Reinforcement Learning-Based Counter Fixed-Wing Drone System Using GNSS Deception

Abstract

As drone intrusions into important facilities have increased, research on drone countermeasures has been conducted to counter drones. In this study, we developed a reinforcement learning (RL)-based counter fixed-wing drone system that can respond to fixed-wing drones in autonomous flight with soft kills. The system redirects fixed-wing drones to a designated target position using the global navigation satellite system (GNSS) deception based on the drone's position and speed measured by RADAR. In this study, to construct an environment for training an RL agent, simplified drone modeling was performed for two types of fixed wing drones, and the RADAR error measured through flight tests was modeled. Subsequently, the Markov decision process (MDP) was defined to enable redirection without prior information regarding fixed-wing drones. After applying the RL agent trained in the defined MDP and environment to the counter fixed-wing drone system, the simulation and flight test results confirmed that redirection was possible for both types of fixed-wing drones.