Global-Local Attention Decomposition for Terrain Encoding in Humanoid Perceptive Locomotion
Shengcheng Fu, Yang Zhang, Zhanxiang Cao, Liyun Yan, Yizhi Chen, Yunpeng Yin, Yue Gao, Tongji University, Shanghai, China · 2026
Read Global-Local Attention Decomposition for Terrain Encoding in Humanoid Perceptive Locomotion by Shengcheng Fu, Yang Zhang, Zhanxiang Cao, Liyun Yan, Yizhi Chen, Yunpeng Yin, Yue Gao, Tongji University, Shanghai, China (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-2606-00637.txt.
Description
Fu and co-authors split ground reads for humanoid walks into world and local watch strands. Reward-taught walks have leapt in skill of late. Sight-led plans still trip on thin step spots and tight rooms. Broad ground sense and fine step picks are both a must there.
Their frame, GLAD, splits those two jobs in a clean way. A world strand pools watch weight to sum the ground round the bot. A local strand thins traits by ground pull and state fit. That split keeps fine step cues from wash-out in the pool. Tests span gaps, step stones, and stairs in sim runs. Set acts spring up, like thin path trails and block dodges. A Unitree G1 bot runs the plan in real rooms by its own lidar.
Catalog record
| CREATOR | Shengcheng Fu, Yang Zhang, Zhanxiang Cao, Liyun Yan, Yizhi Chen, Yunpeng Yin, Yue Gao, Tongji University, Shanghai, China |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.00637