自动驾驶汽车场景测试研究进展综述  

A Survey on the Research Progress of Scenario-Based Testing for Autonomous Vehicles

作　　者：蒋拯民 党少博 李慧云[1,3] 潘毅[1,4] Jiang Zhengmin;Dang Shaobo;Li Huiyun;Pan Yi(Shenzhen Institute of Advanced Technology,Chinese Academy of Sciences,Shenzhen 518055;University of Chinese Academy of Sciences,Beijing 100049;CAS Key Laboratory of Human-Machine Intelligence-Synergy Systems,Shenzhen 518055;Shenzhen Institute of Advanced Technology,Chinese Academy of Sciences(in organization),Shenzhen 518055)

机构地区：[1]中国科学院深圳先进技术研究院,深圳518055 [2]中国科学院大学,北京100049 [3]中国科学院人机智能协同系统重点实验室,深圳518055 [4]中国科学院深圳理工大学(筹),深圳518055

出　　处：《汽车技术》2022年第8期10-22,共13页Automobile Technology

基　　金：广东省基础与应用基础研究基金项目(2020B515130004);澳门科学技术发展基金人工智能重点研发专项(0015/2019/AKP);深圳市科技计划项目(KQTD20200820113106007)。

摘　　要：论述了当前自动驾驶测试技术的研究现状,就测试内容、测试场景、测试方法等主题展开了深入分析梳理。研究表明,场景挖掘已经成为自动驾驶汽车测试的核心技术,但是高维复杂环境下的稀疏小概率风险事件的抽象和建模研究尚处于起步阶段。此外,现有测试方法的区分不清晰,其中虚拟仿真技术的测试置信度不足。未来,业界需要进一步研究理论建模与数据驱动相结合的场景生成方法,推动建立测试场景的定义与规范,并加强场景库建设,此外还应当深入研究随机动态交通流对虚拟仿真的影响,并进一步协调发挥各类测试方法的互补优势。This paper discusses the current situation of test technologies of autonomous driving, presents an in-depth analysis and combing of the subjects including test content, scenario and method. Research shows that the scenario mining has become the core technology of AV testing, but the abstraction and modelling of sparse and small-probability risk events in complex and high-dimensional environment are still in its infancy. In addition, the existing test method can’t distinguish clearly and the test confidence of virtual simulation technology is still insufficient. In the future, more attention should be paid to scenario generation combining theoretical modelling and data-driven methods to promote definition and standardization of test scenario building, and strengthen construction of scenario database. In addition, academia and industry should make in-depth study of impact of real-time random dynamic traffic flow on simulation results and make full use of complementary advantages of various test methods.

关 键 词：自动驾驶汽车 测试场景 测试内容 测试方法 

分 类 号：U463.6[机械工程—车辆工程]