Learning Native Continuation for Action Chunking Flow Policies This work was done during the internship at Spirit AI. † Corresponding authors. ‡ Project leader.
Yufeng Liu 1,2 , Hang Yu 2,4 , Juntu Zhao 1,2 , Bocheng Li 2,5 , Di Zhang 2,4 , Mingzhu Li 2 , Wenxuan Wu 2 , Yingdong Hu 3 , Junyuan Xie 2 , Junliang Guo 2 ‡ , Dequan Wang 1 † , Yang Gao 2,3 † · 2026
Read Learning Native Continuation for Action Chunking Flow Policies This work was done during the internship at Spirit AI. † Corresponding authors. ‡ Project leader. by Yufeng Liu 1,2 , Hang Yu 2,4 , Juntu Zhao 1,2 , Bocheng Li 2,5 , Di Zhang 2,4 , Mingzhu Li 2 , Wenxuan Wu 2 , Yingdong Hu 3 , Junyuan Xie 2 , Junliang Guo 2 ‡ , Dequan Wang 1 † , Yang Gao 2,3 † (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-2602-12978.txt.
Description
Robot arms that act in chunks can stutter at chunk boundaries, breaking smooth motion. Yufeng Liu and eleven co-authors propose Legato, a method that teaches continuation during training for action-chunked flow policies in vision-language-action models. Real-time chunking patches the problem from outside the policy, but can cause spurious mode switching and paths that lack intrinsic smoothness.
Legato instead starts denoising from a schedule-shaped mix of known actions and noise, letting the model see partial action information during training. It reshapes flow dynamics so denoising stays aligned in training and in use, under per-step guidance, with randomised schedule conditioning for varying delays and controllable smoothness. Real-world tests across five manipulation tasks show Legato beating real-time chunking, with about 10% gains in both smoothness and completion time.
Catalog record
| CREATOR | Yufeng Liu 1,2 , Hang Yu 2,4 , Juntu Zhao 1,2 , Bocheng Li 2,5 , Di Zhang 2,4 , Mingzhu Li 2 , Wenxuan Wu 2 , Yingdong Hu 3 , Junyuan Xie 2 , Junliang Guo 2 ‡ , Dequan Wang 1 † , Yang Gao 2,3 † |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-12978