Comparison of Adaptive and Randomized Unscented Kalman Filter Algorithms

The paper deals with state estimation of nonlinear dynamic stochastic systems with a special focus on advanced unscented Kalman filter algorithms. Two algorithms are considered: the adaptive unscented Kalman filter and the randomized unscented Kalman filter. Both algorithms construct one or several sigma-points set used for an approximation of the conditional state moments. While the adaptive algorithm obtains a sigma-point set by optimization of a criterion, the randomized algorithm constructs several sets randomly. In the paper, both algorithms are compared and a recommendation for an application of the algorithms is provided. The algorithms are illustrated in a bearings-only target tracking example.