@article{LIANG2026, 
author = {Xiao LIANG and Pengyu ZHAO and Sitong LIU and Baojin LIU and Longzhi NIE and Chuanjun LI},
title = {Attitude-constrained interactive multi-model factor graph fusion for integrated navigation},
year = {2026},
journal = {Chinese Journal of Aeronautics},
volume = {39},
number = {3},
keywords = {Attitude-constrained position, Factor graph optimization, IMU pre-integration, Interactive multiple models, Multi-sensor integrated navigation},
url = {https://www.sciopen.com/article/10.1016/j.cja.2025.103963},
doi = {10.1016/j.cja.2025.103963},
abstract = {Continuous and highly accurate navigation of long-endurance flight vehicles continues to be a substantial challenge in Global Navigation Satellite Systems (GNSS)-denied environments. Though the integrated navigation based on multiple sensors is used to improve the navigation performance, the existing methods are prone to model mismatch and error accumulation under heterogeneous conditions of sensors. In this paper, a Tightly-coupled Interactive Multi-Model Factor Graph Optimization (TIMMFGO) navigation method is proposed to solve the problem. The developed integrated navigation framework consists of Inertial Navigation Systems (INS), Celestial Navigation Systems (CNS), Radio Navigation Systems (RNS), and Barometric Altimeters (BA). We propose a CNS/INS tightly-coupled graph architecture that integrates star vector observations with INS pre-integration, enabling dynamic compensation of gyroscopic bias while correcting the attitude update accuracy of INS. Then, an Interactive Multi-Model (IMM) adaptive weighting strategy is used to combine the vertical RNS factor with the BA factor for position, which can effectively reduce the altitude bias induced by the spatial configuration constraints of RNS. The simulation demonstrates that compared to the Huber M–estimation–based FGO (HMFGO), Windowing Anomaly-Detection-based FGO (WADFGO) and IMM Unscented Kalman Filter (IMMUKF) methods, the TIMMFGO method improves attitude accuracy by 46.49 %, 25.68 % and 20.67 %, respectively, while correspondingly reducing position accuracy by 29.29 %, 10.79 % and 6.96 %.}
}