Optimal Control with Limited Communication
Division of Engineering and Applied Sciences,
Harvard University, June 1999
Advisor: Roger W. Brockett
To obtain a copy of my thesis, please contact me at: dcv (at) uom DOT gr
 
Control and communication issues are traditionally decoupled in discussions of decision and control problems because this simplifies the analysis and generally works well for classical models. This fundamental assumption deserves re-examination as control applications spread into areas where lack of time on a network shared by sensors, actuators and the controller is as important as lack of computational power. Such areas include the coordinated control of robots, formations of aerial vehicles, micro-actuator arrays and other settings where many systems must share the attention of a decision-maker.  This thesis proposes a model that captures the essential characteristics of control systems with limited communication.  Under our model, controller-plant communication occurs at discrete times and the controller must choose which actuators/sensors to update/read at a particular time. These constraints lead to the need for a theory of sampled-data systems where communication and control are intrinsically coupled.  In this work we formulate such a theory  linking dynamical systems, combinatorics and linear algebra. This theory allows us to revisit classical tracking and stabilization problems, this time with the communication and control aspects interweaved.  In the process, we are lead to a quantitative definition for ``attention'' (in the context of control systems with limited communication) that is both rigorous and intuitive.

The effectiveness of the theory is evaluated using simulations and experiments. We present a limited communication control system consisting of a two-fingered dexterous robot with tactile sensing and vision capabilities.  Application of our theory leads to improved trajectory tracking in manipulation tasks, especially when the controller is challenged to perform near or above its Nyquist rate.