An expected SARSA-based attack detection algorithm for an unmanned aerial vehicle control system

Abstract

With the popularization of UAVs in the civil field, flight safety assurance has become a paramount concern requiring immediate technological and regulatory interventions. UAV flights rely on the measurement of the surrounding environment by sensors, and attacks such as tampering and spoofing of sensor readings can provide incorrect measurements to the UAV control program, which can make the UAV deviate from its intended route and even crash. This paper studies a security problem in Unmanned Aerial Vehicle (UAV) control systems. The UAV adopts an extended Kalman filter (EFK) to fuse all sensor data and generate state estimates. The measurement signal from each sensor may be modified by a malicious attacker. Considering the statistical properties exhibited by the innovation, the issue pertaining to attack detection is constructed as a Partially Observable Markov Decision Process, where the expected SARSA approach is utilized for solving the problem. The detection algorithm is novel and does not depend on detector thresholds. Simulations are provided to illustrate the developed results.