Safety-Critical Control of Discontinuous Systems with Nonsmooth Safe Sets footnoteinfo

Mohammed Alyaseen, Nikolay Atanasov, Jorge Cortes · 2024

Read Safety-Critical Control of Discontinuous Systems with Nonsmooth Safe Sets footnoteinfo by Mohammed Alyaseen, Nikolay Atanasov, Jorge Cortes (2024) 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-2412-15437.txt.

Introduction

Mohammed Alyaseen, Nikolay Atanasov, and Jorge Cortes design controllers that keep jumpy systems inside safe zones with rough edges. Their paper from twenty twenty-four treats dynamics that switch in steps, not smooth flows. The safe zone is built by nesting unions and crossings of simple level sets. Standard safety tools assume smooth motion.

The team shows a trap: a controller that checks only the active safety rule at each point can still be unsafe. To fix this, they add transition functions. These let a path leave one part of the safe zone for another without undue caution. The controller solves a convex problem at each step. It stays feasible and varies smoothly where the dynamics do. A multi-agent reshaping task shows the design at work.

Catalog record

CREATORMohammed Alyaseen, Nikolay Atanasov, Jorge Cortes
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-15437

Browse the full library · About the Tower of Babel

Full text

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

Safety-Critical Control of Discontinuous Systems with Nonsmooth Safe Sets footnoteinfo Mohammed Alyaseen, Nikolay Atanasov, Jorge Cortes Abstract This paper studies the design of controllers for discontinuous dynamics that ensure the safety of non-smooth sets. The safe set is represented by arbitrarily nested unions and intersections of 0-superlevel sets of differentiable functions. We show that any optimization-based controller that satisfies only the point-wise active safety constraints is generally un-safe, ruling out the standard techniques developed for safety of continuous dynamics. This motivates the introduction of the notion of transition functions, which allow us to incorporate even the inactive safety constraints without falling into unnecessary conservatism. These functions allow system trajectories to leave a component of the nonsmooth safe set to transition to a different one. The resulting controller is then defined as the solution to a convex optimization problem, which we show is feasible and continuous wherever the system dynamics is continuous. We illustrate the effectiveness of the proposed design approach in a multi-agent reconfiguration control problem. keywords safety; discontinuous systems; nonsmooth control barrier functions; optimization-based control design † † thanks: [ † † address: University of California San Diego, 9500 Gilman Dr, La Jolla, CA 92093, USA footnoteinfo]This paper was not presented at any IFAC meeting. Corresponding author Mohammed Alyaseen. All authors are with the Contextual Robotics Institute, University of California San Diego. M. Alyaseen is also affiliated with Kuwait University as a holder of a scholarship. This work was partially supported by AFOSR Award FA9550-23-1-0740. 1 Introduction Safety-critical control is a fundamental problem in numerous domains including autonomous driving, power and transportation systems, robotics, and even mitigation of epidemic spreading. This paper tackles the synthesis of easy-to-compute controllers for the objective of safety. Specifically, we address the problem of safety critical control to render a desired nonsmooth set safe under the trajectories of discontinuous dynamics. Discontinuous dynamical systems are convenient for modeling a wide range of behaviors, including motion subject to Coulomb friction, systems with abrupt changes in forces, and dynamics of multi-agent systems under connectivity maintenance or collision avoidance constraints. Likewise, the flexibility enabled by nonsmooth safe sets allows us to encode safety requirements that cannot be expressed merely by a differentiable function, such as keeping the system evolution in or out of a set with nonsmooth boundary, or in some nonsmooth set and out of another simultaneously. Allowing nonsmoothness is also motivated by the fact that dealing with multiple constraints, some of which are active in some regions while different ones are active in others, can be represented as an overall nonsmooth safety requirement. The unique challenge tackled here is constructing an explicit, provably feasible, sufficiently regular, optimization-based safe controller for discontinuous systems with general non-smooth sets defined by nested unions and intersections of smooth component sets. Literature review: Safety-critical control is often formulated mathematically as a control design problem to achieve forward invariance of a desired set of the state space. The notion of control barrier function (CBF) has gained popularity as a tool for invariance control due to its versatility [ 3 , 2 , 23 ] . Broadly speaking, a CBF is a function for which every system state in its 0-superlevel set admits an input value that does not allow the dynamics to leave the superlevel set instantaneously. For convenience, we refer to this point-wise condition as the CBF condition. Achieving control invariance via CBFs is a twofold undertaking. First, one must find a function for which the CBF condition holds. Second, one must design a controller with adequate regularity properties that keeps the the superlevel set of the function invariant. These two problems are in general distinct. Even if a differentiable CBF is available for continuous dynamics, there might not exist a continuous controller rendering its superlevel set invariant, cf. [ 1 ] . In the classical setting, a CBF is a differentiable function whose 0-superlevel set is made control-invariant by employing a Lipschitz controller that satisfies the CBF condition [ 3 ] . A widely used technique for constructing such a controller is quadratic programming (QP), constrained by the CBF condition, which is linear in the input for control-affine dynamics [ 2 ] . The differentiability requirement of the CBF stems from the use of its gradient in the CBF condition, while the Lipschitz requirement on the dynamics and controller ensures existence and uniqueness of solutions for the closed-loop system [ 13 ] . This excludes discontinuous dynamics and non-smooth safe sets, which can be limiting in many applications. However, these two structural assumptions are not indispensable, and indeed many works have explored the extension of CBF techniques beyond them. In one line of work, safety conditions are developed for autonomous hybrid systems and non-smooth barrier functions [ 15 ] . These conditions extend to systems modeled by differential inclusions with control inputs as in [ 8 , 12 ] . The work [ 8 ] provides general conditions for the existence of a continuous safe controller but does not provide an explicit one. Instead, [ 12 ] gives an explicit optimization-based continuous safe controller for non-smooth safe sets defined as an intersection of superlevel sets of differentiable functions. The work [ 9 ] gives sufficient conditions for safety of differential inclusions and general non-smooth barrier functions, and designs an optimization-based controller for sets defined by the intersection of multiple superlevel sets. This controller is safe assuming the optimization is feasible but no conditions are provided for the latter. These methods have been successfully applied for multi-agent network connectivity [ 16 ] . Other works [ 20 ] are not limited to continuous controllers and instead develop Lebesgue-measurable controllers for non-smooth CBFs whose superlevel sets are given as an intersection of multiple differentiable functions. The works mentioned hitherto give explicit safe controllers only for non-smooth sets defined as intersections, that is, conjunctive safety constraints. When the safe set is given as a union of component sets, the safety constraint on the controller changes depending on the component set in which the current state lies, leading to an optimization with disjunctive constraints. Such problems are equivalent to mixed-integer optimization programs [ 10 ] and thus are more challenging to deal with. Disjunctive safety constraints appear in another line of work, that of enforcing signal temporal logic (STL) specifications as barrier functions. The work [ 14 ] models conjunctive signal temporal tasks for Lipschitz dynamics as non-smooth barrier functions and provides a generally discontinuous optimization-based controller to enforce such constraints. The works [ 22 , 24 ] do the same for logical specifications including disjunction. Although the STL-induced CBF is more general than mere time-independent CBFs, the controller developed in those works is generally discontinuous, even with the assumed Lipschitz dynamics, and is provably un-safe for discontinuous systems, as we establish later in the paper. Contributions: We consider the problem of constructing an explicit, provably feasible, optimization-based safe controller for discontinuous systems with non-smooth sets defined by nested unions and intersections of smooth sets. We start by considering the active-component QP controller, an optimization-based controller satisfying only the point-wise active safety constraints. We show that this controller is generally un-safe for discontinuous systems and nonsmooth safe sets, thereby excluding the possibility of using

Read the complete text