Data-Driven Picking Route Optimization in Smart Warehousing: Evidence from a Public WMS Dataset

Abstract

This study investigates wave-level picking route distance in a real footwear-manufacturing warehouse using a public Warehouse Management System (WMS) dataset. Because reliable picking start and completion timestamps are unavailable, the analysis focuses on horizontal travel distance reconstructed from storage-location coordinates. After matching picking records with valid coordinates, the final routable sample contains 7,323 picking waves and 161,347 picking records. Manhattan distance is used to represent aisle-constrained movement. The empirical design combines regression analysis with route simulation. Three regression models evaluate whether order complexity and spatial structure explain baseline route distance, while the simulation compares the recorded baseline sequence, S-shape routing, and a Nearest Neighbor route improved by 2-opt. Adding location dispersion, aisle span, and average depot distance increases model R-squared from 0.715 to 0.843; operator and month fixed effects raise it only slightly to 0.848. In the simulation, S-shape routing produces an average wave-level increase of 4.26%, whereas NN + 2-opt achieves an average wave-level reduction of 16.28% and outperforms the baseline in 80.66% of routable waves. The results show that spatial structure is essential for explaining picking travel and that familiar rule-based policies should be tested against actual layouts before implementation. A simple, transparent heuristic can provide substantial route savings without requiring exact optimization or major automation investment.