Adversarial Sensor Attacks Against Uncertain Cyber-Physical Systems: A Dynamic Output Feedback Approach

Abstract

The design of sensor spoofing attacks for cyber-physical systems (CPSs) has received considerable attention in the literature, as it can reveal the underlying vulnerabilities of the CPS. We present a dynamic output feedback approach for designing stealthy sensor spoofing attacks against CPSs. Unlike the existing works, we consider the case where the attacker has limited knowledge of the victim CPS’s dynamical model, characterized by polytopic uncertainty. It is shown that despite the limited knowledge of the attacker, the proposed stealthy sensor spoofing attack method can provably avoid detection by the onboard detection mechanism, even in the presence of model uncertainties, measurement noises, and disturbances. Furthermore, we show that the resulting attack design is recursively feasible, i.e., the designed attack at the current time step ensures persistent detection constraint satisfaction throughout the attack. Finally, we demonstrate the effectiveness of our approach through an illustrative numerical simulation of a sensor spoofing attack on a quadrotor.