Abstract:

As a pivotal approach supporting the safety verification and commercial implementation of intelligent driving systems, autonomous driving simulation testing has achieved remarkable progress in technical methodologies and application scenarios. Conventional real-world road testing faces critical limitations including prohibitive costs, inadequate coverage of corner case scenarios, and efficiency bottlenecks, rendering it insufficient for safety validation of high-level autonomous driving systems (L4 and above). To address these challenges, simulation testing frameworks have evolved into a multi-layered verification system encompassing mathematical modeling, virtual scenarios, hardware-in-the-loop (HIL), mixed reality, and cloud-based simulation clusters. Specifically, mathematical modeling accelerates algorithm development; virtual scenario simulation enhances the robustness of perception systems, and HIL testing ensures controller reliability. Meanwhile, cloud-based simulation clusters achieve exponential expansion in scenario coverage through large-scale parallel computing. Notably, the FLOWSIM platform leverages fuzzy logic to establish a "genetic-level" human driving behavior model based on real-world driving data, ensuring the accuracy of traffic flow environments in simulated test scenarios. Furthermore, FLOWSIM-MR introduces a virtual-real testing paradigm for autonomous driving based on digital twins. Looking ahead, the maturation of autonomous driving technologies and their testing technologies will be propelled by emerging innovations like generative AI and digital twin systems, driving simulation testing toward higher precision and intelligence. Concurrently, the establishment of international standards (e.g., ISO 34502) and collaborative ecosystems involving governments, academia, and industry will be critical to overcoming the "safety-cost-efficiency" trilemma.

Key words: autonomous driving simulation testing, virtual-real simulation, testing scenario database construction, safety validation of autonomous driving, driving behavior modeling