2-DOF Robot Optimal Control via Artificial Neural Network Reinforcement Learning

In the study optimal control problem a 2-DOF robot is stated. Criterion, which must be minimized, reflected RMS of drives power and length of the load trajectory in the “x- y” plane. The problem was reduced to the minimization of a complex cost function, which includes an artificial neural network (ANN) component. To train ANN reinforcement learning procedure was carried out, where punishment is associated with the cost function value. All the obtained results were analyzed via plots of the robot links motion and trajectory of the load.

I. Introduction

Modern industrial production, which is known as Industry 4.0, is impossible without robots’ exploitation [1]. They increase the capacity of production, making it more flexible, cheaper, and safer. A huge number of problems connected with robots’ applications may be referred as scientific ones. One of them is an optimization of a robot links motion and trajectory planning. For example, the request,,robot optimal control” made in Scopus database brings more than 14000 scientific papers; in 2022, 2021, and 2020 years there are more than 1100 scientific works indexed there.