LeRobot

About LeRobot

LeRobot is an open-source repository from Hugging Face that provides models, datasets, and tools for real-world robotics, written in PyTorch. It can be installed from PyPI with pip. The project aims to lower the barrier to entry for robotics AI work by making shared datasets and pretrained models available to a broad audience.

The repository includes a Python Robot interface that separates control logic from hardware details and lists natively supported robots and teleoperation devices. It also defines the LeRobotDataset format, which stores synchronized video or images with Parquet state and action data, and it includes tools for editing datasets. Policy implementations cover imitation learning, reinforcement learning, vision-language-action models, world models, and reward models. Some reinforcement learning policies are marked as coming soon, and hardware setup and policy setup are described in separate documentation.

Key features

• Unified Robot class interface that separates control logic from hardware specifics
• Native integration with listed robots, teleoperation devices, gamepads, keyboards, and phones
• LeRobotDataset format pairing synchronized MP4 video or images with Parquet state and action data
• Dataset tools to delete episodes, split datasets, add or remove features, and merge datasets
• Policy training launched from a command-line script configuration using lerobot-train
• Implementations of imitation learning, vision-language-action, world model, and reward model policies

LeRobot at a glance