Dual-Factor Constrained Factor Graph Optimization With Hybrid Optimization for Robust GNSS/INS Integration

Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) integrated navigation systems are prone to significant error accumulation in urban environments due to GNSS signal fluctuations, outages, and dynamic vehicle maneuvers such as start-stop maneuvers and sharp turns, which existing methods like Kalman Filter (KF) and conventional Factor Graph Optimization (FGO) struggle to mitigate effectively. To solve this problem, a novel adaptive FGO algorithm featuring dual-factor constraints and hybrid optimization mechanisms is proposed. Initially, a motion state detection model based on inertial measurement unit (IMU) is developed to distinguish stationary and turning states accurately by analyzing real-time features. Subsequently, a dual-factor constraint framework is established to integrate static-dynamic attributes, utilizing zero velocity update (ZUPT) to suppress inertial error during stationary phases and cubic spline interpolation for compensating GNSS latency during turning maneuvers. Furthermore, a hybrid optimization approach that combines the global search capability of the Dogleg method with the local refinement strength of the Levenberg-Marquardt (LM) algorithm is proposed to strike a balance between convergence efficiency and estimation accuracy. Experimental validation in urban environments demonstrates that the proposed method significantly outperforms the KF, traditional FGO and a state-of-the-art deep learning method using Gated Recurrent Units (GRU). Analysis reveals significant improvements in positioning accuracy and velocity estimation stability, particularly in challenging signal conditions, while maintaining improved robustness and trajectory continuity.