CoStream: Composing Simple Behaviors for Generalizable Complex Manipulation
Haonan Chen, Yuxiang Ma, Stephen Tian, Xiaoshen Han, Wenlong Huang, Feiyang Wu, Yunzhu Li, Jiajun Wu, Edward H. Adelson, Yilun Du, [4pt] Harvard University · 2026
Read CoStream: Composing Simple Behaviors for Generalizable Complex Manipulation by Haonan Chen, Yuxiang Ma, Stephen Tian, Xiaoshen Han, Wenlong Huang, Feiyang Wu, Yunzhu Li, Jiajun Wu, Edward H. Adelson, Yilun Du, [4pt] Harvard University (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-2606-26423.txt.
Description
Haonan Chen and colleagues at Harvard tackle long, contact-rich robot tasks, like seating a graphics card into its slot, which need millimeter precision yet must generalize to new jobs. Rigid pipelines are precise but brittle; single learned policies generalize but lack precision.
Their framework, CoStream, composes three simple behaviors instead: one extracts spatial constraints using foundation models, one forecasts trajectories by tracking keypoints in imagined video, and one applies fast tactile corrections. Their outputs combine into one pose command at each control step. Demonstrated on eight real-world tasks, the approach aims to get precision and adaptability from composition rather than retraining.
Catalog record
| CREATOR | Haonan Chen, Yuxiang Ma, Stephen Tian, Xiaoshen Han, Wenlong Huang, Feiyang Wu, Yunzhu Li, Jiajun Wu, Edward H. Adelson, Yilun Du, [4pt] Harvard University |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.26423