We consider networked control problems in which the observation and/or control channel of a feedback loop is adversarial, in contrast with traditional models where it is assumed to be stochastically, but non-maliciously, dropping, modifying, and/or delaying packets. Such strategic channels represent a first approach towards modeling the effects of cyber-attacks, (whereby an intruder is overtaking the underlying communication network and actively trying to jeopardize the control task) on a controlled system.
We study different types of attacks, -- some occurring solely at the level of the communication protocol, others at the physical level -- and compute saddle-node solutions for the resulting dynamic zero-sum game between controller and jammer. Although simple, these models offer significant insight into the influence of an attack's point of entry and stealthiness on its outcomes, and emphasize the role of information asymmetries between controller and jammer. Different parts of this talk are the result of joint work with A. Gupta and T. Basar (UIUC), and V. Ugrinovskii (UNSW, Australia).
Professor Langbort works in the field of Aerospace Information Technology. In particular, he tries to understand the dynamics and control of large-scale distributed systems such as multi-vehicle missions, air traffic management systems, and smart materials. Of particular interest are fundamental questions regarding the role of communication constraints between subsystems, system architecture, tolerance to individual failure, and intrinsic limitations of distributed control algorithms due to segmentation of information. He uses tools from dynamical systems theory, control theory, and convex optimization coupled with concepts from economics and computer science such as organization efficiency and online algorithms.
