基于深度强化学习的个性化跟车控制模型  

Personalized car following control model based on deep reinforcement learning

作　　者：孙腾超 陈焕明[1] SUN Tengchao;CHEN Huanming(School of Mechanical and Electrical Engineering,Qingdao University,Qingdao 266000,Shandong,China)

机构地区：[1]青岛大学机电工程学院,山东青岛266000

出　　处：《农业装备与车辆工程》2024年第5期87-91,共5页Agricultural Equipment & Vehicle Engineering

摘　　要：为解决自动驾驶汽车跟车控制问题,提出并改进了一种基于深度强化学习的个性化跟车控制模型。建立车辆安全时距模型,将其融入车辆运动学模型中;使用深度确定性策略梯度(DDPG)算法训练模型;通过MATLAB和CarSim软件对学习到的控制策略进行联合仿真验证。为使训练结果更加真实可靠,提出将CarSim软件融入到智能体的训练过程;将个性化模块引入模型,使模型可以通过改变参数得到不同的驾驶风格。实验结果表明:在一般车速下,该模型能实现在前车加速或减速情况下控制车辆以一定车速安全行驶,并能够通过改变训练参数实现个性化控制,对自动驾驶车辆跟车过程的研究有一定指导意义。In order to solve the following control problem of autonomous vehicle,a personalized following control model based on deep reinforcement learning was proposed and improved.The vehicle safety time interval model was established and integrated into the vehicle kinematics model.The depth deterministic strategy gradient(DDPG)algorithm was used to train the model.Through MATLAB and CarSim,the learned control strategy was verified by joint simulation.In order to make the training results more real and reliable,CarSim was integrated into the training process of the agent.The personalization module was introduced into the model,so that the model could get different driving styles by changing the parameters.The experimental results showed that the model could control the vehicle to drive safely at a certain speed under the condition of acceleration or deceleration of the vehicle in front,and could realize personalized control by changing the training parameters,which had certain guiding significance for the researched of the following process of the autonomous vehicle.

关 键 词：自动驾驶汽车 个性化跟车模型 深度强化学习 深度确定性策略梯度(DDPG)算法 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] U463.6[自动化与计算机技术—控制科学与工程]