Dangerous Scenarios Accelerated Search Method for Automated Driving System Based on Parallel Architecture

With the increasing complexity of autonomous driving system functionalities and the rapid expansion of application scenarios, the testing process faces significantly higher data dimensionality and computational load. Traditional scenario-exhaustive testing methods incur prohibitively high computational costs when dealing with large-scale parameter spaces, making them inefficient for practical engineering applications. Although current mainstream accelerated testing strategies based on hazardous scenario generation have made progress in improving testing speed, most of them still rely on serial processing, resulting in low resource utilization in multi-core or distributed computing environments. To address this issue, this paper proposes an accelerated testing method for autonomous driving systems that supports multithreaded parallelism. A parallel testing architecture is designed, consisting of a strategy control module, a task allocation module, and an execution module. Through task decomposition and resource scheduling mechanisms, the system efficiently searches hazardous regions in the scenario space and performs algorithm-level validation. Comparative experiments under the same logical scenario are conducted between single-threaded and parallel acceleration modes, demonstrating that the proposed method can significantly improve testing efficiency beyond existing mainstream strategies while meeting the demands of large-scale scenario validation.