Control of a Cable-Driven Parallel Robot via Deep Reinforcement Learning

Abstract

Deep reinforcement learning (DRL) introduces deep neural networks to solve reinforcement learning problems. DRL is suitable for sequential decision-making applications and has been used to solve a variety of robot manipulation tasks recently. This paper applies a DRL algorithm named deep deterministic policy gradient (DDPG) to control a cable-driven parallel robot (CDPR) for the first time. Besides the control of a CDPR, this study investigates the ability of DDPG to learn the optimal tension distribution of the CDPR and the robustness of DDPG to certain model uncertainties of the CDPR. Simulation results show that DDPG can not only control the CDPR but also learn the optimal tension distribution of the CDPR.