Hierarchical Trajectory and Offloading Optimization for Task Delay Minimization in UAV-Assisted MEC

Abstract

This paper addresses the minimization problem of long-term end-to-end delay within a heterogeneous three-tier unmanned aerial vehicle (UAV)-assisted mobile edge computing network featuring two-hop forwarding. The joint optimization of sensor-node-aggregator association, access/backhaul bandwidth allocation, UAV trajectories, and offloading ratios is formulated as a mixed-integer nonlinear stochastic program characterized by complex cross-timeslot queue coupling. By exploiting the inherent asymmetric coupling across network tiers, a weighted weak-coupling surrogate with a provable layering-gap bound is derived. Based on this surrogate, a hierarchical online controller is designed, comprising a queue-aware collection layer, a backhaul coordination layer, and an offloading layer combining closed-form equalization with residual proximal policy optimization. Simulation results validate the effectiveness of the proposed framework, demonstrating significant delay reductions compared to state-of-the-art baselines.