Overview
LeRobot is Hugging Face’s open-source library for robot learning and simulation. It bundles pretrained policies, standardized datasets (LeRobotDataset), simulation environments and end-to-end training pipelines to make reproducible robotics research and engineering more accessible. The project integrates with the Hugging Face Hub for model and dataset sharing.
Key features
- Pretrained policies and example configurations for tasks such as PushT, ALOHA and SimXArm.
- Dataset format and visualization tools to inspect video frames and robot states easily.
- End-to-end tooling for simulation, training, evaluation and publishing to the Hub.
Use cases
- Robotics research and benchmark reproduction across simulated and real environments.
- Engineering pipelines for deploying learned policies on physical robots.
- Educational materials and tutorials for learning robot learning workflows.
Technical highlights
- PyTorch-based implementation compatible with modern ML tooling and the Hugging Face ecosystem.
- Designed for reproducibility with versioned configs, dependency notes and example scripts.
- Apache-2.0 licensed and actively maintained by the community for both research and production use.