Solving the Storage Location Assignment Problem Using Reinforcement Learning

Arne Troch, Erik Mannens, Siegfried Mercelis

Abstract: In this work, we deal with the Storage Location Assignment Problem, often referred to as the SLAP, in an E-commerce Distribution Center (EDC). With E-commerce steadily increasing in popularity over the past decades, it has become a key part of the logistics industry. Due to the direct link with the customer, EDC's are forced into a significantly more complex and dynamic order picking process compared to conventional Bulk Distribution Centers. As a result of these challenges, many traditional approaches such as genetic algorithms and rule-based methods reach only suboptimal solutions. We propose the use of Reinforcement Learning (RL) to solve the SLAP, leading to a solution that adapts to dynamically changing environment parameters during runtime. For this purpose, we define a model that transforms the SLAP into a sequential decision making problem. We validate this novel approach by training a state-of-the-art RL algorithm within this model and comparing its results with a benchmark genetic algorithm approach. We conclude that the RL algorithm achieves promising results, surpassing benchmark performance and nearing optimal performance in a small-scale warehouse environment.