GenTrack: Physical Alignment for Robot-Native Motion Generation and Zero-Shot Humanoid Tracking

2026-08-02

AI Summary

This work addresses the executability gap between text-generated motions and humanoid robots’ physical capabilities, as well as the limited coverage of zero-shot motion tracking. To bridge this gap, the authors propose GenTrack, a framework that enables online co-fine-tuning of a motion generator and a tracker for the first time. The approach alternates between aligning the generator using real-world execution feedback and updating the tracker, while incorporating anchoring constraints and a replay mechanism to mitigate policy drift. Built upon ProtoMotions and the SONIC backbone with group-relative physical alignment, GenTrack significantly enhances the quality of generated motions and zero-shot tracking generalization without requiring additional data collection. Experiments on the Unitree G1 platform demonstrate that the generated motions are more executable, and the tracker achieves broader coverage and higher accuracy across AMASS, LAFAN, and 1,024 out-of-distribution in-the-wild motion prompts.

Abstract

General-purpose humanoid trackers can execute diverse references, but their zero-shot coverage depends on large embodied corpora that are costly to extend. Text-to-motion generators offer scalable supervision, yet models trained on human motion or retargeted data inherit a gap between kinematic plausibility and robot executability. Existing one-way pipelines fix either the generated corpus or the reward tracker. We introduce GenTrack, an online generator--tracker framework that alternates execution-grounded, group-relative generator alignment with tracker training on newly generated references; anchoring and rehearsal constrain drift. On Unitree G1, we evaluate GenTrack with ProtoMotions and SONIC backbones across three zero-shot tracking splits including public AMASS and LAFAN benchmarks, and a private out-of-distribution test set of 1,024 prompt-motion pairs in the wild. The online co-training strategy consistently produces generators that output more robot-executable motions with strong semantic alignment, and trackers with markedly broader zero-shot coverage and improved tracking accuracy, especially on out-of-distribution references. These results demonstrate that joint online post-training effectively narrows the executability gap between retargeted references and robot-native motion, advancing zero-shot humanoid control without additional data collection and beyond the limitations of a static reference pool.

Problem

• zero-shot humanoid control
• robot-native motion generation
• motion executability gap
• embodied motion corpora
• human-to-robot retargeting

Innovation

• online co-training
• robot-native motion generation
• zero-shot humanoid tracking
• execution-grounded alignment
• semantic-motion executability gap