基于自然驾驶数据的跟车危险场景构建

Construction of Car Following Critical Scenarios based on Natural Driving Data

为了切实提高自动驾驶测试效率,本文提出了一种跟车危险场景构建方法。首先基于自然驾驶数据,提取最常见的跟车场景,挖掘前车动力学特征并使用高斯混合模型拟合前车加速度分布,提取边缘加速度;然后结合快速搜索随机树搜索边缘加速度,随机控制前车行为,生成跟车危险场景测试用例;最后,基于仿真云平台进行并行仿真测试。结果表明,提出的跟车危险场景构建方法能够提高危险场景出现概率,极大提升了功能算法的仿真测试效率。

To improve the test efficiency of autonomous vehicles significantly,this article proposes a method for constructing critical car-following scenarios.Firstly,based on the natural driving data,the most common car-following scenarios are extracted,the dynamic characteristics of the front car are mined,and the acceleration distribution of the front car is fitted using the Gaussian Mixture Model(GMM)to extract the edge acceleration.Then,searching the edge acceleration by combining with Rapidly-exploding Random Tree(RRT),which conducts stochastic control of the front car behavior and generates the test case of the car-following critical scenarios.Finally,parallel simulation testing is conducted based on the simulation cloud platform.The results indicate that the method proposed in this article for constructing car-following critical scenarios can effectively improve the probability of critical scenarios occurring.It increases the simulation testing efficiency of functional algorithms greatly.