Return to S. A. Bortoff's publication list.
Abstract: In this presentation, we discuss a new technique for nonlinear observer design. Using spline functions, we show how to construct an observer that has regularly-perturbed linear error dynamics. The observer exists and can be readily computed for a large class of nonlinear systems, including those which do not satisfy the strict differential-geometric conditions necessary for existence of exact linear error dynamics. The design is optimal in the sense that we minimize norm of the regular perturbations that appear in the error dynamics. In this sense, our result is dual to recent advances in approximate feedback linearization, where an approximate feedback linearizing controller is designed by finding a "closest" feedback linearizable system, and constructing the control for it. The observer construction is easily carried out using symbolic mathematical software such as {\em Mathematica}, regardless of the complexity of the nonlinear system. Furthermore, because the resulting observer is based upon spline functions, it is computationally efficient and well-suited for DSP-based implementation.