Summary
HumanTracker is a new benchmark and evaluation framework for humanoid motion tracking, a core capability for teleoperation and whole-body imitation. The authors argue that conventional kinematic error metrics average per-frame pose differences and miss the physical artifacts humans notice most, such as unstable support, foot skating, and mistimed touch-downs. The HumanTracker benchmark provides approximately 153 hours of optical motion trajectories from multiple professional performers, organized into four motion families with text labels for fine-grained diagnosis. In addition, HumanScore is a preference-aligned metric trained on 12K motion pairs containing 24K motions. Experiments across representative state-of-the-art trackers show that HumanScore better predicts human preferences and exposes contact and stability failures that kinematic metrics often miss, making humanoid tracking evaluation both perceptually aligned and scalable. Full paper: arXiv 2608.13555.
Overview
- Field: Computer Vision (CV)
- Authors: Dairu Liu, Zekun Qi, Jiayu Zeng, Ruixi Yu, Yu Guan, Yintianrun Zhang, Xuchuan Chen, Sikai Liang, Zekai Li, Chenghuai Lin, Xinqiang Yu, Wenyao Zhang, He Wang, Li Yi
- arXiv: 2608.13555
Abstract
Humanoid motion tracking is central to teleoperation and whole-body imitation, yet evaluation often disagrees with what people perceive in videos. Kinematic errors average per-frame pose differences but miss the physical artifacts that matter most, particularly unstable support and incorrect contacts such as foot skating and mistimed touch-downs. Meanwhile, widely used test suites are small and lack the diversity needed to stress contact-rich, long-horizon behaviors.
The authors introduce HumanTracker to make humanoid tracking evaluation both perceptually aligned and scalable. Key contributions:
- HumanTracker benchmark: approximately 153 hours of optical motion trajectories from multiple professional performers, organized into four motion families with text labels for fine-grained diagnosis.
- HumanScore: a preference-aligned metric trained on 12K motion pairs containing 24K motions.
Across representative state-of-the-art trackers, HumanScore better predicts human preferences and reveals contact and stability failures that kinematic metrics often miss.
---
*Auto-collected on 2026-08-15*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178633496