A Scalable Compositional Falsification Approach for Identifying Challenging Scenarios in Cyber-Physical Systems

This article presents a scalable approach for identifying system inputs and trajectories that lead to undesirable scenarios in a cyber-physical system (CPS). The proposed approach falls under the broad class of falsification methods. The objective in falsification is to find initial conditions and input signals resulting in trajectories that violate a system property expressed as a formal specification. The existing falsification methods are not suitable for handling systems where the input at a specific time depends on the state history of the system. Autonomous vehicle platoons and certain multi-agent systems that are required to maintain connectivity are some examples of such systems. In the first part of this article, we employ a graph-search-based motion planning algorithm to develop a falsification method capable of identifying falsifying inputs for the aforementioned class of systems. We demonstrate the effectiveness of the approach through a case study involving a vehicle platoon consisting of three autonomous vehicles, wherein one of the vehicles is tasked to execute a lane-change maneuver. In the second part, we extend the proposed approach to analyze a network of three unmanned underwater vehicles, wherein our objective is to identify system inputs resulting in trajectories that lead to a deadlock. The approach leverages the component-based structure inherent in the considered multi-agent system and employs a surrogate model in place of computationally demanding components present in the system. These modifications are shown to improve the scalability of the approach and reduce the computational time in finding falsifying trajectories.