Solution Separation Unscented Kalman Filter

Jindrich Duník, Ondrej Straka, Erik Blasch

Abstract: State estimation of nonlinear stochastic dynamic systems includes the unscented Kalman filter. The paper focuses on the self-assessment ability of state estimation algorithms, which shows to be useful in the problems involving strong nonlinearity of the model. An algorithm with the self-assessment is capable of informing the user that the state estimate may not be credible. A new algorithm, called solution separation unscented Kalman filter (S2UKF) is proposed, which is inspired by the solution separation technique of the Global Navigation Satellite System (GNSS) receiver autonomous integrity monitoring. The algorithm compares a set of sub-solution estimates with a full-solution estimate and informs about non-credible estimates. This ability is demonstrated using a simulation example involving the bearings-only tracking problem.