Data-Driven Key Performance Indicator Modeling for Robotic Mobile Fulfillment Systems

Abstract

This work presents a study on the development and application of data-driven operational efficiency and throughput Key Performance Indicator (KPI) modeling for Robotic Mobile Fulfillment Systems (RMFS). Through rigorous analysis of extensive operational data from an operating RMFS, we demonstrate the efficacy of machine learning approaches in predicting and optimizing the performance of complex warehouse automation systems. The research employs advanced techniques, including gradient boosted bagged tree ensembles and AutoML, to capture complex input interactions and provide parallel predictions across multiple KPIs. Our models achieve a mean R² value of 0.7838 across all templates and KPIs, with particularly strong performance in our top performing metric across templates (mean R² of 0.9660).

The study introduces a novel framework for feature engineering and selection, emphasizing actionable inputs while excluding intermediate variables to enhance model interpretability and practical utility. We validate our approach against novel operating conditions, demonstrating the models’ ability to generalize to unseen scenarios. Interpretability techniques, including SHAP analysis and permutation feature importance, provide valuable insights into system behavior and key performance drivers.

This research establishes a generalizable framework for leveraging data-driven modeling in predicting and optimizing brownfield warehouse automation system behavior. The developed approach offers significant potential for enhancing operational decision-making, system design, and strategic planning in the rapidly evolving field of e-commerce fulfillment.