Verification of Behavior-Based Control Systems in their Physical Environment

Verifying complete behavior-based control systems in their physical environments is currently not possible due to their immense complexities. However, by exploiting the modularity and network structure of behavior-based control systems, the relevant behaviors can be isolated, modeled as hybrid systems, and analyzed by suitable formal verification techniques. In this paper, we present a two-step approach to the formal verification of safety properties of hybrid system models that were obtained this way. The underlying behavior arbitration mechanism allows us to reduce the network complexity by examining the stimulation and inhibition relation between the different behaviors using a high-level model while abstracting from internal computations. By modeling the remaining behavior network in its physical environment as a hybrid system, the safety properties are formally verified using a verification condition generation (VCG) procedure for hybrid programs. The feasibility of the approach is proven by a successful experimental evaluation of the control software of an autonomous mobile robot.