Simulation Testing of Autonomous Driving Systems Based on Safety-Critical Scenario Generation
Autonomous Driving Systems (ADS) have witnessed rapid development in recent years. To ensure the safety and reliability of these complex systems integrating software and hardware, comprehensive testing is imperative prior to their widespread deployment. Although on-road testing most closely resembles real-world operational environments, its high costs and difficulties in covering rare extreme scenarios pose significant limitations. Consequently, simulation testing based on world models has garnered increasing attention, particularly through the design of challenging safety-critical scenarios to evaluate ADS performance. This presentation will present our latest research advancements in this field, including critical scenario modeling methods incorporating road topology features, and innovative techniques leveraging large language models (LLMs) to generate more realistic and diverse safety-critical scenarios.