Learn Global and Optimize Local: A Data-Driven Methodology for Last-Mile Routing

Abstract

In last-mile routing, the optimization is often framed as a Traveling Salesman Problem to minimize travel time and associated cost. However, solutions stemming from this approach do not match the realized paths as drivers deviate due to navigational considerations and preferences. To prescribe routes that incorporate this tacit knowledge, a data-driven model is proposed that aligns well with the hierarchical structure of delivery data wherein each stop belongs to a zone-a geographical area. First, on the global level, a zone sequence is found as a result of a minimization over a cost matrix which balances historical information and distances between zones. Second, within zones, a sequence of stops is determined, such that integrated with the predetermined zone sequence a full solution is obtained.The methodology allows for several enrichments, such as replacing distances by travel times, different cost structures, and intuition while traversing from one zone to another ensuring a seamless connection.These directions are particularly promising as they propel the approach in the top-tier of submissions to the Last-Mile Routing Research Challenge while maintaining the elegant decomposition that ensures a feasible implementation into practice. From a cognitive perspective, the concurrence between prescribed and realized routes reveals that delivery drivers adopt a hierarchical breakdown of the full problem combining intuition and locally optimal decisions when navigating. Furthermore, experimenting with the trade-off between historical information and travel times exposes that drivers rely more on their intuition at the start, whereas at the end, when returning to the station, only travel time is the concern.