DQN-Based Sensor Attack Detection for Quadrotor

Quadrotors are vulnerable to sensor attacks that can result in unpredictable losses. Therefore, it is necessary to research the sensor attack of quadrotors. This paper designs a sensor attack detector based on the deep Q-network (DQN) algorithm. First, the system state is estimated by an extended Kalman filter, and then residuals are generated to respond to the system state. Secondly, we model the quadrotor sensor attack detection process as a partially observable Markov decision process (POMDP). Third, we design an efficient sensor attack detector based on the DQN algorithm. The detector can recognize the stealthy attack designed by the attacker that can cheat the traditional chi-square detection. Finally, the effectiveness of the method is verified by simulation analysis.