As the state dimension increases, the amount of calculation of Unscented Kalman Filter (UKF) algorithm increases rapidly. Besides, UKF is sensitive to the model error, and is not suitable for the system model whose noise doesn't subject to Gaussian distribution. In order to solve these problems, a new robust model-predictive UKF algorithm is proposed. This method incorporates the driving noise into the system state through the augmentation of state dimensions to add the system state information. The model error is restrained by Model Predictive Filter (MPF), and the robustness of the system is enhanced by the robust estimation. Thus the limitation of the traditional UKF algorithm which is sensitive to the model error is overcome. The proposed algorithm is applied to the new integrated SINS/BNTS/CNS navigation system for simulation, and is compared with the adaptive EKF algorithm and the robust adaptive UKF algorithm. The results show that the proposed algorithm can effectively restrain the attitude error and the velocity error, and its filtering capability is superior to that of the adaptive EKF and the robust adaptive UKF.