Research On Path Planning For Indoor Robots Based On Deep Reinforcement Learning

Aim to improve the autonomy of robots,more and more algorithms have been proposed.As a key algorithm for robot navigation,path planning algorithms are particularly important.Although the current traditional path planning algorithm research has also achieved related results,the traditional algorithm lacks an ability of environment perception and environment learning.In this paper,under the background of artificial intelligence related technology theory,further research is carried out on the advanced algorithms to realize robot autonomous path planning.This paper studies the path planning method based on deep reinforcement learning.Using deep learning's powerful learning ability and reinforcement learning's powerful decision-making ability,the deep reinforcement learning method can well realize intelligent path planning.In order to achieve more intelligent robot indoor environment path planning,this paper has made some improvements on the basis of DDPG(Deep Deterministic Policy Gradient)algorithm.(1)Based on the deep reinforcement learning DDPG algorithm,a critic network algorithm improvement based on multi-step state values is proposed to make the training effect more stable and excellent.(2)In the part of neural network model based on deep reinforcement learning,an improved method is also proposed,and RAdam algorithm is introduced to realize more efficient neural network parameter training.(3)Drawing on A3 C and other related theories,on the basis of the improved DDPG algorithm,an asynchronous training method is proposed.(4)On the basis of deep learning,the transfer learning algorithm with faster convergence is introduced and applied to the improved DDPG algorithm to further improve the performance of the algorithm.Related experiments are carried out under the framework of ROS,including simulation experiments and simple indoor field experiments.The object of training is the Turtlebot robot.Experimental results show that the improved algorithm has faster convergence,efficiency and accuracy compared to the ordinary algorithm.