DRL-Based Trajectory Optimization and Computation-Aware Resource Allocation for UAV-Assisted Edge Computing Networks

Unmanned aerial vehicle (UAV) networks face critical challenges in dynamic environments where conventional approaches treat trajectory optimization and resource allocation as separate problems, failing to capture their intricate interdependencies and leading to suboptimal performance, excessive energy consumption, and processing delays. This paper addresses these limitations through a novel hybrid methodology that uniquely integrates deep reinforcement learning with convex optimization for joint optimization. Our innovation lies in two interdependent algorithms: Deep Reinforcement Learning (DRL)-based relay UAV trajectory optimization algorithm (DRL-RUTOA), which leverages Model-Agnostic Meta-Learning for rapid environmental adaptation, and computation-aware multi-UAV trajectory optimization algorithm (CA-MUTOA), which employs a benefit-cost prioritization mechanism for selective computational offloading. Unlike previous approaches, we formulate a unified multi-objective optimization framework that simultaneously balances network throughput, energy efficiency, and computational task management. Simulation results demonstrate that our integrated approach significantly outperforms conventional methods, achieving a 35% improvement in network throughput, 28% reduction in processing delay, and 42% reduction in energy consumption. Additionally, our framework exhibits superior convergence efficiency, requiring only 15 iterations compared to 32-42 iterations for conventional methods, confirming its practical viability for resource-constrained UAV operations in complex mission environments.