Evaluation and Coordination of UAVs in Humanitarian Logistics
Robert van Steenbergen is a PhD student in the Department Industrial Engineering & Business Information Systems. (Co)Promotors are prof.dr.ir. M.R.K. Mes and dr.ir. W.J.A. van Heeswijk from the Faculty of Behavioural, Management and Social Sciences.
In this thesis, we focus on the challenge of effectively operating cargo Unmanned Aerial Vehicles (UAVs) in humanitarian logistics from an operations research perspective. In combination with other vehicle types, we identify in which situations UAVs are beneficial to deploy, and how to coordinate vehicles effectively in a disaster area. We investigate a wide range of disasters to analyze the potential situations in which cargo UAVs can have a positive impact on humanitarian logistics. These disaster cases include a variety of characteristics that humanitarian operations deal with, such as inaccessibility, scarcity of supplies, security (e.g., risks of attacks on vehicles), and uncertainties in demand and travel times.
In addition to cost efficiency, we include aspects such as response times, human suffering, demand coverage, and equality in the analyses, aligning with the non-financial objectives of humanitarian organizations. We develop algorithmic methods (e.g., metaheuristics, mathematical models, and reinforcement learning approaches) to get the most value out of the mixed vehicle fleets and gain insights into the effective deployment of UAVs. To obtain realistic and valid insights, we build disaster scenarios utilizing an extensive range of real-world historical data. In this way, we gain knowledge for future life-saving UAV operations when a disaster occurs. With this knowledge, better decisions can be made regarding the number of UAVs to deploy, where to send them, and how to do this in a way that is both cost-efficient and effective in terms of aiding people in need. To advance the coordination of UAV logistics in humanitarian operations, the main research objective of this thesis is as follows:
The objective is to develop mathematical models and heuristic algorithms to support effective and efficient deployment of Unmanned Aerial Vehicles (UAVs in humanitarian logistics and to obtain quantitative insights into what ways and to what degree cargo UAVs can contribute to the improvement of humanitarian operations in a variety of disaster scenarios.
The thesis is divided into four parts. The first part introduces the research, outlines the context, the motivation, the scope, and the research design, and provides an overview of related academic works. The second part presents a generic simulation-based modeling framework. The framework consists of various essential aspects to model and evaluate humanitarian logistics operations, with a focus on the deployment of UAVs. The elements
are (1) Geographical Information, (2) Population and Demand, (3) Supply Chain Network, (4) Vehicles and UAVs, (5) Logistics Planning and Control, and (6) Graphical User Interface. The flexibility and applicability of the framework are illustrated by analyzing the humanitarian operations of six distinct disaster cases, which occurred on different continents and at different times. The extensive analysis identifies what role UAVs could have played in improving the historical relief operations, and provides directions for further study. We find that UAVs often supply the 10-20% most remote, heaviest affected, and small-demand locations. This allows the other vehicle types to serve the larger-demand and easier-accessible locations, improving cost-efficiency and response times. Part three builds upon these insights, with three chapters that analyze various aspects that require further attention. First, we address the impact of security risks to which humanitarian organizations are exposed when operating in conflict zones. For this, we develop an Adaptive Large Neighborhood Search approach to generate routes that consider both the costs of transportation as well as the exposure to the risk of attacks. The deployment of UAVs results in a decrease of 7% in costs and risks together and 40% safer operations, mainly because UAVs supply the more remote locations in the unsafe disaster area. Second, we address the consequences of travel time uncertainty due to damaged infrastructure and the challenge of humanitarian organizations in dealing with limited resources and limited time. Multiple reinforcement learning approaches are developed to aid as many people as possible with trucks and UAVs. These methods are compared to both exact and heuristic approaches. Both the deployment of UAVs and using the proposed approaches show improvements in the analyzed scenarios. The experiments demonstrate improvements in location visits and demand coverage between 11% and 56% compared to truck-only fleets and lateness reductions of up to 85%. This leads to improving the operational effectiveness, reliability, and equity of service between locations. Third, a multi-period problem is tackled, in which allocation decisions need to be made over time. Trade-offs in the allocation of limited supplies are addressed by considering deprivation costs. In this problem, we study the impact of scarcity of supplies in combination with uncertain demand, a challenge in the initial response phase of an operation. We develop two approximate dynamic programming approaches to handle the dynamics and combinatorics of this supply allocation problem. With the solutions of these approaches, we underline how UAVs are a flexible addition to road transport. UAVs distribute scarce supplies effectively over multiple districts over time, reducing on average both human suffering by 20% and costs by 16%. Part four of the thesis contains the conclusions and insights. We distill the key insights based on both the six disaster cases in the second part and the in-depth analyses performed in the third part. We show that UAVs generally can reduce costs, improve predictability, increase flexibility, reduce human suffering, reduce risks, improve response times, and increase location coverage, not by taking over the whole operation, but by mainly tackling the most challenging and hard-to-reach destinations and creating a less complicated and more reliable operation for the conventional vehicles to serve the remaining majority of beneficiaries.