PHUMA: Physically-Grounded Humanoid Locomotion for Robotics Retargeting

PHUMA is a physically-grounded humanoid locomotion dataset developed by DAVIAN-Robotics, updated in February 2026. It provides motion trajectories specifically designed for humanoid robotics research, focusing on retargeting and locomotion tasks. The data is intended to bridge the gap between raw motion capture and physics-based robot execution.

Use Cases

• Training reinforcement learning models for humanoid walking using physically-grounded trajectories
• Developing motion retargeting algorithms to map human movement to robot skeletons
• Benchmarking physics-based controllers for humanoid stability and gait patterns

Strengths

• Physically-grounded data ensures compatibility with physics engines like MuJoCo or Isaac Gym
• Released under the permissive Apache-2.0 license
• Specialized focus on humanoid-specific locomotion and retargeting

Limitations

• Specific row counts and column schemas are not explicitly detailed in the source documentation
• Narrow focus on humanoid locomotion limits applicability to non-bipedal robotics

Provenance

Source

DAVIAN-Robotics

Collection Method

Physically-grounded motion generation or simulation

Freshness

Last updated February 2026.