Multi-agent Task Assignment in Unmanned Aerial Vehicle Edge Computing based on Deep Learning Approach

Abstract

Unmanned aerial vehicles (UAVs) are used as supportive edge computing for sparsely located user equipment on a large scale. In this work, we propose and address a collaborative edge computing system involving multiple UAVs as agents in deep reinforcement learning (DRL) approach due to the restricted computation and energy capabilities of UAVs. The challenge of task offloading is being tackled to reduce the total delays in execution and energy usage by simultaneously planning the paths, assigning computation tasks, and managing communication resources of UAVs. Additionally, Lyapunov optimization is incorporated for the system stability of mobile edge computing assisted by multiple UAVs. Exploring a multi-agent deep reinforcement learning framework to achieve a combined strategy to manage task allocation, and power management. The sum rate results of the evaluation show that our method for task offloading using multi-UAV and multi-EC achieves superior performance with 25 % increase when compared to other optimization methods and ability to reduce cost and delay.