自动驾驶自动紧急制动关键场景测试案例提取研究

为提取自动驾驶自动紧急制动(autonomous emergency braking system,AEB)关键场景测试案例,依据不同抽象程度依次重构符合自然驾驶规律的功能场景、逻辑场景、具体场景,进而对AEB系统展开测试并求解其关键场景。从自然驾驶数据集NGSIM筛选出车辆跟随实例构建出功能场景,基于场景要素构建功能场景,提取出该场景的关键字要素,采用高斯混合模型拟合自然驾驶数据,获得具有概率密度分布的逻辑场景,基于Gibbs抽样的蒙特卡洛方法生成具体测试实例,并通过重要性抽样方法生成关键场景。最后对生成到的关键场景聚类加速场景生成,并对其AEB系统在Prescan仿真软件进行测试。结果表明,AEB系统可对生成的关键场景实现避免碰撞。

Study of test cases extraction for autonomous driving AEB critical scenarios

In order to extract the test cases of the critical scenarios of the autonomous emergency braking system (AEB), the functional scenes, logical scenes and specific scenarios that conform to the natural driving law are reconstructed according to different levels of abstraction, and then the AEB system is tested and its critical scenarios are solved. The functional scene was constructed by screening out the vehicle following instance from the natural driving dataset NGSIM, extracting the keyword elements of the scene based on the scene elements to build the functional scene, using the Gaussian mixture model to fit the natural driving data, obtaining a logical scene with probability density distribution, generating specific test examples based on the Monte Carlo method of Gibbs sampling, and generating critical scenes through the importance sampling method. Finally, the critical scene clustering is generated to accelerate the scene generation, and the AEB system is tested in the Prescan simulation software. The results show that the AEB system can avoid collisions for the generated critical scenes.