STEP: Warm-Started Visuomotor Policies with Spatiotemporal Consistency Prediction
Jinhao Li, Yuxuan Cong, Yingqiao Wang, Hao Xia, Shan Huang, Yijia Zhang, Ningyi Xu, Guohao Dai · 2026
Read STEP: Warm-Started Visuomotor Policies with Spatiotemporal Consistency Prediction by Jinhao Li, Yuxuan Cong, Yingqiao Wang, Hao Xia, Shan Huang, Yijia Zhang, Ningyi Xu, Guohao Dai (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-2602-08245.txt.
Description
Diffusion policies give robots flexible, multi-sense action spreads, but step-by-step cleaning makes them too slow for fast closed-loop control. Jinhao Li and seven co-authors propose STEP, a lightweight tool that predicts space-time steady warm-start actions for eye-hand policies.
The warm starts sit close to the target action spread and stay in time steady, keeping the diffusion policy's making ability while cutting judgment work. A speed-aware tweak tool then tunes drive excitation based on how fast actions change, preventing execution stalls. Robot manipulation teams gain faster closed-loop control without sacrificing the multimodality that made diffusion policies attractive in the first place.
Catalog record
| CREATOR | Jinhao Li, Yuxuan Cong, Yingqiao Wang, Hao Xia, Shan Huang, Yijia Zhang, Ningyi Xu, Guohao Dai |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-08245