Targeting requirements violations of autonomous driving systems by dynamic evolutionary search (HOP at GECCO'22)

Yixing Luo, Xiao-Yi Zhang, Paolo Arcaini, Zhi Jin, Haiyan Zhao, Fuyuki Ishikawa, Rongxin Wu, Tao Xie

Abstract: Autonomous Driving Systems (ADSs) must satisfy multiple requirements. In some cases, satisfying all of them may not be possible due to environmental conditions. Therefore, ADSs usually make tradeoffs among the requirements, resulting in one or more requirements violations; the correctness of such trade-offs must be evaluated during testing. We propose a new approach, named EMOOD, that can effectively generate test scenarios exposing as many requirements violation combinations as possible. EMOOD first uses a prioritization technique that sorts all combinations to search for according to their criticality. Then, it iteratively applies a many-objective optimization algorithm to find scenarios exposing these combinations. In each iteration, the targeted combination is determined by a technique giving preference to combinations with higher criticality and likelihood to occur. The approach is evaluated on an industrial ADS. This Hot-off-the-Press paper summarizes the paper [1]: Y. Luo, X. Zhang, P. Arcaini, Z. Jin, H. Zhao, F. Ishikawa, R. Wu and T. Xie, "Targeting Requirements Violations of Autonomous Driving Systems by Dynamic Evolutionary Search", 36th International Conference on Automated Software Engineering (ASE 2021).