Constraining Streaming Flow Models for Adapting Learned Robot Trajectory Distributions
Jieting Long, Dechuan Liu, Weidong Cai, Ian Manchester, Weiming Zhi · 2026
Read Constraining Streaming Flow Models for Adapting Learned Robot Trajectory Distributions by Jieting Long, Dechuan Liu, Weidong Cai, Ian Manchester, Weiming Zhi (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-2602-15567.txt.
Description
Long and colleagues give robot motion models a way to obey rules after training. Robot paths are often multi-modal, and streaming flow policies capture that spread well by integrating learned velocity fields straight in action space for smooth, reactive control. But those policies have no built-in way to adapt once trained, so safety limits and task constraints cannot be imposed at run time. The team answers with Constraint-Aware Streaming Flow, or CASF.
Each constraint, in workspace or configuration space, becomes a differentiable distance function turned into a local metric and pulled back into control space. Far from barred regions the metric is just the identity, while near a boundary it gently slows or steers motion away. Tests in simulation and on real robots show smooth, feasible, constraint-respecting paths that keep the multi-modal, reactive character of the base policy and beat post-hoc projection baselines.
Catalog record
| CREATOR | Jieting Long, Dechuan Liu, Weidong Cai, Ian Manchester, Weiming Zhi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-15567