DRL-powered resource coordination for UAV-THz computation offloading

Abstract

Industrial Internet of Things (IIoT) devices are typically equipped with limited resources and they heavily rely on external supports for computation offloading. The advancements in Unmanned Aerial Vehicles (UAVs) have allowed the UAVs to be deployed as edge servers in a UAV-enabled edge computing (UEC) setup. UECs can be used in conjunction with a Ground Base Station (GBS) to service IIoT devices that are just outside of the range of a GBS. However, the longevity of service of a UEC can be an issue due to the limited battery capacity on the UAV, reducing its effectiveness as a solution. In this project, the problem of energy minimization relating to computation offloading optimization is considered along with resource and time constraints. A Terahertz (THz) range communication between devices is used to accommodate a greater system capacity to reduce the latency for data transmissions. Due to the highly dynamic environment and the non-convex nature of the problem, it is not pragmatic to derive a mathematically tractable solution or with conventional algorithms. Thus, a Deep Reinforcement Learning (DRL) approach is used to find a near-optimal solution instead. The Double Deep Q-Network (DDQN) algorithm is explored as a potential solution and its effectiveness is shown in the simulation results with a 25.3% reduction in energy consumption.