Researchers release large-scale HiPHI motion capture dataset

Researchers have introduced the HiPHI dataset, a large-scale motion capture collection designed to advance humanoid robot learning. Covering over 617 hours of precise human movement—including 246 hours of detailed human-object interactions—HiPHI goes beyond previous datasets by incorporating synchronized object data and using a linguistic framework to organize actions. Early tests show reinforcement learning models trained on HiPHI transfer effectively to physical humanoid robots.

• HiPHI captures 617.5 hours of whole-body human motion
• Dataset includes synchronized object trajectories and meshes
• Actions are categorized using the FrameNet linguistic framework
• Contains 245.7 hours of human-object interaction
• Reinforcement learning models improved when trained on HiPHI