Reliability Assessment of an Unscented Kalman Filter by Using Ellipsoidal Enclosure Techniques

The Unscented Kalman Filter (UKF) is widely used for the state, disturbance, and parameter estimation of nonlinear dynamic systems, which both process measurement uncertainties are represented in a probabilistic form. Although UKF can often be shown to more reliable processes than linearization-based Extended (EKF) due enhanced approximation capabilities its underlying probability distribution, it not priori obvious whether strategy selecting sigma points sufficiently accurate handle nonlinearities system dynamics output equations. Such inaccuracies may arise strong combination with large covariances. Then, computationally demanding approaches such as particle filters or representation (multi-modal) densities help (Gaussian) mixture representations possible ways resolve this issue. To detect cases systematic manner that reliably handled by standard EKF UKF, paper proposes computation outer bounds state domains compatible certain percentage confidence under assumption normally distributed states set-based ellipsoidal calculus. practical applicability approach demonstrated variables parameters an unmanned surface vessel (USV).