Integrating Deep Reinforcement Learning for Initialization and Adaptive Pheromone Updates in Ant Colony Optimization for UAV Pathing

Unmanned Aerial Vehicles (UAVs) are indispensable assets for missions in dynamic and complex environments, requiring highly efficient path planning that simultaneously optimizes the often-conflicting objectives of minimizing flight distance, energy consumption, and mission time. While Ant Colony Optimization (ACO) is a recognized and effective metaheuristic for this domain, its performance is significantly constrained by a static, empirically-derived pheromone update mechanism, which prevents the algorithm from adaptively learning or optimally managing the search process. To overcome this critical limitation, this study introduces a novel DRL-Assisted ACO framework where a Deep Reinforcement Learning (DRL) agent is seamlessly integrated with the ACO to strategically determine the optimal paths under multi-objective constraints. This intelligent agent is tasked with learning the optimal, mission-specific pheromone update strategy. It achieves this by observing the performance of generated paths and receiving a sophisticated reward signal meticulously derived from the Analytic Hierarchy Process (AHP), which systematically weights the mission objectives. Validated through a simulated case study conducted in Khartoum State, Su-dan, the DRL-Assisted ACO approach has demonstrably achieved superior performance, exhibiting marked gains in convergence speed and generating paths with a significantly higher overall multi-objective utility score, thereby delivering a robust and adaptive solution essential for high-stakes autonomous UAV operations.