Aiming at the priori estimated covariance of states turns into a non-positive definite matrix resulted from anomalies of observation, and the equivalent weight function in robust estimation can suffer from the ill-conditioned matrix in unscented Kalman filter(UKF), which leads to the accuracy decrease of the filter performance. A singular value decomposition(SVD) aided improved robust UKF algorithm for integration navigation is proposed, which solves the non-positive definite matrix for unscented transformation. The improved robust algorithm takes different strategy by judging the ill-condition matrix. Finally, field data on the vehicle are collected to verify the algorithm. The results show that, the performance of SVD-UKF navigation solution is slightly superior to the EKF algorithm. The improved robust algorithm can greatly weaken the influence of separate, continuous and hybrid observation anomalies for enhancing the reliability and accuracy of navigation solutions.