LeRobot Format: What It Is and How nxted Delivers It
What is the LeRobot dataset format?
LeRobot is Hugging Face’s open robotics library and dataset format. It stores episodes as Parquet (state and action) plus MP4 video and JSON metadata, with first-class tooling and easy sharing on the Hugging Face Hub. It is a friendly default for training and distributing imitation-learning datasets.
What a LeRobot dataset contains
Per-episode observations and actions in Parquet, video in MP4, and JSON metadata (features, fps, splits), designed to load directly into the LeRobot trainer.
- ›Parquet tables for state and action streams
- ›MP4 video per camera
- ›JSON metadata: features, control frequency, splits
- ›Hugging Face Hub integration for sharing
When to choose LeRobot
Pick LeRobot if you use the LeRobot trainer or want the friendliest tooling and easy dataset sharing. See our LeRobot vs RLDS vs HDF5 guide to compare.
How nxted delivers LeRobot data
Every nxted Capture dataset can ship in LeRobot, alongside RLDS and HDF5, with full metadata, a dataset card and a Data Trust Pack.
FAQ
What is LeRobot?
Hugging Face’s open robotics library and dataset format, storing episodes as Parquet plus MP4 video and JSON metadata for imitation learning.
How is LeRobot different from RLDS and HDF5?
LeRobot is Parquet/MP4/JSON with Hub sharing; RLDS is the TFDS format behind Open X-Embodiment; HDF5 is a general scientific container used by ALOHA and robomimic.
Does nxted deliver in LeRobot?
Yes - LeRobot, RLDS and HDF5 are all available, with metadata, a dataset card and a QA report.