自动驾驶汽车的多维评价准则体系  

A Multi-dimensional Evaluation Criterion Scheme for Autonomous Vehicles

作　　者：陆美钰 王宇雷 李蒙 吴坚[2] 邱静[3] 陈虹[1] Lu Meiyu;Wang Yulei;Li Meng;Wu Jian;Qiu Jing;Chen Hong(Tongji University,Shanghai 201804;Jilin University,Changchun 130015;University of Electronic Science and Technology of China,Chengdu 611731)

机构地区：[1]同济大学,上海201804 [2]吉林大学,长春130015 [3]电子科技大学,成都611731

出　　处：《中国汽车》2023年第7期7-17,共11页China Auto

基　　金：国家重点研发计划(编号2022YFB2503405);中国汽车产业创新发展联合基金(编号U1964201);上海市自然科学基金(编号23ZR1467700)。

摘　　要：自动驾驶(AD)技术是一种能够增强或替代人类驾驶员的技术,但同时也给汽车测试和评估(T&E)技术带来了极大的挑战。由于AD技术替代了驾驶员的感知、决策和控制,因此自动驾驶汽车(AVs)的T&E程序必须从独立的人-车二元T&E转变为人-车系统的集成T&E。在复杂的天气、交通和电磁环境下进行公路测试,使得AVs的定量T&E非常困难。为了解决这个问题,本文提出了一种多维度评价准则方案,其中包括运行范围、性能表现和交互与风险保障评价准则。通过形式化方法,即线性时间逻辑(LTL),实现了基于场景的评估方法。在中国GB的典型情景中,本文提出的评价准则方案得到了验证,并且通过基于SCANeR studio软件的模拟实现。Autonomous driving(AD)is an enhancement or replacement for human drivers,while AD has brought a huge challenge to automotive testing and evaluation(T&E)technologies.As AD replaces drivers'perception,decision-making and control,the T&E procedure for autonomous vehicles(AVs)changes from independent T&E of human-vehicle binary to integrated T&E of human-vehicle systems.Complex weather,traffic and electromagnet environments under open-road testing make a quantitative T&E of AVs very difficult.To solve this problem,in this paper,we propose a multi-dimensional evaluation criterion scheme for AVs,including evaluation criteria for operational scope,behavioral performance,interaction and risk protection.Then,a scenario-based evaluation approach is realized by the formal approach,i.e.,linear temporal logic(LTL).The proposed evaluation criterion scheme is verified by a typical scenario in China GB and implemented by simulations based on the SCANeR studio software.

关 键 词：自动驾驶汽车 自动驾驶 评价 运行范围 性能表现 交互与风险保障 形式化方法 

分 类 号：TN9[电子电信—信息与通信工程]