Last-mile logistics is the final echelon of complex supply chain distribution processes and refers to the delivery of goods in urban environments from transportation hubs, e.g. warehouses, fulfillment centres, or local depots, to end consumers. Effective last-mile logistics is vital to the financial sustainability and reliability of companies in the transport sector, e.g. postal, courier-and-delivery, and e-commerce services, and requires solving challenging computational problems to optimise scheduling, routing, and inventory fulfillment decisions that significantly affect fleet, fuel, storage costs and the timeliness of deliveries.
I develop AI-based methods exploiting delivery network topologies and demand patterns, as well as optimality conditions, bounds, and algorithms that enable overcoming computational challenges due to multiple conflicting objectives, hard constraints, and the combinatorial explosion of decision alternatives, to effectively solve last-mile logistics optimisation problems.
State-of-the-art logistics optimisation methods rely on off-the-shelf solvers or bespoke algorithms that are often prone to small input perturbations, e.g. delays due to traffic congestion, parking search, or customer access issues, and offer limited interactivity with practitioners, e.g. logistics operators. This sensitive and opaque nature of optimisation methods limits the ability of practitioners to handle disruptions arising during operation, leading to inflated costs and missed delivery windows under unforeseen demand.
In this context, I pursue two complementary research directions. First, I design and experiment with counterfactual generation approaches that enable effective manual assessments of input parameter modifications to outputs of logistics optimisation methods. Second, I develop graph-based decompositions with configurable components to support practitioners in tailoring logistics optimisation methods to diverse use cases and scenarios.
—
Masters Qualification: Integrated Masters
