Computational Techniques in Optimal State-Estimation—A Tutorial Review

The objective of this tutorial presentation is to review the main computational techniques of the state-estimation problem for linearizable dynamic systems where the design is oriented toward a minimum variance (quadratic loss, gaussian) estimation error. It almost goes without saying that the viewpoints taken and the guidelines given in this paper should not be understood as firm recipes but rather as problem-dependent, sometimes subjective and experience-dependent, recommendations. We treat both the continuous and the discrete estimation problem: the first because it is usually closer to the real process and thus allows more direct physical insight; and the latter because it is often preferred for computation and for realization on large scale or special-purpose digital computers.