A robust and adaptive framework for GNSS/INS integrated navigation based on dual-domain deep learning network during GNSS outages

High-precision vehicle navigation during Global Navigation Satellite System (GNSS) outages is critical for operational safety. To address this, many studies leverage deep learning to predict GNSS measurement increments from Inertial Measurement Unit (IMU) data. These predicted increments are then used to form pseudo-measurements for the filter’s measurement update step. However, the performance of existing deep learning-based solutions is constrained by limitations across their core components, including inadequate network architectures for modeling complex vehicle motion dynamics, insufficient filtering strategies for handling the unknown noise characteristics of pseudo-measurements, and deficient validation methodologies based on simulated instead of authentic outage scenarios. Targeting these deficiencies, this paper introduces a robust and adaptive framework centered on a novel Dual-Domain MixerNet (DDMNet). Distinguishing itself from conventional temporal-only architectures, DDMNet incorporates a Content-aware Frequency Attention mechanism to explicitly extract motion features from both temporal and frequency domains. To address filtering limitations, a Data-driven Robust Adaptive EKF (DRAEKF) is designed to overcome the reliance on empirical noise settings, utilizing a data-driven strategy to dynamically estimate covariance. Uniquely, a GNSS Quality Check mechanism with a look-back strategy is integrated to resolve the critical, yet often neglected, issue of pre-outage measurement degradation. The proposed framework was validated across diverse test scenarios, including datasets with simulated outages and a trajectory featuring an authentic GNSS outage beneath an elevated roadway. In the authentic GNSS outage test, our approach achieved an 80.25% reduction in position Root Mean Square Error (RMSE) compared to pure inertial navigation. The results confirm that the framework provides a highly effective and robust solution for continuous navigation in GNSS outage environments.