Control Seminar

Model Segmentation in Networked Control Systems

Cedric LangbortProfessorUniversity of Illinois at Urban-Champaign

One of the main challenges of control design for networked systems, apart from the limitations introduced by finite-bandwidth communication channels, is the absence of a trusted monolithic model of the plant to be controlled. More often than not, model information is itself distributed among the various networked components, which constrains control design in several ways. If individual components construct their own controller, they must do so with no knowledge of the other subsystems' characteristics, with may result in poor closed-loop performance. In contrast, if a single agent is in charge of control design, it must elicit model information from the subsystems while ensuring that their privacy is protected, contending with their possible incentive to mis-report, and/or, compensating for the limited number of messages it can receive over the network. In this talk, we review some recent work blending ideas from distributed control theory with more classical tools from the theory of communication complexity and mechanism design in Economics, and investigate the trade- offs between quality of the constructed controllers and quantity of model information available through the network. Avenues of future work and some industrial applications will also be mentioned.
Cedric Langbort is currently an assistant professor of Aerospace Engineering at the University of Illinois at Urbana- Champaign (UIUC). He is also associated with the Coordinated Science Laboratory and the Information Trust Institute. He was born in Paris, France in 1977 and studied at the Ecole Nationale Superieure de l'aeronautique et de l'espace – Supaero in Toulouse (Aerospace Engineering degree; M.S. control theory) and the Institut Non-Lineaire in Nice (M.S. dynamical systems) before receiving the Ph.D. degree in Theoretical & Applied Mechanics from Cornell University in January 2005. He then spent a year and a half as a postdoctoral scholar in the Center for the Mathematics of Information at Caltech before joining UIUC. His work is in the field of Aerospace Information Technology with applications to large-scale distributed systems such as multi-vehicle missions, air traffic management systems, or smart materials. Of particular interest are fundamental questions regarding the role of communication constraints between subsystems, system's architecture, tolerance to individual failure, and intrinsic limitations of distributed control algorithms due to segmentation of information.

Sponsored by

Eaton, Ford, GM, Toyota, Whirlpool and the MathWorks