Search papers, labs, and topics across Lattice.
This paper introduces GenTrack, an innovative online generator-tracker framework that enhances humanoid motion generation and tracking by alternately aligning execution-grounded generation with tracker training on newly generated references. The method addresses the gap between kinematic plausibility and robot executability, achieving improved tracking accuracy and broader zero-shot coverage without the need for extensive additional data collection. Evaluations on both standard benchmarks and a private out-of-distribution test set demonstrate that GenTrack significantly narrows the executability gap, leading to more effective humanoid control in diverse scenarios.
Joint online post-training in GenTrack boosts humanoid tracking accuracy and expands zero-shot capabilities without requiring new data.
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.
