Verified Runtime Validation for Partially Observable Hybrid Systems (1811.06502v2)

Abstract: Formal verification provides strong safety guarantees but only for models of cyber-physical systems. Hybrid system models describe the required interplay of computation and physical dynamics, which is crucial to guarantee what computations lead to safe physical behavior (e.g., cars should not collide). Control computations that affect physical dynamics must act in advance to avoid possibly unsafe future circumstances. Formal verification then ensures that the controllers correctly identify and provably avoid unsafe future situations under a certain model of physics. But any model of physics necessarily deviates from reality and, moreover, any observation with real sensors and manipulation with real actuators is subject to uncertainty. This makes runtime validation a crucial step to monitor whether the model assumptions hold for the real system implementation. The key question is what property needs to be runtime-monitored and what a satisfied runtime monitor entails about the safety of the system: the observations of a runtime monitor only relate back to the safety of the system if they are themselves accompanied by a proof of correctness! For an unbroken chain of correctness guarantees, we, thus, synthesize runtime monitors in a provably correct way from provably safe hybrid system models. This paper addresses the inevitable challenge of making the synthesized monitoring conditions robust to partial observability of sensor uncertainty and partial controllability due to actuator disturbance. We show that the monitoring conditions result in provable safety guarantees with fallback controllers that react to monitor violation at runtime.