Aiming at the problem that when the Sage-Husa adaptive Kalman filter ( SHAKF ) algorithm is processing inertial measurement units ( IMU), random errors are likely to accumulate over time and cause filter divergence, an improved Sage-Husa adaptive robustness Kalman filter (MSHARKF) algorithm is proposed. First, build a suitable model for IMU, then combine SHAKF with adaptive robust Kalman filter (ARKF) and incorporate it into the improved time-varying noise estimator, and then introduce the optimal adaptive scale factor (αk ) to the measurement equation Iteratively update, and finally get a new predicted covariance matrix to be substituted into the original equation. The experimental results show that through the Allan variance and root mean square error (RMSE), the static / dynamic data before and after the MEMS-IMU filtering is calculated and the random error noise is reduced to one ten thousandth and one percent of the original data, respectively. Compared with other algorithms in this paper, this method effectively suppresses the filtering divergence of the algorithm, thereby improving the measurement accuracy and long-term stability of the IMU.