Continuous authentication of UAV flight command data using behaviometrics

Abstract

The authentication of flight data in Unmanned Aerial Vehicles (UAVs) is highly critical because processing fake commands by the on-board flight controller can cause fatal consequences. Depending on the criticality level of the UAV mission, multi-layer authentication techniques can be useful to assure higher security levels. This paper proposes a technique for continuous authentication of flight data based on the behavior of the UAV operator, who flies the vehicle in a manual mode. In contrast to one-time authentication, this technique allows for an on-the-fly identification of malicious commands aiming at manipulating, hijacking, or crashing the UAV. The operator behavior is defined by the sequence of flight commands sent to the drone using a standard radio control transmitter. This is based on our assumption that every UAV operator has a distinctive pattern when it comes to controlling a UAV using transmitter's levers or joysticks. To verify this assumption, we captured 22,402 commands from five different operators, who flew a small multicopter UAV using a standard flight transmitter. Machine learning was applied to train a random forest classifier. The results show that the UAV operators can be identified with accuracies between 76% and 88% in a 10-tree configuration. These promising results pave the way for a comprehensive study towards implementing a real-time classifier on the UAV embedded system.