Task Offloading in Hierarchical UAV-Enabled MEC Network: A Trust Region Policy Optimization-Based Approach

With the explosion of multimedia applications and heterogeneous Internet of Things (IoT) devices in service under 5G, the flying ad hoc network (FANET), which acts as an edge server, is overwhelmed by traffic overload. Existing FANET architectures are incomplete for large-scale scenarios, and it is a challenge to optimize unmanned aerial vehicles (UAV) offloading decisions and coordinate within the network. In this article, we propose a novel three-layer computational framework for UAVs, which specifies the functions and roles of UAVs in the system. To analyze model performance quantitatively, the interactions of units and tasks are formulated as Markov decision processes (MDP). Then, we consider delay, energy and resource allocation to minimize the total system cost. To solve the problem, a multi-agent TRPO-based hierarchical computation and offloading (MATRPO-HCO) scheme is proposed to determine the optimal offloading decision for large-scale networks. Simulation results show that the proposed framework is superior to alternative schemes with respect to Quality of Service (QoS) and fairness.