Perception-Limited Smooth Safety Filtering
Lyes Smaili, Soulaimane Berkane

TL;DR
This paper introduces perception-aware smooth safety filters for nonlinear control systems with limited sensing, enabling safer and smoother control actions compared to traditional methods.
Contribution
It proposes two novel perception-aware safety filtering approaches that ensure smooth, Lipschitz-safe control under sensing limitations, with theoretical guarantees and practical validation.
Findings
Enables higher-order tracking control for drones and ground robots.
Provides smoother and more robust safety behaviors than classical CBF filters.
Guarantees forward invariance and uniqueness of the closed-loop system.
Abstract
This paper develops a smooth safety-filtering framework for nonlinear control-affine systems under limited perception. Classical Control Barrier Function (CBF) filters assume global availability of the safety function - its value and gradient must be known everywhere - an assumption incompatible with sensing-limited settings, and the resulting filters often exhibit nonsmooth switching when constraints activate. We propose two complementary perception-aware safety filters applicable to general control-invariant safety sets. The first introduces a smooth perception gate that modulates barrier constraints based on sensing range, yielding a closed-form Lipschitz-safe controller with forward-invariance guarantees. The second replaces the hard CBF constraint with a differentiable penalty term, leading to a smooth unconstrained optimization-based safety filter consistent with CBF principles.…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdaptive Control of Nonlinear Systems · Stability and Control of Uncertain Systems · Advanced Control Systems Optimization
