A Resilient Packet Routing Approach Based on Deep Reinforcement Learning

Flying ad hoc networks (FANETs) are widely used in disaster relief scenarios due to the flexibility and mobility of unmanned aerial vehicles (UAVs). However, in extreme conditions, UAV nodes in FANETs may fail or face interference, making packet routing under these highly dynamic catastrophic scenarios rather challenging. Traditional routing protocols struggle to adapt to such highly dynamic network changes. In this paper, we present the Prioritized Experience Replay Dueling Double DQN (PER-D3QN), a resilient packet routing algorithm utilizing multiagent deep reinforcement learning, to discover routing policies better adapted to these situations. This method leverages the double deep Q-learning mechanism, prioritized experience replay technique, and deep dueling network architecture to enhance learning efficiency and establish better policies. Simulation results show that our proposed packet routing algorithm achieves resilient routing policies under both catastrophic and non-catastrophic conditions, significantly reducing packet loss and end-to-end delay while enhancing overall network performance.