Latent Barrier Steering: Hierarchical Safety for Generative Planning
Latent Barrier Steering: Hierarchical Safety for Generative Planning
This research introduces Latent Barrier Steering (LBS), a hierarchical semantic-to-path framework designed to enhance safety in flow-matching-based generative planning. Unlike existing control-barrier-function (CBF) filters that act locally on generated samples, LBS decouples behavior selection from path certification by steering the generator's behavior latent toward rollouts with larger safety margins before applying a path-space CBF quadratic program only for residual violations. Experiments across navigation and robot end-effector planning tasks demonstrate that LBS preserves safety while improving goal reaching and reducing correction burden compared with traditional path-level safety filters.
The paper introduces Latent Barrier Steering (LBS), a hierarchical framework that decouples behavior selection from path certification to improve safety margins in generative planning for navigation and end-effector tasks, with claims of finite-time recovery guarantees supported by experiments.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
latent steering increases the margin available to the path-space corrector, yielding finite-time recovery guarantees and reduced correction burden. Experiments across navigation and robot end-effector planning tasks show that LBS preserves safety while improving goal reaching and reducing correction burden compared with path-level safety filters.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org