Flying Ad-Hoc Networks (FANETs) introduce ad-hoc networking into the context of flying nodes, allowing real-time communication between these nodes and ground control stations. Due to the nature of this kind of node, the structure of a FANET is dynamic, changing very often. Since it has applications in military scenarios and other mission-critical systems, an agile and reliable network is essential with robust and efficient routing protocols. Nonetheless, maintaining an acceptable network delay generated by the selection of routes remains a considerable challenge, owing to the nodes’ high mobility. This article addresses this problem by proposing a routing scheme based on an improved Q-Learning algorithm to reduce network delay in scenarios with high-mobility, called Q-FANET. This proposal has its performance evaluated and compared with other state-of-the-art methods using the WSNET simulator. The experiments provide evidence that the Q-FANET presents lower delay, a minor increase in packet delivery ratio, and significant lower jitter compared with other reinforcement learning-based routing protocols.

飞行自组织网络 (FANET) 将自组织网络引入飞行节点的环境中，允许这些节点和地面控制站之间进行实时通信。由于这种节点的性质，FANET 的结构是动态的，经常变化。由于它在军事场景和其他关键任务系统中都有应用，敏捷可靠的网络对于强大而高效的路由协议至关重要。尽管如此，由于节点的高移动性，保持由路由选择产生的可接受的网络延迟仍然是一个相当大的挑战。本文通过提出一种基于改进的 Q-Learning 算法的路由方案来解决这个问题，以减少高移动性场景中的网络延迟，称为 Q-FANET。该提议对其性能进行了评估，并与使用 WSNET 模拟器的其他最先进的方法进行了比较。实验证明，与其他基于强化学习的路由协议相比，Q-FANET 具有更低的延迟、数据包传输率的小幅增加和显着更低的抖动。