Towards Scalable and Feature-Driven Validation of Cyber-Physical Systems via Active Exploration of Scenario Spaces
Osama Maqbool, Jürgen Roßmann · 2025
The advent of closely coupled intelligent subsystems has led to highly complex cyber-physical systems (CPS) which can no longer be reliably validated by conventional rule-based and component-level tests alone. Instead, large amount of system-level tests are required that can a) capture the complex interactions between subsystems , b) provide comprehensive test coverage of system-and operational states, and c) identify critical features in system behavior. This paper presents a large scale simulation-based validation methodology that addresses these requirements through a synthesis of three technologies. Firstly, formal and human-readable scenarios are used to systematically define a scenario space spanning the complete operational domain of the system. Secondly, digital twins (DT) of the system are used for flexible and realistic virtual tests that closely mirror the full system complexity. Finally, active learning is employed to efficiently explore the scenario space with scalable batch simulations, focusing particularly on different types of critical behaviors. The methodology is demonstrated on an aerospace rendezvous and docking maneuver to deliver estimates of system performance with a special focus around critical behaviors. Results show that our method identifies more critical scenarios than plateaus reached by random methods while requiring a fraction of the computational cost.