End-to-End Imitation Learning for Physical Manipulation

LeRobot is a Hugging Face open-source ecosystem designed to lower the barrier of entry to robot learning. It provides pre-trained policies, datasets, and lightweight wrappers for physical manipulators (such as standard low-cost robotic arms).

This project focuses on executing end-to-end imitation learning tasks (e.g. pick-and-place) on physical hardware. By uploading training datasets directly to the ecosystem and using standard imitation learning algorithms (ACT, Diffusion Policy), we evaluate the model's spatial adaptability and robustness during real-world tasks.