1. College of Geo?exploration Science and Technology, Jilin University, Changchun 130026, China; 

 Traditional Kalman filter is usually difficult for adaptive changes in the complex navigation environ? ment of integrated GNSS / INS, which leads to the accuracy decreasing of parameter estimation or even the filter di? vergence. In this paper, Sage??Husa adaptive filter and robust filter based on innovation vector are introduced to process navigation data with forward and backward fusion. The comparison with traditional Kalman filter indicate that the estimation accuracy of horizontal attitude, heading angle, three?dimensional velocity and position is im? proved by 37. 38% , 64. 23% , 23. 51% , 29. 06% on average, and the optimal improvement is 58. 77% , 95. 17% , 36. 61% , 49. 20% , respectively. The adaptive, robust and forward and backward Kalman filters im? prove the accuracy and stability of parameter estimation, and can be used for post?processing of integrated naviga? tion data in complex urban navigation environment.