Functional Uncertainty in Real-Time Safety-Critical Systems

Safety-critical cyber-physical systems increasingly use components that are unable to provide deterministic guarantees of the correctness of their functional outputs; rather, they characterize each outcome of a computation with an associated uncertainty regarding its correctness. The problem of assuring correctness in such systems is considered. A model is proposed in which components are characterized by bounds on the degree of uncertainty under both worst-case and typical circumstances; the objective is to assure safety under all circumstances while optimizing for performance for typical circumstances. A problem of selecting components for execution in order to obtain a result of a certain minimum uncertainty as soon as possible, while guaranteeing to do so within a specified deadline, is considered. An optimal semi-adaptive algorithm for solving this problem is derived. The scalability of this algorithm is investigated via simulation experiments comparing this semi-adaptive scheme with a purely static approach.