基于博弈论的自动驾驶驶离专用车道换道决策模型  

A Game Theory-based Lane Change Decision Model for Automated Vehicle Leaving the Freeway Dedicated Lane for Automated Driving

作　　者：陆春意 何赏璐 高彬彬 曹从咏[1] 范钲文 LU Chunyi;HE Shanglu;GAO Binbin;CAO Congyong;FAN Zhengwen(School of Automation,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学自动化学院,南京210094

出　　处：《交通信息与安全》2024年第4期144-153,174,共11页Journal of Transport Information and Safety

基　　金：国家自然科学基金项目(52102380);中国博士后科学基金特别资助项目(2021T140325);南京市重大科技专项项目(202209008);江苏省科技计划项目(BZ2023023)资助。

摘　　要：在高速公路内侧车道设置自动驾驶专用车道的场景下,自动驾驶车辆驶离专用车道,并换道至匝道的过程中,频繁的换道行为会带来自动驾驶车和人工驾驶车交互冲突,对分流区产生交通扰动,进而带来安全风险以及影响交通效率。面向不同车道属性和自动驾驶混合车流,构建了基于博弈论和模型预测控制的自动驾驶换道决策模型。根据自动驾驶车辆到高速公路主线驶离点的距离和换道途径车道上可接受换道间隙的分布,引入换道危机感以量化自动驾驶车辆换道的难易程度。利用加速度变化率来量化混合车流中人工驾驶车辆风格,构建自动驾驶车和人工驾驶车不同类型车辆的模型预测控制成本函数。通过当前时刻的交通状况预测求解自动驾驶车辆下1个时刻的最优加速度,用斯塔克尔伯格博弈描述自动驾驶车辆在换道过程中与人工驾驶车辆的交互,以人工驾驶车辆的收益最大化为前提,选择最大的自动驾驶车辆收益,求解获得自动驾驶车辆的最优换道时间。搭建了基于Python和SUMO的联合仿真实验平台,设置4种不同车流密度的专用车道和混合车流交通场景,并与SUMO默认换道模型等2类模型进行了对比。结果表明:在所有不同车流密度的场景下,所研究的模型会选择合适的换道策略以保证速度损失的最小化,在速度方面均优于所对比的换道模型,换道过程的平均速度分别增加1.44%和11.81%,有效提升了自动驾驶车驶离专用车道的效率。Under the condition where there is dedicated lane for autonomous vehicles on the freeway,autonomous vehicles need to conduct a consequence lane-changing maneuvers to leave the dedicated lane as well as the freeway mainline.These series of actions would bring the conflicts between autonomous vehicles and human-driving vehi-cles,which would increase the risk of collisions but decrease the traffic efficiency of diversion area.This study de-veloped a lane-changing decision model for autonomous vehicles based on game theory and model predictive con-trol,considering the impact of different types of lane and mixed traffic conditions.A measure,lane changing crisis,is proposed for autonomous vehicles,which is calculated based on the distance from automated vehicle to the off-ramp and the distribution of acceptable lane change gap in adjacent lanes.The driving style of human-driving ve-hicle is valued by the variation of acceleration.The cost functions in the model predictive control method for differ-ent types of vehicles are formulated.The optimal acceleration of autonomous vehicles at the next time interval could be predicted based on the current traffic state.The interaction between autonomous vehicle and human-driv-ing vehicle is described by the Stackelberg game.The lane changing decision is made to maximize the benefit of au-tonomous vehicle when the benefit of human-driving vehicle is the largest.And the optimal acceleration of autono-mous vehicle at the next time interval is also achieved.A simulation platform integrating Python and SUMO is con-structed.Four experiment scenarios with different traffic densities under different lane types and mixed traffic condi-tions are set.And the proposed model has been also compared with the other two lane-changing models,e.g.the de-fault model within SUMO.The results indicate that the proposed model could consistently make optimal lane-chang-ing decision to minimize the loss of speed.And compared with the other lane-changing models,the proposed model could increase the averag

关 键 词：智能交通 混合车流 换道决策模型 斯塔克尔伯格博弈 自动驾驶专用车道 

分 类 号：U491[交通运输工程—交通运输规划与管理]