Linear observer synthesis for nonlinear systems using Koopman Operator framework

In this paper we develop a new approach for observer synthesis for discrete time autonomous nonlinear systems based on Koopman operator theoretic framework. Koopman operator is a linear but an infinite-dimensional operator that governs the time evolution of system outputs in a linear fashion. We exploit this property to synthesize an observer form which enables the use of Luenberger/Kalman-like linear observers for nonlinear estimation. Using the techniques for Koopman eigenvalue/eigenfunction/mode computation, we describe a numerical procedure to construct such an observer form which is often valid in a large portion of state space or even globally. We numerically compare our approach with Extended Kalman Filter and report superior performance.