自主车辆行为决策的安全验证方法  

Security validation approach for autonomous vehicle behavior decision

作　　者：刘秉政[1] 曹凯[1] 马建军[2] 

机构地区：[1]山东理工大学交通与车辆工程学院,山东淄博255091 [2]中国人民解放军驻上海航天局,上海201109

出　　处：《山东理工大学学报（自然科学版）》2011年第6期6-12,共7页Journal of Shandong University of Technology:Natural Science Edition

基　　金：国家自然科学基金资助项目(61074140);山东省自然科学基金资助项目(ZR2010FM007)

摘　　要：在各种不同类型交通事件的触发下,自主车辆的行为过程在连续和离散模式间动态切换,表现出复杂的混合系统的动态行为.采用形式验证技术研究自主车辆行为决策的安全验证方法.依据自主车辆环境感应器测得的原始数据,对其他车辆的初始状态进行估计,并根据自主车辆和其他车辆的不确定初始状态将其抽象为马尔科夫链.然后,利用离线产生的马尔科夫链进行车辆随机可达集的在线计算.最后,根据其他车辆的随机可达集以及自主车辆在规划路径引导下的随机可达集,计算碰撞概率.仿真例证表明该方法对于冲突概率的计算具有良好的时效性.Under triggered by different types of traffic events,the behavior of autonomous vehicle switched constantly between continuous and discrete modes and showed the complex dynamic behavior of hybrid systems.In this paper,a security validation approach for decision-making behavior of autonomous vehicle was proposed based on the formal verification.According to raw data measured by environment sensor of autonomous vehicle,the initial state of other vehicles was estimated,and the initial states of all vehicles were Abstracted to Markov chains.Then,the offline generated Markov chains were applied to the online computation of the stochastic reachable sets.Ultimately,the crash probability was approximated based on the stochastic reachable sets of other vehicle and the autonomous vehicle along planning trajectory.Simulation examples show that the proposed method has a good timeliness for the calculation of conflict probability.

关 键 词：自主车辆 混合系统 随机可达集 马尔科夫链 在线安全验证 

分 类 号：U298[交通运输工程—交通运输规划与管理]