To solve the problem of poor robustness in unscented Kalman filter (UKF), considering the tightly integrated characteristics of the GPS/INS model, a mixed square root UKF based on the interacting multiple model (IMM-SRUKF, interacting multiple models-square root UKF)was presented. The algorithm adopted not only an interactive multi-model structure to overcome the uncertainty factors of the model, but also the method of square root filter to solve the problem that the covariance matrix is difficult to maintain positive definite. Moreover, taking into account the fact that the internal filter with linear / nonlinear model did not match,  the idea of hybrid filter was introduced to optimize the SRUKF(square root UKF). The new algorithm was applied to the tightly integrated model. Simulation results show that the algorithm has better robustness at the price of proper time complexity and can be applied to complex navigation cases.