On using the output progression in unscented Kalman filtering

By calculating the state from past output samples, Newton observers (NO) provide an intuitive approach to state estimation in nonlinear systems. However, due to their sensitivity to measurement noise, they failed to become established. Unscented Kalman filters (UKF), on the other hand, have proven to provide good noise attenuation and perform well if the underlying system is highly nonlinear. These benefits, however, come at the cost of a, sometimes long, settling time. We propose a modified filtering algorithm that builds upon both the unscented transformation used in UKF and the rationale of the NO. By leveraging the information contained in the output progression, the new filter obtains the immediacy of the NO and improved steady-state noise suppression. State estimation with the proposed algorithm is investigated both in a large-scale simulation study and in experiments with a physical gantry crane model.