Reinforcement Learning for Optimal Robotic Arm Control in Ball-Balancing Tasks

Abstract

Reinforcement learning (RL) stands out as a promising method for developing better control strategies in robotic systems, particularly in tasks involving complex dynamics such as ball-balancing in automated manufacturing. In automated manufacturing processes, efficient control of robotic arms is crucial for maintaining precision and productivity. This paper presents the implementation of the Deep Deterministic Policy Gradient (DDPG) algorithm for optimal control of a robotic arm in ball-balancing tasks. The efficacy of the proposed DDPG algorithm is demonstrated through comparison with the existing Soft Actor-Critic (SAC) algorithm. Through simulations, the results obtained from DDPG are analysed and contrasted with those from SAC. The smooth trajectory achieved with the DDPG algorithm showcases its ability to generate precise and consistent control commands, leading to stable and smooth motion of the robotic arm during ball balancing. This research contributes to advancing automation in manufacturing by presenting a robust RL framework for optimizing robotic arm control in complex tasks.