From Robust Networks to AI-Ready Parcel Locker Systems: A Multilayer Perspective on Design, Demand and Resilience

The growth of e-commerce is intensifying pressure on last-mile delivery, making robust and equitable parcel locker networks a key urban challenge. Modeling city-scale locker systems as two-mode multilayer networks, and analysing them through Dual and Coverage centrality, reveals hidden vulnerabilities and "bridge"lockers that classical coverage-based siting overlooks, as shown for Milan, Rome, and Naples. In this discussion paper we argue that this network-centric view is a natural backbone for AI-ready parcel locker systems. We distill the main lessons from multilayer analysis, such as bipartite modelling, hub-bridge separation, and robustness curves, and outline how to couple them with modern AI methods: (i) learning-based demand models to replace potential-access weights with behaviourally grounded flows; (ii) graph representation learning and graph neural networks to approximate centrality and failure impact in real time; and (iii) reinforcement learning and bandit formulations for sequential locker siting and capacity decisions. Our contribution is conceptual: we sketch an integrated research agenda where multilayer centrality, AI-driven demand estimation, and learning-based policy search jointly inform adaptive, data-driven, and resilience-aware locker planning.