False Data Injection of a Robotic Navigation System: Concepts and Simulations

Abstract

Today, robotic vehicles (RV) play an essential role in both modern industry and the military. More and more RVs become autonomously operated and are prone to cyber-attacks of their embedded control systems. These attacks can be detected and prevented with the appropriate tools but can also be almost undetectable if executed correctly. In this work, a type of cyber-attack, namely false data injection is performed to spoof an Extended Kalman Filter (EKF) and a conventional fault detection mechanism is used to attempt to identify the attack. A two-wheeled differentially steered robotic vehicle based on the TurtleBot 3 ‘Burger’ is simulated in MATLAB. The navigation system is attacked by injecting false data into the measurement device - a LiDAR sensor. The false data injection alters the range and bearing measurements and aims to change the measured heading angle of the robot and alter its trajectory. The attack is then tuned to make it undetectable by the implemented counter measures. In this thesis, it is demonstrated that false data injection can be executed in such a way that it is almost impossible for the fault detection mechanism to perceive an attack. The simulations have shown that the key to stealth is to inject the false data in a slow and steady manner to deceive the EKF by steering its output. The deviation introduced is mostly unnoticeable as long as it is in the same order of magnitude of the noise, which is part of the robot’s navigation system. In contrast, sudden and severe changes of measurement data are detected easily by the EKF and the fault detection mechanism.