Multi-Agent Collaborative Optimization of UAV Trajectory and Latency-Aware DAG Task Offloading in UAV-Assisted MEC

Abstract

The domain of UAV-assisted Multi-Access Edge Computing (MEC) emerges as a novel frontier, characterized by the seamless integration of edge computing capabilities with UAV to facilitate advanced computational services for Terminal Devices (TDs). This research tackles two critical aspects in UAV-assisted MEC frameworks: the strategic formulation of UAV flight paths and the refinement of execution latency for Directed Acyclic Graph (DAG) tasks. We introduce an innovative solution employing Deep Reinforcement Learning (DRL), coined as the Twin Delayed Deep Deterministic Policy Gradient for UAV Trajectory Planning and Task Offloading (TD3-TT) algorithm. This algorithm harmonizes UAV flight planning, DAG task delegation, and scheduling hierarchies, thereby enabling UAV to efficiently undertake task offloading and processing concurrently along their designated optimal trajectories. Through this approach, the latency within the computational network is significantly diminished. A thorough examination of simulation outcomes reveals that the TD3-TT algorithm exhibits notable convergence and robustness, surpassing conventional benchmarks and markedly reducing the execution latency of DAG tasks.

Keywords

• Deep reinforcement learning
• directed acyclic graph
• unmanned aerial vehicle
• multi-access edge computing