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

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.