Battery Management for Lifelong Operation of Warehouse Robots Under Dynamic Demand

Abstract

Effective battery management is crucial for sustaining continuous operations in warehouses employing fleets of battery-powered robots. This paper addresses the dual challenge of optimizing robot availability while managing energy constraints through a two-pronged optimization framework. We first formulate a Mixed-Integer Linear Programming (MILP) model with goal programming that simultaneously optimizes battery charging schedules and dynamic demand fulfillment, preventing overcharging while ensuring operational efficiency. Second, recognizing MILP's computational limitations for large fleets, we develop a modified rule-based algorithm with adaptive thresholds and time-slot scheduling for scalable deployment. The framework includes analytical derivations of minimum charging station requirements based on battery dynamics - bridging infrastructure planning with operational optimization - and implements a receding horizon strategy that enables real-time adaptation while maintaining collision-free navigation through integration with lifelong Multi-Agent Path Finding (MAPF). Comprehensive simulations under fluctuating demand scenarios demonstrate that both approaches outperform traditional threshold-based methods and recent optimization techniques, offering a practical spectrum from optimal to near-optimal solutions for warehouse automation systems. Note to Practitioners - This paper was motivated by the need for efficient energy management in autonomous warehouse systems, where fleets of battery-powered robots perform logistics tasks. In practice, robot downtime caused by suboptimal charging policies can disrupt operations and increase costs. Existing solutions often rely on fixed schedules or simplistic heuristics that fail to adapt to real-time demand or battery conditions. This work presents a flexible framework that balances operational efficiency and computational scalability. The optimization-based method is well-suited for high-accuracy planning in environments with sufficient computational resources, while the adaptive rule-based approach offers a responsive and easily deployable alternative for large-scale or time-sensitive operations. Practitioners can leverage these strategies to decide how many charging stations to deploy, when individual robots should recharge, and how to sustain continuous task fulfillment. Although the proposed methods are validated in simulation, future work will explore real-time deployment and robustness under fluctuating demand and environmental uncertainties in live warehouse settings. © 2004-2012 IEEE.