DRL Optimizes Battery Charging for Autonomous Warehouse Robots

Managing battery charging for Autonomous Mobile Robots (AMRs) in warehouses is a significant operational challenge that directly impacts order processing efficiency and overall throughput. Traditional fixed-rule charging heuristics often fall short in dynamic environments and struggle with coordinating multiple AMRs, leading to inefficiencies. This research introduces a novel solution using a Deep Reinforcement Learning (DRL) framework based on Proximal Policy Optimization (PPO). The PPO-based model is designed for multi-block warehouses with fixed charging stations. It dynamically learns two critical decisions: which charging station an AMR should use and for how long, explicitly factoring in anticipated queuing times at stations. This intelligent decision-making allows AMRs to optimize their charging schedules in real-time. Extensive experiments comparing the DRL framework against state-of-the-art DRL and traditional heuristic approaches demonstrate its superiority. The PPO framework increased order-completion rates by up to 6% and substantially reduced the total time spent on recharging operations. The model also proved robust across various warehouse configurations and fluctuating order arrival rates, offering valuable operational insights into its policy.