Evolutionary Behavior Tree Generation for Dynamic Scenario Creation in Testing of Automated Driving Systems.

Automated driving systems face significant chal-lenges in ensuring safety due to the complex driving environment and algorithmic errors. Scenario-based tests offer a flexible, efficient, and cost-effective approach for validating the security and safety of these systems compared to impractical real road testing. Addressing the nned for critical scenario acquisition, this paper presents an approach that utilizes Evolutionary Algorithms (EAs) to derive a set of critical scenarios from a single initial scenario. The proposed method combines scenario exploration and generation and starts with an initial scenario set, allowing for the use of any random scenario. Through the mechanisms of mu-tation, crossover, and selection inherent in EAs, a diverse range of new scenarios is generated until the criticality metrics indicate that certain scenarios meet the predefined criteria, prompting the termination of the EA process. Finally, an evaluation regarding the different possible scenario outcomes is done.