Sign in to view source links and access this dataset
Description
Motion capture data from the LAFAN1 library retargeted for Unitree humanoid robots (H1, H1_2, and G1) by author lvhaidong in 2025. The data was generated using numerical optimization techniques, including Interaction Mesh and Inverse Kinematics (IK), to ensure natural movement while maintaining end-effector pose and joint velocity constraints.
Use Cases
Training motion synthesis models for Unitree H1 robots using joint position and velocity constraints
Evaluating retargeting algorithms against the provided Interaction Mesh optimizations
Developing natural gait controllers for G1 humanoid models based on human motion capture
Strengths
Supports three specific Unitree humanoid models: H1, H1_2, and G1
Optimized via Interaction Mesh and IK to minimize foot slippage
Derived from the established LAFAN1 motion capture benchmark
Limitations
Kinematic-only data which excludes dynamic stability and torque constraints
Unknown total record count and file format specifications
Restricted to the specific hardware configurations of the H1 and G1 series
Provenance
Source
LAFAN1 motion capture dataset
Collection Method
Synthetic/Numerical optimization using Interaction Mesh and IK
Freshness
Last updated April 2025.
Geography
United States
The retargeting only accounts for kinematics; users must implement their own physics-based balancing or torque control as these are not included in the dataset.