CS3AS HRC episodes (LeRobot v2.1)
- episodes: 1
- frames: 345
- fps: 30
- tasks: safeskin
- robot:
dobot_cs3as_pgea
Parquet columns include observation.state, observation.skins, observation.skin_analog, observation.skin_distance, observation.skin_approach, and action. Hand-eye video (if recorded) is observation.images.handeye (H.264 / avc1 at 30 FPS). Native data/episodes/*.npz stays at the 125 Hz robot bus; this Hub folder is undersampled to the camera clock so the LeRobot visualizer can play image + state together.
Visualize
Local player (cameras play next to joints / SafeSkin; --serve so the browser can load mp4s):
python hrc/examples/view_episodes.py data/lerobot/skins --camera --serve
python hrc/examples/visualize_episode.py data/lerobot/skins --gui --camera
Hugging Face LeRobot dataset visualizer after upload.
YOUR_ORG is not a CS3AS constant. It is your Hugging Face username (personal account) or organization slug. Print it with:
huggingface-cli whoami
# or: hf auth whoami
Use the name field for a personal dataset (alice/cs3as-hrc). Use an org you can write to for a shared dataset (my-lab/cs3as-hrc).
huggingface-cli upload-folder --repo-type dataset YOUR_ORG/cs3as-hrc data/lerobot/skins
Then paste YOUR_ORG/cs3as-hrc into the visualizer, or open:
https://huggingface.co/spaces/lerobot/visualize_dataset?path=/YOUR_ORG/cs3as-hrc/episode_0
meta/info.json follows LeRobot v2.1 (data_path / video_path templates). Convert to v3 with python -m lerobot.scripts.convert_dataset_v21_to_v30 if a newer toolchain requires it.
- Downloads last month
- 118