Rolling-Horizon Optimization of Mobile parcel Locker deployment for last-Mile delivery

Abstract

Global e-commerce now generates nearly 5,900 parcels every second, yet the final leg of that journey remains the most expensive, inefficient, and environmentally damaging segment of the delivery chain. Home delivery, still the dominant delivery method, faces significant pressure from urban congestion, failed deliveries, and a disproportionately high share of total logistics costs. Out-of-home delivery solutions such as automated parcel lockers (APLs) have emerged as a compelling alternative, consolidating orders at shared pickup points and reducing the dependency on door-to-door deliveries. Among these solutions, mobile automated parcel lockers (MAPLs) further improve operational flexibility by allowing locker units to be repositioned dynamically in response to shifting demand. Despite the growing adoption of parcel locker systems in both practice and research, the literature reveals a persistent gap: while fixed APL systems have received considerable modeling attention, the MAPL literature remains limited, particularly for dynamic settings with uncertain demand. In real operations, retailers have no visibility into future demand and must make daily deployment and assignment decisions under a fundamental trade-off between postponing orders for consolidation gains and serving them quickly to protect service level. To address this, a rolling-horizon optimization framework is developed for MAPL deployment and order assignment under dynamically revealed demand. Three myopic mixed-integer linear programming (MILP) strategies are formulated and evaluated using a computational case study based on Doha, Qatar: a Same-Day Delivery model, a Flexible Daily Deployment model that permits postponement within a maximum allowable lead-time, and a Minimum Weighted Lead-Time model that balances delivery cost against delivery lead-time. The framework is then extended into a Forecast-aided rolling-Horizon Deployment (FHD) model, where the model looks ahead over a defined horizon using probabilistic demand predictions while committing only to current-day decisions before re-optimizing as new information arrives. This allows the operational value of forecast information to be quantified as the percentage gap between FHD performance and the full-information benchmark. The analysis examines how forecast horizon length and forecast accuracy translate into measurable gains in deployment efficiency and reductions in third-party delivery reliance, relative to both myopic and full-information benchmarks. The findings are translated into managerial insights, offering both researchers and logistics providers actionable guidance for last-mile decision-making under demand uncertainty.

Date of Award	2026
Original language	American English
Awarding Institution	HBKU College of Science and Engineering