Algorithm Selection for Allocating Pods Within Robotic Mobile Fulfillment Systems: A Hyper-Heuristic Approach

Abstract

Robotic Mobile Fulfillment Systems (RMFS) are an example of warehouse automation. Nonetheless, the complexity of RMFS is such that tackling the entire problem at once is unfeasible. So, this work focuses on a component known as the Pod Allocation Problem (PAP). We analyze the performance of 20 hyper-heuristic models over 60 instances and compare them against a baseline of six low-level heuristics. Our data revealed three key insights. First, sequence-based hyper-heuristics outperformed low-level heuristics in 22% of the models we tested. Second, we noted that, under certain conditions, even the worst-performing heuristics can lead to successful hyper-heuristics. For example, when simulating for 24 hours, the best hyper-heuristic uses the worst heuristic for 80% of the time, and yields a solution with a throughput time 1.4% better than that of the best heuristic. Finally, the resulting model is affected by simulation time, sequence ordering, and heuristic subset.

Keywords

• Robotic mobile fulfillment system
• hyper-heuristics
• heuristics
• pod allocation
• warehouse
• Kiva system