Control of Cable Driven Parallel Robots Through Deep Reinforcement Learning

Abstract

Cable-driven parallel robots (CDPRs) pose significant challenges for precise control due to their complex cable dynamics and environmental uncertainties. This paper presents the implementation of Deep Reinforcement Learning (DRL) techniques to control a planar CDPR using the Deep Deterministic Policy Gradient (DDPG) algorithm. Leveraging a cable robot simulator for Reinforcement Learning (RL) allows for safe exploration, faster iterations, and cost-effective training by efficiently handling the robot's complex dynamics. Additionally, we utilized the MuJoCo physics engine to accurately simulate the nonlinear behavior of the cable robot, offering a robust platform for training and validating RL-based control strategies. The performance of this approach is evaluated using prescribed paths, and the results effectively demonstrate strong tracking accuracy and the overall effectiveness of the control strategy.