Using Contextual Reinforcement Learning to Design FANET Defence Protocols to Combat Grey Hole Attacks

Flying ad-hoc networks (FANETs) are collections of Unmanned Aerial Vehicles (UAVs) or nodes which communicate information using multi-hop routing protocols. This lack of dependence on fixed infrastructure has advantages in search and rescue operations where natural disasters have destroyed fixed infrastructure. The typical protocols used in these ad-hoc networks are not secured against grey hole attacks, where malicious nodes seek to undermine the operation of the network by dropping packets. This paper details the next phase of my research which investigates how reinforcement learning models with context can improve routing by recognising and reacting to different types of malicious behaviour through progressive interactions. In particular, the context predicts the malicious type of the node, and reinforcement learning provides the optimal response which is tailored to that specific type. By adopting this approach for RL-based network protocols, I hope to show that their generality can be enhanced, enabling them to respond optimally to threats beyond their initial training scope.