Deep Reinforcement Learning for Routing and Scheduling Optimization in Fanet-Assisted Iot Networks

In areas with limited wireless coverage, IoT has emerged as a promising solution for supporting user data traffic while ensuring service guarantees. With the rapid advancement of wireless communication technologies and smart devices, the Internet of Things (IoT) is poised to enable more automated and intelligent services and applications in the future. However, the transmission rate and range of IoT devices are constrained by their limited power, which can lead to network congestion when handling a large volume of data packets, consequently increasing end-to-end latency and posing challenges for providing satisfactory services to delay-sensitive applications. Flying ad hoc networks (FANETs) have emanated over the last few years for numerous civil and military applications. Therefore, we proposes leveraging FANETs to assist IoT in data transmission, with the goal of meeting the stringent latency requirements of delaysensitive data. In this work, a FANET-assisted IoT network model is proposed, wherein FANETs function as relay routers to facilitate the transmission of data packets from IoT nodes to access points, thereby alleviating network congestion. To minimize the maximum end-to-end latency, we formulate a joint problem of IoT node scheduling and FANETs routing. Given the non-convex nature and strong coupling of the parameters in this problem, and based on the network topology, we propose a Graph attention networks-based (GAT-based) Multi-Agent Deep Reinforcement Learning (MADRL) solution to optimize the maximum end-to-end delay. Simulation results demonstrate that the proposed GAT-based MADRL algorithm outperforms traditional reinforcement learning algorithms in terms of end-toend latency, cumulative reward and learning efficiency.