With the advancement of multi-source sensor fusion technologies, integrating the Global Positioning System (GPS) with the Inertial Navigation System (INS) has become a practical solution for navigation and positioning in unmanned vehicles. However, in challenging environments such as tunnels and forests, GPS signals are often unavailable, and standalone INS suffers from inherent drift and accumulated measurement noise, resulting in significant degradation in positioning accuracy. To address these issues, this paper proposes an Incremental regularized bidirectional long short-term memory (IncRBiLSTM) framework to enhance navigation robustness without relying on auxiliary sensors. The proposed model leverages bidirectional temporal feature extraction and incremental learning to continuously adapt to dynamic environments, while regularization techniques mitigate overfitting and improve generalization. Additionally, an empirical mode decomposition–lifted wavelet transform (EMD–LWT)-based denoising algorithm is employed to preprocess raw Inertial Measurement Unit(IMU) signals, effectively suppressing noise and enhancing signal quality. The proposed method is validated using real-world vehicular trajectory data collected in campus driving scenarios. Experimental results show that the proposed framework reduces Root Mean Square Error (RMSE) by approximately 71.98% and Mean Absolute Error(MAE) by 77.70% compared to existing methods, confirming its effectiveness and robustness for accurate navigation during GPS signal interruptions.
Wang et al. (Sun,) studied this question.