Deep Reinforcement Learning for Task Allocation in UAV-enabled Mobile Edge Computing

Abstract

Mobile edge computing (MEC) is a new technology for reducing the network delay in dealing with computing tasks. Considering unmanned aerial vehicle (UAV) with advantage of strong mobility and reliability, a UAV-assisted MEC system is always adopted and researched to provide computing service in wide-range areas. However, since the problems of task allocation are highly nonconvex, it is difficult to reach the optimal solution. In this paper, deep reinforcement learning is used for solving the problem of task allocation in the UAV-assisted MEC system. With the modeling of the UAV-assisted MEC system, the problems of task allocation are formulated. The Markov Decision Process (MDP) is developed for the nonconvexity during solving the task allocation problem. Since the MDP has continuous action space, a dual-delay depth deterministic strategy gradient algorithm is suggested for obtaining the joint optimal strategy of task allocation. Experiment results show better performance of the proposed method compared with other optimization approaches.