Reusable Specification Patterns for Verification of Resilience in Autonomous Hybrid Systems

Abstract

AbstractAutonomous hybrid systems are systems that combine discrete and continuous behavior with autonomous decision-making, e.g., using reinforcement learning. Such systems are increasingly used in safety-critical applications such as self-driving cars, autonomous robots or water supply systems. Thus, it is crucial to ensure their safety and resilience, i.e., that they function correctly even in the presence of dynamic changes and disruptions. In this paper, we present an approach to obtain formal resilience guarantees for autonomous hybrid systems using the interactive theorem prover KeYmaera X. Our key ideas are threefold: First, we derive a formalization of resilience that is tailored to autonomous hybrid systems. Second, we present reusable patterns for modeling stressors, detecting disruptions, and specifying resilience as a service level response in the differential dynamic logic (d$$\mathcal {L}$$ L ). Third, we combine these concepts with an existing approach for the safe integration of learning components using hybrid contracts, and extend it towards dynamic adaptations to stressors. By combining reusable patterns for stressors, observers, and adaptation contracts for learning components, we provide a systematic approach for the deductive verification of resilience of autonomous hybrid systems with reduced specification effort. We demonstrate the applicability of our approach with two case studies, an autonomous robot and an intelligent water distribution system.