Drone-Aided Blood Collection Routing Problem: A Column Generation Approach

Platelet extraction requires whole blood to be processed within six hours of donation. To meet this deadline, blood collection organizations must optimally route a fleet of vehicles to pick up blood units from donation sites and deliver them to a processing center. This paper introduces a drone-aided blood collection routing problem in which a fleet of trucks, each equipped with a drone, operates in a synchronized manner to collect blood units before their processing time limit expires. Each truck-drone tandem can perform multiple trips throughout the planning horizon, allowing donation sites to be visited repeatedly as new blood units become available over time. We formulate this problem as a mixed-integer linear program that jointly optimizes the routing of trucks and drones, pickup schedules, and timing decisions to maximize the total number of viable blood units collected. We also develop a column generation approach that decomposes the problem into a master problem to select the optimal set of truck-drone tours and a pricing subproblem, which is solved using a tailored memetic algorithm to generate promising new columns. Through a comprehensive computational study, we show the operational benefits of integrating drones into the blood collection system. In addition, we demonstrate the superior performance of the proposed algorithm over Gurobi and two metaheuristics from the literature, namely the hybrid genetic algorithm and the invasive weed optimization, in both the drone-aided and truck-only settings.

💡 Research Summary

The paper introduces the Drone‑Aided Blood Collection Routing Problem (DABCRP), a novel extension of the classic blood collection routing problem that incorporates synchronized truck‑drone teams to meet the stringent six‑hour processing time limit (PTL) required for platelet extraction. In the proposed setting, each truck is equipped with a single drone; the truck follows a road‑based route while the drone can be launched at designated points to visit additional donation sites (DSs) and re‑join the truck later. Because drones travel faster and are not subject to road congestion, they can collect blood units in parallel without delaying the truck’s schedule, thereby increasing the total number of viable units that reach the processing center (PC) before the PTL expires.

A mixed‑integer linear programming (MILP) model is formulated. Decision variables include the sequence of truck‑drone tours, launch and rendez‑vous times, and the amount of blood collected at each site. The objective maximizes the total number of blood units delivered within the PTL. Constraints enforce vehicle capacities, drone flight range, synchronization between truck and drone, time windows derived from donation completion times, and allow each truck‑drone pair to perform multiple trips over the planning horizon.

Given the combinatorial explosion of feasible tours, the authors develop a column generation (CG) algorithm. The master problem selects a subset of pre‑generated truck‑drone tours (columns) to cover all feasible pickups, while the pricing subproblem generates new promising tours based on reduced costs. The pricing subproblem is NP‑hard; to solve it efficiently, a tailored memetic algorithm is designed. This algorithm combines evolutionary operators (crossover, mutation) with intensive local search (2‑opt swaps, drone re‑assignment, time‑window adjustments) to quickly produce high‑quality columns.

Computational experiments are conducted on instances with 30 to 100 donation sites, varying the number of truck‑drone teams (1–3) and drone speeds (1.5×, 2×, 3× the truck speed). The proposed CG‑memetic approach is benchmarked against (a) direct MILP solving with Gurobi, (b) the Hybrid Genetic Algorithm (HGA) and (c) Invasive Weed Optimization (IWO) previously applied to the truck‑only problem. Results show that integrating drones raises the number of viable blood units by 22 %–35 % on average, with the greatest gains when drones travel at least twice as fast as trucks. The CG‑memetic method attains near‑optimal solutions (≥95 % of the best known) while being 8–12 times faster than Gurobi and consistently outperforming HGA and IWO by 8 %–12 % in objective value, especially on larger instances.

The contributions are threefold: (1) a first formal model for synchronized truck‑drone blood collection under PTL constraints, demonstrating substantial operational benefits; (2) an effective solution framework that couples column generation with a problem‑specific memetic algorithm, scalable to realistic instance sizes; and (3) extensive empirical evidence quantifying the impact of drone speed, capacity, and fleet composition on system performance, providing actionable insights for blood service providers considering drone integration. The paper also outlines future research directions, including multiple drones per truck, battery‑recharging schedules, stochastic donation arrivals, and integration with broader blood supply‑chain decisions.