Multimodal critical-scenarios search method for test of autonomous vehicles

作     者：Tianyue Feng Lihao Liu Xingyu Xing Junyi Chen 

作者机构：Tongji UniversityShanghaiChina 

出 版 物：《Journal of Intelligent and Connected Vehicles》 (智能网联汽车（英文）)

年 卷 期：2022年第5卷第3期

页      面：167-176页

核心收录：

学科分类：0809[工学-电子科学与技术（可授工学、理学学位）] 08[工学] 

基　　金：the National Key Research and Development Program of China.(2021YFB2501205) 

主　　题：Autonomous driving system Virtual test Scenario Optimization algorithm 

摘      要：Purpose–The purpose of this paper is to search for the critical-scenarios of autonomous vehicles(AVs)quickly and comprehensively,which is essential for verification and validation(V&V).Design/methodology/approach–The author adopted the index F1 to quantitative critical-scenarios’coverage of the search space and proposed the improved particle swarm optimization(IPSO)to enhance exploration ability for higher *** with the particle swarm optimization(PSO),there were three *** the initial phase,the Latin hypercube sampling method was introduced for a uniform distribution of *** the iteration phase,the neighborhood operator was adapted to explore more modals with the particles divided into *** the convergence phase,the convergence judgment and restart strategy were used to explore the search space by avoiding local *** with the Monte Carlo method(MC)and PSO,experiments on the artificial function and critical-scenarios search were carried out to verify the efficiency and the application effect of the ***–Results show that IPSO can search for multimodal critical-scenarios comprehensively,with a stricter threshold and fewer samples in the experiment on critical-scenario search,the coverage of IPSO is 14%higher than PSO and 40%higher than ***/value–The critical-scenarios’coverage of the search space is firstly quantified by the index F1,and the proposed method has higher search efficiency and coverage for the critical-scenarios search of AVs,which shows application potential for V&V.