| Title | : | Stealthy Attacks on Autonomous Systems |
| Speaker | : | Dr. Sibin Mohan (Associate Professor, The George Washington University) |
| Details | : | Fri, 5 Jul, 2024 10:00 AM @ SSB 233 |
| Abstract: | : | Autonomous Systems (AVs) rely on the sensing and modeling of the real-world to carry out their missions. The perception models in AVs use the observations from these sensors to reason about the vehicle’s state and correct for deviations, even attacks. In practice, even the most meticulously designed control systems always operate under a certain amount of noise because of the unavoidable observational/measurement errors involved in both sensor measurement and the modeling of complex vehicular dynamics. All of this inherently creates a space that can be exploited by an adversary without risk of detection. In this talk, I will present new methods that can exploit this space, use a software-only attack. Our system, Requiem, presents a blackbox attack — i.e., there is no knowledge required about the internal details of the system — the only requirement is that the state estimation function be “learnable†from observation of the inputs and outputs. The final result of a Requiem-based attack causes a deviation in the physical system’s trajectory while the system itself believes that it is following the original mission parameters — an attack that is hard to detect or defend against. |
