Enhancing Robotic Control: A TD3-Based Approach for Planar Continuum Robots

Abstract

Reinforcement learning (RL) has proven to be a potent paradigm for tackling intricate control challenges, demonstrating significant success across diverse domains. This paper introduces an innovative approach harnessing the Twin Delayed DDPG (TD3) algorithm for the control of a planar continuum robot. The primary goal is to enhance performance and learning efficiency compared to conventional algorithms, as exemplified by the Deep Deterministic Policy Gradient (DDPG). The experimental investigation centers on a thorough evaluation of average reward, success rate, and convergence speed for both TD3 and DDPG, providing insights into their respective capabilities. Results unequivocally illustrate TD3's superiority over DDPG, showcasing expedited convergence and superior overall rewards, with a noteworthy 9.95% improvement in the success rate. The paper delves into the implications of these findings, emphasizing their significance in the realm of robotic control applications and the advancements ushered in by this research.