Research And Application Of Deep Reinforcement Learning In Painting Robot Control

When controlling the robot,the designer usually designs the control system according to the described control object model and the environment model,so as to meet the control requirements.However,some tasks and environments are difficult to model and they will change with time,making it difficult to achieve the control goal.As a branch of artificial intelligence technology,deep reinforcement learning plays a great role in the field of game and machine game.Researchers have applied it to the design of robot control system and achieved good results.Deep Deterministic Policy Gradient is the most commonly used depth reinforcement learning algorithm in robot control field,but it has the same disadvantages as other depth reinforcement learning algorithms,including hard to reproduce,parameter debugging,low training efficiency and har to design the reward function.This paper studies the application and implementation of DDPG algorithm in the field of drawing robot control,and focuses on solving the accuracy,efficiency and implementation of DDPG algorithm for drawing robot control.This paper improves the reward function and experience pool mechanism of the DDPG algorithm,and builds a two-axis manipulator model on the two-dimensional plane to verify the control effect of the improved DDPG algorithm.Firstly,in this paper,the angle error and its derivative,the action value of the last step and the position point error of the manipulator end actuator are considered in the design of the reward function.Secondly,in view of the problem of learning efficiency,a buffer pool is added to the experience pool of the DDPG algorithm,and the filtered high value experience is stored in the buffer pool,and the experience is extracted from the buffer pool to participate in the neural network training.Finally,in order to verify the control effectiveness of the improved algorithm,the algorithm before and after the improvement is compared in the experiment.The experiment shows that the control effect of the improved DDPG algorithm is obviously improved.This paper designs a set of drawing robot control system based on the improved DDPG algorithm,and builds a simulation model in SIMULINK.In addition,according to the drawing robot control requirements,the reward function is redesigned.In this simulation model,the improved DDPG algorithm is compared with several mainstream continuous control algorithms.The results show that the improved DDPG algorithm performs better.In order to verify the performance of the improved DDPG algorithm in the drawing robot control system,the three-dimensional drawing robot simulation model,image feature point recognition and extraction system,location point map system and trajectory planning system are designed in SIMULINK.Finally,the improved DDPG algorithm is used to control the robot painting,to verify the effectiveness of the control system.The implementation of deep reinforcement learning algorithms is a problem that has always plagued researchers.In order to solve the problem of the implementation of the improved DDPG algorithm on the real robot,a system solution based on model migration is designed in this paper.The hardware system is built and the communication model is constructed,and programming of upper computer and manipulator control panel.Use actual collected data and simulation data to train the agent,while using real manual teaching data to fine-tune the model.Finally,the drawing experiment is completed on this system,and it is found that the control strategy based on reinforcement learning algorithm can successfully complete the robot control task,and has good adaptability.