Simulation-Based Debugging of Formal Environment Models*

Abstract

Logic-based formal models of robot environments are often used to aid the generation and verification of robotic plans. They are however often simplified and rather abstract compared to the real world that the robot acts in. This can lead to considerable discrepancies between the behavior of the formal model and that of physics-based simulation engines. These discrepancies are not always apparent to the designer. In this paper we propose a new methodology to make these discrepancies explicit by combining formal verification and simulation. Our approach is able to find relevant discrepancies, while only requiring a small number of simulations.