Embedding Reinforcement Learning in Simulation

Abstract

Reinforcement learning (RL) usage in the industrial domain is on the rise since the shift towards Industry4.0 systems. The need for faster adaptive systems is encouraging manufacturers to invest more in artificial intelligence (AI) technologies. Our aim is to add better intelligence into simulation tools by embedding RL capabilities. Discrete event simulations have been used for decision support in manufacturing systems for decades. New simulation tools such as AnyLogic have improved significantly in the past few years. In this paper, we built a RL library for AnyLogic simulation models. The RL library is developed for model designers, who may not be experts in the field of RL. We applied our RL library on a real-world use case model for truck dispatching problem. The results show the benefits of using RL in real-world problems to find better dispatching policies. Additionally, the visualization capabilities of AnyLogic enabled us to explain the RL agent's reaction to changes in the system.