Tier 4 Deeply open
What the publisher claims: open-source, open datasets
The one-liner: An open ecosystem layer: formats, datasets, training code and evaluation in one permissive repository.
The six questions
Code, datasets and pretrained policies are all downloadable from GitHub and the Hub.
Apache 2.0 code; per-dataset licences are mostly permissive but heterogeneous.
Runs on commodity hardware behind a standard Robot interface covering dozens of arms and humanoids.
Dataset format, collection workflow and training recipes are documented, but provenance and quality vary by contributor.
A skilled group can train comparable policies from published recipes and public datasets, with results varying by embodiment.
huggingface/lerobot shows 341 merged PRs including 97 external, with community robot contributions and public roadmaps.
License
Apache 2.0, with parts derived from Diffusion Policy under MIT OSI-approved — license source ↗
- Individual Hub datasets and policies carry their own licences; check per repository
What is actually released
- library code
- LeRobotDataset format
- thousands of Hub datasets
- policy implementations and training recipes (ACT, Diffusion, pi0, SmolVLA, GR00T)
- simulation benchmarks (LIBERO, MetaWorld)
Why it matters
Shows that openness in robotics can be delivered as shared infrastructure rather than as a single open model.
Sources
- https://raw.githubusercontent.com/huggingface/lerobot/main/LICENSE
- https://raw.githubusercontent.com/huggingface/lerobot/main/README.md
Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.