Gym-DC: A Distribution Centre Reinforcement Learning Environment

Saeid Rezaei, Federico Toffano, Kenneth N. Brown

Abstract: Distribution centres in supply chains receive shipments and forward them to transport providers for the next part of their journey to their final destinations. In some Physical Internet proposals, distribution centres will be autonomous. The decision system should choose a transport provider for each packet. Reinforcement learning is a well-established method for learning policies by acting in an environment and observing states. Coupled with Deep Learning, it has shown significant results in competitive environments like board games. To develop and evaluate Reinforcement Learning solutions for managing a distribution center on the Physical Internet, we need a simulated environment that should be as close as possible to real-world conditions. We present Gym-DC - the first framework for Reinforcement Learning research for distribution centres and Physical Internet hubs, based on the OpenAI Gym.