Hugging Face Expands LeRobot with Unitree G1 Humanoid Support and Faster Robotics Workflows

Hugging Face published LeRobot v0.5.0 on March 9, 2026, calling it the project’s biggest release yet. The update expands the open-source robotics stack across hardware, policies, datasets, simulation, and tooling, with more than 200 merged pull requests and over 50 new contributors since v0.4.0.

The headline hardware addition is full Unitree G1 humanoid support. Hugging Face says the integration covers locomotion, manipulation, teleoperation, and whole-body control, making G1 the first humanoid platform supported by LeRobot. The release also adds OpenArm and OpenArm Mini, Earth Rover, OMX Robot, and new CAN bus motor-controller support through RobStride and Damiao.

On the model side, LeRobot v0.5.0 adds Pi0-FAST autoregressive vision-language-action models, Real-Time Chunking for more responsive inference, Wall-X, X-VLA, SARM, and PEFT support for large VLAs. The update is aimed at both simulation training and real-hardware deployment, which is important because open robotics stacks often fragment when teams move between those two settings.

Dataset and training throughput also improved. Hugging Face says streaming video encoding removes wait time between recorded episodes, image training is up to 10x faster, and encoding is 3x faster. The release also adds subtask annotation, image-to-video conversion, and additional dataset-editing tools that should reduce preprocessing friction for large-scale robot data collection.

Another major change is EnvHub, which lets developers load simulation environments directly from the Hugging Face Hub instead of wiring them locally. Hugging Face also integrated NVIDIA IsaacLab-Arena for GPU-accelerated manipulation environments, giving LeRobot a clearer path from community-shared environments to large-scale reinforcement learning and policy evaluation.

The codebase now requires Python 3.12+, migrates to Transformers v5, supports third-party policy plugins, and updates PyTorch bounds for NVIDIA Blackwell GPUs. Taken together, the release shows Hugging Face trying to make embodied AI development look more like the modern model ecosystem: shared components, Hub-native distribution, faster data pipelines, and broader humanoid support inside one open framework.