Note

This is a Hugging Face dataset. For large datasets, ensure huggingface_hub>=1.1.3 to avoid rate limits. Learn more in the Hugging Face integration docs.

Hugging Face

GR00T-X-Embodiment-Sim → FiftyOne (Native Multimodal MCAP)#

preview

A 42-episode subset of nvidia/PhysicalAI-Robotics-GR00T-X-Embodiment-Sim, six episodes from each of seven embodiments, converted to native multimodal MCAP episodes. Each episode carries the camera streams as H.264 foxglove.CompressedVideo, per-part robot state and action telemetry with timeline plot channels, and the task instruction.

Installation#

pip install fiftyone

Usage#

import fiftyone as fo
import fiftyone.utils.huggingface as fouh

dataset = fouh.load_from_hub(
    "Voxel51/GR00T-X-Embodiment-Sim",
    name="GR00T-X-Embodiment-Sim",
    persistent=True,
)
fo.launch_app(dataset)

What you get#

  • 42 .mcap episodes across single_panda_gripper, bimanual_panda_gripper, bimanual_panda_hand, gr1_arms_waist, gr1_full_upper_body, sim_behavior_r1_pro, and unitree_g1

  • Telemetry segmented into named channels per the source modality.json (for example /joint-position-state, /end-effector-position-absolute-state, /gripper-qpos-action), each with a timeline plot channel

  • Per-episode fields: embodiment, task, source_task, episode_index, num_frames, duration

  • Camera streams transmuxed from the source H.264 without re-encoding

License & attribution#

The source dataset is released by NVIDIA under CC-BY-4.0; this subset is distributed under the same license. Changes from the source: episode subsetting and conversion from LeRobot parquet/mp4 format to MCAP.