huggingface/lerobot

LeRobot is an open-source Python library from Hugging Face that makes it easier for researchers and hobbyists to build AI-powered robots. The problem it solves is that robotics research is fragmented: different robots speak different languages, datasets are stored in incompatible formats, and training AI models from scratch requires deep expertise. LeRobot provides a common platform that ties all of this together. At its core, the library offers three things. First, a unified programming interface for controlling robots, whether that is a cheap hobby arm like the SO-100 or a full humanoid like Reachy2, the same Python code works across all supported hardware. Second, a standardized dataset format called LeRobotDataset that stores synchronized video footage and sensor readings so that thousands of recorded robot sessions can be shared, streamed, and reused. Third, a collection of ready-to-use AI policies, meaning the decision-making brains you slot into a robot. These range from simple imitation learning (the robot copies human demonstrations) to advanced Vision-Language-Action models that understand natural language commands. You would use LeRobot if you want to train a robot arm to pick up objects, share robot demonstration data with other researchers, fine-tune an existing AI policy on your own hardware, or run standard benchmarks like LIBERO to compare your models. It is aimed at anyone who finds robotics interesting but does not want to rebuild all the plumbing from scratch. The tech stack is Python and PyTorch.