Formal safety verification of non-deterministic systems based on probabilistic reachability computation

Abstract

In this paper, we develop a formal safety verification method based on analytic probabilistic reachability computation, which can estimate the probability of controlled non-deterministic systems entering unsafe states subject to disturbances. Specifically, we employ stochastic differential equations (SDEs) to describe the dynamics of the system and resort to a regularized indicator function to express the collision probability between the state trajectory of the system and unsafe states. We proceed to formulate this collision probability as the viscosity solution to a second-order variational-inequality and provide a rigorous proof for such a novel interpretation. Moreover, we discuss the ENO-Godunov scheme for solving the deduced variational-inequality, which obviates the need for Monte-Carlo simulations and the optimality condition along a complex boundary. The developed framework offers a structured approach to identify potential risks in safety critical systems and maintains a user-friendly implementation. Lastly, we demonstrate the above application in a safety verification problem related to maritime navigation.