LeRobot

Making AI for Robotics more accessible. Framework + ACT, SmolVLA. Hugging Face.

Overview

LeRobot is a PyTorch framework for end-to-end robot learning (imitation + RL). It provides datasets, policy architectures (ACT, SmolVLA), training scripts, and deployment tools. SmolVLA is a 450M-parameter VLA model. Datasets include DROID-100, ALOHA, SO-100.

Policy Models

ACT — Action Chunking Transformer, ~1.5h on A100 for 100K steps
SmolVLA — 450M VLA, ~5h on A100 for 20K steps

Official Links

huggingface.co/docs/lerobot — Documentation
github.com/huggingface/lerobot — Code
Hugging Face Hub: lerobot — Datasets & models

Citation

See the project repository for BibTeX and references.