Summary
LeRobot v0.5.0 has been released as the project’s largest update to date. The version adds its first full-body-control integration for the Unitree G1 humanoid robot and introduces six new policy approaches: Pi0-FAST, an autoregressive vision-language-action model; Real-Time Chunking; Wall-X; X-VLA; SARM; and parameter-efficient fine-tuning. Training and inference have also been improved through streaming video encoding and image-training speeds that are reportedly up to 10 times faster. A new EnvHub feature allows simulation environments to be loaded directly from Hugging Face Hub. The release also upgrades infrastructure requirements to Python 3.12 or later and Transformers v5, with support for NVIDIA IsaacLab-Arena. LeRobot remains an open-source robotics project developed by Hugging Face.
LeRobot v0.5.0 Released with Humanoid Robot Support
LeRobot v0.5.0 has been released as the project’s largest update to date.
Key Highlights
- Unitree G1 humanoid robot: LeRobot now supports a humanoid system with full-body control for the first time.
- Six new policy approaches:
- Pi0-FAST, an autoregressive vision-language-action model
- Real-Time Chunking
- Wall-X
- X-VLA
- SARM
- Parameter-efficient fine-tuning (PEFT)
- Performance improvements: Streaming video encoding is supported, while image-training speed has increased by a reported factor of 10.
- EnvHub: Simulation environments can now be loaded directly from Hugging Face Hub.
- Infrastructure updates: The release requires Python 3.12 or later, uses Transformers v5, and supports NVIDIA IsaacLab-Arena.
Project: https://github.com/huggingface/lerobot
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177168842