HumanTracker: Towards Comprehensive and Human-Aligned Motion Tracking Benchmark
HumanTracker is a benchmark containing approximately 153 hours of optical motion trajectories from multiple professional performers, organized into four motion families with text labels. It introduces HumanScore, a preference-aligned metric trained on 12K motion pairs containing 24K motions. The benchmark aims to make humanoid tracking evaluation perceptually aligned and scalable.
Humanoid motion tracking evaluation often disagrees with human perception. Kinematic errors miss physical artifacts like unstable support and incorrect contacts. Existing test suites are small and lack diversity. HumanTracker addresses these issues with a large benchmark and HumanScore, a preference-aligned metric. Across state-of-the-art trackers, HumanScore better predicts human preferences and reveals contact and stability failures that kinematic metrics miss.
HumanScore is trained on 12K motion pairs (24K motions) to align with human preferences. It outperforms kinematic metrics in predicting human judgments and detecting contact/stability failures. The benchmark's 153 hours of motion data across four families enables fine-grained diagnosis via text labels.
This benchmark could become a standard for evaluating humanoid motion tracking in teleoperation and whole-body imitation. Its focus on perceptual alignment may drive adoption in robotics companies developing humanoid control systems.
Provides a scalable evaluation tool that could reduce development time for humanoid robotics by identifying perceptual failures early. May influence procurement decisions for motion tracking solutions.
Expect follow-up research using HumanTracker to improve tracker architectures. Potential integration into robotics development pipelines. Watch for adoption by major humanoid robotics labs and benchmark leaderboards.