自动驾驶系统交通规则符合性仿真验证方法

随着自动驾驶技术的不断发展,高级别自动驾驶车辆逐步在限定区域开展实际道路测试,确保和提高自动驾驶系统安全驾驶能力是当前研究、测试和工程开发的热点难点。面对自动驾驶车辆将长期与人类驾驶车辆混行,并与其他交通参与者遵守同样交通规则的现实需要,提出一种验证和测试自动驾驶系统交通规则符合性的方法,以期降低多车混行条件下的交通安全风险。针对各类交通法律法规语义自动解析技术瓶颈,提出规范化-逻辑化两阶段交通规则数字化模型,基于改进谓词度量时序逻辑框架(Metric Temporal Logic,MTL),将自然语言交通规则转换为命题、逻辑连接词和时序算子组成的逻辑编码,生成了自动驾驶系统可理解、可执行、可验证的数字化交通规则,并构建了交通规则命题的分级分类体系。提出了一套基于自动驾驶车辆高精度运动轨迹的交通规则符合性验证算法,并搭建仿真试验平台,在高速公路交通场景下开展了试验验证。理论分析与试验表明：精简命题空间、新增时序算子和谓词逻辑词等改进有效提高了原有MTL框架的时间表现能力,解决了时序逻辑性不足等问题,大幅提高了交通规则数字化转换效率,对地方性交通法规和未来交通法规修订提供了良好的兼容性。提出的交通规则符合性验证方法及试验平台可以有效测试自动驾驶系统对现有交通规则的遵守能力,相关成果对提高自动驾驶系统安全性能和未来混行交通安全管控水平具有重要意义。

Compliance Validation of Traffic Rules for Automated Driving System

With the development of automated driving, road tests have gradually been conducted for high-level automated driving vehicles in limited areas. Assuring and improving the safe driving capability of self-driving systems is a popular topic in current research, testing, and development. To reduce the traffic safety risk under mixed traffic conditions, a method of verifying and testing the compliance of automated driving vehicles with the same traffic rules is proposed. Aiming at the technical bottleneck of automatic semantic analysis of various traffic laws and rules, this paper proposes a two-stage digital model of normalization-logic traffic rules based on improved predicate metric temporal logic (MTL). Natural language traffic rules were transformed into logical codes constituting propositions, logical connectives, and time-series operators. In addition, digital traffic rules that can be understood, executed, and verified in automated driving systems were generated. A classification and grading system for traffic rule propositions was constructed. Furthermore, a set of traffic rule compliance verification algorithms based on high-precision trajectories of automated driving vehicles were proposed, a simulation test platform was built, and verification was performed in a highway traffic scenario. Theoretical analysis and test results show that improvements such as simplifying the proposition space, adding time series operators, and predicate logic words effectively improve the time representation ability of the original MTL framework, solve the problem of time series logic, and greatly improve the efficiency of the digital transformation of traffic rules. Additionally, the method is compatible with local traffic laws and future traffic law revisions. The proposed traffic rule compliance verification method and test platform can effectively test the ability of an automated driving system to comply with the existing traffic rules. These results are significant for improving the safety performance of automated driving systems and the level of hybrid traffic safety control in the future.