Distributionally Robust and Learning-Based Vehicle Routing for Disaster Blood Logistics

In humanitarian disasters, delivering blood and blood products rapidly and reliably is critical but logistically complex. This paper proposes a hybrid framework that integrates Distributionally Robust Optimization (DRO) for offline anticipative planning with Contextual Bandits for online adaptive vehicle dispatch. The system coordinates a heterogeneous fleet of ambulances, drones, and robots to navigate uncertain demand, infrastructure degradation, and time-sensitive deliveries. We formulate a two-level approach: a robust vehicle routing problem accounting for moment-based uncertainty in demand, and an online dispatch policy that learns context-aware allocation decisions in real time. Extensive simulations evaluate the proposed method against DRO-only, CB-only, and stochastic routing baselines. Results show that the hybrid approach outperforms all others in robustness, service rate, and delay, while maintaining reasonable computational efficiency.