DRL-Based Joint Task Scheduling and Trajectory Planning Method for UAV-Assisted MEC Scenarios

Abstract

The proliferation of Internet of Things (IoT) devices has resulted in a massive increase in data generation, necessitating robust solutions for real-time data processing and analysis. The integration of Unmanned Aerial Vehicles (UAVs) with MEC systems presents a promising enhancement, providing dynamic, mobile edge computing capabilities that can adapt to changing conditions and demands. However, efficiently managing the task offloading and trajectory planning of UAVs in such scenarios poses significant challenges, particularly in maximizing coverage while minimizing time and energy consumption. In this context, this paper proposes a novel UAV-based MEC model utilizing the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) reinforcement learning algorithm to optimize task scheduling and UAV trajectory in real-time. Our model enables UAVs to function as mobile edge servers, dynamically processing and routing tasks between terminal devices and edge servers based on real-time device demands. By incorporating a sophisticated reward function that prioritizes the minimization of system costs-including energy, time, and throughput maximization-our approach not only enhances the operational efficiency of UAV-assisted MEC systems but also improves the quality and responsiveness of edge computing services. After training the MADDPG model in the simulated environment, the experiments show that our approach significantly improved the performance of UAV-assisted MEC systems in real-time scenarios, converging after 50 learning episodes and achieving 90% task completion rate.

Keywords

• Mobile edge computing
• UAV
• deep deterministic policy gradient
• task offloading
• trajectory planning