Deep Reinforcement Learning for Task Offloading and Resource Allocation in UAV Cluster-Assisted Mobile-Edge Computing

Abstract

The combination of mobile-edge computing (MEC) and uncrewed aerial vehicles (UAVs) has important implications for the future development of the Internet of Things (IoT). Additional computing power and extensive network coverage enable users to experience better quality of service even when terrestrial base stations (BSs) scarce or destroyed. In this article, computational offloading and resource allocation for a UAV cluster-assisted MEC system are investigated. The cluster consists of a mobile UAV as the cluster head (ACH) and multiple fixed-position UAVs as cluster members (ACMs), where the ACH offloads the computational tasks generated by BS and assigns them to the ACM for collaborative processing. Since the positions of user equipment (UE) and UAV, as well as the speed and angle of ACH flight, are highly continuous, we construct a Markov decision process (MDP) and propose an offloading algorithm that combines a deep deterministic policy gradient algorithm with a priority experience replay mechanism (PER-DDPG) in order to jointly optimize the user association and UE task offloading rate to minimize the system cost. Simulation results show that compared with the computational unloading algorithms based on actor-critical (AC), deep Q network (DQN), and deep deterministic policy gradient (DDPG), respectively, the proposed PER-DDPG algorithm has good convergence and robustness, and can obtain an optimal unloading strategy with low latency and low power consumption.