EE Colloquium: Control in the Presence of Large Uncertainties

Abstract: The concept of control and uncertainty are at the foundation of almost everything we do. Control helps us make things behave in a timely and prescribed manner, while the concept of uncertainty can make achieving our objectives even harder. Recent and popular studies have used the idea that a better description of the dynamics of a system helps one achieve a more reliable control of that system. However, several real-world systems can prove difficult to model and these systems can possess a completely different dynamic equation in a different phase of operation.  Furthermore, the effectiveness of control is only as good as the quality of information (data) that is used when executing a control action – poor data input into a well-designed controller can lead to undesirable performance. In this talk, I will present effective methods of reliable control for systems that their mathematical models are either limited or completely unavailable. I will outline scenarios where such methods can outperform controllers that were developed using a good understanding of the system’s dynamics. Furthermore, I will present methods of filtering (estimating) input data that are more effective than the well-known Gaussian-based Kalman filtering techniques. The methodologies I will discuss can be applied generally to a wide range of systems, and the filtering of data that can be characterized by a wide range of probability distributions.