事故数据驱动的自动驾驶场景用例泛化生成

Accident Data-Driven Autonomous Driving Scenario Use Cases Generalization Generation

为了实现自动驾驶功能的安全测试及加速测试,提高测试场景用例的覆盖度,本文提出一种适用于自动驾驶测试的逻辑场景组合泛化生成方法。以高速公路双车道变道场景为例,基于国家车辆事故深度调查体系(NAIS)中真实的交通事故案例,将事故场景解构为风险元素并按照连续型风险元素和离散型风险元素分别进行统计分析、分层及标识化处理;引入组合测试方法理论,利用PICT工具实现相关风险元素的重构泛化,得到472例高速公路双车道变道逻辑场景用例。结果表明:通过组合测试方法生成的逻辑场景用例可以提升对双因素互相影响组合场景97.13%的覆盖度,实现对逻辑场景用例兼顾区分度和覆盖度的泛化生成。

In order to realize the safety testing and accelerated testing of automatic driving functions and im-prove the coverage of test scenarios, this paper proposes a generalization generation method of log-ical scenario combination suitable for automatic driving testing. Taking the two-lane lane change scene of expressway as an example, based on the real traffic accident cases in the National Automo-bile Accident in-Depth Investigation System (NAIS), the accident scene is deconstructed into risk elements, and statistical analysis, stratification and identification processing are carried out ac-cording to continuous risk elements and discrete risk elements. The combination test method the-ory was introduced, and the relevant risk elements were reconstructed and generalized using PICT tools, and 472 cases of expressway two-lane lane changing logic scenarios were obtained. The re-sults show that the logical scenario cases generated by the combined testing method can improve the coverage of the two-factor interaction combined scenario by 97.13%, and realize the general-ized generation of logical scenario cases with both discrimination and coverage.