A hybrid GNSS/INS integrated navigation method combining GRU-SA and covariance-adaptive factor graph optimization during GNSS outages

Abstract An integrated navigation system that combines the global navigation satellite system (GNSS) and the inertial navigation system (INS) provides accurate navigation and positioning information for carriers. However, the complex traffic environment leads to frequent GNSS outages. The positioning accuracy of the GNSS/INS integrated navigation system decreases due to the lack of global positioning information. In order to address this problem, we propose a hybrid GNSS/INS integrated navigation method for GNSS outage conditions, combining the GNSS pseudo-measurement regression model and the covariance-adaptive factor graph optimization (CAFGO). The GNSS pseudo-measurement regression model is prone to gradient vanishing caused by high-dimensional data input from high-frequency INS data, leading to a decline in long-sequence feature dependency. To overcome this limitation, an enhanced GNSS pseudo-measurement regression model integrating gated recurrent units (GRU) with self-attention (SA) is proposed, which adaptively weights GRU outputs to enhance focus on critical temporal features. Meanwhile, to address the fluctuations in the measurement covariance matrix of GNSS pseudo-measurements caused by the uncertainty of neural networks, we propose a CAFGO method, which adaptively adjusts the covariance of GNSS factors in the optimization process based on model prediction errors. Real-world experiment results indicate that the proposed method effectively improves the positioning accuracy of the GNSS/INS integrated navigation system during GNSS outages. In the 60 s and 120 s GNSS outages comparison experiment, the positioning root mean square error of the proposed method decreased by 70.40% and 49.69% compared to the GRU-factor graph optimization method.