Improved IMU/GNSS EKF Fusion Using XGBoost Machine Learning Algorithm

For years, Inertial Measurement Unit (IMU) and Global Positioning System (GPS) have been playing a crucial role in navigation systems. The emergence of inexpensive IMU sensors has offered a lightweight alternative, yet they suffer from larger errors that build up gradually, leading to drift errors in navigation. Traditionally, IMUs are combined with GPS to ensure stable and accurate navigation. However, this integration fails during challenging GPS environments. This work introduces an inventive approach to alleviate navigation drifts by eradicating IMU errors through the utilization of the Extreme Gradient Boosting (XGBoost) algorithm. Unlike traditional approaches relying on high-end IMUs, this paper utilizes Inverse Kinematics to obtain pseudo-clean IMU training data from PVA values estimated by the EKF when GPS signals are available. The proposed approach is evaluated using both simulated and real-world datasets across various GPS outage durations, demonstrating significant enhancements in PVA estimation.