Latent Barrier Steering adds hierarchical safety to generative planners
Original title: Latent Barrier Steering: Hierarchical Safety for Generative Planning
On navigation and robot end-effector tasks, LBS kept safety while improving goal reaching versus path-level filters.
Researchers introduce Latent Barrier Steering, a hierarchical framework for safe flow-matching generative planning. Local control-barrier filters can fail when a new constraint blocks an entire behavior mode rather than a small local violation. LBS first steers the behavior latent toward rollouts with a larger safety margin, then applies a path-space CBF quadratic program only for residual violations. Experiments on navigation and robot end-effector planning report preserved safety, better goal reaching, and a lower correction burden than path-level filters.
Unconfirmed: Safety and goal-reaching gains are reported only for the paper’s navigation and end-effector experiments.
- Latent steering is claimed to increase the margin available to the path-space corrector.
- The authors state finite-time recovery guarantees and a reduced correction burden.
- Compared filters act during sampling or after prediction on generated samples.
From the original article
1 min read at arxiv.orgWe introduce Latent Barrier Steering (LBS), a hierarchical semantic-to-path framework for safe flow-matching-based generative planning. Existing control-barrier-function (CBF) safety filters for diffusion and flow-matching planners act locally on generated samples, either during sampling or after prediction.
By Renhao Zhang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.