Advisor: Dr. Dimitri Mavris

Milestone: PhD Thesis Final Examination (Defense)

Degree Program: Aerospace Engineering

Title: Methodology for Supporting Military Strategy Development at the Operational Level of Warfare Using Reinforcement Learning

Abstract: Advances in military technology have expanded the capabilities available to commanders, increasing the complexity of operational decision-making and the number of potential courses of action. As technological advantages become increasingly transient, the effective employment of available capabilities is critical to operational success. Traditional approaches to military planning, including human wargaming and modeling and simulation, provide valuable decision support but face limitations in systematically exploring large decision spaces. In particular, simulation-based approaches often rely on predefined decision-making rules that constrain the strategies available for evaluation. Artificial intelligence and reinforcement learning offer an opportunity to develop decision-making policies through repeated interactions with simulated environments, enabling the exploration of strategies without explicitly prescribing their underlying logic. This thesis investigates the application of deep reinforcement learning to operational-level air campaign planning using an agent-based simulation environment. The research focuses on aircraft allocation, target prioritization, and mission sequencing across multi-cycle campaigns. A central challenge is the combinatorial nature of the aircraft assignment problem at scale, which is compounded by variable force compositions, spatially distributed threats, and evolving battlefield conditions. To address this challenge, a Deep Q-Network-based formulation is developed that aggregates battlefield information into spatial zones and decomposes the joint assignment problem into a sequence of individual aircraft allocation decisions. This approach enables the use of fixed-dimensional state and action representations while accommodating variations in the number and distribution of battlefield entities. Experiments investigate the effects of spatial aggregation and alternative neural network architectures on learning effectiveness and operational performance. Additionally, a capability-based decision-making formulation is proposed to investigate the potential for generalization across aircraft types without requiring extensive modifications to the underlying model. The results demonstrate that such an approach is viable and provides a foundation for further research into operational strategy exploration, with additional insight into the strategies that the agent finds.

Date and time: 2026-11-17, 2PM

Location: Weber CoVE

Committee:
Dr. Dimitri Mavris (advisor), School of Aerospace Engineering
Dr. Kyriakos Vamvoudakis, School of Aerospace Engineering
Dr. Jenna Jordan, School of International Affairs
Dr. Olivia Pinon-Fischer, School of Aerospace Engineering
Dr. Michael Salpukas, Raytheon Technologies