Manipulating Drone Dynamic State Estimation to Compromise Navigation

To address fast-increasing security incidents of drones, in this paper, we investigate fundamental solutions to defend against unsophisticated commercial drones. We have examined several common flight control and navigation algorithms and identified weaknesses in the prevalent EKF-based dynamic state estimation schemes. Exploiting the unique circumstances and characteristics of these drone algorithms, we have developed a theoretical framework of False Data Injection (FDI) specific to drone dynamic state estimation, different from existing FDI attacks on static state estimation. We have further defined two FDI attacks that allow compromised magnetometer “measurements” to pass the drone anomaly detection for misleading a drone's dynamic state estimation, such that its navigation and flight control are misguided. We have evaluated two practical scenarios to show the effectiveness of the proposed attacks with simulations on the popular ArduPilot flight control system. Our evaluation shows that the proposed attacks on these common algorithms can seriously affect drone navigation, stability, and power consumption. We believe this work is the first FDI attack against common drone dynamic state estimation schemes, which opens a new area of research. We are currently expanding our investigation on exploiting other related issues in this direction.