Runtime verification for stochastic systems

Abstract

We desire a capability for the safety monitoring of complex, mixed hardware/software systems, such as a semi-autonomous car. The field of runtime verification has developed many tools for monitoring the safety of software systems in real time. However, these tools do not allow for uncertainty in the system's state or failure, both of which are essential for the problems we care about. In this thesis I propose a capability for monitoring the safety criteria of mixed hardware/software systems that is robust to uncertainty and hardware failure. I start by framing the problem as runtime verification of stochastic, faulty, hidden-state systems. I solve this problem by performing belief state estimation over a novel set of models that combine Büchi automata, for modeling safety requirements, with probabilistic hierarchical constraint automata, for modeling mixed hardware/software systems. This method is innovative in its melding of safety monitoring techniques from the runtime verification community with probabilistic mode estimation techniques from the field of model-based diagnosis. I have verified my approach by testing it on automotive safety requirements for a model of an actuator component. My approach shows promise as a real-time safety monitoring tool for such systems.