LeRobot v0.5.0: Scaling Every Dimension

Summary

LeRobot v0.5.0 is a major release that expands supported hardware, policies, datasets, simulation, and its software foundation, following more than 200 merged PRs and more than 50 new contributors since v0.4.0. It adds full Unitree G1 humanoid support for locomotion, manipulation, teleoperation, and whole-body control, alongside OpenArm, Earth Rover, CAN-bus motor controllers, and other integrations. The policy additions include Pi0-FAST's autoregressive VLAs with FAST action tokenization, Real-Time Chunking for continuously blended predictions, Wall-X, X-VLA, SARM, and PEFT fine-tuning. For data and simulation workflows, streaming video encoding removes waits between recording episodes, while EnvHub loads environments from the Hugging Face Hub and IsaacLab-Arena supplies GPU-accelerated simulation. The codebase now requires Python 3.12, uses Transformers v5, supports installable third-party policy plugins, and adds remote Rerun visualization plus installation and documentation improvements.

Takeaways

1. 1

   Full Unitree G1 support covers locomotion, manipulation, teleoperation, and whole-body control, marking LeRobot's first humanoid integration.

2. 2

   Pi0-FAST adds autoregressive Vision-Language-Action modeling with FAST action tokenization, while Real-Time Chunking blends new predictions with in-progress actions for more responsive inference.

3. 3

   Streaming video encoding processes frames during capture and can automatically use hardware video encoders, eliminating the wait between recording episodes.