))
+            end = max(lo, min(end, hi))
+            if end < start:
+                start, end = end, start
+            if not text:
+                continue
+            cleaned.append({"text": text, "start": start, "end": end})
+        cleaned.sort(key=lambda s: s["start"])
+        if dedupe:
+            return self._dedupe_starts_to_distinct_frames(cleaned, record)
+        return cleaned
+
+    def _describe_episode(
+        self, record: EpisodeRecord, task: str, window: tuple[float, float] | None = None
+    ) -> str:
+        """Grounding pass: free-form chronological description of the (windowed) video."""
+        prompt = self._with_causal_rules(load_prompt("plan_subtask_describe").format(episode_task=task))
+        text = self._vlm_field(self._video_message(record, prompt, window=window), "description")
+        return text.strip() if isinstance(text, str) and text.strip() else ""
+
+    @staticmethod
+    def _dedupe_starts_to_distinct_frames(
+        spans: list[dict[str, Any]], record: EpisodeRecord
+    ) -> list[dict[str, Any]]:
+        """Bump same-frame subtask starts onto distinct frames.
+
+        Two consecutive VLM spans whose ``start`` rounds to the same
+        source frame (after :func:`snap_to_frame`) would otherwise emit
+        two ``style=subtask`` rows at the identical persistent
+        timestamp. The training-time renderer's ``active_at(t,
+        style=subtask)`` resolver can't disambiguate that and raises
+        ``Ambiguous resolver for style='subtask'``.
+
+        Walk the (sorted-by-start) spans, snap each to its frame, and
+        if the snapped frame is already taken push the span onto the
+        next unused frame so both subtasks survive on distinct
+        timestamps. If the episode ends before a free frame is found,
+        the trailing span is dropped with a warning — better than
+        poisoning the render.
+        """
+        if not spans:
+            return spans
+        frames = record.frame_timestamps
+        if not frames:
+            return spans
+        used: set[float] = set()
+        out: list[dict[str, Any]] = []
+        for span in spans:
+            ts = snap_to_frame(span["start"], frames)
+            if ts in used:
+                next_ts = next((f for f in frames if f > ts and f not in used), None)
+                if next_ts is None:
+                    logger.warning(
+                        "episode %d: subtask %r snapped to occupied frame "
+                        "%.3f and no free later frame exists — dropping",
+                        record.episode_index,
+                        span.get("text"),
+                        ts,
+                    )
+                    continue
+                ts = next_ts
+            used.add(ts)
+            new_span = {**span, "start": ts}
+            if float(new_span.get("end", ts)) < ts:
+                new_span["end"] = ts
+            out.append(new_span)
+        return out
+
+    def _generate_plan(
+        self,
+        record: EpisodeRecord,  # noqa: ARG002  (kept for signature stability)
+        subtask_spans: Sequence[dict[str, Any]],
+        *,
+        refresh_t: float | None = None,
+        interjection: str | None = None,  # noqa: ARG002
+        task: str | None = None,  # noqa: ARG002
+    ) -> str | None:
+        """Deterministic plan = numbered list of *still-todo* subtasks.
+
+        No VLM call: a plain numbered list keeps the plan aligned with the
+        upcoming subtasks (the old VLM "compact hierarchical plan" prompt
+        cost a round-trip per episode/refresh and could diverge).
+
+            1. <subtask 1>
+            2. <subtask 2>
+
+        On a refresh at ``refresh_t`` (from ``run_plan_updates`` on
+        interjections, and ``run_episode`` at each boundary), only subtasks
+        starting at or after ``refresh_t`` are included — so it always
+        describes what's left.
+        """
+        if not subtask_spans:
+            return None
+        remaining = [
+            s for s in subtask_spans if refresh_t is None or float(s.get("start", 0.0)) >= float(refresh_t)
+        ]
+        if not remaining:
+            # Past the last subtask boundary on a late refresh — nothing
+            # left to plan; emit None so the caller skips the row.
+            return None
+        return "\n".join(f"{i}. {span.get('text', '').strip()}" for i, span in enumerate(remaining, start=1))
+
+    def _generate_memory(
+        self,
+        record: EpisodeRecord,
+        prior_memory: str,
+        completed: str,
+        remaining: Sequence[str],
+        *,
+        task: str | None = None,
+    ) -> str:
+        prompt = load_prompt("plan_memory").format(
+            episode_task=(task if task is not None else record.episode_task),
+            prior_memory=prior_memory or "(none)",
+            completed_subtask=completed,
+            remaining_subtasks=", ".join(remaining) if remaining else "(none)",
+        )
+        memory = self._vlm_field(self._text_message(prompt), "memory")
+        return memory.strip() if isinstance(memory, str) else ""

src/lerobot/annotations/steerable_pipeline/prompts/__init__.py

@@ -0,0 +1,33 @@
+#!/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.
+"""Prompt templates loaded as plain text.
+
+One file per use site. Templates use ``str.format(**vars)`` substitution; we
+intentionally avoid jinja2 here so the templates remain inspectable in
+plain editors and roundtrip cleanly through ``ruff format``.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+_DIR = Path(__file__).parent
+
+
+def load(name: str) -> str:
+    """Read prompt template ``name.txt`` from the ``prompts/`` directory."""
+    path = _DIR / f"{name}.txt"
+    return path.read_text(encoding="utf-8")

src/lerobot/annotations/steerable_pipeline/prompts/interjections_initial_speech.txt

@@ -0,0 +1,12 @@
+The user just asked the robot: "{episode_task}".
+
+Generate a short verbal acknowledgement the robot would speak back before
+beginning the task. Style: compact, confident, friendly.
+
+Examples (Hi Robot, Shi 2025): "Sure, I won't put cheese on it.",
+"OK, starting with the sponge.", "Got it.".
+
+Prefer very short replies: "Got it.", "On it.", "OK."
+
+Output strictly valid JSON:
+  {{ "text": "<the spoken acknowledgement>" }}

src/lerobot/annotations/steerable_pipeline/prompts/interjections_interjection.txt

@@ -0,0 +1,46 @@
+You are generating training data for a Hi Robot-style hierarchical
+robot policy. The robot in this demonstration has ALREADY executed
+every step shown in the video — we cannot retroactively change the
+action stream. To keep training data consistent with the video, the
+"interjection" must align with what the robot is *about to do next* in
+the demonstration, framed as a natural mid-task user request.
+
+The episode's overall task: "{episode_task}".
+
+The images above show roughly {window_seconds:.1f} seconds straddling a
+subtask boundary in the demonstration:
+
+- Subtask the robot just finished: "{prev_subtask}"
+- Subtask the robot is about to start: "{next_subtask}"
+- Time into episode: {timestamp:.2f}s
+
+Write ONE compact interjection the user would naturally say at this
+moment to prompt / confirm / encourage the robot to do "{next_subtask}".
+Keep it like a mid-task coaching cue, not a full instruction paragraph.
+Also write the robot's compact verbal acknowledgement.
+
+Hard rules:
+
+- The interjection MUST be consistent with the next subtask. The user
+  cannot ask for something different from what the robot then does in
+  the video. If you're tempted to say "actually skip X" or "do Y
+  instead", DO NOT — those would contradict the demonstration.
+- The interjection must reference an object, location, or action that
+  is plausible given the visible scene and the next subtask text.
+- One short phrase or sentence each. Conversational, not robotic.
+- Prefer direct cues: "{next_subtask}, please."; "Now {next_subtask}."
+- Keep robot speech very short: "OK.", "On it.", "Doing that."
+
+Style examples (vary the phrasing — don't reuse these verbatim):
+  - "Now go ahead and {next_subtask}."
+  - "Great, can you {next_subtask} next?"
+  - "{next_subtask}, please."
+  - "Before you continue, please {next_subtask}."
+  - "Looking good — {next_subtask} now."
+  - "Okay, {next_subtask}."
+
+Output strictly valid JSON:
+  {{
+    "interjection": "<short cue from the user, asking for the next subtask>",
+    "speech":       "<short robot acknowledgement>"
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/plan_memory.txt

@@ -0,0 +1,36 @@
+You are updating the robot's compressed semantic memory at the boundary of
+a completed subtask.
+
+Reference (verbatim from MEM, Torne 2026):
+"Remove or compress information in the language memory whenever
+appropriate. Keep ONLY the minimal set of relevant information for future
+task execution. Specific object attributes (colors, precise quantities of
+each item) get discarded when their details won't affect subsequent
+actions. Functional outcomes (where items went, how many) are preserved."
+
+Episode task: "{episode_task}"
+Previous memory: {prior_memory}
+Just-completed subtask: "{completed_subtask}"
+Remaining subtasks (for relevance judgement only): {remaining_subtasks}
+
+Write the memory as a short FIRST-PERSON, PAST-TENSE narrative of what the
+robot has accomplished so far — the running story it would tell itself.
+
+Authoring rules:
+- First person, past tense. Every sentence starts with "I": "I picked
+  up...", "I opened...", "I moved to...".
+- One or two short sentences. Extend the previous memory with the
+  just-completed subtask; do not rewrite it from scratch.
+- Keep WHAT happened (functional outcomes — where items went, how many),
+  drop HOW (grasp details, motions).
+- Compress completed steps and drop object attributes (colors, exact
+  counts) once they no longer affect the remaining subtasks.
+
+Example (MEM, Torne 2026):
+  Before: "I prepared the pot and got the potatoes, milk, and butter. I
+           moved to the drawer."
+  After:  "I prepared the pot and got the ingredients. I opened the
+           drawer with the masher."
+
+Output strictly valid JSON:
+  {{ "memory": "<one or two short first-person past-tense sentences>" }}

src/lerobot/annotations/steerable_pipeline/prompts/plan_subtask_describe.txt

@@ -0,0 +1,27 @@
+You are watching a teleoperated robot demonstration from a single
+camera. The user asked the robot to: "{episode_task}"
+
+This is an OBSERVATION pass. Watch the entire clip and describe, in
+chronological order, ONLY what the robot physically does — the concrete
+motions, approaches, contacts, grasps, releases, and relocations you can
+actually SEE in the frames.
+
+Hard rules:
+- Describe only motion visible in the video. Do NOT use the task
+  instruction to guess steps that aren't shown. The instruction is the
+  goal; the video is ground truth.
+- Do NOT segment into named subtasks yet and do NOT output JSON beyond
+  the single field below. Just narrate what happens.
+- Give an approximate timestamp (in seconds) for each distinct event,
+  e.g. "0.0-1.4s: the base drives forward toward the stove".
+- Do NOT invent objects, grasps, destinations, or steps. If the robot
+  only does one thing (e.g. it just navigates and the clip ends), say
+  exactly that and nothing more.
+- Be concrete and literal. "the gripper closes on the mug" — not "the
+  robot prepares to make coffee".
+
+Output strictly valid JSON:
+
+  {{
+    "description": "<chronological, timestamped description of ONLY what is visible>"
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/plan_subtasks.txt

@@ -0,0 +1,112 @@
+You are labeling a teleoperated robot demonstration.
+
+The user originally asked: "{episode_task}"
+
+You are shown the entire demonstration as a single video. Watch the
+whole clip, then segment it into a list of consecutive atomic subtasks
+the robot performs.
+
+{observation_block}GROUNDING — read this first, it overrides everything below:
+- Label ONLY what the robot actually does in the video. Every subtask
+  you emit must correspond to motion you can SEE in specific frames.
+- Do NOT invent, anticipate, or pad. If the robot only does one thing
+  (e.g. it just navigates to a location and the clip ends), emit
+  EXACTLY ONE subtask. Many demonstrations are a single atomic skill.
+- ``max_steps`` below is a hard CEILING, not a target. Emitting fewer
+  subtasks than the ceiling is not just allowed, it is expected for
+  short / atomic demonstrations. One correct subtask is far better
+  than several invented ones.
+- If the video does not clearly show the action implied by the task,
+  describe what you actually see — do NOT fabricate the task's steps
+  from the instruction text. The instruction tells you the goal; the
+  VIDEO is the ground truth for what happened.
+
+Authoring rules — Hi Robot atom granularity, pi0.7-style short prompts:
+
+- Each subtask = one COMPOSITE atomic skill the low-level policy can
+  execute end-to-end. A "skill" bundles its own approach motion with
+  its terminal action — do NOT split the approach off as its own
+  subtask. The whole-arm policy already learns to reach as part of
+  every manipulation primitive.
+- Write each subtask as an IMPERATIVE COMMAND, starting with one of
+  these verbs (extend only when none fits):
+    pick up <obj>           — approach + grasp + lift in one subtask
+    put <obj> on/in <loc>   — transport + release in one subtask
+    place <obj> on/in <loc> — synonym of "put"; pick one and stay consistent
+    push <obj>              — contact + linear shove
+    pull <obj>              — contact + linear retract
+    turn <knob/dial/handle> — rotary actuation
+    press <button>          — single-press contact
+    open <drawer/door/lid>  — full open motion
+    close <drawer/door/lid> — full close motion
+    pour <src> into <dst>   — tilt + flow
+    insert <obj> into <slot>— alignment + push-fit
+    go to <loc>             — ONLY when no grasp / actuation follows
+                             (e.g. a pure relocation between phases).
+                             If the next subtask grasps something at
+                             that location, drop "go to ..." and just
+                             write "pick up ..." instead.
+- Forbidden ultra-fine splits — the VLM is NOT allowed to emit these
+  as standalone subtasks; fold them into the parent composite:
+    "move to X"   → fold into "pick up X" (or whatever follows)
+    "reach for X" → fold into "pick up X"
+    "grasp X"     → fold into "pick up X"
+    "lift X"      → fold into "pick up X" (or "put X on Y" if it's
+                    the transport phase of a place)
+    "release X"   → fold into "put X on Y" (or "place X in Y")
+- Keep it SHORT — a verb phrase, not a sentence. Drop articles
+  ("the", "a") and adverbs ("carefully", "slowly"). Add a "how"
+  detail (which hand, which grasp point) ONLY when it is needed to
+  disambiguate. Every subtask must begin with one of the verbs
+  above (no leading nouns, no "then", no "first").
+- NEVER use third person. Never write "the robot", "the arm", "the
+  gripper moves", "it picks up" — the robot is implied. Command it,
+  do not describe it.
+- Use the exact object nouns from the task above. If the task says
+  "cube", every subtask says "cube" — never switch to "block". If it
+  says "box", never switch to "bin"/"container". Keep vocabulary
+  consistent across the whole episode.
+- Good: "pick up blue cube", "put blue cube in box", "open drawer",
+  "turn red knob", "press start button", "go to sink".
+- Bad: "move to blue cube" (approach as its own subtask — forbidden,
+  must be folded into "pick up blue cube"); "the robot arm moves
+  towards the blue cube" (third person, too long); "carefully pick
+  up the cube" (adverb, article); "release the yellow block"
+  ("block" when the task said "cube", and "release" must be folded
+  into a "put"/"place" subtask).
+- Subtasks are non-overlapping and cover the full episode in order.
+  Choose the cut points yourself based on what you see in the video
+  (gripper open/close events, contact, regrasps, transitions).
+- Each subtask spans at least {min_subtask_seconds} seconds. If a
+  candidate span would be shorter, merge it into its neighbour
+  rather than emitting it.
+- Do not exceed {max_steps} subtasks total. Fewer, larger composites
+  are preferred over many micro-steps.
+- Every subtask's [start_time, end_time] must lie within
+  [0.0, {episode_duration}] seconds.
+
+SPECIAL CASES — verb disambiguation (each rule is narrowly visual and
+fires ONLY on the spatial situation it names; it must not change how you
+label any other situation):
+- STACK vs PUT: if an object is placed ON TOP OF another specific object
+  (not on a flat table / shelf / counter), use "stack ... on ...", not
+  "put". "stack blue book on green book", NOT "put blue book on table".
+- INSERT vs PUT: if an object goes INTO a fitted slot / hole / socket /
+  receptacle (push-fit), use "insert ... into ...", not "put".
+- RETRIEVE/PICK-UP vs PUT (direction): watch the gripper. If it CLOSES
+  on the object and the object moves WITH the hand, it is "pick up" /
+  "retrieve" (object leaves its location). If the gripper OPENS and the
+  object stays where the hand left it, it is "put" / "place" (object
+  arrives at a location). Decide by which way the object moves, not by
+  where the hand ends up.
+- POUR vs PUT: only use "pour" when the source is tilted and contents
+  flow out; moving a full container without tilting is "put"/"place".
+
+Output strictly valid JSON of shape:
+
+  {{
+    "subtasks": [
+      {{"text": "<short imperative verb phrase>", "start": <float>, "end": <float>}},
+      ...
+    ]
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/plan_task_aug_axes.txt

@@ -0,0 +1,67 @@
+You are generating structured augmentations of a robot task instruction
+for training a language-conditioned policy. Unlike free-form rephrasing,
+your variants follow a NAMED 5-axis taxonomy — each axis omits or varies
+a specific element of the task while preserving its meaning.
+
+Original task: "{base_task}"
+
+Produce variants along five named axes. Each axis has a target count.
+The whole batch should expose the policy to maximum linguistic diversity
+WITHOUT changing what the robot is supposed to do.
+
+Axes and target counts:
+
+  synonym_paraphrase ({n_synonym}):
+    Different wording / verbs / sentence structure. ALL information
+    from the original task is preserved — same object, same arm
+    specification if present, same orientation if present, same grasp
+    if present.
+
+  omit_arm ({n_omit_arm}):
+    Drop the left/right/both arm specification from the task. Skip
+    entirely (emit 0 entries) if the original task does NOT mention an
+    arm. Do not invent an arm specification just to omit it.
+
+  omit_orientation ({n_omit_orientation}):
+    Drop orientation cues (upright, sideways, facing the user,
+    long-edge-first, etc.). Skip entirely if no orientation cue is
+    present in the original task.
+
+  omit_grasp_method ({n_omit_grasp_method}):
+    Drop the grip / grasp method specification (pinch, wrap, hold by
+    the rim, etc.). Skip entirely if no grasp method is mentioned.
+
+  combined_omissions ({n_combined}):
+    Combine TWO of the above omissions simultaneously (e.g. drop both
+    arm and orientation). Skip entirely if fewer than two of (arm,
+    orientation, grasp_method) appear in the original task.
+
+Hard rules:
+- Each variant MUST preserve the core action, the target object, AND
+  the goal / destination. Do not change which object is involved, where
+  it goes, or the high-level action. "Navigate to the stove" may become
+  "go to the stove" or "head over to the stove" — it must NEVER become
+  "wander around the kitchen", "explore the room", or anything that
+  drops or generalises the stove destination. If you cannot vary the
+  wording without changing the goal, emit fewer variants.
+- Only the FIVE listed elements (wording, arm, orientation, grasp
+  method, or a combination) may be varied or omitted. The verb's
+  meaning, the object, and the destination are fixed.
+- Each variant is plain prose, no markdown, no quotes, no list numbers.
+- Each variant must be DISTINCT from every other variant in the entire
+  output, both within and across axes. Near-duplicates are not allowed.
+- If an axis cannot reach its target count because the original task
+  lacks the omittable element, emit fewer entries — do NOT pad the
+  axis with paraphrases that belong to a different axis.
+- Variants should not all start with verbs — vary sentence structure
+  (some imperative, some polite request, some question).
+
+Output strictly valid JSON of shape:
+
+  {{
+    "synonym_paraphrase": ["<v1>", "<v2>", ...],
+    "omit_arm": ["<v1>", "<v2>", ...],
+    "omit_orientation": ["<v1>", ...],
+    "omit_grasp_method": ["<v1>", ...],
+    "combined_omissions": ["<v1>", ...]
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/plan_task_rephrasings.txt

@@ -0,0 +1,32 @@
+You are generating training data for a Hi Robot-style policy. We need
+{n} alternative phrasings of the same robot task so the policy sees
+diverse user prompts during training instead of the same canonical
+string repeated every frame.
+
+Original task:
+"{base_task}"
+
+Generate exactly {n} alternative phrasings of the same task. Vary:
+
+- formality (casual / polite / curt)
+- verbosity (mostly short imperative; occasional polite request)
+- word choice (synonyms, different verbs)
+- sentence structure (imperative / question / suggestion)
+
+Hard rules:
+- Each phrasing MUST preserve the exact meaning of the original task.
+  Do not change which object is involved, the destination, or the
+  action. Do not add extra steps. Do not invent new objects.
+- Each phrasing must be a short phrase or sentence, plain prose, no
+  markdown, no quotes, no list numbers.
+- Phrasings must be distinct — no near-duplicates.
+- Output exactly {n} entries.
+
+Output strictly valid JSON:
+  {{
+    "rephrasings": [
+      "<phrasing 1>",
+      "<phrasing 2>",
+      ...
+    ]
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/plan_video_task.txt

@@ -0,0 +1,17 @@
+The video above shows a robot manipulation episode in full. Look at
+the entire video and describe in ONE concise sentence what the robot
+is doing.
+
+Rules:
+- One sentence, in natural English, like a user instruction.
+- Capture the goal of the demonstration, not low-level motions.
+  Example: "place the yellow cube into the red bin" — not "move the
+  end-effector down 5cm and close the gripper".
+- 4 to 15 words. Plain prose, no markdown, no bullets, no quotes.
+- Do not invent objects or actions that aren't visible.
+- Do not output anything other than the JSON object below.
+
+Output strictly valid JSON:
+  {{
+    "task": "<single concise sentence describing what the robot does in this video>"
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/vqa.txt

@@ -0,0 +1,32 @@
+You are generating a frame-grounded visual question/answer pair for
+chain-of-thought training. Reference: ECoT (Zawalski 2024) and Steerable
+Policies — both train policies on grounded features such as bounding box
+pixel coordinates, keypoints, counts, attributes, and spatial relations.
+
+The frame shows a robot working on: "{episode_task}".
+
+Question types and the EXACT answer JSON shape required for each:
+
+  bbox       => {{"detections": [{{"label": "<obj>", "bbox_format": "xyxy",
+                                    "bbox": [x1, y1, x2, y2]}}, ...]}}
+                bbox is in pixel coordinates (x_min, y_min, x_max, y_max).
+                ECoT example: "a white cup [124, 25, 176, 113]".
+
+  keypoint   => {{"label": "<point>", "point_format": "xy",
+                  "point": [x, y]}}
+
+  count      => {{"label": "<obj>", "count": <int>,
+                  "note": "<optional short note>"}}
+
+  attribute  => {{"label": "<obj>", "attribute": "<color|shape|state|...>",
+                  "value": "<observed value>"}}
+
+  spatial    => {{"subject": "<obj>", "relation": "<left_of|right_of|on|in|"
+                  "above|below|near>", "object": "<obj>"}}
+
+Generate a question of type "{question_type}". Output strictly valid JSON:
+
+  {{
+    "question": "<short, frame-grounded question>",
+    "answer":   <object whose shape matches the schema above>
+  }}

src/lerobot/annotations/steerable_pipeline/reader.py

@@ -0,0 +1,216 @@
+#!/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.
+"""Datatrove-shaped reader.
+
+The reader walks ``data/chunk-*/file-*.parquet`` and yields one record per
+episode containing:
+
+- ``episode_index``: int
+- ``frame_timestamps``: tuple[float, ...]
+- ``frame_indices``: tuple[int, ...]
+- ``episode_task``: str (canonical task from ``meta/tasks.parquet``)
+- ``data_path``: pathlib.Path of the source parquet shard
+- ``frames_df``: pandas.DataFrame slice for the episode (only loaded on demand)
+
+This shape lets each module operate per-episode without loading all parquet
+rows into memory at once.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Iterator, Sequence
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+import pyarrow.parquet as pq
+
+from lerobot.datasets.io_utils import load_tasks
+from lerobot.datasets.utils import DEFAULT_TASKS_PATH
+
+
+@dataclass
+class EpisodeRecord:
+    """Per-episode record yielded by the reader."""
+
+    episode_index: int
+    episode_task: str
+    frame_timestamps: tuple[float, ...]
+    frame_indices: tuple[int, ...]
+    data_path: Path
+    row_offset: int  # row offset within the parquet file where this episode starts
+    row_count: int  # number of rows for this episode
+
+    # Memoized parquet slice — populated on first ``frames_df()`` call so
+    # repeat queries from different modules don't re-read the whole shard.
+    _frames_df_cache: Any = field(default=None, init=False, repr=False, compare=False)
+
+    def frames_df(self):  # type: ignore[no-untyped-def]
+        """Lazy-load the pandas slice for this episode (memoized)."""
+        if self._frames_df_cache is None:
+            import pandas as pd  # noqa: PLC0415  - deferred for optional dataset extra
+
+            table = pq.read_table(self.data_path)
+            df: pd.DataFrame = table.to_pandas()
+            self._frames_df_cache = df.iloc[self.row_offset : self.row_offset + self.row_count].reset_index(
+                drop=True
+            )
+        return self._frames_df_cache
+
+
+def reconstruct_subtask_spans(
+    rows: Sequence[dict[str, Any]],
+    *,
+    episode_end_t: float | None = None,
+) -> list[dict[str, Any]]:
+    """Turn ``style="subtask"`` rows into ``{text, start, end}`` spans.
+
+    Each span's ``end`` is the next span's ``start``. The final span's
+    ``end`` defaults to its own ``start`` (zero-duration) — pass
+    ``episode_end_t`` to extend it to the episode's last frame instead,
+    which is what downstream consumers (memory, interjection boundary
+    selection) expect.
+
+    Used by the ``plan`` module (plan-update pass) and the
+    ``interjections`` module (interjection anchoring), which both need the
+    same span shape.
+    """
+    sorted_rows = sorted(
+        (r for r in rows if r.get("style") == "subtask"),
+        key=lambda r: float(r["timestamp"]),
+    )
+    spans: list[dict[str, Any]] = []
+    for r in sorted_rows:
+        t = float(r["timestamp"])
+        if spans:
+            spans[-1]["end"] = t
+        spans.append({"text": r.get("content") or "", "start": t, "end": t})
+    if spans and episode_end_t is not None and float(episode_end_t) > spans[-1]["start"]:
+        spans[-1]["end"] = float(episode_end_t)
+    return spans
+
+
+def snap_to_frame(t: float, frame_timestamps: Sequence[float]) -> float:
+    """Snap an arbitrary float to the nearest exact source frame timestamp.
+
+    Modules use this when emitting event-style rows so the row's
+    timestamp matches a real parquet frame: event rows must land on an
+    exact frame, otherwise the per-frame event lookup the writer does
+    would never match them.
+    """
+    if not frame_timestamps:
+        return float(t)
+    nearest = min(frame_timestamps, key=lambda f: abs(f - t))
+    return float(nearest)
+
+
+def _load_tasks_lookup(root: Path) -> dict[int, str]:
+    """Map ``task_index -> task`` from ``meta/tasks.parquet``.
+
+    Returns an empty dict when the file is absent — the task description is
+    derived later from the video if needed. Reuses the library-level
+    :func:`lerobot.datasets.io_utils.load_tasks`, which returns the tasks
+    frame indexed by task string with a ``task_index`` column.
+    """
+    if not (root / DEFAULT_TASKS_PATH).exists():
+        return {}
+    tasks = load_tasks(root)
+    return {int(idx): str(task) for task, idx in zip(tasks.index, tasks["task_index"], strict=True)}
+
+
+def iter_episodes(root: Path, *, only_episodes: tuple[int, ...] | None = None) -> Iterator[EpisodeRecord]:
+    """Yield :class:`EpisodeRecord` for every episode under ``root/data/``.
+
+    Episodes are yielded in ascending ``episode_index`` order. The reader does
+    not assume a specific chunk/file layout: it scans every ``*.parquet``
+    under ``data/`` and groups by ``episode_index``.
+    """
+    tasks = _load_tasks_lookup(root)
+    data_dir = root / "data"
+    parquet_files = sorted(data_dir.rglob("*.parquet"))
+
+    only_set = set(only_episodes) if only_episodes is not None else None
+
+    for path in parquet_files:
+        yield from _iter_one_path(path, tasks, only_set)
+
+
+def _iter_one_path(path: Path, tasks: dict[int, str], only_set: set[int] | None) -> Iterator[EpisodeRecord]:
+    table = pq.read_table(path)
+    names = table.column_names
+    if "episode_index" not in names:
+        return
+    episode_col = table.column("episode_index").to_pylist()
+    timestamp_col = (
+        table.column("timestamp").to_pylist() if "timestamp" in names else [0.0] * len(episode_col)
+    )
+    frame_col = (
+        table.column("frame_index").to_pylist() if "frame_index" in names else list(range(len(episode_col)))
+    )
+    task_col = table.column("task_index").to_pylist() if "task_index" in names else None
+
+    def _build(
+        ep: int,
+        start: int,
+        end: int,
+        task_idx: int | None,
+        ts_buf: list[float],
+        fi_buf: list[int],
+    ) -> EpisodeRecord | None:
+        if only_set is not None and ep not in only_set:
+            return None
+        task = tasks.get(task_idx, "") if task_idx is not None else ""
+        return EpisodeRecord(
+            episode_index=ep,
+            episode_task=task,
+            frame_timestamps=tuple(ts_buf),
+            frame_indices=tuple(fi_buf),
+            data_path=path,
+            row_offset=start,
+            row_count=end - start,
+        )
+
+    cur_ep: int | None = None
+    start_offset = 0
+    ts_buf: list[float] = []
+    fi_buf: list[int] = []
+    cur_task_idx: int | None = None
+
+    for i, ep in enumerate(episode_col):
+        if cur_ep is None:
+            cur_ep = ep
+            start_offset = i
+            ts_buf = [timestamp_col[i]]
+            fi_buf = [frame_col[i]]
+            cur_task_idx = task_col[i] if task_col is not None else None
+            continue
+        if ep != cur_ep:
+            rec = _build(cur_ep, start_offset, i, cur_task_idx, ts_buf, fi_buf)
+            if rec is not None:
+                yield rec
+            cur_ep = ep
+            start_offset = i
+            ts_buf = [timestamp_col[i]]
+            fi_buf = [frame_col[i]]
+            cur_task_idx = task_col[i] if task_col is not None else None
+        else:
+            ts_buf.append(timestamp_col[i])
+            fi_buf.append(frame_col[i])
+
+    if cur_ep is not None:
+        rec = _build(cur_ep, start_offset, len(episode_col), cur_task_idx, ts_buf, fi_buf)
+        if rec is not None:
+            yield rec

src/lerobot/annotations/steerable_pipeline/staging.py

@@ -0,0 +1,92 @@
+#!/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.
+"""Per-episode staging.
+
+Each module writes its raw output as a JSONL file under
+``<staging_dir>/episode_{ep:06d}/<module>.jsonl``. The writer reads back this
+staging tree and partitions rows into the two language columns.
+
+JSONL is preferred over parquet here because the staging artifact is meant to
+be human-inspectable, easy to diff between prompt iterations, and trivially
+appended to. The final dataset format is parquet; staging is just an
+intermediate.
+"""
+
+from __future__ import annotations
+
+import json
+from collections.abc import Iterable
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+ModuleName = str
+
+_MODULES: tuple[ModuleName, ...] = (
+    "plan",
+    "interjections",
+    "vqa",
+)
+
+
+@dataclass
+class EpisodeStaging:
+    """Filesystem layout for a single episode's staged module outputs."""
+
+    root: Path
+    episode_index: int
+
+    @property
+    def episode_dir(self) -> Path:
+        return self.root / f"episode_{self.episode_index:06d}"
+
+    def path_for(self, module: ModuleName) -> Path:
+        if module not in _MODULES:
+            raise ValueError(f"Unknown module {module!r}; expected one of {_MODULES}")
+        return self.episode_dir / f"{module}.jsonl"
+
+    def write(self, module: ModuleName, rows: Iterable[dict[str, Any]]) -> Path:
+        path = self.path_for(module)
+        path.parent.mkdir(parents=True, exist_ok=True)
+        # Atomic replace: a crash mid-write would otherwise leave a
+        # half-written JSONL file that ``read()`` would then fail to
+        # parse. Write to a sibling .tmp and rename so the target path
+        # only ever points at a complete file.
+        tmp_path = path.with_suffix(path.suffix + ".tmp")
+        with tmp_path.open("w", encoding="utf-8") as f:
+            for row in rows:
+                f.write(json.dumps(row, ensure_ascii=False, sort_keys=True))
+                f.write("\n")
+        tmp_path.replace(path)
+        return path
+
+    def read(self, module: ModuleName) -> list[dict[str, Any]]:
+        path = self.path_for(module)
+        if not path.exists():
+            return []
+        out: list[dict[str, Any]] = []
+        with path.open(encoding="utf-8") as f:
+            for line in f:
+                line = line.strip()
+                if line:
+                    out.append(json.loads(line))
+        return out
+
+    def read_all(self) -> dict[ModuleName, list[dict[str, Any]]]:
+        return {m: self.read(m) for m in _MODULES}
+
+    def has(self, module: ModuleName) -> bool:
+        return self.path_for(module).exists()

src/lerobot/annotations/steerable_pipeline/validator.py

@@ -0,0 +1,332 @@
+#!/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.
+"""Pre-write validation against staged outputs.
+
+Runs after all three modules have written their per-episode artifacts but
+*before* the writer rewrites parquet shards. The validator never touches
+parquet; it only inspects the staging tree and the source frame timestamps
+exposed by :class:`EpisodeRecord`.
+
+Checks (per the plan's "Intermediate staging and validation" section):
+
+- exact timestamp alignment against source frame timestamps
+- no orphan speech / interjection pairs
+- plan / memory emission consistency (events have a paired persistent row)
+- VQA assistant ``content`` is valid JSON (one of bbox / keypoint / count /
+  attribute / spatial)
+- every row maps to its correct column under :func:`column_for_style`
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from collections.abc import Iterable, Sequence
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+from lerobot.datasets.language import (
+    LANGUAGE_EVENTS,
+    LANGUAGE_PERSISTENT,
+    column_for_style,
+    is_view_dependent_style,
+    validate_camera_field,
+)
+
+from .reader import EpisodeRecord
+from .staging import EpisodeStaging
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class ValidationReport:
+    """Outcome of one validation pass across all episodes."""
+
+    errors: list[str] = field(default_factory=list)
+    warnings: list[str] = field(default_factory=list)
+    episodes_checked: int = 0
+
+    @property
+    def ok(self) -> bool:
+        return not self.errors
+
+    def add_error(self, message: str) -> None:
+        self.errors.append(message)
+
+    def add_warning(self, message: str) -> None:
+        self.warnings.append(message)
+
+    def summary(self) -> str:
+        return f"checked={self.episodes_checked} errors={len(self.errors)} warnings={len(self.warnings)}"
+
+
+VQA_ANSWER_SHAPES: dict[str, set[str]] = {
+    "bbox": {"detections"},
+    "keypoint": {"label", "point_format", "point"},
+    "count": {"label", "count"},
+    "attribute": {"label", "attribute", "value"},
+    "spatial": {"subject", "relation", "object"},
+}
+
+
+def classify_vqa_answer(payload: Any) -> str | None:
+    """Best-effort classification of a VQA answer payload to a question type."""
+    if not isinstance(payload, dict):
+        return None
+    keys = set(payload.keys())
+    for kind, required in VQA_ANSWER_SHAPES.items():
+        if required.issubset(keys):
+            return kind
+    return None
+
+
+@dataclass
+class StagingValidator:
+    """Walks the staging tree and produces a :class:`ValidationReport`."""
+
+    timestamp_atol: float = 0.0  # exact-match by default
+    dataset_camera_keys: tuple[str, ...] | None = None
+    """Known ``observation.images.*`` keys on the dataset. When set, the
+    validator additionally enforces that every view-dependent row's
+    ``camera`` field references one of these keys. Pass ``None`` (default)
+    to skip that cross-check (e.g. in unit tests with no real dataset)."""
+
+    def validate(
+        self,
+        records: Sequence[EpisodeRecord],
+        staging_dir: Path,
+    ) -> ValidationReport:
+        report = ValidationReport()
+        for record in records:
+            self._validate_episode(record, staging_dir, report)
+            report.episodes_checked += 1
+        return report
+
+    def _validate_episode(
+        self,
+        record: EpisodeRecord,
+        staging_dir: Path,
+        report: ValidationReport,
+    ) -> None:
+        staging = EpisodeStaging(staging_dir, record.episode_index)
+        staged = staging.read_all()
+        all_rows: list[dict[str, Any]] = []
+        for module_name, rows in staged.items():
+            for row in rows:
+                row = {**row, "_module": module_name}
+                all_rows.append(row)
+
+        frame_ts = set(record.frame_timestamps)
+
+        events: list[dict[str, Any]] = []
+        persistent: list[dict[str, Any]] = []
+        for row in all_rows:
+            self._check_column_routing(row, report, record.episode_index)
+            self._check_camera_field(row, report, record.episode_index, self.dataset_camera_keys)
+            # ``_check_column_routing`` already recorded any unknown-style error;
+            # don't let the same ``column_for_style`` lookup raise here uncaught.
+            try:
+                column = column_for_style(row.get("style"))
+            except ValueError:
+                continue
+            if column == LANGUAGE_PERSISTENT:
+                persistent.append(row)
+            else:
+                events.append(row)
+
+        for row in events:
+            self._check_event_timestamp_alignment(row, frame_ts, report, record.episode_index)
+
+        self._check_speech_interjection_pairs(events, report, record.episode_index)
+        self._check_plan_memory_consistency(persistent, events, report, record.episode_index)
+        self._check_vqa_json(events, report, record.episode_index)
+        self._check_vqa_uniqueness_per_frame_camera(events, report, record.episode_index)
+
+    def _check_camera_field(
+        self,
+        row: dict[str, Any],
+        report: ValidationReport,
+        episode_index: int,
+        dataset_camera_keys: Sequence[str] | None,
+    ) -> None:
+        """Enforce the camera invariant + that the key matches the dataset's cameras."""
+        style = row.get("style")
+        camera = row.get("camera")
+        try:
+            validate_camera_field(style, camera)
+        except ValueError as exc:
+            report.add_error(f"ep={episode_index} module={row.get('_module')}: {exc}")
+            return
+        if is_view_dependent_style(style) and dataset_camera_keys and camera not in dataset_camera_keys:
+            report.add_error(
+                f"ep={episode_index} module={row.get('_module')}: camera {camera!r} on style "
+                f"{style!r} is not one of the dataset's video keys {sorted(dataset_camera_keys)!r}"
+            )
+
+    def _check_vqa_uniqueness_per_frame_camera(
+        self,
+        events: Iterable[dict[str, Any]],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        """Ensure at most one (vqa, user) and one (vqa, assistant) per (t, camera)."""
+        counts: dict[tuple[float, str, str], int] = {}
+        for row in events:
+            if row.get("style") != "vqa":
+                continue
+            ts = row.get("timestamp")
+            camera = row.get("camera")
+            role = row.get("role")
+            if ts is None or camera is None or role is None:
+                continue  # other validators flag these
+            key = (float(ts), str(camera), str(role))
+            counts[key] = counts.get(key, 0) + 1
+        for (ts, camera, role), n in counts.items():
+            if n > 1:
+                report.add_error(
+                    f"ep={episode_index}: {n} duplicate vqa rows at t={ts} "
+                    f"camera={camera!r} role={role!r}; expected at most one per (t, camera, role)"
+                )
+
+    def _check_column_routing(
+        self,
+        row: dict[str, Any],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        style = row.get("style")
+        module = row.get("_module")
+        try:
+            target_col = column_for_style(style)
+        except ValueError:
+            report.add_error(f"ep={episode_index} module={module}: unknown style {style!r}")
+            return
+        if module == "plan" and target_col != LANGUAGE_PERSISTENT:
+            report.add_error(
+                f"ep={episode_index} module=plan emitted style {style!r} that routes to {target_col} (must be persistent)"
+            )
+        if module in {"interjections", "vqa"} and target_col != LANGUAGE_EVENTS:
+            report.add_error(
+                f"ep={episode_index} module={module} emitted style {style!r} that routes to {target_col} (must be events)"
+            )
+
+    def _check_event_timestamp_alignment(
+        self,
+        row: dict[str, Any],
+        frame_ts: set[float],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        ts = row.get("timestamp")
+        if ts is None:
+            report.add_error(f"ep={episode_index}: event row missing timestamp: {row!r}")
+            return
+        if self.timestamp_atol == 0.0:
+            if float(ts) not in frame_ts:
+                report.add_error(
+                    f"ep={episode_index}: event row timestamp {ts!r} does not match any source frame timestamp"
+                )
+        else:
+            if not any(abs(float(ts) - f) <= self.timestamp_atol for f in frame_ts):
+                report.add_error(
+                    f"ep={episode_index}: event row timestamp {ts!r} not within {self.timestamp_atol}s of any frame"
+                )
+
+    def _check_speech_interjection_pairs(
+        self,
+        events: Iterable[dict[str, Any]],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        speech_ts: dict[float, int] = {}
+        interjection_ts: dict[float, int] = {}
+        for row in events:
+            ts = row.get("timestamp")
+            if ts is None:
+                continue
+            ts_f = float(ts)
+            if row.get("style") is None and row.get("role") == "assistant":
+                speech_ts[ts_f] = speech_ts.get(ts_f, 0) + 1
+            if row.get("style") == "interjection":
+                interjection_ts[ts_f] = interjection_ts.get(ts_f, 0) + 1
+
+        for ts in interjection_ts:
+            if ts not in speech_ts:
+                report.add_error(f"ep={episode_index}: interjection at t={ts} has no paired speech atom")
+
+    def _check_plan_memory_consistency(
+        self,
+        persistent: Sequence[dict[str, Any]],
+        events: Sequence[dict[str, Any]],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        plan_ts = sorted({float(r["timestamp"]) for r in persistent if r.get("style") == "plan"})
+        memory_ts = sorted({float(r["timestamp"]) for r in persistent if r.get("style") == "memory"})
+        subtask_ts = sorted({float(r["timestamp"]) for r in persistent if r.get("style") == "subtask"})
+        interjection_ts = sorted(
+            {
+                float(r["timestamp"])
+                for r in events
+                if r.get("style") == "interjection" and r.get("timestamp") is not None
+            }
+        )
+
+        if persistent and not plan_ts:
+            report.add_warning(f"ep={episode_index}: persistent rows present but no plan emitted")
+        # every interjection should have a same-timestamp plan refresh
+        for ts in interjection_ts:
+            if ts not in set(plan_ts):
+                report.add_error(
+                    f"ep={episode_index}: interjection at t={ts} has no co-timestamped plan update"
+                )
+        # memory should be emitted at subtask boundaries (subset relation)
+        if memory_ts and subtask_ts:
+            mem_set = set(memory_ts)
+            sub_set = set(subtask_ts)
+            stray = sorted(mem_set - sub_set)
+            if stray:
+                report.add_warning(f"ep={episode_index}: memory rows at {stray} not at any subtask boundary")
+
+    def _check_vqa_json(
+        self,
+        events: Iterable[dict[str, Any]],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        for row in events:
+            if row.get("style") != "vqa" or row.get("role") != "assistant":
+                continue
+            content = row.get("content")
+            if content is None:
+                report.add_error(
+                    f"ep={episode_index}: VQA assistant row at t={row.get('timestamp')} has null content"
+                )
+                continue
+            try:
+                payload = json.loads(content)
+            except (TypeError, ValueError) as exc:
+                report.add_error(
+                    f"ep={episode_index}: VQA assistant content not valid JSON at t={row.get('timestamp')}: {exc}"
+                )
+                continue
+            shape = classify_vqa_answer(payload)
+            if shape is None:
+                report.add_error(
+                    f"ep={episode_index}: VQA assistant payload at t={row.get('timestamp')} does not match any known shape: keys={list(payload) if isinstance(payload, dict) else type(payload).__name__}"
+                )

src/lerobot/annotations/steerable_pipeline/vlm_client.py

@@ -0,0 +1,617 @@
+#!/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.
+"""Shared Qwen-VL client.
+
+The pipeline uses a single shared VLM across modules. vLLM is preferred when
+available (high throughput, JSON-guided decoding); transformers is the
+fallback. A ``stub`` backend is used for unit tests so fixtures never call
+into a real model.
+
+The client speaks one method, :meth:`VlmClient.generate_json`, which:
+
+- accepts a list of OpenAI/HF-style multimodal messages,
+- requests JSON output from the server,
+- batches requests transparently,
+- and reprompts once on a JSON parse failure with an inline correction
+  message before raising.
+"""
+
+from __future__ import annotations
+
+import atexit
+import base64
+import io
+import json
+import os
+import shlex
+import signal
+import subprocess
+import sys
+import threading
+import time
+import urllib.request
+from collections.abc import Callable, Sequence
+from concurrent.futures import ThreadPoolExecutor
+from dataclasses import dataclass
+from typing import Any, Protocol
+
+from .config import VlmConfig
+
+
+class VlmClient(Protocol):
+    """Protocol every backend must implement."""
+
+    def generate_json(
+        self,
+        messages_batch: Sequence[Sequence[dict[str, Any]]],
+        *,
+        max_new_tokens: int | None = None,
+        temperature: float | None = None,
+    ) -> list[Any]:
+        """Generate one JSON-decoded response per messages list."""
+
+
+@dataclass
+class StubVlmClient:
+    """Deterministic stub used in unit tests.
+
+    A test passes a callable that maps the *last user message text* (or, if
+    that is empty, the full message list) to a JSON-serializable response.
+    """
+
+    responder: Callable[[Sequence[dict[str, Any]]], Any]
+
+    def generate_json(
+        self,
+        messages_batch: Sequence[Sequence[dict[str, Any]]],
+        *,
+        max_new_tokens: int | None = None,
+        temperature: float | None = None,
+    ) -> list[Any]:
+        return [self.responder(list(messages)) for messages in messages_batch]
+
+
+def _strip_to_json(text: str) -> Any:
+    text = text.strip()
+    # Strip <think>...</think> blocks (Qwen3 Thinking style)
+    while "<think>" in text and "</think>" in text:
+        start = text.find("<think>")
+        end = text.find("</think>", start) + len("</think>")
+        text = (text[:start] + text[end:]).strip()
+    # Strip ```json ... ``` fences from chat-tuned backbones
+    if text.startswith("```"):
+        first = text.find("\n")
+        last = text.rfind("```")
+        if first != -1 and last != -1 and last > first:
+            text = text[first + 1 : last].strip()
+    try:
+        return json.loads(text)
+    except (ValueError, json.JSONDecodeError):
+        pass
+    # Fall back to extracting the first balanced {...} block.
+    obj_text = _extract_first_json_object(text)
+    if obj_text is None:
+        raise json.JSONDecodeError("No JSON object found", text, 0)
+    return json.loads(obj_text)
+
+
+def _extract_first_json_object(text: str) -> str | None:
+    """Return the first balanced ``{...}`` substring, ignoring braces in
+    string literals. Returns ``None`` if no balanced block is found."""
+    start = text.find("{")
+    if start < 0:
+        return None
+    depth = 0
+    in_string = False
+    escape = False
+    for i in range(start, len(text)):
+        ch = text[i]
+        if escape:
+            escape = False
+            continue
+        if ch == "\\":
+            escape = True
+            continue
+        # Note: ``escape`` is always False here — the ``if escape`` branch
+        # above already handled and reset it.
+        if ch == '"':
+            in_string = not in_string
+            continue
+        if in_string:
+            continue
+        if ch == "{":
+            depth += 1
+        elif ch == "}":
+            depth -= 1
+            if depth == 0:
+                return text[start : i + 1]
+    return None
+
+
+@dataclass
+class _GenericTextClient:
+    """Wraps any text-generation callable in JSON-mode + one-retry semantics."""
+
+    generate_text: Callable[[Sequence[Sequence[dict[str, Any]]], int, float], list[str]]
+    config: VlmConfig
+
+    def generate_json(
+        self,
+        messages_batch: Sequence[Sequence[dict[str, Any]]],
+        *,
+        max_new_tokens: int | None = None,
+        temperature: float | None = None,
+    ) -> list[Any]:
+        max_tok = max_new_tokens if max_new_tokens is not None else self.config.max_new_tokens
+        temp = temperature if temperature is not None else self.config.temperature
+        raw = self.generate_text(messages_batch, max_tok, temp)
+        out: list[Any] = []
+        for messages, text in zip(messages_batch, raw, strict=True):
+            try:
+                out.append(_strip_to_json(text))
+                continue
+            except (ValueError, json.JSONDecodeError):
+                pass
+            retry = list(messages) + [
+                {"role": "assistant", "content": text},
+                {
+                    "role": "user",
+                    "content": (
+                        "Your previous reply was not valid JSON. "
+                        "Reply with strictly valid JSON, no prose, no fences."
+                    ),
+                },
+            ]
+            retry_text = self.generate_text([retry], max_tok, temp)[0]
+            try:
+                out.append(_strip_to_json(retry_text))
+            except (ValueError, json.JSONDecodeError):
+                # After retry: log preview and return None instead of crashing
+                # the whole pipeline. Modules treat None as "skip".
+                preview = retry_text.strip().replace("\n", " ")[:200]
+                print(
+                    f"[vlm] WARNING: failed to parse JSON after retry; preview: {preview!r}",
+                    flush=True,
+                )
+                out.append(None)
+        return out
+
+
+def make_vlm_client(config: VlmConfig) -> VlmClient:
+    """Build the shared VLM client.
+
+    Only the ``openai`` backend is supported for now. The shipped workflow
+    is Hugging Face Jobs (``examples/annotations/run_hf_job.py``): it boots
+    a vLLM server inside the ``vllm/vllm-openai`` image and the pipeline
+    talks to it over the OpenAI-compatible API (``--vlm.backend=openai``,
+    optionally auto-spawning the server via ``auto_serve`` /
+    ``serve_command``). The former in-process ``vllm`` / ``transformers``
+    backends were removed to keep the support surface to the HF Jobs path.
+
+    For ``stub``, construct :class:`StubVlmClient` directly with a responder
+    callable; it is rejected here to make accidental misuse obvious.
+    """
+    if config.backend == "openai":
+        return _make_openai_client(config)
+    if config.backend == "stub":
+        raise ValueError(
+            "Use StubVlmClient(...) directly for the stub backend; make_vlm_client builds real clients."
+        )
+    if config.backend in {"vllm", "transformers"}:
+        raise ValueError(
+            f"backend={config.backend!r} (in-process local model) is not supported for now — "
+            "only backend='openai' (the Hugging Face Jobs flow) is. Run the pipeline via "
+            "examples/annotations/run_hf_job.py, which serves the model with vLLM in the "
+            "vllm/vllm-openai image and talks to it over the OpenAI-compatible API."
+        )
+    raise ValueError(f"Unknown VLM backend: {config.backend!r}")
+
+
+def _make_openai_client(config: VlmConfig) -> VlmClient:
+    """Backend that talks to any OpenAI-compatible server.
+
+    Compatible with ``vllm serve``, ``transformers serve``,
+    ``ktransformers serve``, and hosted endpoints. By default the server
+    is expected to be already running. Set ``auto_serve=True`` to have
+    this client spawn one (default: ``transformers serve``), wait until
+    it's ready, and tear it down on process exit.
+
+    Image blocks ``{"type":"image", "image":<PIL.Image>}`` are
+    auto-converted to ``image_url`` data-URLs. Video blocks
+    ``{"type":"video", "video":[<PIL>...]}`` are forwarded as
+    multi-frame ``video_url`` items where supported.
+    """
+    try:
+        from openai import OpenAI  # type: ignore[import-not-found]
+    except ImportError as exc:
+        raise ImportError(
+            "openai package is required for backend='openai'. Install with `pip install openai`."
+        ) from exc
+
+    api_base = config.api_base
+    api_key = config.api_key
+    auto_serve = config.auto_serve
+    api_bases: list[str] = [api_base]
+
+    print(
+        f"[lerobot-annotate] backend=openai model={config.model_id} "
+        f"api_base={api_base} auto_serve={auto_serve}",
+        flush=True,
+    )
+    if auto_serve:
+        if config.parallel_servers > 1:
+            print(
+                f"[lerobot-annotate] spawning {config.parallel_servers} parallel servers",
+                flush=True,
+            )
+            api_bases = _spawn_parallel_inference_servers(config)
+        elif _server_is_up(api_base):
+            print(f"[lerobot-annotate] reusing server already up at {api_base}", flush=True)
+        else:
+            print("[lerobot-annotate] no server reachable; spawning one", flush=True)
+            api_base = _spawn_inference_server(config)
+            api_bases = [api_base]
+            print(f"[lerobot-annotate] server ready at {api_base}", flush=True)
+
+    clients = [OpenAI(base_url=base, api_key=api_key) for base in api_bases]
+    # round-robin counter for parallel mode
+    rr_counter = {"i": 0}
+
+    # ``mm_processor_kwargs`` is a vllm-specific extra; transformers serve
+    # rejects it with HTTP 422. Send it only when explicitly opted in via
+    # an env var (e.g. ``LEROBOT_OPENAI_SEND_MM_KWARGS=1`` for vllm).
+    send_mm_kwargs = os.environ.get("LEROBOT_OPENAI_SEND_MM_KWARGS", "").lower() in {"1", "true", "yes"}
+
+    rr_lock = threading.Lock()
+
+    def _one_call(messages: Sequence[dict[str, Any]], max_tok: int, temp: float) -> str:
+        api_messages, mm_kwargs = _to_openai_messages(messages)
+        kwargs: dict[str, Any] = {
+            "model": config.model_id,
+            "messages": api_messages,
+            "max_tokens": max_tok,
+            "temperature": temp,
+        }
+        extra_body: dict[str, Any] = {}
+        if send_mm_kwargs and mm_kwargs:
+            extra_body["mm_processor_kwargs"] = {**mm_kwargs, "do_sample_frames": True}
+        if config.chat_template_kwargs:
+            extra_body["chat_template_kwargs"] = config.chat_template_kwargs
+        if extra_body:
+            kwargs["extra_body"] = extra_body
+        with rr_lock:
+            chosen = clients[rr_counter["i"] % len(clients)]
+            rr_counter["i"] += 1
+        response = chosen.chat.completions.create(**kwargs)
+        return response.choices[0].message.content or ""
+
+    def _gen(batch: Sequence[Sequence[dict[str, Any]]], max_tok: int, temp: float) -> list[str]:
+        if len(batch) <= 1 or config.client_concurrency <= 1:
+            return [_one_call(messages, max_tok, temp) for messages in batch]
+        # Parallel fan-out — vllm batches these on the server side.
+        max_workers = min(config.client_concurrency, len(batch))
+        with ThreadPoolExecutor(max_workers=max_workers) as pool:
+            futures = [pool.submit(_one_call, messages, max_tok, temp) for messages in batch]
+            return [f.result() for f in futures]
+
+    return _GenericTextClient(_gen, config)
+
+
+def _bind_serve_port(cmd: str, port: int) -> str:
+    """Bind a serve command to ``port``: substitute a ``{port}`` placeholder
+    if present, else append ``--port`` when the command omits it (leaving an
+    explicit ``--port`` untouched). Shared by the single- and parallel-server
+    paths so a serve_command never reaches the server with a literal
+    ``{port}``."""
+    if "{port}" in cmd:
+        return cmd.replace("{port}", str(port))
+    if "--port" not in cmd:
+        return f"{cmd} --port {port}"
+    return cmd
+
+
+def _spawn_parallel_inference_servers(config: VlmConfig) -> list[str]:
+    """Spawn ``config.parallel_servers`` independent vllm replicas.
+
+    Each replica:
+    - is pinned to a single GPU via ``CUDA_VISIBLE_DEVICES``
+    - listens on ``serve_port + i``
+    - is shut down via the same atexit hook as the single-server path
+
+    Returns the list of ``api_base`` URLs the client should round-robin
+    across.
+    """
+    n = config.parallel_servers
+    api_bases: list[str] = []
+    procs: list[subprocess.Popen] = []
+    ready_events: list[threading.Event] = []
+    # Multiple readiness signals — uvicorn's own banner is suppressed at
+    # ``--uvicorn-log-level warning``, so we also accept vllm's own
+    # "Starting vLLM API server" line and the route-listing line. The
+    # HTTP probe below is the ultimate fallback.
+    ready_markers = (
+        "Uvicorn running",
+        "Application startup complete",
+        "Starting vLLM API server",
+        "Available routes are",
+    )
+    # Single lock for all server-stream threads so multibyte chars from
+    # different servers don't interleave and tear UTF-8 sequences.
+    print_lock = threading.Lock()
+
+    base_cmd = config.serve_command or (
+        f"vllm serve {shlex.quote(config.model_id)} "
+        f"--tensor-parallel-size 1 "
+        f"--max-model-len {config.max_model_len or 32768} "
+        f"--uvicorn-log-level warning"
+    )
+
+    num_gpus = config.num_gpus if config.num_gpus > 0 else n
+    for i in range(n):
+        port = config.serve_port + i
+        gpu = i % num_gpus
+        env = os.environ.copy()
+        env["CUDA_VISIBLE_DEVICES"] = str(gpu)
+        cmd = _bind_serve_port(base_cmd, port)
+        api_base = f"http://localhost:{port}/v1"
+        api_bases.append(api_base)
+        print(f"[server-{i}] launching on GPU {gpu} port {port}: {cmd}", flush=True)
+        proc = subprocess.Popen(
+            shlex.split(cmd),
+            stdout=subprocess.PIPE,
+            stderr=subprocess.STDOUT,
+            text=True,
+            bufsize=1,
+            env=env,
+        )
+        procs.append(proc)
+        ready = threading.Event()
+        ready_events.append(ready)
+
+        def _stream(idx: int, p: subprocess.Popen, ev: threading.Event) -> None:
+            # Read whole lines and emit each line atomically under the
+            # shared print_lock so output from N servers stays readable.
+            assert p.stdout is not None
+            for line in iter(p.stdout.readline, ""):
+                with print_lock:
+                    sys.stdout.write(f"[server-{idx}] {line}")
+                    if not line.endswith(("\n", "\r")):
+                        sys.stdout.write("\n")
+                    sys.stdout.flush()
+                if any(m in line for m in ready_markers):
+                    ev.set()
+
+        threading.Thread(target=_stream, args=(i, proc, ready), daemon=True).start()
+
+        def _probe(idx: int, base: str, ev: threading.Event, p: subprocess.Popen) -> None:
+            while not ev.is_set() and p.poll() is None:
+                if _server_is_up(base):
+                    print(f"[server-{idx}] ready (http probe)", flush=True)
+                    ev.set()
+                    return
+                time.sleep(2)
+
+        threading.Thread(target=_probe, args=(i, api_base, ready, proc), daemon=True).start()
+
+    def _shutdown() -> None:
+        for i, p in enumerate(procs):
+            if p.poll() is None:
+                print(f"[server-{i}] stopping pid={p.pid}", flush=True)
+                p.send_signal(signal.SIGINT)
+        for p in procs:
+            try:
+                p.wait(timeout=15)
+            except subprocess.TimeoutExpired:
+                p.kill()
+                p.wait(timeout=5)
+
+    atexit.register(_shutdown)
+
+    deadline = time.monotonic() + config.serve_ready_timeout_s
+    while any(not ev.is_set() for ev in ready_events) and time.monotonic() < deadline:
+        for i, p in enumerate(procs):
+            if p.poll() is not None:
+                raise RuntimeError(
+                    f"[server-{i}] inference server exited unexpectedly with rc={p.returncode}"
+                )
+        time.sleep(2)
+    if any(not ev.is_set() for ev in ready_events):
+        raise RuntimeError(f"[server] not all replicas became ready within {config.serve_ready_timeout_s}s")
+    print(f"[lerobot-annotate] all {n} servers ready: {api_bases}", flush=True)
+    return api_bases
+
+
+def _server_is_up(api_base: str) -> bool:
+    """Return True if ``api_base/models`` answers 200 within 2 seconds."""
+    url = api_base.rstrip("/") + "/models"
+    # ``api_base`` is the user-configured local-server URL we just spawned
+    # or the user passed in via ``--vlm.api_base``; the bandit B310 warning
+    # is for arbitrary user-controlled URLs with file:/ schemes which
+    # cannot reach this code path.
+    try:
+        with urllib.request.urlopen(url, timeout=2) as resp:  # noqa: S310  # nosec B310
+            return resp.status == 200
+    except Exception:  # noqa: BLE001
+        return False
+
+
+def _spawn_inference_server(config: VlmConfig) -> str:
+    """Spawn ``transformers serve`` (or ``serve_command``), wait until it
+    accepts ``/v1/models``, and register a shutdown hook.
+
+    Streams the server's stdout/stderr to the parent terminal in
+    real-time on a background thread so users can see model-load
+    progress and errors as they happen.
+
+    Returns the full ``api_base`` URL the OpenAI client should use.
+    """
+    cmd = config.serve_command
+    if not cmd:
+        cmd = (
+            f"transformers serve {shlex.quote(config.model_id)} "
+            f"--port {config.serve_port} --continuous-batching"
+        )
+    # Bind the single server to ``serve_port`` (what ``api_base`` below
+    # targets): substitute a literal ``{port}`` placeholder, else append
+    # ``--port``. Without this a serve_command carrying ``{port}`` would
+    # reach the server unsubstituted and fail to parse.
+    cmd = _bind_serve_port(cmd, config.serve_port)
+    api_base = f"http://localhost:{config.serve_port}/v1"
+    print(f"[server] launching: {cmd}", flush=True)
+    proc = subprocess.Popen(
+        shlex.split(cmd),
+        stdout=subprocess.PIPE,
+        stderr=subprocess.STDOUT,
+        text=True,
+        bufsize=1,
+    )
+
+    # Watch the server output for the uvicorn readiness banner. This is
+    # more reliable than polling /v1/models because transformers serve
+    # rescans its cache on every model-list request, which can exceed
+    # the urllib timeout and trigger an infinite probe loop.
+    ready_event = threading.Event()
+    # See _spawn_parallel_inference_servers for why we accept these.
+    ready_markers = (
+        "Uvicorn running",
+        "Application startup complete",
+        "Starting vLLM API server",
+        "Available routes are",
+    )
+
+    def _probe() -> None:
+        while not ready_event.is_set() and proc.poll() is None:
+            if _server_is_up(api_base):
+                print("[server] ready (http probe)", flush=True)
+                ready_event.set()
+                return
+            time.sleep(2)
+
+    threading.Thread(target=_probe, daemon=True).start()
+
+    def _stream_output() -> None:
+        # Read raw chunks instead of iterating lines so tqdm progress
+        # bars (which overwrite using \r) flush in real time.
+        assert proc.stdout is not None
+        buf = ""
+        prefix_started = False
+        while True:
+            ch = proc.stdout.read(1)
+            if ch == "":
+                # process exited; flush any tail
+                if buf:
+                    sys.stdout.write(buf)
+                    sys.stdout.flush()
+                return
+            if not prefix_started:
+                sys.stdout.write("[server] ")
+                prefix_started = True
+            sys.stdout.write(ch)
+            sys.stdout.flush()
+            buf += ch
+            if ch in ("\n", "\r"):
+                if any(marker in buf for marker in ready_markers):
+                    ready_event.set()
+                buf = ""
+                prefix_started = False
+
+    threading.Thread(target=_stream_output, daemon=True).start()
+
+    def _shutdown() -> None:
+        if proc.poll() is None:
+            print(f"[server] stopping pid={proc.pid}", flush=True)
+            proc.send_signal(signal.SIGINT)
+            try:
+                proc.wait(timeout=15)
+            except subprocess.TimeoutExpired:
+                proc.kill()
+                proc.wait(timeout=5)
+
+    atexit.register(_shutdown)
+
+    deadline = time.monotonic() + config.serve_ready_timeout_s
+    while time.monotonic() < deadline:
+        if proc.poll() is not None:
+            raise RuntimeError(
+                f"[server] inference server exited unexpectedly with rc={proc.returncode}. "
+                f"See [server] log lines above for the cause."
+            )
+        if ready_event.wait(timeout=2):
+            return api_base
+    proc.terminate()
+    raise RuntimeError(f"[server] did not become ready within {config.serve_ready_timeout_s}s")
+
+
+def _to_openai_messages(
+    messages: Sequence[dict[str, Any]],
+) -> tuple[list[dict[str, Any]], dict[str, Any]]:
+    """Convert internal messages to OpenAI chat format.
+
+    Returns ``(api_messages, mm_kwargs)``. Multimodal-processor kwargs
+    (``fps`` from ``video_url`` blocks) are extracted out so the caller
+    can pass them via ``extra_body.mm_processor_kwargs`` rather than
+    inside the content blocks (which transformers serve rejects).
+
+    File-URL video blocks are inlined as base64 data URLs.
+    """
+    out_messages: list[dict[str, Any]] = []
+    mm_kwargs: dict[str, Any] = {}
+    for message in messages:
+        content = message.get("content")
+        if not isinstance(content, list):
+            out_messages.append({"role": message["role"], "content": content})
+            continue
+        out_blocks: list[dict[str, Any]] = []
+        for block in content:
+            block_type = block.get("type") if isinstance(block, dict) else None
+            if block_type == "text":
+                out_blocks.append({"type": "text", "text": block.get("text", "")})
+            elif block_type == "image":
+                out_blocks.append(
+                    {"type": "image_url", "image_url": {"url": _pil_to_data_url(block["image"])}}
+                )
+            elif block_type == "video":
+                frames = block.get("video", [])
+                for img in frames:
+                    out_blocks.append({"type": "image_url", "image_url": {"url": _pil_to_data_url(img)}})
+            elif block_type == "video_url":
+                video_url = dict(block["video_url"])
+                url = video_url.get("url", "")
+                if url.startswith("file://"):
+                    video_url["url"] = _file_to_data_url(url[len("file://") :])
+                out_blocks.append({"type": "video_url", "video_url": video_url})
+                fps = block.get("fps")
+                if fps is not None:
+                    mm_kwargs["fps"] = fps
+            else:
+                out_blocks.append(block)
+        out_messages.append({"role": message["role"], "content": out_blocks})
+    return out_messages, mm_kwargs
+
+
+def _file_to_data_url(path: str) -> str:
+    """Read a local video file and return a base64 ``data:video/mp4`` URL."""
+    with open(path, "rb") as f:
+        b64 = base64.b64encode(f.read()).decode("ascii")
+    return f"data:video/mp4;base64,{b64}"
+
+
+def _pil_to_data_url(image: Any) -> str:
+    """Encode a PIL.Image as a base64 data URL."""
+    buf = io.BytesIO()
+    image.save(buf, format="PNG")
+    b64 = base64.b64encode(buf.getvalue()).decode("ascii")
+    return f"data:image/png;base64,{b64}"

src/lerobot/annotations/steerable_pipeline/writer.py

@@ -0,0 +1,341 @@
+#!/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.
+"""Final parquet rewrite.
+
+For every episode the writer:
+
+1. reads the staged module outputs,
+2. partitions them into a persistent slice (PERSISTENT_STYLES) and an event
+   slice (EVENT_ONLY_STYLES + style=None tool-call atoms),
+3. sorts each slice deterministically,
+4. broadcasts the persistent slice across every frame in the episode,
+5. for each frame, materializes the sublist of event rows whose timestamp
+   exactly equals that frame's timestamp,
+6. drops the legacy ``subtask_index`` column,
+7. writes the parquet shard back in place.
+
+The writer does NOT add a dataset-level ``tools`` column. Tool *calls* are
+emitted per-row via the existing ``tool_calls`` field on the v3.1 row
+struct for every speech atom. The tool *schema* (the description
+of the ``say`` function and its parameters) is a fixed code constant —
+``SAY_TOOL_SCHEMA`` below — and downstream chat-template consumers import
+it directly rather than reading a redundant per-row column.
+
+Invariants enforced here (and re-checked by the validator):
+
+- per-episode persistent slice is byte-identical across every frame;
+- ``language_events`` rows on a frame all have ``timestamp == frame_ts``
+  (timestamps come straight from the source parquet — never recomputed);
+- every row passes ``column_for_style(style)``.
+"""
+
+from __future__ import annotations
+
+import logging
+from collections import defaultdict
+from collections.abc import Sequence
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+from lerobot.datasets.io_utils import write_table_one_row_group_per_episode
+from lerobot.datasets.language import (
+    EVENT_ONLY_STYLES,
+    LANGUAGE_EVENTS,
+    LANGUAGE_PERSISTENT,
+    PERSISTENT_STYLES,
+    column_for_style,
+    validate_camera_field,
+)
+
+from .reader import EpisodeRecord
+from .staging import EpisodeStaging
+
+logger = logging.getLogger(__name__)
+
+
+# Tool schema constants live in lerobot.datasets.language — single
+# source of truth. Re-exported here so existing imports
+# (``from lerobot.annotations.steerable_pipeline.writer import SAY_TOOL_SCHEMA``)
+# keep working.
+from lerobot.datasets.language import DEFAULT_TOOLS, SAY_TOOL_SCHEMA  # noqa: F401, E402
+
+
+def _row_persistent_sort_key(row: dict[str, Any]) -> tuple:
+    return (float(row["timestamp"]), row.get("style") or "", row.get("role") or "")
+
+
+def _row_event_sort_key(row: dict[str, Any]) -> tuple:
+    # events are bucketed per-frame, but within a frame we still want determinism
+    return (
+        row.get("style") or "",
+        row.get("role") or "",
+        row.get("camera") or "",
+    )
+
+
+def _normalize_row(row: dict[str, Any], style: str | None, *, with_timestamp: bool) -> dict[str, Any]:
+    """Coerce a staged row into the language-column struct shape.
+
+    Key order matches ``PERSISTENT_ROW_FIELDS`` / ``EVENT_ROW_FIELDS`` — the
+    writer infers the parquet struct schema from insertion order, so
+    ``timestamp`` (persistent rows only) sits between ``style`` and ``camera``.
+    """
+    camera = row.get("camera")
+    validate_camera_field(style, camera)
+    out: dict[str, Any] = {
+        "role": str(row["role"]),
+        "content": None if row.get("content") is None else str(row["content"]),
+        "style": style,
+    }
+    if with_timestamp:
+        out["timestamp"] = float(row["timestamp"])
+    out["camera"] = None if camera is None else str(camera)
+    out["tool_calls"] = _normalize_tool_calls(row.get("tool_calls"))
+    return out
+
+
+def _normalize_persistent_row(row: dict[str, Any]) -> dict[str, Any]:
+    """Coerce a staged row into the persistent column's struct shape."""
+    style = row.get("style")
+    if style not in PERSISTENT_STYLES:
+        raise ValueError(
+            f"persistent slice contains row with non-persistent style {style!r}; "
+            "row would be misrouted under column_for_style()"
+        )
+    if "timestamp" not in row:
+        raise ValueError(f"persistent row missing timestamp: {row!r}")
+    if "role" not in row:
+        # Friendly error from the writer instead of a raw KeyError below;
+        # the validator doesn't check ``role`` yet.
+        raise ValueError(f"persistent row missing role: {row!r}")
+    return _normalize_row(row, style, with_timestamp=True)
+
+
+def _normalize_event_row(row: dict[str, Any]) -> dict[str, Any]:
+    """Coerce a staged row into the event column's struct shape (no timestamp)."""
+    style = row.get("style")
+    if style is not None and style not in EVENT_ONLY_STYLES:
+        raise ValueError(
+            f"event slice contains row with style {style!r}; expected None or one of {EVENT_ONLY_STYLES}"
+        )
+    if column_for_style(style) != LANGUAGE_EVENTS:
+        raise ValueError(f"event row with style {style!r} would not route to language_events")
+    if "role" not in row:
+        raise ValueError(f"event row missing role: {row!r}")
+    return _normalize_row(row, style, with_timestamp=False)
+
+
+def _normalize_tool_calls(value: Any) -> list[Any] | None:
+    if value is None:
+        return None
+    if not isinstance(value, list):
+        raise ValueError(f"tool_calls must be a list or None, got {type(value).__name__}")
+    return list(value)
+
+
+def _validate_atom_invariants(row: dict[str, Any]) -> None:
+    """At-least-one of content/tool_calls; style=None implies tool_calls."""
+    has_content = row.get("content") is not None
+    has_tools = row.get("tool_calls") is not None
+    if not (has_content or has_tools):
+        raise ValueError(f"row has neither content nor tool_calls: {row!r}")
+    if row.get("style") is None and not has_tools:
+        raise ValueError(f"style=None requires tool_calls: {row!r}")
+
+
+def _validate_speech_atom(row: dict[str, Any]) -> None:
+    """Speech atoms: role=assistant, style=None, content=None, say tool call."""
+    if row.get("style") is not None:
+        return  # not a speech atom
+    if row.get("role") != "assistant":
+        raise ValueError(f"speech atom must have role=assistant: {row!r}")
+    if row.get("content") is not None:
+        raise ValueError(f"speech atom must have content=null: {row!r}")
+    tool_calls = row.get("tool_calls")
+    if not tool_calls or not isinstance(tool_calls, list):
+        raise ValueError(f"speech atom must have non-empty tool_calls list: {row!r}")
+    first = tool_calls[0]
+    if not isinstance(first, dict):
+        raise ValueError(f"speech atom tool_calls[0] must be a dict: {row!r}")
+    if first.get("type") != "function":
+        raise ValueError(f"speech atom tool_calls[0].type must be 'function': {row!r}")
+    fn = first.get("function") or {}
+    if fn.get("name") != "say":
+        raise ValueError(f"speech atom tool_calls[0].function.name must be 'say': {row!r}")
+    args = fn.get("arguments") or {}
+    if not isinstance(args, dict) or "text" not in args or not isinstance(args["text"], str):
+        raise ValueError(f"speech atom must carry 'text' string in arguments: {row!r}")
+
+
+@dataclass
+class LanguageColumnsWriter:
+    """Rewrite ``data/chunk-*/file-*.parquet`` with the two language columns."""
+
+    drop_existing_subtask_index: bool = True
+
+    def write_all(
+        self,
+        records: Sequence[EpisodeRecord],
+        staging_dir: Path,
+        root: Path,
+    ) -> list[Path]:
+        episodes_by_path: dict[Path, list[EpisodeRecord]] = defaultdict(list)
+        for record in records:
+            episodes_by_path[record.data_path].append(record)
+
+        written: list[Path] = []
+        for path, eps in episodes_by_path.items():
+            self._rewrite_one(path, eps, staging_dir, root)
+            written.append(path)
+        return written
+
+    def _rewrite_one(
+        self,
+        path: Path,
+        episodes: Sequence[EpisodeRecord],
+        staging_dir: Path,
+        root: Path,
+    ) -> None:
+        table = pq.read_table(path)
+        n_rows = table.num_rows
+
+        # Ensure we cover every episode in the file. Episodes that don't have
+        # staging artifacts are passed through with empty annotation lists —
+        # this keeps the writer idempotent and safe for partial reruns.
+        staged_per_ep: dict[int, dict[str, list[dict[str, Any]]]] = {}
+        for record in episodes:
+            staging = EpisodeStaging(staging_dir, record.episode_index)
+            staged_per_ep[record.episode_index] = staging.read_all()
+
+        persistent_by_ep: dict[int, list[dict[str, Any]]] = {}
+        events_by_ep_ts: dict[int, dict[float, list[dict[str, Any]]]] = {}
+
+        for ep_index, ep_staged in staged_per_ep.items():
+            persistent_rows: list[dict[str, Any]] = []
+            event_rows: list[dict[str, Any]] = []  # carry timestamp until bucketed
+            for _module_name, rows in ep_staged.items():
+                for row in rows:
+                    style = row.get("style")
+                    if column_for_style(style) == LANGUAGE_PERSISTENT:
+                        persistent_rows.append(row)
+                    else:
+                        event_rows.append(row)
+
+            persistent_rows.sort(key=_row_persistent_sort_key)
+            normalized_persistent = []
+            for r in persistent_rows:
+                _validate_atom_invariants(r)
+                _validate_speech_atom(r)
+                normalized_persistent.append(_normalize_persistent_row(r))
+            persistent_by_ep[ep_index] = normalized_persistent
+
+            buckets: dict[float, list[dict[str, Any]]] = defaultdict(list)
+            for r in event_rows:
+                _validate_atom_invariants(r)
+                _validate_speech_atom(r)
+                ts = float(r["timestamp"])
+                buckets[ts].append(_normalize_event_row(r))
+            for ts in list(buckets.keys()):
+                buckets[ts].sort(key=_row_event_sort_key)
+            events_by_ep_ts[ep_index] = buckets
+
+        episode_col = (
+            table.column("episode_index").to_pylist() if "episode_index" in table.column_names else None
+        )
+        ts_col = table.column("timestamp").to_pylist() if "timestamp" in table.column_names else None
+        if episode_col is None or ts_col is None:
+            raise ValueError(f"{path} is missing 'episode_index' or 'timestamp' — required by the writer.")
+
+        per_row_persistent: list[list[dict[str, Any]]] = []
+        per_row_events: list[list[dict[str, Any]]] = []
+        for i in range(n_rows):
+            ep = episode_col[i]
+            ts = float(ts_col[i])
+            per_row_persistent.append(persistent_by_ep.get(ep, []))
+            buckets = events_by_ep_ts.get(ep, {})
+            per_row_events.append(buckets.get(ts, []))
+
+        new_table = self._materialize_table(
+            table, per_row_persistent, per_row_events, drop_old=self.drop_existing_subtask_index
+        )
+        # Re-emit one row group per episode (a bulk pq.write_table would collapse
+        # them into one). Write to a sibling tmp path and atomically rename so a
+        # crash mid-write can't leave a half-written shard.
+        tmp_path = path.with_suffix(path.suffix + ".tmp")
+        write_table_one_row_group_per_episode(new_table, tmp_path)
+        tmp_path.replace(path)
+
+    def _materialize_table(
+        self,
+        table: pa.Table,
+        persistent: list[list[dict[str, Any]]],
+        events: list[list[dict[str, Any]]],
+        *,
+        drop_old: bool,
+    ) -> pa.Table:
+        cols = []
+        names = []
+        for name in table.column_names:
+            if drop_old and name == "subtask_index":
+                continue
+            if name in (LANGUAGE_PERSISTENT, LANGUAGE_EVENTS):
+                continue  # we'll re-add canonical versions
+            # Strip any legacy ``tools`` column previously emitted by older
+            # writers — the schema no longer uses it (constant lives in
+            # SAY_TOOL_SCHEMA / DEFAULT_TOOLS).
+            if name == "tools":
+                continue
+            cols.append(table.column(name))
+            names.append(name)
+
+        # We let pyarrow infer struct/list schema rather than passing the
+        # canonical type from `lerobot.datasets.language` directly: that type
+        # uses `pa.json_()` for the `tool_calls` element type, which
+        # `pa.array(..., type=...)` cannot materialize from Python lists on
+        # current pyarrow versions. The inferred schema round-trips through
+        # parquet and `LeRobotDataset` correctly — `tests/datasets/test_language.py`
+        # exercises the same flow.
+        persistent_arr = pa.array(persistent)
+        events_arr = pa.array(events)
+
+        cols.extend([persistent_arr, events_arr])
+        names.extend([LANGUAGE_PERSISTENT, LANGUAGE_EVENTS])
+
+        return pa.Table.from_arrays(cols, names=names)
+
+
+def speech_atom(timestamp: float, text: str) -> dict[str, Any]:
+    """Build a canonical speech tool-call atom for the events column."""
+    return {
+        "role": "assistant",
+        "content": None,
+        "style": None,
+        "timestamp": float(timestamp),
+        "camera": None,
+        "tool_calls": [
+            {
+                "type": "function",
+                "function": {
+                    "name": "say",
+                    "arguments": {"text": text},
+                },
+            }
+        ],
+    }

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

@@ -436,17 +436,18 @@ 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
 
         Stops on DeviceNotConnectedError, logs other errors and continues.
         """
-        if self.stop_event is None:
+        stop_event = self.stop_event
+        if stop_event is None:
             raise RuntimeError(f"{self}: stop_event is not initialized before starting read loop.")
 
         failure_count = 0
-        while not self.stop_event.is_set():
+        while not stop_event.is_set():
             try:
                 raw_frame = self._read_from_hardware()
                 processed_frame = self._postprocess_image(raw_frame)
@@ -484,6 +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")
@@ -465,32 +466,38 @@ 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.
         """
-        if self.stop_event is None:
+        stop_event = self.stop_event
+        if stop_event is None:
             raise RuntimeError(f"{self}: stop_event is not initialized before starting read loop.")
 
         failure_count = 0
-        while not self.stop_event.is_set():
+        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
@@ -522,6 +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
@@ -532,7 +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]:
         """
@@ -557,25 +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).
@@ -592,24 +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

@@ -246,11 +246,12 @@ class ZMQCamera(Camera):
         """
         Internal loop run by the background thread for asynchronous reading.
         """
-        if self.stop_event is None:
+        stop_event = self.stop_event
+        if stop_event is None:
             raise RuntimeError(f"{self}: stop_event is not initialized.")
 
         failure_count = 0
-        while not self.stop_event.is_set():
+        while not stop_event.is_set():
             try:
                 frame = self._read_from_hardware()
                 capture_time = time.perf_counter()
@@ -292,6 +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/common/control_utils.py

@@ -17,12 +17,9 @@ from __future__ import annotations
 ########################################################################################
 # Utilities
 ########################################################################################
-import logging
 import time
-import traceback
 from contextlib import nullcontext
 from copy import copy
-from functools import cache
 from typing import TYPE_CHECKING, Any
 
 import numpy as np
@@ -43,34 +40,6 @@ from lerobot.robots import Robot
 from lerobot.types import PolicyAction
 
 
-@cache
-def is_headless():
-    """
-    Detects if the Python script is running in a headless environment (e.g., without a display).
-
-    This function attempts to import `pynput`, a library that requires a graphical environment.
-    If the import fails, it assumes the environment is headless. The result is cached to avoid
-    re-running the check.
-
-    Returns:
-        True if the environment is determined to be headless, False otherwise.
-    """
-    try:
-        import pynput  # noqa
-
-        return False
-    except Exception:
-        print(
-            "Error trying to import pynput. Switching to headless mode. "
-            "As a result, the video stream from the cameras won't be shown, "
-            "and you won't be able to change the control flow with keyboards. "
-            "For more info, see traceback below.\n"
-        )
-        traceback.print_exc()
-        print()
-        return True
-
-
 def predict_action(
     observation: dict[str, np.ndarray],
     policy: PreTrainedPolicy,
@@ -122,59 +91,6 @@ def predict_action(
     return action
 
 
-def init_keyboard_listener():
-    """
-    Initializes a non-blocking keyboard listener for real-time user interaction.
-
-    This function sets up a listener for specific keys (right arrow, left arrow, escape) to control
-    the program flow during execution, such as stopping recording or exiting loops. It gracefully
-    handles headless environments where keyboard listening is not possible.
-
-    Returns:
-        A tuple containing:
-        - The `pynput.keyboard.Listener` instance, or `None` if in a headless environment.
-        - A dictionary of event flags (e.g., `exit_early`) that are set by key presses.
-    """
-    # Allow to exit early while recording an episode or resetting the environment,
-    # by tapping the right arrow key '->'. This might require a sudo permission
-    # to allow your terminal to monitor keyboard events.
-    events = {}
-    events["exit_early"] = False
-    events["rerecord_episode"] = False
-    events["stop_recording"] = False
-
-    if is_headless():
-        logging.warning(
-            "Headless environment detected. On-screen cameras display and keyboard inputs will not be available."
-        )
-        listener = None
-        return listener, events
-
-    # Only import pynput if not in a headless environment
-    from pynput import keyboard
-
-    def on_press(key):
-        try:
-            if key == keyboard.Key.right:
-                print("Right arrow key pressed. Exiting loop...")
-                events["exit_early"] = True
-            elif key == keyboard.Key.left:
-                print("Left arrow key pressed. Exiting loop and rerecord the last episode...")
-                events["rerecord_episode"] = True
-                events["exit_early"] = True
-            elif key == keyboard.Key.esc:
-                print("Escape key pressed. Stopping data recording...")
-                events["stop_recording"] = True
-                events["exit_early"] = True
-        except Exception as e:
-            print(f"Error handling key press: {e}")
-
-    listener = keyboard.Listener(on_press=on_press)
-    listener.start()
-
-    return listener, events
-
-
 def sanity_check_dataset_name(repo_id, policy_cfg):
     """
     Validates the dataset repository name against the presence of a policy configuration.

src/lerobot/common/train_utils.py

@@ -15,12 +15,14 @@
 # limitations under the License.
 from pathlib import Path
 
+from huggingface_hub import HfApi, snapshot_download
 from torch.optim import Optimizer
 from torch.optim.lr_scheduler import LRScheduler
 
 from lerobot.configs.train import TrainPipelineConfig
 from lerobot.optim import (
     load_optimizer_state,
+    load_optimizer_state_dict,
     load_scheduler_state,
     save_optimizer_state,
     save_scheduler_state,
@@ -34,6 +36,7 @@ from lerobot.utils.constants import (
     TRAINING_STATE_DIR,
     TRAINING_STEP,
 )
+from lerobot.utils.hub import find_latest_hub_checkpoint
 from lerobot.utils.io_utils import load_json, write_json
 from lerobot.utils.random_utils import load_rng_state, save_rng_state
 
@@ -49,8 +52,19 @@ def get_step_checkpoint_dir(output_dir: Path, total_steps: int, step: int) -> Pa
     return output_dir / CHECKPOINTS_DIR / step_identifier
 
 
-def save_training_step(step: int, save_dir: Path) -> None:
-    write_json({"step": step}, save_dir / TRAINING_STEP)
+def save_training_step(
+    step: int, save_dir: Path, num_processes: int | None = None, batch_size: int | None = None
+) -> None:
+    state: dict = {"step": step}
+    # num_processes and batch_size are recorded so a resumed run can detect a changed world size or
+    # batch size: the sampler's resume offset is computed from the (num_processes, batch_size) that
+    # produced `step`, since both scale how many sampler positions a step consumes (see
+    # compute_sampler_state).
+    if num_processes is not None:
+        state["num_processes"] = num_processes
+    if batch_size is not None:
+        state["batch_size"] = batch_size
+    write_json(state, save_dir / TRAINING_STEP)
 
 
 def load_training_step(save_dir: Path) -> int:
@@ -58,6 +72,16 @@ def load_training_step(save_dir: Path) -> int:
     return training_step["step"]
 
 
+def load_training_num_processes(checkpoint_dir: Path) -> int | None:
+    """World size recorded at checkpoint time, or None for checkpoints written before it was stored."""
+    return load_json(checkpoint_dir / TRAINING_STATE_DIR / TRAINING_STEP).get("num_processes")
+
+
+def load_training_batch_size(checkpoint_dir: Path) -> int | None:
+    """Per-process batch size recorded at checkpoint time, or None for older checkpoints."""
+    return load_json(checkpoint_dir / TRAINING_STATE_DIR / TRAINING_STEP).get("batch_size")
+
+
 def update_last_checkpoint(checkpoint_dir: Path) -> Path:
     last_checkpoint_dir = checkpoint_dir.parent / LAST_CHECKPOINT_LINK
     if last_checkpoint_dir.is_symlink():
@@ -75,6 +99,10 @@ def save_checkpoint(
     scheduler: LRScheduler | None = None,
     preprocessor: PolicyProcessorPipeline | None = None,
     postprocessor: PolicyProcessorPipeline | None = None,
+    num_processes: int | None = None,
+    batch_size: int | None = None,
+    model_state_dict: dict | None = None,
+    optim_state_dict: dict | None = None,
 ) -> None:
     """This function creates the following directory structure:
 
@@ -100,9 +128,22 @@ def save_checkpoint(
         scheduler (LRScheduler | None, optional): The scheduler to save the state from. Defaults to None.
         preprocessor: The preprocessor/pipeline to save. Defaults to None.
         postprocessor: The postprocessor/pipeline to save. Defaults to None.
+        num_processes (int | None, optional): Distributed world size to record for sample-exact
+            resume. Defaults to None (not recorded).
+        batch_size (int | None, optional): Per-process batch size to record for sample-exact
+            resume. Defaults to None (not recorded).
+        model_state_dict: Pre-gathered full (unsharded) model state dict. Required under FSDP,
+            where `policy.state_dict()` would return sharded tensors; the caller gathers it via a
+            cross-rank collective and passes it here so rank 0 can write it directly. It holds
+            FSDP's fp32 master weights and is saved as-is (the loader casts to the policy dtype on
+            read). When None (DDP / single-GPU), the model is saved the normal way. Defaults to None.
+        optim_state_dict: Pre-gathered full (unsharded) optimizer state dict. Required under FSDP
+            (gathered alongside `model_state_dict` via `gather_fsdp_state_dicts`); saved in the same
+            safetensors format as the single-GPU path. When None, `optimizer.state_dict()` is used.
+            Defaults to None.
     """
     pretrained_dir = checkpoint_dir / PRETRAINED_MODEL_DIR
-    policy.save_pretrained(pretrained_dir)
+    policy.save_pretrained(pretrained_dir, state_dict=model_state_dict)
     cfg.save_pretrained(pretrained_dir)
     if cfg.peft is not None:
         # When using PEFT, policy.save_pretrained will only write the adapter weights + config, not the
@@ -112,7 +153,15 @@ def save_checkpoint(
         preprocessor.save_pretrained(pretrained_dir)
     if postprocessor is not None:
         postprocessor.save_pretrained(pretrained_dir)
-    save_training_state(checkpoint_dir, step, optimizer, scheduler)
+    save_training_state(
+        checkpoint_dir,
+        step,
+        optimizer,
+        scheduler,
+        num_processes=num_processes,
+        batch_size=batch_size,
+        optim_state_dict=optim_state_dict,
+    )
 
 
 def save_training_state(
@@ -120,6 +169,9 @@ def save_training_state(
     train_step: int,
     optimizer: Optimizer | None = None,
     scheduler: LRScheduler | None = None,
+    num_processes: int | None = None,
+    batch_size: int | None = None,
+    optim_state_dict: dict | None = None,
 ) -> None:
     """
     Saves the training step, optimizer state, scheduler state, and rng state.
@@ -131,19 +183,23 @@ def save_training_state(
             Defaults to None.
         scheduler (LRScheduler | None, optional): The scheduler from which to save the state_dict.
             Defaults to None.
+        num_processes (int | None, optional): Distributed world size to record. Defaults to None.
+        batch_size (int | None, optional): Per-process batch size to record. Defaults to None.
+        optim_state_dict: Pre-gathered full optimizer state dict (for FSDP). Saved instead of
+            `optimizer.state_dict()` when provided. Defaults to None.
     """
     save_dir = checkpoint_dir / TRAINING_STATE_DIR
     save_dir.mkdir(parents=True, exist_ok=True)
-    save_training_step(train_step, save_dir)
+    save_training_step(train_step, save_dir, num_processes=num_processes, batch_size=batch_size)
     save_rng_state(save_dir)
     if optimizer is not None:
-        save_optimizer_state(optimizer, save_dir)
+        save_optimizer_state(optimizer, save_dir, optim_state_dict=optim_state_dict)
     if scheduler is not None:
         save_scheduler_state(scheduler, save_dir)
 
 
 def load_training_state(
-    checkpoint_dir: Path, optimizer: Optimizer, scheduler: LRScheduler | None
+    checkpoint_dir: Path, optimizer: Optimizer, scheduler: LRScheduler | None, load_optimizer: bool = True
 ) -> tuple[int, Optimizer, LRScheduler | None]:
     """
     Loads the training step, optimizer state, scheduler state, and rng state.
@@ -153,6 +209,10 @@ def load_training_state(
         checkpoint_dir (Path): The checkpoint directory. Should contain a 'training_state' dir.
         optimizer (Optimizer): The optimizer to load the state_dict to.
         scheduler (LRScheduler | None): The scheduler to load the state_dict to (can be None).
+        load_optimizer (bool, optional): Whether to load the optimizer state from disk. Defaults to
+            True. Set to False under FSDP, where the sharded optimizer state must be loaded after
+            `accelerator.prepare()` via `load_fsdp_optimizer_state` (the optimizer is returned
+            untouched here).
 
     Raises:
         NotADirectoryError: If 'checkpoint_dir' doesn't contain a 'training_state' dir
@@ -167,8 +227,119 @@ def load_training_state(
 
     load_rng_state(training_state_dir)
     step = load_training_step(training_state_dir)
-    optimizer = load_optimizer_state(optimizer, training_state_dir)
+    if load_optimizer:
+        optimizer = load_optimizer_state(optimizer, training_state_dir)
     if scheduler is not None:
         scheduler = load_scheduler_state(scheduler, training_state_dir)
 
     return step, optimizer, scheduler
+
+
+def gather_fsdp_state_dicts(model, optimizer) -> tuple[dict, dict]:
+    """Gather the full (unsharded) model and optimizer state dicts under FSDP.
+
+    `model.state_dict()` and `FSDP.optim_state_dict(...)` are cross-rank collectives, so this must be
+    called on *every* rank with the prepared (FSDP-wrapped) `model` and `optimizer`. With
+    `rank0_only=True` and `offload_to_cpu=True`, every rank runs the all-gather but only rank 0
+    materializes the full dicts (the others get empty dicts) and they are kept on CPU to bound GPU
+    memory. The returned optimizer state dict is keyed by parameter FQNs and is world-size
+    independent; `load_fsdp_optimizer_state` reshards it on resume.
+
+    Returns:
+        (model_state_dict, optim_state_dict): full dicts on rank 0, empty dicts on other ranks.
+    """
+    from torch.distributed.fsdp import (
+        FullOptimStateDictConfig,
+        FullStateDictConfig,
+        FullyShardedDataParallel as FSDP,  # noqa F401
+        StateDictType,
+    )
+
+    state_cfg = FullStateDictConfig(offload_to_cpu=True, rank0_only=True)
+    optim_cfg = FullOptimStateDictConfig(offload_to_cpu=True, rank0_only=True)
+    with FSDP.state_dict_type(model, StateDictType.FULL_STATE_DICT, state_cfg, optim_cfg):
+        model_state_dict = model.state_dict()
+        optim_state_dict = FSDP.optim_state_dict(model, optimizer)
+    return model_state_dict, optim_state_dict
+
+
+def load_fsdp_optimizer_state(model, optimizer, checkpoint_dir: Path) -> None:
+    """Load the FSDP optimizer state (saved as safetensors) and reshard it into the optimizer.
+
+    This is a cross-rank collective and must be called on every rank *after* `accelerator.prepare()`
+    with the prepared (FSDP-wrapped) `model` and `optimizer`. The saved state is the full,
+    world-size-independent optimizer state (keyed by parameter FQNs); `FSDP.optim_state_dict_to_load`
+    reshards it to the current FSDP topology, so resume on a different number of GPUs works.
+    """
+    from torch.distributed.fsdp import (
+        FullOptimStateDictConfig,
+        FullStateDictConfig,
+        FullyShardedDataParallel as FSDP,  # noqa F401
+        StateDictType,
+    )
+
+    # Every rank reads the same full state from the (shared) checkpoint dir, so rank0_only=False.
+    full_osd = load_optimizer_state_dict(checkpoint_dir / TRAINING_STATE_DIR)
+    state_cfg = FullStateDictConfig(rank0_only=False)
+    optim_cfg = FullOptimStateDictConfig(rank0_only=False)
+    with FSDP.state_dict_type(model, StateDictType.FULL_STATE_DICT, state_cfg, optim_cfg):
+        sharded_osd = FSDP.optim_state_dict_to_load(model=model, optim=optimizer, optim_state_dict=full_osd)
+    optimizer.load_state_dict(sharded_osd)
+
+
+def push_checkpoint_to_hub(
+    checkpoint_dir: Path,
+    repo_id: str,
+    *,
+    private: bool | None = None,
+) -> None:
+    """Upload a saved checkpoint directory to the Hub under checkpoints/<name>/.
+
+    Called once per save step when save_checkpoint_to_hub is enabled, so a
+    timed-out or crashed run still leaves recoverable checkpoints on the Hub.
+    The model repo is created idempotently, and the commit is tagged with the
+    checkpoint step so a checkpoint can be recovered with
+    --policy.pretrained_revision=<step> instead of a commit sha.
+    """
+    api = HfApi()
+    api.create_repo(repo_id=repo_id, repo_type="model", private=private, exist_ok=True)
+    commit = api.upload_folder(
+        folder_path=str(checkpoint_dir),
+        repo_id=repo_id,
+        repo_type="model",
+        path_in_repo=f"checkpoints/{checkpoint_dir.name}",
+        commit_message=f"checkpoint {checkpoint_dir.name}",
+    )
+    api.create_tag(
+        repo_id=repo_id,
+        tag=checkpoint_dir.name,
+        revision=commit.oid,
+        repo_type="model",
+        exist_ok=True,
+    )
+
+
+def resolve_resume_checkpoint(repo_id: str, output_dir: Path) -> Path:
+    """Download the latest checkpoint of a Hub training repo into a local run dir.
+
+    The symmetric counterpart to `push_checkpoint_to_hub`: given a model repo holding
+    `checkpoints/<step>/{pretrained_model,training_state}` subtrees, download the highest-numbered step
+    into `output_dir/checkpoints/<step>/`, recreate the local `last` symlink, and return that local
+    checkpoint dir. Used to resume training from the Hub on a machine (or HF Jobs pod) that does not
+    have the original local run dir.
+    """
+    latest = find_latest_hub_checkpoint(repo_id)
+    if latest is None:
+        raise FileNotFoundError(
+            f"No checkpoint found in '{repo_id}' under '{CHECKPOINTS_DIR}/'. "
+            "Was the run trained with --save_checkpoint_to_hub?"
+        )
+    snapshot_download(
+        repo_id=repo_id,
+        repo_type="model",
+        allow_patterns=f"{latest}/*",
+        local_dir=str(output_dir),
+    )
+    checkpoint_dir = output_dir / latest
+    update_last_checkpoint(checkpoint_dir)
+    return checkpoint_dir