For almost thirty years, control by eigenvalue-eigenvector assignment and quadratic optimization (including LQR, LQG, LQR-LTR) have been the most popular controller design methods for multi-input, multi-output systems. However, these methods can only guarantee that the control system state vector as a whole behaves well. Very little is said about the transient behavior of individual state variables. This proposal introduces the idea of Covariance Control, which is a method for ensuring a "good behavior" of each state variable separately. We believe that covariance control can become more powerful and useful than the more popular state-space control methods.
