PathCovAVTest: Testing for Violations of Autonomous Driving Systems Based on Path Coverage of Traffic Regulation Scenarios

ABSTRACT Autonomous vehicles play a crucial role in alleviating traffic congestion and eliminating traffic accidents. To ensure the safety and reliability of autonomous driving systems, comprehensive testing must be conducted before their deployment on public roads. Currently, testing methods primarily focus on simple scenarios involving safety violations, generating test cases based on traffic accident and traffic regulation violation scenarios. However, under complex traffic environments and driving conditions, the automatic generation of traffic regulation violation test cases to identify regulatory violations by autonomous vehicles remains insufficiently explored. In response, we propose a method for testing violations by autonomous vehicles—PathCovAVTest—which generates violation scenarios through traffic regulation scenario models to evaluate both safety violations and traffic regulation violations of the autonomous driving system. First, we design a fitness function for a genetic algorithm based on a Petri net–based traffic regulation scenario model. This function considers the safety of collisions between autonomous vehicles and other traffic participants, as well as the similarity between vehicle trajectory paths and path state sequences. The evolutionary process then produces test cases that represent traffic regulations, aimed at uncovering violations by autonomous vehicles. Simulation experiments conducted on Baidu Apollo, an industrial‐grade platform, demonstrate that PathCovAVTest can effectively identify 16 types of violations committed by autonomous vehicles. Furthermore, compared with baseline methods, PathCovAVTest detects more traffic regulation violations by autonomous driving systems and improves the efficiency of generating unique violation scenarios.