Research on the Development of a Scenario and Evaluation Criteria for the Evaluation of Autonomous Vehicle AI Safe Driving Ability based on the Road Traffic Law

Recently, research on the verification of autonomous vehicle performance is being actively conducted worldwide. This study aimed to go beyond simple performance verification and explore the range within which autonomous vehicles can be socially accepted in mixed-traffic situations with human drivers. To this end, the study sought to develop scenarios for evaluating the safe driving capabilities of autonomous vehicles to determine whether they can comply with road traffic laws in the same manner as human drivers in such mixed-traffic environments. The scenarios developed in this study include realistic aspects such as the level of traffic accident risk, compliance with laws, yielding behavior, and maintaining safety in situations where human drivers are present. During the scenario development process, evaluation items based on the Road Traffic Law were identified, and a new scenario layer was created to apply these items by incorporating the Road Traffic Law into the layer. In addition, each scenario item was coded to ensure that the scenarios could be expressed in a simplified and standardized format. In total, 42 scenarios based on road traffic laws were developed to evaluate the driving capabilities of autonomous vehicles. After constructing a virtual environment using Software in the Loop Simulation (SILS), we verified the compliance of autonomous vehicles with the laws for each scenario. Furthermore, quantitative evaluation criteria and tolerance ranges were derived to assess autonomous vehicles based on the abstract Road Traffic Laws. Through this study, we established a foundation for evaluating the safe driving capabilities of autonomous vehicles in basic road regulation situations, which differs from the existing evaluation systems that focus on the physical functions and performance aspects of sensors. This is expected to serve as foundational data for future development of a realistic test and evaluation system for assessing the driving capabilities and behavior of autonomous vehicles.