Some new Questions in Distributed Control

Since control scientists started investigating the influence of specific information patterns in optimal control in the sixties, "distributed control" has almost become synonymous with the problem of designing controllers with externally-imposed observation and communication structures. While the difficulty of this question certainly justifies the extensive attention it has received until now, the current state of technology, coupled with the benefits we expect from network-centric decision systems, have made it necessary to consider other types of theoretical problems, whose models and goals, although also involving group decision-making under partial information, do not readily fit the traditional framework outlined above.
In this talk, I will give motivation and results for some of these new distributed control problems. In particular, I will be interested in questions such as:

— "Can control design itself be distributed between different agents and, if so, what does it mean in terms of privacy, reconfigurability, and incentive compatibility?"
— "What are the consequences of segmentation of information about a plant's model (as opposed to plant observation) and disparity of control authority on the quality of group decision?"
These issues will be tackled through a combination of tools from control theory, distributed algorithms, and communication complexity. At the end of the talk, if time permits, I will also present an online multiplayer game currently under development at the University of Illinois, which we plan to use as a testbed to learn how groups of human subjects take decision in the presence of segmentation of information. We hope that these observations will inform our theoretical models and help start the development of rigorous methods for the "distributed control" of mixed teams of humans and machines.

Some of the work presented here is joint with Dr. Jean-Charles Delvenne (UCL, Belgium), Takashi Tanaka, Jerome Barral, and Robert Wilson (UIUC).