Deep reinforcement learning for layout planning – An MDP-based approach for the facility layout problem

Abstract

Deep Reinforcement Learning (DRL) has demonstrated operational excellence in several production-related problems. This paper applies DRL to facility layout problems (FLP) using Proximal Policy Optimisation, Advantage Actor-Critic and Deep Q-Networks. We show that the proposed approach produces an improved arrangement of facilities. The contribution of this work is the proof of concept that DRL can optimise layouts with respect to material handling costs using only an image representation of the layout and a reward signal. The approach shows potential to generalise to new layouts without the need to model or train, thus significantly speeding up layout design procedures.