Dr. Todd Humphreys from The University of Texas at Austin presented Secure Perception for Autonomous Systems
The next few decades will see pervasive autonomous control systems become critical to the world economy—from autonomous cars and aircraft to smart homes, smart cities, and vast energy, communication, and financial networks controlled at multiple scales. Protecting these systems from malicious attacks is a matter of urgent societal interest. The study of secure autonomous control has made important advances over the past few years, but these constitute not solutions so much as problem framing and an emerging consensus that traditional fault detection and mitigation fails when confronted with a deliberate attacker: outlaws are different from outliers; fraud is different from faults. Moreover, the vast majority of this early literature focuses only on cyber attacks—infiltration of the communications networks over which sensor measurements y and control commands u are conveyed.
My research focuses on a category of attacks that has seen scant treatment in the secure control literature. Like cyber attacks, these attacks are hard to detect and can be executed from a distance, but unlike cyber attacks, they are effective even against control systems whose communications networks are secure, and so can be considered a more menacing long-term threat. Moreover, this category is subject to realistic physical (as opposed to questionable artificial) constraints on how y can be manipulated, which offers hope for substantial theoretical progress toward effective attack detection and survival. These are field attacks: attacks on the physical fields—electromagnetic, acoustic, pressure, etc.—measured by system sensors. I’m especially interested in field attacks against control systems performing collision avoidance and precise navigation and timing functions, as these are of special importance to the rise of autonomous vehicles. My students and I in the UT Radionavigation Laboratory have demonstrated the surprising potency of a field attack targeting the GPS sensor of an autonomous helicopter: in a live demonstration the target helicopter became remotely controllable almost as if caught in a tractor beam. More recently, we launched a field attack against the semi-autonomous navigation system of an $80M superyacht, driving it several kilometers off course without raising alarms.
Dr. Humphreys specializes in the application of optimal estimation techniques to problems in satellite navigation, orbital and attitude dynamics, and signal processing. He directs the Radionavigation Laboratory at UT-Austin, where software-defined GPS receivers are developed as a platform for GPS technology innovation and study of the ionosphere and neutral atmosphere. His recent focus has been on defending against intentional GPS spoofing and jamming. In 2008 he co-founded Coherent Navigation, a startup that hardens GPS by, among other things, exploiting telephony signals from the Iridium satellite constellation. Dr. Humphreys joined the faculty of the Cockrell School of Engineering in Fall 2009.