A simulation-based deep learning approach to operational robot fleet sizing in robotic mobile fulfillment systems

초록

This study proposes a data-driven decision support framework for estimating robot fleet size in robotic mobile fulfillment systems (RMFS) using simulation and deep learning. Unlike the common view that treats fleet sizing as a one-time strategic decision embedded in layout design or capital planning, we frame it as a recurring operational problem. Warehouses often maintain conservative fleet sizes to ensure stability, yet deploying all robots can be wasteful and may even degrade performance. Fluctuating order volumes and internal conditions, therefore, call for agile tools that support responsive fleet adjustments, with such flexibility being a core strength of RMFS. To address this need, we develop a discrete-event simulation that systematically varies RMFS factors—customer orders, layout configurations, and operating strategies—to generate a training dataset. For each scenario, the optimal fleet size that minimizes makespan is identified, which is often smaller than the maximum available. A multilayer perceptron model is then constructed, incorporating feature engineering and a customized loss function tailored to the problem, and trained to predict the simulator-derived optimal fleet size. Finally, we present an illustrative application in which warehouse managers can use the model to adapt fleet deployment in daily operations, yielding measurable savings in both robot usage and fulfillment time compared with full-fleet deployment. © 2026 Elsevier Ltd

키워드