PHUMA
收藏资源简介:
PHUMA是一个大规模的人形机器人运动数据集,它通过仔细的数据整理和基于物理的重新定位,解决了现有数据集中常见的物理问题,如漂浮、穿透和脚滑。PHUMA数据集包含来自各种来源的76000个运动剪辑,共计73小时,涵盖了广泛的运动类型,如跑步、跳跃和蹲下。该数据集通过物理感知的整理过程和物理约束的重新定位方法构建,以确保运动的物理可行性和可靠性。PHUMA旨在推动人形机器人运动模仿的研究,以实现稳定和类似人类的运动。
PHUMA is a large-scale humanoid robot motion dataset. It addresses prevalent physical anomalies—such as floating limbs, spatial penetration, and foot slippage—that frequently occur in existing datasets, via rigorous data curation and physics-based retargeting. The PHUMA dataset contains 76,000 motion clips from diverse sources, totaling 73 hours of content, and covers a wide spectrum of motion types including running, jumping, and squatting. Built through a physics-aware curation pipeline and physics-constrained retargeting method, the dataset ensures the physical feasibility and reliability of all included motions. PHUMA is designed to advance research on humanoid robot motion imitation, with the objective of achieving stable and human-like robotic movements.




