面向自动紧急转向场景的自动驾驶测试用例生成方法  

Method of automatic driving testing case generation for AES scene

作　　者：饶聪波 赵津[2] 刘畅 孙念怡 RAO Congbo;ZHAO Jin;LIU Chang;SUN Nianyi(Key Laboratory of Advanced Manufacturing Technology of the Ministry of Education,Guizhou University,Guiyang 550025,China;School of Mechanical Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学现代制造技术教育部重点实验室,贵州贵阳550025 [2]贵州大学机械工程学院,贵州贵阳550025

出　　处：《现代电子技术》2024年第16期130-136,共7页Modern Electronics Technique

基　　金：贵州省高层次创新人才(GCC[2023]016);贵州省智能网联车辆协同感知科技创新人才团队(CXTD[2022]009)。

摘　　要：为了验证智能汽车的安全性,需要生成大量用例用于系统测试。对此,提出一种基于临界安全距离模型的用例生成方法,该方法用于生成测试自动紧急转向(AES)系统的场景用例。首先通过对临界安全距离模型进行分析,识别出影响行车安全的关键参数;然后,从自然驾驶数据集High D中提取这些参数,并采用核密度估计方法构建AES测试场景的描述模型。使用蒙特卡洛(MC)方法对描述模型进行抽样,生成与自然驾驶行为参数特征相符的测试用例。同时,为了解决MC方法生成用例中风险及危险场景匮乏的问题,进一步引入重要性抽样(IS)方法,以提升风险用例和危险用例的生成比例。实验结果表明:所提方法能够高效地生成用于AES系统的测试用例;与MC方法相比,IS方法在风险用例上平均增加207.9%,在危险用例上平均增加272.6%,从而显著提高了测试效率。In order to verify the safety of intelligent vehicles,a large number of cases need to be generated for the system testing.On this basis,a case generation method based on the critical safety distance model is proposed,which is used to generate scene cases for testing the autonomous emergency steering(AES)system.The critical safe distance model is analyzed to identify the key parameters that affect the driving safety.Further,these parameters are extracted from the automatic driving dataset High D,and the kernel density estimation method is used to construct the descriptive model for the AES testing scene.The Monte Carlo(MC)method is used to sample the descriptive model,and generate testing cases that match the feature parameters of automatic driving behaviors.In order to improve the shortages of risk and dangerous scenes in cases generated by the MC method,the importance sampling(IS)method is further introduced to increase the generation proportion of risk and dangerous cases.The experimental results show that the proposed method can efficiently generate testing cases for the AES system.In comparison with the MC method,the IS method has an average increase of 207.9%in risk cases and 272.6%in dangerous cases,significantly improving the testing efficiency.

关 键 词：自动紧急转向 自动驾驶 测试用例 临界安全距离 High D数据集 核密度估计 蒙特卡洛法 重要性抽样 

分 类 号：TN911.23-34[电子电信—通信与信息系统] TP391.9[电子电信—信息与通信工程]