Executive Summary
This thesis develops and evaluates a reinforcement learning (RL)-enabled decision-support framework integrated into the Logistics Battle Command (LBC) simulation to enhance military sustainment in contested environments. The research demonstrates that RL can improve logistics performance while maintaining doctrinal alignment and reproducibility, addressing limitations of traditional deterministic planning tools. A Deep Q-Network (DQN) agent was embedded to make adaptive provider-selection decisions for critical commodities.
Why It Matters
This research is crucial for defense analysts as it explores how AI-driven decision support can enhance military logistics resilience and adaptability in contested operational environments, directly impacting force sustainment and operational effectiveness.
Key Takeaways
- Reinforcement learning (RL) can significantly improve adaptive decision-making in contested military logistics by integrating with existing simulation platforms like LBC.
Source evidence · PDF page 28
This thesis addresses this gap by developing and evaluating a reinforcement-learning–enabled decision-support framework integrated into the LBC simulation environment. The research investigates whether reinforcement learning (RL) can improve the performance of contested logistics decisions while preserving doctrinal alignment, governance constraints, and experimental reproducibility.
- There is an inherent tension between strict adherence to military doctrine and the adaptive optimization achieved by RL agents, requiring careful balancing for operational deployment.
Source evidence · PDF page 198
The alignment profile reflects a structural tension between adherence to doctrinal rules and adaptive optimization under survivability-oriented incentives. The Heuristic policy maximizes doctrinal compliance by design; its decisions follow predefined priority structures independent of evolving network dynamics. That rule-based decision structure produces high alignment scores but limits responsiveness to distributional stress.
- Hybrid decision-making systems, combining RL with guardrail mechanisms, can ensure mission critical overrides while allowing RL to retain primary authority in most situations.
Source evidence · PDF page 221
Under deterministic GUI execution, using seed 42, the hybrid mechanism triggered 12 overrides across 3,119 total decisions, corresponding to an intervention rate of 0.38%. All overrides occurred under consumer-critical conditions, and RL retained authority in 99.62% of decisions. This execution path constitutes authoritative behavioral evidence for Chapter V.
Strategic Relevance
This thesis directly addresses the strategic imperative for resilient military logistics in an era of multi-domain operations and contested environments. By integrating AI into sustainment decision-making, it enhances the ability of forces to operate effectively despite adversarial interference, contributing to overall operational endurance and strategic flexibility.