汽车工程 ›› 2024, Vol. 46 ›› Issue (9): 1600-1607.doi: 10.19562/j.chinasae.qcgc.2024.09.007

• • 上一篇    

自动驾驶拟人连续交互测试场景生成方法

朱冰1,范天昕1,赵健1,张培兴1(),宋东鉴1,薛越1,赵文博2   

  1. 1.吉林大学,汽车底盘集成与仿生全国重点实验室,长春 130025
    2.国汽(北京)智能网联汽车研究院有限公司,北京 102600
  • 收稿日期:2024-07-01 修回日期:2024-07-24 出版日期:2024-09-25 发布日期:2024-09-19
  • 通讯作者: 张培兴 E-mail:zhangpeixing@jlu.edu.cn
  • 基金资助:
    国家自然科学基金(U22A20247);国家重点研发计划项目(2022YFB2503402);中国博士后科学基金(2023M741354);国家资助博士后研究人员计划(GZC20230945)

Generation Method for Anthropomorphic Continuous Interactive Test Scenarios of Automated Driving

Bing Zhu1,Tianxin Fan1,Jian Zhao1,Peixing Zhang1(),Dongjian Song1,Yue Xue1,Wenbo Zhao2   

  1. 1.Jilin University,National Key Laboratory of Automotive Chassis Integration and Bionics,Changchun 130025
    2.China Intelligent and Connected Vehicles (Beijing) Research Institute Co. ,Ltd. ,Beijing 102600
  • Received:2024-07-01 Revised:2024-07-24 Online:2024-09-25 Published:2024-09-19
  • Contact: Peixing Zhang E-mail:zhangpeixing@jlu.edu.cn

摘要:

基于场景的仿真测试方法是自动驾驶汽车安全性验证的重要手段,然而当前测试场景生成方法多输出独立场景片段,如何模拟真实人类驾驶过程生成具有一定挑战的连续交互测试场景已成为自动驾驶测试评价亟须攻克的难题。本文提出了一种自动驾驶拟人连续交互测试场景生成方法。首先建立自动驾驶拟人连续交互测试场景生成架构,并基于HighD数据集进行车辆运动行为分析;在此基础上,基于轨迹相似性特征分析被测自动驾驶汽车当前行为,并通过状态转移矩阵预测其未来轨迹;基于轨迹交互规则确定测试场景中交通车未来行为类型,通过Transform网络架构生成交通车拟人连续交互轨迹;最后,在仿真环境中对生成测试场景的危险性、拟人性等关键性能进行评估,证明了本文方法的有效性。

关键词: 自动驾驶, 连续交互测试场景, 轨迹预测, 拟人交互轨迹生成

Abstract:

Scenario-based simulation test method is an important means of automated driving vehicle safety verification; however, current test scenarios generation methods are mostly for independent scenarios. How to simulate the human real driving process to generate continuous interactive test scenario with challenges has become a problem that needs to be solved urgently in automated driving test evaluation. In this paper, an automated driving anthropomorphic continuous interactive test scenarios generation method is proposed. Firstly, the architecture for anthropomorphic continuous interactive test scenarios generation is established, and the vehicle motion behavior analysis is conducted based on the HighD dataset. On this basis, the current behavior of tested automated driving vehicle based on the trajectory similarity feature is analyzed, and the prediction of the future trajectory through the state transfer matrix is realized. Then, the type of the future behaviors of the traffic vehicles based on the trajectory interaction rules are determined, and the specific trajectory is generated by Transform network. Finally, the key performance indicators such as danger and anthropomorphism of the generated test scenarios are evaluated in simulation test environment, which proves the effectiveness of the method proposed in this paper.

Key words: automated driving, continuous interactive test scenario, trajectory prediction, anthropomorphic interactive trajectory generation