Navigation for UAV Pair-Supported Relaying in Unknown IoT Systems with Deep Reinforcement Learning

Unmanned aerial vehicles (UAVs) have recently been regarded as a promising technology in Internet of things (IoT). UAVs functioned as intermediate relay nodes are capable of establishing uninterrupted and high-quality communication links between remotely deployed IoT devices and the destination. Multiple UAVs are required to be deployed due to their limited onboard energy. We study a UAV pair-supported relaying in unknown IoT systems, which consists of transmitter and receiver. Our goal is that transmitter gathers the data from each device then transfers the information to receiver, and receiver finally transmits the information to the destination, while meeting the constraint that the amount of information received from each device reaches a certain threshold. This is an optimization problem with highly coupled variables, such as trajectories of transmitter and receiver. On account of no prior knowledge of the environment, a dueling double deep Q network (dueling DDQN) algorithm is proposed to solve the problem. Whether it is in the phase of transmitter’s receiving information or the phase of transmitter’s forwarding information to receiver, the effectiveness and superiority of the proposed algorithm is demonstrated by extensive simulationsin in comparison to some base schemes under different scenarios.