LeRobot v0.4.0 introduces major upgrades to the open-source robotics learning framework, including a completely overhauled dataset infrastructure (v3.0) with chunked episodes and streaming for massive datasets like OXE. The release adds support for LIBERO and Meta-World simulation environments, integrates advanced VLA models (PI0.5 and GR00T N1.5), and introduces a new plugin system for simplified hardware integration. Additional improvements include dataset editing tools, multi-GPU training via Accelerate, a new processor pipeline for data handling, phone-based teleoperation, and a comprehensive Robot Learning Course.
Table of contents
TL;DRTable-of-ContentsDatasets: Ready for the Next Wave of Large-Scale Robot LearningSimulation Environments: Expanding Your Training GroundsCodebase: Powerful Tools For EveryonePolicies: Unleashing Open-World GeneralizationRobots: A New Era of Hardware Integration with the Plugin SystemThe Hugging Face Robot Learning CourseFinal thoughts from the teamSort: