METHOD OF MULTI-CRITERIA ROUTING IN FANETS USING DEEP REINFORCEMENT LEARNING

Abstract

This paper presents the development and analysis of a novel approach to intelligent routing in highly dynamic Flying Ad hoc Networks (FANETs) for Unmanned Aerial Vehicles (UAVs). Due to high nodal mobility and the resulting constant network topology changes, traditional routing protocols, such as AODV, become inefficient as the channel information they collect quickly becomes stale. This necessitates the application of intelligent methods capable of real-time learning and adaptive decision-making. To ensure efficient communication and coordinated swarm control, the study utilizes Multi-Agent Reinforcement Learning (MARL) based on the Q-Learning algorithm. The key scientific novelty lies in modifying the traditional multi-objective reward function, which historically focused solely on latency and energy efficiency. Specifically, the conventional delay metric ( ) has been replaced by the Age of Information ( ) metric. AoI measures data freshness (the time elapsed since packet generation) rather than just delivery time, which is critical for real-time swarm control commands. Furthermore, to enhance system cyber resilience and reliability, a Trust Metric ( ) has been integrated into the reward structure. This enables the algorithm to select not only an energy-efficient path but also a reliable (secure) path, actively preventing the use of potentially compromised relay nodes. The final multi-objective reward function rigorously balances data freshness, energy efficiency, load balancing, and trust, thereby ensuring network stability and security in a dynamic environment. The proposed approaches are suitable for building scalable control systems for robotic agent swarms, autonomous intelligent gateways, and mobile ground stations, aiming to significantly enhance the security and overall stability of FANETs in critical mission scenarios.