The emergence of remote control applications, where plant-to-control and control-to-plant communications are conducted over a heterogeneous network, has necessitated the development of new conceptual and algorithmic tools to accommodate the nontraditional limitations imposed by the new paradigm. In wireless control applications, for example, both the sensors and the controls are power limited, requiring sparing uses of this resource, which leads in turn to a limited use of sensor and control actions. An emerging issue for the sensors and the controllers in such a paradigm is not only what to send (that is, how to shape the sensor or control signals), but also when to send. Another issue that arises, particularly when the communication medium is shared, is how to distribute the available network resources (such as bandwidth) between the tasks of sensor measurement and control (that is, when to measure and when to control), as well as across different users (agents) on the network. Yet a third issue is how to cope with unreliability of the transmission medium, when the information on whether the signals sent have reached their intended destinations or not is at best only partially available.

This plenary talk will introduce a paradigm for estimation and control that captures the salient features of networked control systems as above. It will identify a number of representative problems that fall in this framework, for which closed-form or computationally feasible solutions can be obtained. One of these is the problem of remote recursive estimation of a stochastic process where an observation post can transmit only infrequent information on the process to a remote site that is charged with the estimation task. Given a constraint on the number of transmissions, what is the optimal joint transmission-estimation policy under some predetermined criterion? How would this policy change if the transmission medium is unreliable? A second problem is one where the plant/actuator and controller are separated, and the controller can send only infrequent signals to the actuator. Again given an upper bound on the number of transmissions, what is the optimal joint transmission-controller policy for a given performance criterion? In both cases, and in the context of specific models, the optimal solutions involve threshold-type policies. The talk will also include a discussion on a third model that captures a tradeoff between measurement and control, which requires the development of dynamic scheduling policies. Some other extensions will also be discussed, and remaining challenging issues will be identified.