汽车工程 ›› 2023, Vol. 45 ›› Issue (8): 1408-1416.doi: 10.19562/j.chinasae.qcgc.2023.08.011

所属专题: 智能网联汽车技术专题-感知&HMI&测评2023年

• • 上一篇    下一篇

基于场景降维及采样方法的场景库优化方法研究

朱向雷1,2,吴志新2,张宇飞3,赵帅1,2,李克秋1,孙博华3()   

  1. 1.天津大学智能与计算学部,天津  300000
    2.中国汽车技术研究中心有限公司,天津  300000
    3.吉林大学,汽车仿真与控制国家重点实验室,长春  130000
  • 收稿日期:2023-01-04 修回日期:2023-02-24 出版日期:2023-08-25 发布日期:2023-08-17
  • 通讯作者: 孙博华 E-mail:bohuasun@jlu.edu.cn
  • 基金资助:
    吉林省自然科学基金(20220101213JC);国家自然科学基金(青年基金)(52102457);中国博士后科学基金(面上资助)(2021M691207);四川省自然科学基金(2023NSFSC1002)和长春市科技发展计划项目(21QC09)

Research on Scenario Library Optimization Method Based on Scenario Dimension Reduction and Sampling Method

Xianglei Zhu1,2,Zhixin Wu2,Yufei Zhang3,Shuai Zhao1,2,Keqiu Li1,Bohua Sun3()   

  1. 1.College of Intelligence and Computing,Tianjin University,Tianjin  300000
    2.China Automotive Technology & Research Center (CATARC) Co. ,Ltd. ,Tianjin  300000
    3.Jilin University,The State Key Laboratory of Automotive Simulation and Control,Changchun  130000
  • Received:2023-01-04 Revised:2023-02-24 Online:2023-08-25 Published:2023-08-17
  • Contact: Bohua Sun E-mail:bohuasun@jlu.edu.cn

摘要:

本文采用场景降维及采样的方法对场景库进行优化。首先将场景元素分类,并基于层次分析法对其重要性权重值求解,根据权重值的大小进行场景元素离散化,构建场景空间;然后通过场景空间自身的属性计算场景危险度,通过自然驾驶数据库中场景的发生概率计算场景空间中场景的发生概率,构造重要性函数,通过自然驾驶数据库的引导在人为构建的场景空间中筛选出关键场景组成测试场景库,同时为加快搜索效率,采用多起点优化算法与泛洪填充算法进行采样搜索;最后,根据场景风险评估方法验证了关键场景的有效性。结果表明,本文所提出的场景库优化方法能够筛选出用于自动驾驶测试的关键场景,提高自动驾驶测试的效率与现实意义。

关键词: 自动驾驶汽车, 场景库优化, 场景发生概率, 场景危险度, 关键场景筛选

Abstract:

In this paper, scenario dimension reduction and sampling methods are used to optimize the scenario library. Firstly, scenario elements are classified, with their importance weights solved by the Analytic Hierarchy Process, based on which the scenario elements are discretized to construct the scenario space. Then, the risk degree of scenarios are calculated by the attributes of the scenario space itself and the occurrence probability of scenarios in the scenario space are calculated using the Natural Driving Database to construct the importance function. The critical scenarios are screened out from the artificially constructed scenario space through the guidance of the Natural Driving Database to form the testing scenario database. At the same time, to speed up the search process, the multi-starting optimization algorithm and the flood-filling algorithm are used for sampling search. Finally, the effectiveness of the critical scenarios is verified by the scenario risk assessment method. The results show that the scenario library optimization method proposed in this paper can screen out critical scenarios for automatic driving test, and improve the efficiency and practical significance of automatic driving testing.

Key words: automated vehicles, scenario library optimization, scenario occurrence probability, scenario risk degree, critical scenarios screening