The sigma points whose center and covariance were predictive value and predictive variance would deviate from true value while UKF applied for large-scale outdoor SLAM. This paper advantage of an iteration mechanism in update equations of unscented SLAM Updated state estimate and covariancere applied to new sigma points while the observation obtained for one instant. Filtering parameters calculated after nscented of these sigma points, the state estimate value close to the true one, and efficiency of data association and accuracy of map building. The simulation results show that this proposed method could reduce error approximately 19% in x while 21% in y while accelerate convergence speed of SLAM.