结合人工势场的Q-learning无人驾驶汽车路径规划算法★刘晓晨1，郑孝遥1，沈晨2（1.安徽师范大学计算机与信息学院，安徽芜湖241002；2.安徽工程大学电气工程学院，安徽芜湖241002）摘要：基于强化学习算法规划路径常用栅格法来描述环境，但存在路径过于贴近障碍物、非最短路径等与实际应用场景不符的情况。针对此问题，提出了一种结合人工势场知识的Q-learning无人驾驶汽车路径规划算法，引入障碍物的斥力场值来优化选择状态时的奖励值，同时增加无人驾驶汽车的斜向运动。仿真实验表明，与现有的算法相比，在消耗时间有所增加的情况下，结合人工势场的Q-learning无人驾驶汽车路径规划算法能够找到一条更符合实际情境的更优路径。关键词：强化学习；无人驾驶汽车；路径规划；人工势场；Q学习算法中国分类号：TP391.99文献标识码：A文章编号：1003-0107（2022）12-0001-05PathPlanningforDriverlessCarusingQ-learningAlgorithmcombinedwithArtificialPotentialFieldLIUXiaochen1，ZHENGXiaoyao1，SHENChen2（1.SchoolofComputerandInformation,AnhuiNormalUniversity，Wuhu241002，China；2.SchoolofElectricalEngineering,AnhuiPolytechnicUniversity，Wuhu241002，China）Abstract：Thegridmethodisoftenusedtodescribetheenvironmentinpathplanningbasedonreinforcementlearningalgorithm.However，therearesomesituationsthatisinconsistentwiththeactualapplicationscene，suchasthepathistooclosetoobstaclesornon-shortest.Tosolvethisproblem，aQ-learningdriverlesscarpathplanningalgorithmcombinedwithartificialpo-tentialenergyfieldknowledgeisproposed，byintroducingtherepulsionfieldvalueofobsta-cles，therewardvalueintheselectionstateisoptimized,andtheobliquemotionofthedriver-lesscarisincreasedatthesametime.Simulationexperimentsshowthatcomparedwiththeex-istingalgorithms，theQ-learningpathplanningalgorithmfordriverlesscarcombinedwithartifi-cialpotentialfieldcanfindabetterpaththatismoreinlinewiththeactualsituationwhenthetimeconsumptionisslightlyincreased.Keywords：reinforcementlearning；driverlesscar；pathplanning；artificialpotentialfield；Q-learningCLCnumber：TP391.99Documentcode：AArticleID：1003-0107(2022)12-0001-05结合人工势场的Q-learning无人驾驶汽车路径规划算法刘辰炜，等...
