Swanlund Endowed Chair
CAS Professor of Electrical & Computer Engineering, Center for Advanced Study
Research Professor, Information Trust Institute
Research Professor, Coordinated Science Laboratory
University of Illinois Urbana-Champaign
Interaction between information/communication and control (with “control” interpreted in a broader context, including strategic decision making in teams and games) has been a dominating research topic for several decades. This interaction is in general a complex one because, viewed as separate decision units, each one could help the other to achieve an overall better performance: more information generally leads to better control performance, and a judicious use of control could improve the information content of transmitted messages. These dual roles are not always aligned, however, making the derivation of optimal solutions to team problems generally much more challenging than obtaining for example saddle-point solutions to similarly structured games. Regardless of these difficulties, which are inherent to stochastic decision problems with nonclassical information, the common element in all these problems has been to find a satisfactory answer to the question of “what to send”, or equivalently “how to shape the information/sensor and control signals” so as to collectively meet a targeted objective. With the emergence of remote control applications, where the plant-control and control-plant communications are conducted over a heterogeneous network, or applications that involve distributed agents over large networks, some nontraditional and nonstandard constraints have been imposed on designs, prompted by limitations on power usage and other available resources. The questions that are now being asked are not only “what to send”, but also “when to send”, given some constraints on the number of transmissions (which could include sensor signals, control signals, or communication between agents) and also given the unreliability of the transmission medium where the information on whether the signals sent have reached their intended destinations or not is at best only partially available. Following a brief overview of the classical paradigm of “what to send” and “how to shape”, this talk will discuss a framework wherein also the question of “when to send” can be given a precise meaning and addressed along with the former. Solutions to these problems involve threshold-type policies with online dynamic scheduling and offline computation. This is a rich paradigm with relevance not only to remote control but also to multi-agent teams and games.