Adaptive unscented particle filter based on predicted residual

Abstract

In order overcome the particle degradation and non-adjusted online in the traditional particle filter algorithm, an adaptive un scented particle filter algorithm based on predicted residual is proposed. The algorithm adopts a new proposal distribution combing the unscented kalman filter with the adaptive factor. The algorithm uses Unscented Kalman filter to generate a proposal distribution, in which the covariance of the predicted measurement, the cross-covariance of the state and measurement and the covariance of the state update are online adjusted by predicted residual as adaptive factor. Simulation experiments results of nonlinear state estimation demonstrate that the adaptive unscented particle filter is more adaptive and accuracy is also improved.