Maximum End-to-End Latency Minimization in UAV-Assisted IoT Networks.

With the advancement in wireless communications and smart device technologies, the Internet of Things (IoT) will make future services and applications more automatic and intelligent. However, the transmission rate and range of IoT devices are restricted by the limited battery power, which makes it challenging to provide satisfying services for delay-sensitive applications. Thanks to the high flexibility of unmanned aerial vehicles (UAVs), we propose a UAV-assisted IoT network model, where the UAV acts as the relay to help IoT nodes transmit their packets to the access point to prevent the network from overloading. With the aim of minimizing the maximum endto-end latency, a joint UAV trajectory design and IoT device scheduling problem is formulated. As the problem cannot be formulated with explicit expressions, it is intractable to solve the problem with traditional model-driven optimization algorithms. Therefore, we adopt deep reinforcement learning to solve this problem, and propose a soft actor-critic (SAC) based algorithm for its high sampling efficiency. Numerical results demonstrate that the maximum network end-to-end latency is substantially reduced with the assistance of the UAV. Moreover, the proposed SAC algorithm is shown to possess higher sampling efficiency than the deep Q-network