考虑交互博弈的无信号交叉路口自动驾驶车辆决策规划研究  

Interactive Decision-making and Planning for Autonomous Driving Vehicles in Unsignalized Intersection

作　　者：徐灿 赵万忠[2] 李琳 张瑞军 王春燕[2] 陈锋 XU Can;ZHAO Wanzhong;LI Lin;ZHANG Ruijun;WANG Chunyan;CHEN Feng(School of Automotive and Transportation Engineering,Hefei University of Technology,Hefei 230009;College of Energy and Power Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016;Zhejiang VIE Science&Technology Company limited.,Zhuji 311835)

机构地区：[1]合肥工业大学汽车与交通工程学院,合肥230009 [2]南京航空航天大学能源与动力学院,南京210016 [3]浙江万安科技股份有限公司,诸暨311835

出　　处：《机械工程学报》2023年第14期202-212,共11页Journal of Mechanical Engineering

基　　金：国家自然科学基金资助项目(52072175,51775007)。

摘　　要：为了解决无信号交叉路口自动驾驶车辆决策保守,与周围车辆交互性差的问题,提出一种考虑交互博弈的无信号交叉路口自动驾驶车辆决策规划算法。该方法分为以下几步,首先基于运动学模型及道路约束对自车和周围车辆进行初步的运动预测,并建立两车的交互动作空间,得到车辆可能的行驶域。其次,建立一个新颖的危险度评估方法,能评估两车在任意位置、任意姿态、任意速度下的碰撞危险度,用于各交互动作状态行为值的求解。进一步基于斯塔克伯格主从博弈求出两车的均衡动作策略,该策略即为当前交通环境下考虑交互得到的最优动作。最后,通过Prescan/Simulink构建交叉路口场景进行联合仿真,来验证该算法的合理性。结果表明所提出的考虑交互博弈的算法在保证安全性的基础上,能相对于基于决策树和无交互的方法分别提高7.3%和12.4%的效率,并能在多车复杂工况下与周围车辆进行灵活交互。In order to solve the driving conservatism and the poor interaction of autonomous vehicles in unsignalized intersection,an interactive decision-making and planning algorithm is proposed.It can be divided into the following steps.Firstly,the motion of surrounding vehicles is tentatively predicted by the kinematic model-based method with road constraints.Based on this,the interactive action space of ego vehicle and the surrounding vehicle is built to cover the feasible driving range.Then,a novel threat assessment function is formulated that can evaluate the colliding risk of vehicles under any position,posture,and velocity.It is used for the solution of the state-action value.Furthermore,the Stackelberg Game is utilized to obtain the equilibrium action of ego vehicle and the interactive vehicle,which is the optimal action with considering the interaction in current traffic.Finally,the proposed algorithm is validated by the constructed intersection scenario in Prescan/Simulink.The results reveal that the proposed interactive algorithm can improve the efficiency by 7.3%and 12.4%compared to the decision-tree and the no-interaction method respectively while ensuring safety.Besides,it can also realize agile interaction in multi-vehicle complex interaction.

关 键 词：自动驾驶 决策规划 交互博弈 危险度评估 无信号路口 

分 类 号：U471[机械工程—车辆工程]