智能网联汽车自动驾驶行为决策方法研究  

Research on Decision-making Method for Autonomous Driving Behavior of Connected and Automated Vehicle

作　　者：徐泽洲 曲大义 洪家乐[2] 宋晓晨 XU Zezhou;QU Dayi;HONG Jiale;SONG Xiaochen(Institute of Urban Transportation, Qingdao 266071;Qingdao University of Technology, Qingdao 266520,China)

机构地区：[1]青岛市城市规划设计研究院,山东青岛266071 [2]青岛理工大学,山东青岛266520

出　　处：《复杂系统与复杂性科学》2021年第3期88-94,共7页Complex Systems and Complexity Science

基　　金：国家自然科学基金(51678320)。

摘　　要：针对在交叉口自动驾驶车辆与其他车辆直行冲突的问题,构建自动驾驶汽车行为决策模型,采用深度强化学习对自动驾驶汽车通过道路交叉口进行训练,让自动驾驶汽车自主决策学习,实现复杂场景的快速控制,并与非支配排序遗传算法对比验证自动驾驶汽车的稳定性。仿真结果表明采用深度确定性策略梯度算法的自动驾驶车辆行为决策方法能够更好地输出速度确保了油门及刹车值的平稳变化,有效解决了自动驾驶汽车的安全和舒适问题。Aiming at the problem of direct conflict between autonomous vehicles and other human-driven vehicles at intersections,an autonomous vehicle behavior decision model is built,and deep reinforcement learning is used to train autonomous vehicles when passing road intersections,allowing autonomous vehicles to make autonomous decisions and achieve fast control of complex scenarios,and the comparison with the non-dominated sorting genetic algorithm-Ⅱverifies the stability of the autonomous vehicle.The simulation results show that the autonomous vehicle beha-vior decision-making method using the depth deterministic strategy gradient algorithm has better output speed to ensure the smooth changes of the throttle and brake values,and effectively solve the safety and comfort problems of autonomous vehicles.

关 键 词：智能网联 自动驾驶 深度强化学习 行为决策 仿真分析 

分 类 号：U463[机械工程—车辆工程] TP18[交通运输工程—载运工具运用工程]