Zhuang, Xin-YuXin-YuZhuangWang, I-LinI-LinWangLee, Chia-YenChia-YenLee2025-09-152025-09-152025-0603608352https://www.scopus.com/pages/publications/105002838243https://scholars.lib.ntu.edu.tw/handle/123456789/732061Urban last-mile delivery systems face increasing challenges from rising e-commerce demand, frequent delivery failures, and sustainability concerns. This paper presents a novel integration of decentralized smart lockers into crowdshipping operations, uniquely leveraging their excess capacity as ad hoc transshipment points to improve delivery networks. Specifically, parcels are transferred via smart lockers by one or more crowdshippers, reducing trip detours and expanding geographical coverage. Unlike prior studies, our approach eliminates the need for time-synchronized parcel handovers, significantly enhancing operational flexibility. A mixed-integer programming (MIP) model is developed to optimize driver-parcel assignments and routing for the entire system without imposing a single-transshipment assumption. However, to address scalability challenges in large instances, we introduce a rolling-horizon framework and two tailored column-generation algorithms—complete (CCG) and greedy (GCG)—which assume at most one transshipment per parcel. In experiments with 900 drivers and 300 parcels, the CCG achieves exact solutions in 20 min under this assumption, while the GCG demonstrates a 12.1% cost reduction with a 1–2% optimality gap, requiring significantly less computation time. Although the MIP and rolling-horizon models can only solve smaller instances, they validate the effectiveness of the algorithms. This study provides practical and scalable solutions for overcoming last-mile delivery challenges.enfalseColumn generationCrowdshippingInteger programmingLast mile deliverySmart locker[SDGs]SDG11journal article10.1016/j.cie.2025.1111372-s2.0-105002838243