A TCN-BiLSTM and ANR-IEKF Hybrid Framework for Sustained Vehicle Positioning During GNSS Outages

Area of Science:

• Robotics and Autonomous Systems
• Navigation and Positioning

Background:

• Integrated Global Navigation Satellite System and Inertial Navigation System (GNSS/INS) performance degrades in urban canyons due to signal blockage.
• Reliable, continuous positioning is critical for autonomous driving applications.

Purpose of the Study:

• To develop a robust hybrid framework for high-precision navigation in GNSS-denied environments.
• To enhance the reliability of GNSS/INS systems during signal outages.

Main Methods:

• A novel framework combining a Temporal Convolutional Network and Bidirectional Long Short-Term Memory (TCN-BiLSTM) for pseudo-GNSS generation.
• An Adaptive Noise-Regulated Iterated Extended Kalman Filter (ANR-IEKF) for fusing pseudo-GNSS and INS data.
• Real-world vehicle dataset validation including straight-line and turning scenarios.

Main Results:

• The ANR-IEKF + TCN-BiLSTM framework demonstrated superior positioning accuracy and robustness compared to baseline models.
• Achieved over 50% enhancement in positioning accuracy during 70-second GNSS outages.
• Reduced positioning errors to approximately 3.4 meters against strong deep learning baselines.

Conclusions:

• The proposed framework offers a reliable solution for maintaining high-precision navigation in GNSS-denied environments.
• The adaptive Kalman filter effectively enhances robustness by dynamically adjusting noise statistics.
• The TCN-BiLSTM model successfully generates accurate pseudo-GNSS measurements during signal outages.