An Improved Magnetic Field SLAM Algorithm Based on Iterative Extended Kalman Filter and Gaussian Process Regression

This paper introduces an approach for magnetic field simultaneous localization and mapping by leveraging Reduced-Rank Gaussian Process Regression. The proposed algorithm aims to improve the efficiency and accuracy of magnetic field-based localization in environments with spatial variations. The methodology involves representing the magnetic field potential as a sum of basis functions. The use of Reduced-Rank Gaussian Process Regression facilitates a streamlined representation, enabling faster computation and reduced storage requirements. Then, two estimation methods are designed: an Extended Kalman Filter and Iterative Extended Kalman Filter methods to estimate the states of the dynamic model. Simulation results have demonstrated the effectiveness of the proposed approaches in estimating the true dynamic states, with slight improvement of the Iterative Extended Kalman Filter accuracy at certain magnetic field length scales, compared to the Extended Kalman Filter design.