Integration von Simulation und Reinforcement Learning zur Portalrobotersteuerung

Industry 4.0 brings on one side a lot of new opportunities and on the other side numerous challenges, e.g. the autonomous interaction of machines and transportation vehicles. Simulation combined with methods of the Artificial Intelligence (AI), like reinforcement learning (RL), are a cornerstone of the enablement of this required autonomy. This paper deals with the cooperation of simulation and RL in order to enable an autonomous control of gantry robots. It presents an adequate architecture for the interaction of simulation and RL-agents and discusses the specific requirements this interaction has to meet in order to get RLagents trained in the context of gantry robot control. Further, trained RL-agents, and their performance in different settings are presented.