Application of Evolutionary Algorithms and Criticality Metrics for the Verification and Validation of Automated Driving Systems at Urban Intersections

This paper proposes an approach to identify relevant parameter combinations within Logical Scenarios for the verification and validation (V&V) of automated driving systems (ADS). One approach to potentially reach the goal of safe ADS is scenario-based testing. This faces the challenge of which scenarios are relevant for a safety argument. Based on previous work, an evolutionary algorithm (EA) is used to identify safety critical scenario parameter combinations. This approach is now adopted with multiple safety metrics in an urban setting and is evaluated in a simulation environment. An automated driving function is virtually tested in an exemplary Logical Scenario including dynamic objects at a recreated intersection in the city of Hamburg. Based on this application, capabilities and limitations of the used search method and criticality metrics are analyzed. Furthermore, the benefits of this approach to identify critical combinations of parameters in Logical Scenarios are discussed.