The strategic integration of manned aircraft and Unmanned Aircraft Systems (UAS) enables informed and expedient decision-making during initial stages of wildfire response. Initial attack fire surveillance and suppression efforts, while linked, are often optimized independently due to the computational intractability of the combined model. An initial attack operation, best described as a multi-agent partially observable semi-Markov decision process (MA-POSMDP), is here decomposed into separate surveillance and suppression Markov decision processes (MDPs) operating on different, but constant, time scales. A hierarchical planner employing an iterative combination of collision-avoidance inspired primitive surveillance actions and suppression macro-actions, is introduced. The proposed general framework, which supports the larger category of Manned-Unmanned [aircraft] Teaming (MUMT) manned-agent problems, is exemplified by a set of multi-rotor UAS surveying a wildfire while a manned helicopter suppresses the wildfire with a water bucket. Monte Carlo solvers using UTC with domain-specific extensions, to include geometric constraints for reducing the vast suppression action space, are applied. The hierarchical planner outperforms traditional firefighting techniques and myopic baselines when simulated in abstracted and actual wildfire case studies. We also validate the early and informed dispatching of additional suppression assets using regression models to support initial attack operations and ensure containment to established thresholds.
The intelligent organization and utilization of heterogeneous aeromedical evacuation (MEDEVAC) assets across non-contiguous battlefields during evolving high-intensity combat considers critical factors to include flight regulation requirements, historical casualty flows, and evacuation asset characteristics. Associated dynamic resource allocation models quickly become intractable when integrating even a reduced number of MEDEVAC assets in select fixed locations, given the various combinations of aircraft, aircrews, and MEDEVAC request execution strategies. A general approach is presented applying a three-stage hierarchical framework with structural, high, and low-level planners using multi-agent semi-Markov decision processes (MSMDPs) with Monte Carlo search solutions. The structural planner dynamically allocates heterogeneous MEDEVAC assets to centralized decision nodes using generated casualty data and various asset combinations applied to the high-level planner. The optimized infrastructure informs high-level and low-level planners facilitating MEDEVAC requests through centralized decision nodes to aircraft and aircrews. We apply the presented hierarchical framework to a simulated battle on the Hawaiian island chain using HH-60M and V-280 aircraft to evacuate patients. The framework outcomes are compared against traditional MEDEVAC allocation policies. The overall model is further used to validate the integration of Army watercraft as over-water Ambulance Exchange Points (AXPs) to enhance patient outcomes.