Unitree Releases Open-Source Full-Body Motion Dataset for Humanoid Robots, Powered by Pinocchio

Unitree Robotics is excited to announce the release of the LAFAN1_Retargeting_Dataset, an open-source dataset designed to make humanoid robot movements more natural and fluid. Hosted on their official Hugging Face page (link), this dataset supports three Unitree humanoid models: H1, H1-2, and G1.

To achieve this breakthrough, Unitree leveraged Pinocchio, the core robotics dynamics library of the Maestro Consortium. By combining LAFAN1 motion capture data with numerical optimization techniques based on Interaction Mesh and Inverse Kinematics (IK), Unitree refined motion trajectories while ensuring compliance with end-effector pose constraints, joint position, and velocity limits—all while preventing foot slippage.

The results are showcased in an impressive simulation video. While currently in simulation, the true milestone will be deploying these optimized motions on real humanoid robots.

This dataset, built upon cutting-edge tools like Pinocchio, provides a powerful resource for researchers and developers aiming to push the boundaries of humanoid robotics.

Links

Explore the Dataset: Unitree Hugging Face Page

Watch the Simulation Video: YouTube Video