Model-Free Dynamic Control of a 3-DoF Delta Parallel Robot for Pick-and-Place Application based on Deep Reinforcement Learning

This article investigates the performance of the Delta robot using model-free controller. Since it is difficult to obtain an accurate dynamic model of the robot and identification methods are complex and challenging. Therefore, in this article, a new approach is presented which checks the performance of the Delta robot in pick-and-place operation using model-free dynamic controllers. For the purposes of this paper, Deep Reinforcement Learning (DRL) and also the Deep Deterministic Policy Gradient (DDPG) network is used. The DDPG algorithm is implemented for the first time on the Delta robot, and according to the obtained results, it reveals that it can provide a favorable approach for the pick-and-place operation. In this method, based on the science of DRL, a reward function based on the pick-and-place operation is also defined, which gives negative reward (punishment) or a positive reward according to the way the robot performs. It is noteworthy that according to the obtained results, it is possible to generalize other points for the pick-and-place operation. It should be noted that the robot was able to learn the target after a long episode and its graph converged after approximately 18,000 episodes.