基于改进DDQN的无人车路径规划算法

An Improved DDQN Path Planning Algorithm for Unmanned Vehicle

针对DDQN算法在路径规划方面存在收敛速度慢和路径质量低等问题,基于DDQN算法研究了一种无人车路径规划算法。首先,通过获得多个时刻的奖励值,将这些奖励累加并均值处理从而充分利用奖励值信息;然后,通过优化斥力生成的方向改进人工势场法,并用改进的人工势场法代替随机探索提升收敛速度;最后,通过判断路径与障碍物的关系移除冗余节点,并使用贝塞尔曲线对路径进行平滑处理提升路径质量。仿真结果表明,在20×20的两种环境中,改进的DDQN算法相比原始DDQN算法收敛速度分别提升69.01%和55.88%,路径长度分别缩短21.39%和14.33%,并且路径平滑度更高。将改进的DDQN算法部署在无人车上进行检验,结果表明无人车能够较好完成路径规划任务。

Aiming at the problems of slow convergence speed and low path quality in the DDQN algorithm,a path planning algorithm for unmanned vehicles based on DDQN was studied.First,the reward value information is made full use of by obtaining the reward values at a plurality of times,accumulating the rewards,and averaging the rewards.Then,the artificial potential field method is improved by optimizing the direction of repulsion generation,and the improved artificial potential field method is used to replace the random exploration to improve the convergence rate.Finally,the redundant nodes are removed by judging the relationship between the path and obstacles,and the Bessel curve is used to smooth the path to improve the quality of the path.Simulation results show that the convergence rate of the improved DDQN algorithm is improved by 69.01%and 55.88%,and the path length is shortened by 21.39%and 14.33%,respectively,compared with the original DDQN in the two environments of 20×20,and the path smoothness is higher.The improved DDQN algorithm is deployed on the unmanned vehicle to test,and the results show that the unmanned vehicle can complete the path planning task well.