A survey on scenario-based test and evaluation methods for autonomous vehicles

With the rapid advancement of autonomous driving technology, traditional mileage-based test methods are no longer suitable for the evaluation needs of autonomous vehicles. In contrast, scenario-based test methods can better simulate various driving environments and provide more comprehensive and accurate evaluations. Consequently, scenario-based test and evaluation methods have become a key area of research. Through an in-depth review of existing scenario-based test and evaluation methods for autonomous vehicles, the following work is carried out. First, the sources of test scenario data are introduced, the structure of scenario databases is analyzed, and their application in the test process is discussed. Next, a comprehensive review of data-driven, mechanism-based, and knowledge-based methods is provided, with these methods being categorized according to their specific application domains. Following this, various test methods are summarized, including software-in-the-loop (SIL), model-in-the-loop (MIL), hardware-in-the-loop (HIL), driving simulator testing, vehicle-in-the-loop (VIL), closed site test, and open road test, along with their key technologies. Then, the autonomous vehicle evaluation system is elaborated in detail, covering aspects such as classification systems, evaluation frameworks, and comprehensive evaluation. Finally, areas requiring further research in scenario-based test and evaluation for autonomous vehicles are outlined, with ideas for future studies being provided.