Uniform Feasibility For Smoothed Backup Control Barrier Functions

Anil Alan, Bart De Schutter · 2025

Read Uniform Feasibility For Smoothed Backup Control Barrier Functions by Anil Alan, Bart De Schutter (2025) 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-2511-13499.txt.

Introduction

Anil Alan and Bart De Schutter prove feasibility guarantees for safety filters built from smoothed backup control barrier functions. Safe sets are often defined by the lowest of several smooth functions, which creates a nonsmooth edge. Replacing that minimum with a soft log-sum-exp smoothing is common, but the smoothed version needs proof that it still works as a barrier.

Under a strict safety condition, they show the smooth function qualifies as a barrier for a range of smoothing values and give an explicit lower bound for compact sets. Tail conditions extend the result to unbounded sets. Applied to backup methods, safety of the backup set and its boundary behavior suffice for feasibility.

Catalog record

CREATORAnil Alan, Bart De Schutter
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-13499

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Uniform Feasibility For Smoothed Backup Control Barrier Functions Anil Alan, Bart De Schutter Abstract We study feasibility guarantees for safety filters developed using Control Barrier Functions (CBFs) when a safe set is defined using the pointwise minimum of continuously differentiable functions, a construction that is common for the backup CBF method and typically nonsmooth. We replace the minimum by its log–sum–exp ( soft-min ) smoothing and show that, under a strict safety condition, the smooth function becomes a CBF (or extended CBF) for a range of the smoothing parameter. For compact safe sets, we derive an explicit lower bound on the smoothing parameter that makes the smooth function a CBF and hence renders the corresponding safety filter feasible. For unbounded sets, we introduce tail conditions under which the smooth function satisfies an extended CBF condition uniformly. Finally, we apply these results to backup CBFs. We show that safety of a compact (terminal) backup set under a backup controller, together with a condition ensuring safety of the backup trajectories on the relevant boundary of the safe set, is sufficient for feasibility for backup CBFs. These results provide a recipe for a priori feasibility guarantees for smooth inner approximations of nonsmooth safe sets without the need for additional online certification. I Introduction Safety is the main requirement for controlling modern autonomous systems. In safety-critical control, the objective is to provide a formal guarantee that, starting from a designated set, trajectories never leave this set during operation. Control Barrier Functions (CBFs) embed such guarantees as constraints in online optimization—often by penalizing the deviation from a desired control law, which leads to a design paradigm called safety filter [ 1 ] . The practicality of this framework ultimately hinges on the feasibility of the resulting optimization problem, which in turn requires a certificate that the constraint function is a valid CBF. Converse safety theorems study conditions such that a safe set admits a CBF, therefore they can be used to address the certification problem when the safe set is represented by a continuously differentiable function. Converse results mostly rely on compactness assumption of the set, and span stability-like conditions [ 18 ] , separation certificate via Meyer functions [ 24 ] , and robust variants [ 20 ] . Alternatives avoid compactness assumption at the expense of other restrictive conditions such as global Lipschitz constant [ 14 ] or bounded Lie derivatives [ 7 ] . Recently, a relaxed formulation of CBF, named extended CBF (eCBF), has been proposed to remove these assumptions from the converse side of necessity (of the existence of an eCBF), while retaining safety guarantees on the direct side of sufficiency [ 15 ] . While converse safety theorems enable CBF certification, they presuppose the existence of a control law that keeps the system within the set, or equivalently, the set must be viable 1 1 1 A set is viable if, from every point in it, the system can apply some admissible control that keeps the trajectory inside the set for all future time. under some control input in the sense of Aubin’s viability theory [ 2 ] . Ideally, one would find the largest viable set (viability kernel), but related computations are generally intractable in high dimensions due to the curse of dimensionality [ 16 ] . The backup control barrier function (BCBF) framework provides a scalable alternative [ 8 , 4 , 23 , 3 ] . Here, a backup controller is used to propagate states forward into a smaller viable “backup set,” and the set of all states that reach this backup set (at a terminal time) without leaving the safe region is implicitly defined via a minimum over time-indexed constraints, which leads to a nonsmooth. Since identifying a suitably small viable backup set can be achieved through convex relaxations or semidefinite programming, e.g., via LMIs [ 12 ] , this approach avoids direct computation of the viability kernel. The paradigm has been extended to data-driven settings via Koopman operators [ 6 ] , to reduced-order models [ 17 ] , to robust formulations [ 22 ] , and to the multiple backup controllers setting [ 11 ] . However, feasibility guarantees in these BCBF-based methods often rely on ad hoc assumptions—such as presuming the existence of suitable class- \mathcal{K} bounds without constructive certificates [ 8 , 4 ] , or imposing global compactness [ 9 ] , or bounded Lie derivatives [ 21 ] —which limits systematic deployment. Contributions. We bridge BCBFs with converse safety certification to find feasibility guarantees for CBF-based safety filters. We do this by smoothing the nonsmooth pointwise minimum via the soft-min (log–sum–exp) operator and provide a range of parameters such that converse safety theorems in [ 15 ] can be employed for CBF/eCBF certification. In particular: (i) For compact safe sets, uniform continuity can be invoked to provide CBF conditions. (ii) For noncompact sets, we introduce explicit tail conditions under which the soft-min uniformly satisfies an eCBF condition. (iii) In the BCBF setting, we show that safety of a compact backup set under a backup controller, combined with a condition enforcing safety along the relevant boundary of the safe set, is sufficient to certify the nonsmooth backup CBF as a CBF. This relaxes the global compactness assumption in [ 9 ] . While the smoothing idea also appeared in [ 19 ] for BCBFs, the approach of [ 19 ] requires computing an inner feasible subset online via an optimization problem for given design parameters. In contrast, our results provide an a priori and uniform feasibility guarantee that eliminates the need for additional online certification. Organization. Section II reviews notation, (e)CBFs, converse safety, and the BCBF setup. Section III presents the soft-min construction and uniform thresholds for compact and noncompact cases. Section IV implements these results for BCBFs. Section V concludes the paper. II Background II-A Notation We write \mathbb{I}_{N}\triangleq\{1,\cdots,N\} for an integer N\geq 1 . The set of continuously differentiable functions is denoted by C^{1} . For h:\mathbb{R}^{n}\to\mathbb{R} and h\in C^{1} , its gradient is \nabla h:\mathbb{R}^{n}\to\mathbb{R}^{n} , and the transpose of its gradient is denoted as \nabla^{\top}h . A value y\in\mathbb{R} is a regular value of h if \nabla h(x)\neq 0 for every x with h(x)=y . For a vector field f:\mathbb{R}^{n}\to\mathbb{R}^{n} , L_{f}h(x)\triangleq\nabla^{\top}\!h(x)f(x) denotes the Lie derivative of h with respect to (w.r.t.) f . For a set \mathcal{S}\subseteq\mathbb{R}^{n} , \partial\mathcal{S} and {\rm Int}(\mathcal{S}) denote its boundary and interior. A continuous function \alpha:[0,a)\to\mathbb{R}_{\geq 0} belongs to class- \mathcal{K} if it is strictly increasing and \alpha(0)=0 . If a=\infty and \alpha(r)\to\infty as r\to\infty , then \alpha\in\mathcal{K}_{\infty} . A continuous function \gamma:\mathbb{R}_{\geq 0}\times\mathbb{R}_{\geq 0}\to\mathbb{R}_{\geq 0} belongs to class- \mathcal{K}_{\infty}\mathcal{K} if \gamma(\cdot,s)\in\mathcal{K}_{\infty} for all s\geq 0 and \gamma(r,\cdot)\in\mathcal{K} for all r\geq 0 . II-B Control Barrier Functions and Converse Safety Theorems Consider a system \dot{x}=f(x,u),\quad x(0)=x_{0}, (1) with x\in\mathbb{R}^{n} and u\in\mathbb{U}\subseteq\mathbb{R}^{m} , where f:\mathbb{R}^{n}\times\mathbb{R}^{m}\to\mathbb{R}^{n} is locally Lipschitz. We aim at safety , that the closed-loop system with a state-feedback controller u=k(x) stays in a predefined set \mathcal{S} . We will consider sets defined as \begin{split}\mathcal{S}\triangleq\{x\in\mathbb{R}^{n}~|~h(x)\geq 0\},\quad h:\mathbb{R}^{n}\to\mathbb{R},~~h\in C^{1}.\end{split} (2) Control Barrier Functions (CBFs) can be used to construct controllers with a safety guarantee. Definition 1 (CBF, [ 1 ] )

Read the complete text