Automated Scenario Generation for Regression Testing of Autonomous Vehicles

Autonomous vehicles are technologically feasible and are becoming a reality. However, before they can be launched, thorough testing is necessary. In this paper, we present a novel approach to automatically generate test scenarios for regression testing of autonomous vehicle systems as a black box in a virtual simulation environment. To achieve this we focus on the problem of generating the motion of other traffic participants without loss of generality. We combine the combinatorial interaction testing approach with a simple trajectory planner as a feasibility checker to generate efficient test sets with variable coverage. The underlying constraint satisfaction problem is solved with a simple backtracking algorithm.