结合LSTM的强化学习动态环境路径规划算法

LSTM Combined with Reinforcement Learning Dynamic Environment Path Planning Algorithm

在路径规划领域已经涌现出了诸多的优秀的经典算法,但这些传统方法往往基于静态环境,对于动态可变环境缺乏处理能力.本文提出一种结合LSTM强化学习动态环境路径规划算法.首先,本文以环境图像作为输入,最大限度了保证了原始的信息来源.而后构建了自动编码器用来对环境图像进行特征降维,降低了整体模型的复杂程度.最后采用深度强化学习算法DDPG进行路径规划,其中Actor部分采用LSTM的网络构建,使Actor在决策时可以参考前序信息,做到有预测的避开动态障碍.最后通过实验证明了本文算法的可行性和高效性.

Many excellent classical algorithms have emerged in the field of path planning,but these traditional methods are often based on static environment and lack processing power for dynamic variable environment.This paper proposes a path planning algorithm for dynamic environment based on LSTM reinforcement learning.First of all,this paper takes the environment image as the input to ensure the original information source to the maximum extent.Then an Autoencoder is built to reduce the dimension of environment image,which reduces the complexity of the whole model.At last,the deep reinforcement learning algorithm DDPG is used for path planning,and the Actor part uses LSTM network,so that the Actor can refer to the prior information and make decisions with the prediction of environment change.Finally,the feasibility and efficiency of the proposed algorithm are proved by experiments.