Control of an Inverted Pendulum by Reinforcement Learning Method in PLC Environment

The aim of this study is to implement Q-learning algorithm to move an inverted pendulum from the downright position to upright position in a PLC environment. Instead of using classical control algorithms that need a linear model of the system to be controlled, we used model-free control algorithm, i.e. Q-learning, and relaxed the linearity assumption. We demonstrate that reinforcement learning can be successfully used in industrial machine learning applications to learn complex control policies without having a detailed model of the controlled system. An experimental set up is designed using PLC controlled mechanical parts, and the code is written in PLC. After about three hours of learning stage, the Q learning algorithm successfully moved inverted pendulum from downright position to upright position and keep it in balanced upright position.