无信号灯交叉口左转场景的自适应协同决策优化方法

打开文本图片集

关键词：深度强化学习；自动驾驶决策；无信号交叉口；多策略粒子群优化中图分类号：TP399 文献标志码：A 文章编号：1001-3695（2026）04-021-1148-06doi：10.19734/j. issn.1001-3695.2025.08.0272

Adaptive cooperative decision-making method for unsignalized intersection left-turn scenarios

Ning Mi 1 ， Zhang Weiwei 2，3 ， Zhang Xuechao’，Liang Fengjun'，Wang Yujiang1+ （1.SchoolofMechanicalandutomotieEnginering，uangxi UniverstyfSienceandTechologyizhouGuangxi54566ina; 2.ColegofoiedesUesi8;lfcluoiegi University of Engineering Science，Shanghai201620，China）

Abstract：This paper proposedan improved decision-making method named MSPSO-TD3to enhance the precisionand safety ofleft-turndecisionsatcomplexunsignalizedurbanintersections.Thisapproach integratedMSPSOwiththe TD3algorithm. Firstly，it introducedanMSPSO moduletoimproveglobal explorationeficiency，andutilizeda dynamicparameternetwork anda multi-subgroupparalelsearch mechanism toprovide high-qualitycandidateactionsforT3.Secondly，itemployeda dual-Criticnetwork during thedecisionfusionstage tojointlyevaluatecandidateactionsandselettheoptimalone.Italsoapplied MSPSOregularizationto enhance the global optimalityof thepolicy.In CARLA simulation experiments，theproposed method achieved an 80% success rate in high-density traffic scenarios. It reduced the average passage time to 5.37 s and decreased the average acceleration to 2.24m/s2 . The MSPSO-TD3 method demonstrates significant improvements in traffic efficiency，safety，and comfort，showing strong decision-making optimization capabilities in complex scenarios.

Key words：deepreinforcementlearning；automatic decision-making；unsignalized intersection；multi-swarm particleswarm optimization

0 引言

无信号灯交叉路口由于缺乏明确的优先交通规则且交互复杂，成为城市道路系统中交通事故和拥堵的高发区域[1，2]。（剩余15997字）