Humanoid-GPT: Scaling Data and Structure for Zero-Shot Motion Tracking

Zekun Qi, Xuchuan Chen, Dairu Liu, Chenghuai Lin, Yunrui Lian, Sikai Liang, Zhikai Zhang, Yu Guan, Jilong Wang, Wenyao Zhang, Xinqiang Yu, He Wang, Li Yi · 2026

Read Humanoid-GPT: Scaling Data and Structure for Zero-Shot Motion Tracking by Zekun Qi, Xuchuan Chen, Dairu Liu, Chenghuai Lin, Yunrui Lian, Sikai Liang, Zhikai Zhang, Yu Guan, Jilong Wang, Wenyao Zhang, Xinqiang Yu, He Wang, Li Yi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-03985.txt.

Description

Qi and co-authors introduce Humanoid-GPT for whole-body robot control. It is a GPT-style Transformer with causal attention. Past trackers used shallow MLP nets with scarce data. They faced a trade-off between agility and broad skill. This model scales both data and structure at once.

Pretraining used a two-billion-frame corpus of retargeted motion. It unifies major motion-capture sets with large in-house recordings. The result is one generative model for motion tracking. It tracks fast, complex moves, the team reports. It also shows strong zero-shot transfer to unseen motions. New control tasks were handled with no task-specific training, they add.

Catalog record

CREATORZekun Qi, Xuchuan Chen, Dairu Liu, Chenghuai Lin, Yunrui Lian, Sikai Liang, Zhikai Zhang, Yu Guan, Jilong Wang, Wenyao Zhang, Xinqiang Yu, He Wang, Li Yi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.03985

Browse the full library · About the Tower of Babel