AQR-FANET: An Anticipatory Q-Learning-based Routing Protocol for FANETs

Flying Ad Hoc Network (FANET) is the main network paradigm for connecting multi-UAVs in the air. The routing protocol for a FANET must be robust towards increasing node dynamics so that it can achieve good performance even in high-speed environments. Conventional ad hoc routing protocols do not scale well with increasing node mobilities whereas state-of-the-art Q-Learning-based routing protocols lack proper exploration necessary for optimal route selection and discovery. This paper proposes AQR-FANET, a proactive and anticipatory Q-Learning-based routing protocol for FANETs that focuses on improving routing performance in high-speed FANETs. AQR-FANET features an improved route selection method, using a softmax-based routing policy, to automatically balance exploration and exploitation. This mechanism allows the protocol to discover more optimal routes in high-speed environments. The route selection is further enhanced by adjusting the Q-values with cross-layer metrics, including current and estimated future signal-to-noise ratio (SNR) based on mobility predictions. The evaluations demonstrate the superiority of AQR-FANET as it achieves a significant 28% average improvement over state-of-the-art Q-Learning-based routing protocols in terms of PDR, while balancing the maximum end-to-end delay. The combination of better exploration mechanism and SNR predictions allow the protocol to scale better with increasing UAV speeds than other existing routing protocols.