PHUMA: Physically-Grounded Humanoid Locomotion Dataset
Kyungmin Lee, Sibeen Kim, Minho Park, Hyunseung Kim, Dongyoon Hwang, Hojoon Lee, Jaegul Choo
2025-11-04
Summary
This paper introduces a new dataset called PHUMA designed to help robots learn to walk and move like humans, focusing on making the movements realistic and stable.
What's the problem?
Teaching robots to walk like people is hard because good data of human movement is expensive and limited. While researchers have tried using videos from the internet to create more data, these videos often contain errors like characters floating in the air, limbs passing through objects, or feet sliding instead of properly stepping, making it difficult for robots to learn correctly.
What's the solution?
The researchers created PHUMA, a large dataset of human motion captured from videos, but they carefully cleaned and adjusted the data. They used rules based on physics to make sure the movements were realistic, preventing things like floating or feet sliding. This ensures the robot learns to move in a way that’s actually possible in the real world.
Why it matters?
This is important because it allows robots to learn more natural and stable walking patterns without needing expensive, professionally recorded motion capture data. By using a larger, more reliable dataset like PHUMA, robots can better imitate human movement and perform a wider variety of tasks.
Abstract
Motion imitation is a promising approach for humanoid locomotion, enabling agents to acquire humanlike behaviors. Existing methods typically rely on high-quality motion capture datasets such as AMASS, but these are scarce and expensive, limiting scalability and diversity. Recent studies attempt to scale data collection by converting large-scale internet videos, exemplified by Humanoid-X. However, they often introduce physical artifacts such as floating, penetration, and foot skating, which hinder stable imitation. In response, we introduce PHUMA, a Physically-grounded HUMAnoid locomotion dataset that leverages human video at scale, while addressing physical artifacts through careful data curation and physics-constrained retargeting. PHUMA enforces joint limits, ensures ground contact, and eliminates foot skating, producing motions that are both large-scale and physically reliable. We evaluated PHUMA in two sets of conditions: (i) imitation of unseen motion from self-recorded test videos and (ii) path following with pelvis-only guidance. In both cases, PHUMA-trained policies outperform Humanoid-X and AMASS, achieving significant gains in imitating diverse motions. The code is available at https://davian-robotics.github.io/PHUMA.