Application of Deep Reinforcement Learning in Control of Drawing Robots

摘要

Abstract In this paper, the drawing robot studied is after the system extracts the edge of the image, convert its pixels to actual physical coordinates, then control the robot to track these coordinate values and realize image drawing. In many cases, the edges extracted by the edge detection algorithm have a lot of unnecessary details and edge lines of many branch points. Here, some techniques are used to remove these redundant and branch points. When the robot is tracking the trajectory, the quality of its control algorithm directly determines its performance. In this paper, the control system is designed using the deep deterministic strategy gradient (DDPG) algorithm in reinforcement learning, focusing on the setting of its reward function. Here we propose a reward function. After simulation testing, the reward function setting meets the system requirements, the controller can achieve high-precision and high-speed trajectory tracking effect, and it can also suppress chattering caused by external interference and model errors.