feat(envs): Add NVIDIA IsaacLab-Arena Lerobot (#2699)

* adding Isaaclab Arena from collab * adding into lerobot-eval * minor modification * added bash script for env setup * setups * fix applauncher not getting the arguments * data conversion, train and eval smolvla * fixed imports * clean-up * added test suits & clean up - wip * fixed video recording * clean-up * hub integration working * clean-up * added kwargs * Revert "added kwargs" This reverts commit 9b445356385d0707655cf04d02be058b25138119. * added kwargs * clean-up * cleaned unused function * added logging * docs * cleaned up IsaaclabArenaEnv * clean-up * clean-up * clean up * added tests * minor clean-up * fix: support for state based envs * feat(envs): Add NVIDIA IsaacLab Arena integration with LeRobot for policy evaluation at scale * feat(envs): Add IsaacLab Arena integration for policy evaluation Integrate NVIDIA IsaacLab Arena with LeRobot to enable GPU-accelerated simulation through the EnvHub infrastructure. This enables: - Training imitation learning policies (PI0, SmolVLA, etc.) - Evaluating trained policies in with IsaacLab Arena The implementation adds: - IsaaclabArenaEnv config with Arena-specific parameters - IsaaclabArenaProcessorStep for observation processing - Hub loading from nvkartik/isaaclab-arena-envs repository - Video recording support Available environments include GR1 microwave manipulation, Galileo pick-and-place, G1 loco-manipulation, and button pressing tasks. Datasets: nvkartik/Arena-GR1-Manipulation-Task Policies: nvkartik/pi05-arena-gr1-microwave, nvkartik/smolvla-arena-gr1-microwave * added isaaclab arena wrapper and corresponding tests * added error handling * renamed wrapper file: isaaclab_arena to isaaclab * added extra kwarg changes * adjustments for hub envs * correct class name in test file * fixed parsing of env_kwargs * tested end to end * removed unused code * refactor design * shifted IsaacLab to hub * removed IsaacLab tests * docs: Add LW-BenchHub evaluation instructions * docs: Add LW-BenchHub evaluation instructions * docs diet * minor edits to texts * IL Arena commit hash * update links * minor edits * fix numpy version after install of lerobot * links update * valideated on vanilla brev * docs: Add LW-BenchHub evaluation instructions * remove kwargs from all make_env calls * remove kwargs from all make_env calls * fix LW table and indentations * remove environment list from docs * docs: Update lw-benchhub eval config in envhub docs * removing kwargs * removed extra line * ensure pinocchio install for lightwheel + add lightwheel website link * remove env_kwargs * no default empty value for hub_path * not using assert method * remove env_processor defaults * revert and adding default "" value for hub_path * pinning down packages versions * explicit None value for hub_path * Update src/lerobot/configs/eval.py Co-authored-by: Jade Choghari <chogharijade@gmail.com> Signed-off-by: Lior Ben Horin <liorbenhorin@gmail.com> * corrected formatting * corrected job_name var in config * updated docs and namespace * updated namespace * updated docs * updated docs * added hardware requirements * updated docs --------- Signed-off-by: Lior Ben Horin <liorbenhorin@gmail.com> Co-authored-by: lbenhorin <lbenhorin@nvidia.com> Co-authored-by: Lior Ben Horin <liorbenhorin@gmail.com> Co-authored-by: Jade Choghari <chogharijade@gmail.com> Co-authored-by: Steven Palma <imstevenpmwork@ieee.org> Co-authored-by: tianheng.wu <tianheng.wu@lightwheel.ai>
2026-05-24 04:59:47 +00:00 · 2026-01-02 20:36:24 +01:00
parent 9701b9c273
commit fc296548cb
10 changed files with 704 additions and 15 deletions
@@ -59,6 +59,8 @@
    title: Environments from the Hub
  - local: envhub_leisaac
    title: Control & Train Robots in Sim (LeIsaac)
+  - local: envhub_isaaclab_arena
+    title: NVIDIA IsaacLab Arena Environments
  - local: libero
    title: Using Libero
  - local: metaworld
@@ -0,0 +1,474 @@
+# NVIDIA IsaacLab Arena & LeRobot
+
+LeRobot EnvHub now supports **GPU-accelerated simulation** with IsaacLab Arena for policy evaluation at scale.
+Train and evaluate imitation learning policies with high-fidelity simulation — all integrated into the LeRobot ecosystem.
+
+<img
+  src="https://huggingface.co/nvidia/isaaclab-arena-envs/resolve/main/assets/Gr1OpenMicrowaveEnvironment.png"
+  alt="IsaacLab Arena - GR1 Microwave Environment"
+  style={{ maxWidth: "100%", borderRadius: "8px", marginBottom: "1rem" }}
+/>
+
+[IsaacLab Arena](https://github.com/isaac-sim/IsaacLab-Arena) integrates with NVIDIA IsaacLab to provide:
+
+- 🤖 **Humanoid embodiments**: GR1, G1, Galileo with various configurations
+- 🎯 **Manipulation & loco-manipulation tasks**: Microwave opening, pick-and-place, button pressing
+- ⚡ **GPU-accelerated rollouts**: Parallel environment execution on NVIDIA GPUs
+- 🖼️ **RTX Rendering**: Evaluate vision-based policies with realistic rendering, reflections and refractions.
+- 📦 **LeRobot-compatible datasets**: Ready for training with GR00T Nx, PI0, SmolVLA, ACT, Diffusion policies
+- 🔄 **EnvHub integration**: Load environments from HuggingFace Hub with one line
+
+## Installation
+
+### Prerequisites
+
+Hardware requirements are shared with Isaac Sim, and are detailed in [Isaac Sim Requirements](https://docs.isaacsim.omniverse.nvidia.com/5.1.0/installation/requirements.html).
+
+- NVIDIA GPU with CUDA support
+- NVIDIA driver compatible with IsaacSim 5.1.0
+- Linux (Ubuntu 22.04 / 24.04)
+
+### Setup
+
+```bash
+# 1. Create conda environment
+conda create -y -n lerobot-arena python=3.11
+conda activate lerobot-arena
+conda install -y -c conda-forge ffmpeg=7.1.1
+
+# 2. Install Isaac Sim 5.1.0
+pip install "isaacsim[all,extscache]==5.1.0" --extra-index-url https://pypi.nvidia.com
+
+# Accept NVIDIA EULA (required)
+export ACCEPT_EULA=Y
+export PRIVACY_CONSENT=Y
+
+# 3. Install IsaacLab 2.3.0
+git clone https://github.com/isaac-sim/IsaacLab.git
+cd IsaacLab
+git checkout v2.3.0
+./isaaclab.sh -i
+cd ..
+
+# 4. Install IsaacLab Arena
+git clone https://github.com/isaac-sim/IsaacLab-Arena.git
+cd IsaacLab-Arena
+git checkout release/0.1.1
+pip install -e .
+cd ..
+
+
+# 5. Install LeRobot
+git clone https://github.com/huggingface/lerobot.git
+cd lerobot
+pip install -e .
+cd ..
+
+
+# 6. Install additional dependencies
+pip install onnxruntime==1.23.2 lightwheel-sdk==1.0.1 vuer[all]==0.0.70 qpsolvers==4.8.1
+pip install numpy==1.26.0 # Isaac Sim 5.1 depends on numpy==1.26.0, this will be fixed in next release
+```
+
+## Evaluating Policies
+
+### Pre-trained Policies
+
+The following trained policies are available:
+
+| Policy                      | Architecture | Task          | Link                                                                     |
+| :-------------------------- | :----------- | :------------ | :----------------------------------------------------------------------- |
+| pi05-arena-gr1-microwave    | PI0.5        | GR1 Microwave | [HuggingFace](https://huggingface.co/nvidia/pi05-arena-gr1-microwave)    |
+| smolvla-arena-gr1-microwave | SmolVLA      | GR1 Microwave | [HuggingFace](https://huggingface.co/nvidia/smolvla-arena-gr1-microwave) |
+
+### Evaluate SmolVLA
+
+```bash
+pip install -e ".[smolvla]"
+pip install numpy==1.26.0 # revert to numpy version is 1.26
+```
+
+```bash
+lerobot-eval \
+    --policy.path=nvidia/smolvla-arena-gr1-microwave \
+    --env.type=isaaclab_arena \
+    --env.hub_path=nvidia/isaaclab-arena-envs \
+    --rename_map='{"observation.images.robot_pov_cam_rgb": "observation.images.robot_pov_cam"}' \
+    --policy.device=cuda \
+    --env.environment=gr1_microwave \
+    --env.embodiment=gr1_pink \
+    --env.object=mustard_bottle \
+    --env.headless=false \
+    --env.enable_cameras=true \
+    --env.video=true \
+    --env.video_length=10 \
+    --env.video_interval=15 \
+    --env.state_keys=robot_joint_pos \
+    --env.camera_keys=robot_pov_cam_rgb \
+    --trust_remote_code=True \
+    --eval.batch_size=1
+```
+
+### Evaluate PI0.5
+
+```bash
+pip install -e ".[pi]"
+pip install numpy==1.26.0 # revert to numpy version is 1.26
+```
+
+<Tip>PI0.5 requires disabling torch compile for evaluation:</Tip>
+
+```bash
+TORCH_COMPILE_DISABLE=1 TORCHINDUCTOR_DISABLE=1 lerobot-eval \
+    --policy.path=nvidia/pi05-arena-gr1-microwave \
+    --env.type=isaaclab_arena \
+    --env.hub_path=nvidia/isaaclab-arena-envs \
+    --rename_map='{"observation.images.robot_pov_cam_rgb": "observation.images.robot_pov_cam"}' \
+    --policy.device=cuda \
+    --env.environment=gr1_microwave \
+    --env.embodiment=gr1_pink \
+    --env.object=mustard_bottle \
+    --env.headless=false \
+    --env.enable_cameras=true \
+    --env.video=true \
+    --env.video_length 15 \
+    --env.video_interval 15 \
+    --env.state_keys=robot_joint_pos \
+    --env.camera_keys=robot_pov_cam_rgb \
+    --trust_remote_code=True \
+    --eval.batch_size=1
+```
+
+<Tip>
+  To change the number of parallel environments, use the ```--eval.batch_size```
+  flag.
+</Tip>
+
+### What to Expect
+
+During evaluation, you will see a progress bar showing the running success rate:
+
+```
+Stepping through eval batches:   8%|██████▍    | 4/50 [00:45<08:06, 10.58s/it, running_success_rate=25.0%]
+```
+
+### Video Recording
+
+To enable video recording during evaluation, add the following flags to your command:
+
+```bash
+--env.video=true \
+--env.video_length=15 \
+--env.video_interval=15
+```
+
+For more details on video recording, see the [IsaacLab Recording Documentation](https://isaac-sim.github.io/IsaacLab/main/source/how-to/record_video.html).
+
+<Tip>
+When running headless with `--env.headless=true`, you must also enable cameras explicitly for camera enabled environments:
+
+```bash
+--env.headless=true --env.enable_cameras=true
+```
+
+</Tip>
+
+### Output Directory
+
+Evaluation videos are saved to the output directory with the following structure:
+
+```
+outputs/eval/<date>/<timestamp>_<env>_<policy>/videos/<task>_<env_id>/eval_episode_<n>.mp4
+```
+
+For example:
+
+```
+outputs/eval/2026-01-02/14-38-01_isaaclab_arena_smolvla/videos/gr1_microwave_0/eval_episode_0.mp4
+```
+
+## Training Policies
+
+To learn more about training policies with LeRobot, please refer to training documentation:
+
+- [SmolVLA](./smolvla)
+- [Pi0.5](./pi05)
+- [GR00T N1.5](./groot)
+
+Sample IsaacLab Arena datasets are available on HuggingFace Hub for experimentation:
+
+| Dataset                                                                                                   | Description                | Frames |
+| :-------------------------------------------------------------------------------------------------------- | :------------------------- | :----- |
+| [Arena-GR1-Manipulation-Task](https://huggingface.co/datasets/nvidia/Arena-GR1-Manipulation-Task-v3)      | GR1 microwave manipulation | ~4K    |
+| [Arena-G1-Loco-Manipulation-Task](https://huggingface.co/datasets/nvidia/Arena-G1-Loco-Manipulation-Task) | G1 loco-manipulation       | ~4K    |
+
+## Environment Configuration
+
+### Full Configuration Options
+
+```python
+from lerobot.envs.configs import IsaaclabArenaEnv
+
+config = IsaaclabArenaEnv(
+    # Environment selection
+    environment="gr1_microwave",      # Task environment
+    embodiment="gr1_pink",            # Robot embodiment
+    object="power_drill",             # Object to manipulate
+
+    # Simulation settings
+    episode_length=300,               # Max steps per episode
+    headless=True,                    # Run without GUI
+    device="cuda:0",                  # GPU device
+    seed=42,                          # Random seed
+
+    # Observation configuration
+    state_keys="robot_joint_pos",     # State observation keys (comma-separated)
+    camera_keys="robot_pov_cam_rgb",  # Camera observation keys (comma-separated)
+    state_dim=54,                     # Expected state dimension
+    action_dim=36,                    # Expected action dimension
+    camera_height=512,                # Camera image height
+    camera_width=512,                 # Camera image width
+    enable_cameras=True,              # Enable camera observations
+
+    # Video recording
+    video=False,                      # Enable video recording
+    video_length=100,                 # Frames per video
+    video_interval=200,               # Steps between recordings
+
+    # Advanced
+    mimic=False,                      # Enable mimic mode
+    teleop_device=None,               # Teleoperation device
+    disable_fabric=False,             # Disable fabric optimization
+    enable_pinocchio=True,            # Enable Pinocchio for IK
+)
+```
+
+### Using Environment Hub directly for advanced usage
+
+Create a file called `test_env_load_arena.py` or [download from the EnvHub](https://huggingface.co/nvidia/isaaclab-arena-envs/blob/main/tests/test_env_load_arena.py):
+
+```python
+import logging
+from dataclasses import asdict
+from pprint import pformat
+import torch
+import tqdm
+from lerobot.configs import parser
+from lerobot.configs.eval import EvalPipelineConfig
+
+
+@parser.wrap()
+def main(cfg: EvalPipelineConfig):
+    """Run zero action rollout for IsaacLab Arena environment."""
+    logging.info(pformat(asdict(cfg)))
+
+    from lerobot.envs.factory import make_env
+
+    env_dict = make_env(
+        cfg.env,
+        n_envs=cfg.env.num_envs,
+        trust_remote_code=True,
+    )
+    env = next(iter(env_dict.values()))[0]
+    env.reset()
+    for _ in tqdm.tqdm(range(cfg.env.episode_length)):
+        with torch.inference_mode():
+            actions = env.action_space.sample()
+            obs, rewards, terminated, truncated, info = env.step(actions)
+            if terminated.any() or truncated.any():
+                obs, info = env.reset()
+    env.close()
+
+
+if __name__ == "__main__":
+    main()
+```
+
+Run with:
+
+```bash
+python test_env_load_arena.py \
+    --env.environment=g1_locomanip_pnp \
+    --env.embodiment=gr1_pink \
+    --env.object=cracker_box \
+    --env.num_envs=4 \
+    --env.enable_cameras=true \
+    --env.seed=1000 \
+    --env.video=true \
+    --env.video_length=10 \
+    --env.video_interval=15 \
+    --env.headless=false \
+    --env.hub_path=nvidia/isaaclab-arena-envs \
+    --env.type=isaaclab_arena
+```
+
+## Troubleshooting
+
+### CUDA out of memory
+
+Reduce `batch_size` or use a GPU with more VRAM:
+
+```bash
+--eval.batch_size=1
+```
+
+### EULA not accepted
+
+Set environment variables before running:
+
+```bash
+export ACCEPT_EULA=Y
+export PRIVACY_CONSENT=Y
+```
+
+### Video recording not working
+
+Enable cameras when running headless:
+
+```bash
+--env.video=true --env.enable_cameras=true --env.headless=true
+```
+
+### Policy output dimension mismatch
+
+E.g. ensure `action_dim` matches your policy:
+
+```bash
+--env.action_dim=36
+```
+
+### libGLU.so.1 Errors during Isaac Sim initialization
+
+Ensure you have the following dependencies installed, this is likely to happen on headless machines.
+
+```bash
+sudo apt update && sudo apt install -y libglu1-mesa libxt6
+```
+
+## See Also
+
+- [EnvHub Documentation](./envhub.mdx) - General EnvHub usage
+- [IsaacLab Arena GitHub](https://github.com/isaac-sim/IsaacLab-Arena)
+- [IsaacLab Documentation](https://isaac-sim.github.io/IsaacLab/)
+
+## LightWheel LW-BenchHub
+
+[LightWheel AI](https://www.lightwheel.ai) are bringing `Lightwheel-Libero-Tasks` and `Lightwheel-RoboCasa-Tasks` with 268 tasks to the LeRobot ecosystem.
+LW-BenchHub collects and generates large-scale datasets via teleoperation that comply with the LeRobot specification, enabling out-of-the-box training and evaluation workflows.
+With the unified interface provided by EnvHub, developers can quickly build end-to-end experimental pipelines.
+
+### Install
+
+Assuming you followed the [Installation](#installation) steps, you can install LW-BenchHub with:
+
+```bash
+conda install pinocchio -c conda-forge -y
+git clone https://github.com/LightwheelAI/lw_benchhub
+cd lw_benchhub
+pip install -e .
+```
+
+For more detailed instructions, please refer to the [LW-BenchHub Documentation](https://docs.lightwheel.net/lw_benchhub/usage/Installation).
+
+### Lightwheel Tasks Dataset
+
+LW-BenchHub datasets are available on HuggingFace Hub:
+
+| Dataset                                                                                                       | Description             | Tasks | Frames |
+| :------------------------------------------------------------------------------------------------------------ | :---------------------- | :---- | :----- |
+| [Lightwheel-Tasks-X7S](https://huggingface.co/datasets/LightwheelAI/Lightwheel-Tasks-X7S)                     | X7S LIBERO and RoboCasa | 117   | ~10.3M |
+| [Lightwheel-Tasks-Double-Piper](https://huggingface.co/datasets/LightwheelAI/Lightwheel-Tasks-Double-Piper)   | Double-Piper LIBERO     | 130   | ~6.0M  |
+| [Lightwheel-Tasks-G1-Controller](https://huggingface.co/datasets/LightwheelAI/Lightwheel-Tasks-G1-Controller) | G1-Controller LIBERO    | 62    | ~2.7M  |
+| [Lightwheel-Tasks-G1-WBC](https://huggingface.co/datasets/LightwheelAI/Lightwheel-Tasks-G1-WBC)               | G1-WBC RoboCasa         | 32    | ~1.5M  |
+
+For training policies, refer to the [Training Policies](#training-policies) section.
+
+### Evaluating Policies
+
+#### Pre-trained Policies
+
+The following trained policies are available:
+
+| Policy                   | Architecture | Task                           | Layout     | Robot           | Link                                                                                  |
+| :----------------------- | :----------- | :----------------------------- | :--------- | :-------------- | :------------------------------------------------------------------------------------ |
+| smolvla-double-piper-pnp | SmolVLA      | L90K1PutTheBlackBowlOnThePlate | libero-1-1 | DoublePiper-Abs | [HuggingFace](https://huggingface.co/LightwheelAI/smolvla-double-piper-pnp/tree/main) |
+
+#### Evaluate SmolVLA
+
+```bash
+lerobot-eval \
+  --policy.path=LightwheelAI/smolvla-double-piper-pnp \
+  --env.type=isaaclab_arena \
+  --rename_map='{"observation.images.left_hand_camera_rgb": "observation.images.left_hand", "observation.images.right_hand_camera_rgb": "observation.images.right_hand", "observation.images.first_person_camera_rgb": "observation.images.first_person"}' \
+  --env.hub_path=LightwheelAI/lw_benchhub_env \
+  --env.kwargs='{"config_path": "configs/envhub/example.yml"}' \
+  --trust_remote_code=true \
+  --env.state_keys=joint_pos \
+  --env.action_dim=12 \
+  --env.camera_keys=left_hand_camera_rgb,right_hand_camera_rgb,first_person_camera_rgb \
+  --policy.device=cuda \
+  --eval.batch_size=10 \
+  --eval.n_episodes=100
+```
+
+### Environment Configuration
+
+Evaluation can be quickly launched by modifying the `robot`, `task`, and `layout` settings in the configuration file.
+
+#### Full Configuration Options
+
+```yml
+# =========================
+# Basic Settings
+# =========================
+disable_fabric: false
+device: cuda:0
+sensitivity: 1.0
+step_hz: 50
+enable_cameras: true
+execute_mode: eval
+episode_length_s: 20.0 # Episode length in seconds, increase if episodes timeout during eval
+
+# =========================
+# Robot Settings
+# =========================
+robot: DoublePiper-Abs # Robot type, DoublePiper-Abs, X7S-Abs, G1-Controller or G1-Controller-DecoupledWBC
+robot_scale: 1.0
+
+# =========================
+# Task & Scene Settings
+# =========================
+task: L90K1PutTheBlackBowlOnThePlate # Task name
+scene_backend: robocasa
+task_backend: robocasa
+debug_assets: null
+layout: libero-1-1 # Layout and style ID
+sources:
+  - objaverse
+  - lightwheel
+  - aigen_objs
+object_projects: []
+usd_simplify: false
+seed: 42
+
+# =========================
+# Object Placement Retry Settings
+# =========================
+max_scene_retry: 4
+max_object_placement_retry: 3
+
+resample_objects_placement_on_reset: true
+resample_robot_placement_on_reset: true
+
+# =========================
+# Replay Configuration Settings
+# =========================
+replay_cfgs:
+  add_camera_to_observation: true
+  render_resolution: [640, 480]
+```
+
+### See Also
+
+- [LW-BenchHub GitHub](https://github.com/LightwheelAI/LW-BenchHub)
+- [LW-BenchHub Documentation](https://docs.lightwheel.net/lw_benchhub/)
@@ -38,6 +38,8 @@ class EvalPipelineConfig:
    seed: int | None = 1000
    # Rename map for the observation to override the image and state keys
    rename_map: dict[str, str] = field(default_factory=dict)
+    # Explicit consent to execute remote code from the Hub (required for hub environments).
+    trust_remote_code: bool = False

    def __post_init__(self) -> None:
        # HACK: We parse again the cli args here to get the pretrained path if there was one.
@@ -12,4 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.

-from .configs import AlohaEnv, EnvConfig, PushtEnv  # noqa: F401
+from .configs import AlohaEnv, EnvConfig, HubEnvConfig, PushtEnv  # noqa: F401
@@ -13,7 +13,7 @@
 # limitations under the License.

 import abc
-from dataclasses import dataclass, field
+from dataclasses import dataclass, field, fields
 from typing import Any

 import draccus
@@ -68,6 +68,22 @@ class EnvConfig(draccus.ChoiceRegistry, abc.ABC):
        raise NotImplementedError()


+@dataclass
+class HubEnvConfig(EnvConfig):
+    """Base class for environments that delegate creation to a hub-hosted make_env.
+
+    Hub environments download and execute remote code from the HF Hub.
+    The hub_path points to a repository containing an env.py with a make_env function.
+    """
+
+    hub_path: str | None = None  # required: e.g., "username/repo" or "username/repo@branch:file.py"
+
+    @property
+    def gym_kwargs(self) -> dict:
+        # Not used for hub environments - the hub's make_env handles everything
+        return {}
+
+
@EnvConfig.register_subclass("aloha")
@dataclass
 class AlohaEnv(EnvConfig):
@@ -368,3 +384,71 @@ class MetaworldEnv(EnvConfig):
            "obs_type": self.obs_type,
            "render_mode": self.render_mode,
        }
+
+
+@EnvConfig.register_subclass("isaaclab_arena")
+@dataclass
+class IsaaclabArenaEnv(HubEnvConfig):
+    hub_path: str = "nvidia/isaaclab-arena-envs"
+    episode_length: int = 300
+    num_envs: int = 1
+    embodiment: str | None = "gr1_pink"
+    object: str | None = "power_drill"
+    mimic: bool = False
+    teleop_device: str | None = None
+    seed: int | None = 42
+    device: str | None = "cuda:0"
+    disable_fabric: bool = False
+    enable_cameras: bool = False
+    headless: bool = False
+    enable_pinocchio: bool = True
+    environment: str | None = "gr1_microwave"
+    task: str | None = "Reach out to the microwave and open it."
+    state_dim: int = 54
+    action_dim: int = 36
+    camera_height: int = 512
+    camera_width: int = 512
+    video: bool = False
+    video_length: int = 100
+    video_interval: int = 200
+    # Comma-separated keys, e.g., "robot_joint_pos,left_eef_pos"
+    state_keys: str = "robot_joint_pos"
+    # Comma-separated keys, e.g., "robot_pov_cam_rgb,front_cam_rgb"
+    # Set to None or "" for environments without cameras
+    camera_keys: str | None = None
+    features: dict[str, PolicyFeature] = field(default_factory=dict)
+    features_map: dict[str, str] = field(default_factory=dict)
+    kwargs: dict | None = None
+
+    def __post_init__(self):
+        if self.kwargs:
+            # dynamically convert kwargs to fields in the dataclass
+            # NOTE! the new fields will not bee seen by the dataclass repr
+            field_names = {f.name for f in fields(self)}
+            for key, value in self.kwargs.items():
+                if key not in field_names and key != "kwargs":
+                    setattr(self, key, value)
+            self.kwargs = None
+
+        # Set action feature
+        self.features[ACTION] = PolicyFeature(type=FeatureType.ACTION, shape=(self.action_dim,))
+        self.features_map[ACTION] = ACTION
+
+        # Set state feature
+        self.features[OBS_STATE] = PolicyFeature(type=FeatureType.STATE, shape=(self.state_dim,))
+        self.features_map[OBS_STATE] = OBS_STATE
+
+        # Add camera features for each camera key
+        if self.enable_cameras and self.camera_keys:
+            for cam_key in self.camera_keys.split(","):
+                cam_key = cam_key.strip()
+                if cam_key:
+                    self.features[cam_key] = PolicyFeature(
+                        type=FeatureType.VISUAL,
+                        shape=(self.camera_height, self.camera_width, 3),
+                    )
+                    self.features_map[cam_key] = f"{OBS_IMAGES}.{cam_key}"
+
+    @property
+    def gym_kwargs(self) -> dict:
+        return {}
@@ -20,11 +20,11 @@ import gymnasium as gym
 from gymnasium.envs.registration import registry as gym_registry

 from lerobot.configs.policies import PreTrainedConfig
-from lerobot.envs.configs import AlohaEnv, EnvConfig, LiberoEnv, PushtEnv
+from lerobot.envs.configs import AlohaEnv, EnvConfig, HubEnvConfig, IsaaclabArenaEnv, LiberoEnv, PushtEnv
 from lerobot.envs.utils import _call_make_env, _download_hub_file, _import_hub_module, _normalize_hub_result
 from lerobot.policies.xvla.configuration_xvla import XVLAConfig
 from lerobot.processor import ProcessorStep
-from lerobot.processor.env_processor import LiberoProcessorStep
+from lerobot.processor.env_processor import IsaaclabArenaProcessorStep, LiberoProcessorStep
 from lerobot.processor.pipeline import PolicyProcessorPipeline


@@ -73,6 +73,26 @@ def make_env_pre_post_processors(
    if isinstance(env_cfg, LiberoEnv) or "libero" in env_cfg.type:
        preprocessor_steps.append(LiberoProcessorStep())

+    # For Isaaclab Arena environments, add the IsaaclabArenaProcessorStep
+    if isinstance(env_cfg, IsaaclabArenaEnv) or "isaaclab_arena" in env_cfg.type:
+        # Parse comma-separated keys (handle None for state-based policies)
+        if env_cfg.state_keys:
+            state_keys = tuple(k.strip() for k in env_cfg.state_keys.split(",") if k.strip())
+        else:
+            state_keys = ()
+        if env_cfg.camera_keys:
+            camera_keys = tuple(k.strip() for k in env_cfg.camera_keys.split(",") if k.strip())
+        else:
+            camera_keys = ()
+        if not state_keys and not camera_keys:
+            raise ValueError("At least one of state_keys or camera_keys must be specified.")
+        preprocessor_steps.append(
+            IsaaclabArenaProcessorStep(
+                state_keys=state_keys,
+                camera_keys=camera_keys,
+            )
+        )
+
    preprocessor = PolicyProcessorPipeline(steps=preprocessor_steps)
    postprocessor = PolicyProcessorPipeline(steps=postprocessor_steps)

@@ -98,7 +118,6 @@ def make_env(
        hub_cache_dir (str | None): Optional cache path for downloaded hub files.
        trust_remote_code (bool): **Explicit consent** to execute remote code from the Hub.
            Default False — must be set to True to import/exec hub `env.py`.
-
    Raises:
        ValueError: if n_envs < 1
        ModuleNotFoundError: If the requested env package is not installed
@@ -112,19 +131,35 @@ def make_env(
    """
    # if user passed a hub id string (e.g., "username/repo", "username/repo@main:env.py")
    # simplified: only support hub-provided `make_env`
+    # TODO: (jadechoghari): deprecate string API and remove this check
    if isinstance(cfg, str):
+        hub_path: str | None = cfg
+    elif isinstance(cfg, HubEnvConfig):
+        hub_path = cfg.hub_path
+    else:
+        hub_path = None
+
+    # If hub_path is set, download and call hub-provided `make_env`
+    if hub_path:
        # _download_hub_file will raise the same RuntimeError if trust_remote_code is False
-        repo_id, file_path, local_file, revision = _download_hub_file(cfg, trust_remote_code, hub_cache_dir)
+        repo_id, file_path, local_file, revision = _download_hub_file(
+            hub_path, trust_remote_code, hub_cache_dir
+        )

        # import and surface clear import errors
        module = _import_hub_module(local_file, repo_id)

        # call the hub-provided make_env
-        raw_result = _call_make_env(module, n_envs=n_envs, use_async_envs=use_async_envs)
+        env_cfg = None if isinstance(cfg, str) else cfg
+        raw_result = _call_make_env(module, n_envs=n_envs, use_async_envs=use_async_envs, cfg=env_cfg)

        # normalize the return into {suite: {task_id: vec_env}}
        return _normalize_hub_result(raw_result)

+    # At this point, cfg must be an EnvConfig (not a string) since hub_path would have been set otherwise
+    if isinstance(cfg, str):
+        raise TypeError("cfg should be an EnvConfig at this point")
+
    if n_envs < 1:
        raise ValueError("`n_envs` must be at least 1")

@@ -46,7 +46,7 @@ def _convert_nested_dict(d):


 def preprocess_observation(observations: dict[str, np.ndarray]) -> dict[str, Tensor]:
-    # TODO(aliberts, rcadene): refactor this to use features from the environment (no hardcoding)
+    # TODO(jadechoghari, imstevenpmwork): refactor this to use features from the environment (no hardcoding)
    """Convert environment observation to LeRobot format observation.
    Args:
        observation: Dictionary of observation batches from a Gym vector environment.
@@ -98,11 +98,19 @@ def preprocess_observation(observations: dict[str, np.ndarray]) -> dict[str, Ten

    if "robot_state" in observations:
        return_observations[f"{OBS_STR}.robot_state"] = _convert_nested_dict(observations["robot_state"])
+
+    # Handle IsaacLab Arena format: observations have 'policy' and 'camera_obs' keys
+    if "policy" in observations:
+        return_observations[f"{OBS_STR}.policy"] = observations["policy"]
+
+    if "camera_obs" in observations:
+        return_observations[f"{OBS_STR}.camera_obs"] = observations["camera_obs"]
+
    return return_observations


 def env_to_policy_features(env_cfg: EnvConfig) -> dict[str, PolicyFeature]:
-    # TODO(aliberts, rcadene): remove this hardcoding of keys and just use the nested keys as is
+    # TODO(jadechoghari, imstevenpmwork): remove this hardcoding of keys and just use the nested keys as is
    # (need to also refactor preprocess_observation and externalize normalization from policies)
    policy_features = {}
    for key, ft in env_cfg.features.items():
@@ -302,7 +310,7 @@ def _import_hub_module(local_file: str, repo_id: str) -> Any:
    return module


-def _call_make_env(module: Any, n_envs: int, use_async_envs: bool) -> Any:
+def _call_make_env(module: Any, n_envs: int, use_async_envs: bool, cfg: EnvConfig | None) -> Any:
    """
    Ensure module exposes make_env and call it.
    """
@@ -311,7 +319,11 @@ def _call_make_env(module: Any, n_envs: int, use_async_envs: bool) -> Any:
            f"The hub module {getattr(module, '__name__', 'hub_module')} must expose `make_env(n_envs=int, use_async_envs=bool)`."
        )
    entry_fn = module.make_env
-    return entry_fn(n_envs=n_envs, use_async_envs=use_async_envs)
+    # Only pass cfg if it's not None (i.e., when an EnvConfig was provided, not a string hub ID)
+    if cfg is not None:
+        return entry_fn(n_envs=n_envs, use_async_envs=use_async_envs, cfg=cfg)
+    else:
+        return entry_fn(n_envs=n_envs, use_async_envs=use_async_envs)


 def _normalize_hub_result(result: Any) -> dict[str, dict[int, gym.vector.VectorEnv]]:
@@ -18,7 +18,7 @@ from dataclasses import dataclass
 import torch

 from lerobot.configs.types import PipelineFeatureType, PolicyFeature
-from lerobot.utils.constants import OBS_IMAGES, OBS_STATE
+from lerobot.utils.constants import OBS_IMAGES, OBS_STATE, OBS_STR

 from .pipeline import ObservationProcessorStep, ProcessorStepRegistry

@@ -152,3 +152,78 @@ class LiberoProcessorStep(ObservationProcessorStep):
            result[mask] = axis * angle.unsqueeze(1)

        return result
+
+
+@dataclass
+@ProcessorStepRegistry.register(name="isaaclab_arena_processor")
+class IsaaclabArenaProcessorStep(ObservationProcessorStep):
+    """
+    Processes IsaacLab Arena observations into LeRobot format.
+
+    **State Processing:**
+    - Extracts state components from obs["policy"] based on `state_keys`.
+    - Concatenates into a flat vector mapped to "observation.state".
+
+    **Image Processing:**
+    - Extracts images from obs["camera_obs"] based on `camera_keys`.
+    - Converts from (B, H, W, C) uint8 to (B, C, H, W) float32 [0, 1].
+    - Maps to "observation.images.<camera_name>".
+    """
+
+    # Configurable from IsaacLabEnv config / cli args: --env.state_keys="robot_joint_pos,left_eef_pos"
+    state_keys: tuple[str, ...]
+
+    # Configurable from IsaacLabEnv config / cli args: --env.camera_keys="robot_pov_cam_rgb"
+    camera_keys: tuple[str, ...]
+
+    def _process_observation(self, observation):
+        """
+        Processes both image and policy state observations from IsaacLab Arena.
+        """
+        processed_obs = {}
+
+        if f"{OBS_STR}.camera_obs" in observation:
+            camera_obs = observation[f"{OBS_STR}.camera_obs"]
+
+            for cam_name, img in camera_obs.items():
+                if cam_name not in self.camera_keys:
+                    continue
+
+                img = img.permute(0, 3, 1, 2).contiguous()
+                if img.dtype == torch.uint8:
+                    img = img.float() / 255.0
+                elif img.dtype != torch.float32:
+                    img = img.float()
+
+                processed_obs[f"{OBS_IMAGES}.{cam_name}"] = img
+
+        # Process policy state -> observation.state
+        if f"{OBS_STR}.policy" in observation:
+            policy_obs = observation[f"{OBS_STR}.policy"]
+
+            # Collect state components in order
+            state_components = []
+            for key in self.state_keys:
+                if key in policy_obs:
+                    component = policy_obs[key]
+                    # Flatten extra dims: (B, N, M) -> (B, N*M)
+                    if component.dim() > 2:
+                        batch_size = component.shape[0]
+                        component = component.view(batch_size, -1)
+                    state_components.append(component)
+
+            if state_components:
+                state = torch.cat(state_components, dim=-1)
+                state = state.float()
+                processed_obs[OBS_STATE] = state
+
+        return processed_obs
+
+    def transform_features(
+        self, features: dict[PipelineFeatureType, dict[str, PolicyFeature]]
+    ) -> dict[PipelineFeatureType, dict[str, PolicyFeature]]:
+        """Not used for policy evaluation."""
+        return features
+
+    def observation(self, observation):
+        return self._process_observation(observation)
@@ -509,7 +509,12 @@ def eval_main(cfg: EvalPipelineConfig):
    logging.info(colored("Output dir:", "yellow", attrs=["bold"]) + f" {cfg.output_dir}")

    logging.info("Making environment.")
-    envs = make_env(cfg.env, n_envs=cfg.eval.batch_size, use_async_envs=cfg.eval.use_async_envs)
+    envs = make_env(
+        cfg.env,
+        n_envs=cfg.eval.batch_size,
+        use_async_envs=cfg.eval.use_async_envs,
+        trust_remote_code=cfg.trust_remote_code,
+    )

    logging.info("Making policy.")

@@ -17,7 +17,7 @@
 import json
 import tempfile
 from collections.abc import Callable
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from pathlib import Path
 from typing import Any

@@ -1884,7 +1884,7 @@ class FeatureContractAddStep(ProcessorStep):
    """Adds a PolicyFeature"""

    key: str = "a"
-    value: PolicyFeature = PolicyFeature(type=FeatureType.STATE, shape=(1,))
+    value: PolicyFeature = field(default_factory=lambda: PolicyFeature(type=FeatureType.STATE, shape=(1,)))

    def __call__(self, transition: EnvTransition) -> EnvTransition:
        return transition