Dynamic Trajectory Control and User Association for Unmanned-Aerial-Vehicle-Assisted Mobile Edge Computing: A Deep Reinforcement Learning Approach

Mobile edge computing (MEC) has become an effective framework for latency-sensitive and computation-intensive applications by deploying computing resources at network edge. The unmanned aerial vehicle (UAV)-assisted MEC leverages UAV mobility and communication advantages to enable services in dynamic environments, where frequent adjustments to flight trajectories and user association are required due to dynamic factors such as time-varying task requirements, user mobility, and communication environment variation. This paper addresses the joint optimization problem of UAV flight trajectory control and user association in dynamic environments, which explicitly incorporates the constraints governed by UAV flight dynamics. The joint problem is formulated as a non-convex optimization formulation that involves continuous–discrete hybrid decision variables. To overcome the inherent complexity of this problem, a novel proximal policy optimization-based dynamic control (PPO-DC) algorithm is developed. This algorithm aims to reduce the weighted combination of delay and energy consumption by dynamically controlling the UAV trajectory and user association. The numerical results validate that the PPO-DC algorithm successfully enables real-time UAV trajectory control under flight dynamics constraints, ensuring feasible and efficient flight trajectory. Compared to the state-of-the-art hybrid-action deep reinforcement learning (DRL) algorithms or metaheuristics, the PPO-DC achieves notable improvements in system performance by simultaneously lowering system delay and energy consumption.