ball tree优化的自动驾驶仿真测试场景生成方法

Ball tree optimized automatic driving simulation test scenario generation method

基于场景的仿真测试方法可以有效加速自动驾驶汽车的测试进程,但是传统的采样方法面对高维度采样空间时无法维持高效性,提出了一种ball tree优化的仿真测试场景采样方法,并基于Carla模拟器构建了仿真测试场景自动化生成框架验证算法的有效性。分别使用随机采样方法、基于KD tree结构的最近邻采样方法与基于ball tree结构的最近邻采样方法进行场景参数采样,并生成不同天气要素下的仿真测试场景进行验证。最后将仿真过程与人工方法进行对比。结果表明,提出方法相对于人工方法具有11.38倍场景制作速度的提升,且相对于KD tree结构的采样方法的场景生成速度提升了27.97%。

The scenario-based simulation test method can effectively accelerate the test process of autonomous vehicles,but the traditional sampling method cannot maintain high efficiency in the face of high-dimensional sampling space.This paper proposed a ball tree optimized simulation test scene sampling method.And based on the Carla simulator,it built a simulation test scenario automatic generation framework to verify the effectiveness of the algorithm.The proposed method used the random sampling method,the nearest neighbor sampling method based on the KD tree structure and the nearest neighbor sampling method based on the ball tree structure to sample scene parameters and generated extreme weather simulation test scenarios for verification.Finally,the results show that the proposed method increases by 11.38 times in scene production speed by comparing the simulation process with the manual method and show that the scene generation speed has increased by 27.97%by comparing that with the KD tree structure sampling method.