RESEARCH ON MODELING AND OBJECT TRACKING FOR ROBOT ARM BASED ON DEEP REINFORCEMENT LEARNING

Abstract

Robotic arms (RAs) are progressively advancing and extensively utilized across several industrial and research domains. In conjunction with that advancement, researchers have incorporated numerous intelligent algorithms to facilitate the RA’s obstacle avoidance strategy. The paper proposes the Deep Reinforcement Learning (DRL) method to control the RA to maneuver the ball on the workspace to a specified location. Based on Resnet18-Unet model, 2D images will be analyzed to get the coordinate of the object’s center. The input’s data is utilized to train the Proximal Policy Optimization (PPO). Throughout this training procedure, the RA will discern suitable moves to maneuver the ball to the designated location. Based on the action’s reward, RA’s operation aims for maximum total reward results. The simulation results are completely conducted by PyBullet environment.