面向决策规划系统测试的具体场景自动化生成方法

An Automatic Generation Method of Concrete Scenario for Decision-Making and Planning System Test

为解决自动驾驶决策规划系统仿真测试中场景文件生成效率低的问题,基于场景表示的3个抽象层次,提出了一种具体场景自动化生成方法。基于层级模型构建功能场景,并提取功能场景关键字,映射至参数空间以生成逻辑场景,最后通过关键字对参数空间进行采样,自动生成OpenSCENARIO格式的具体场景文件。仿真验证结果表明,自动生成的场景文件能够满足高等级自动驾驶测试要求,且相比于手动编写方式,该方法能减少场景文件的编写工作,提高生成效率。

In order to increase the efficiency of scenario file generation for virtual tests of autonomous vehicles’ decision-making and planning systems, an automatic generation method of concrete scenario is proposed according to three abstraction levels of scenario representation. Firstly, the functional scenario is constructed based on a hierarchical model.Secondly, keywords are extracted from the functional scenario and mapped to the parameter space to generate logic scenario. Finally, based on the keywords, concrete scenarios are automatically sampled from the parameter space and compiled into the OpenSCENARIO file format. The results show that the generated scenario files can satisfy the test requirements of high-level autonomous vehicles. Compared with the manual compilation method, the proposed method can increase the efficiency of scenario generation by reducing the coding work.