An Extended Kalman Filter for Magnetic Field SLAM Using Gaussian Process Regression

Area of Science:

• Robotics
• Geophysics
• Computer Science

Background:

• Odometry systems suffer from position drift, impacting localization accuracy.
• Magnetic field variations offer a potential source for accurate mapping and localization.
• Existing magnetic field SLAM methods using Rao-Blackwellized particle filters (RBPF) are computationally intensive.

Purpose of the Study:

• To develop a computationally efficient algorithm for simultaneous localization and mapping (SLAM) using ambient magnetic field variations.
• To reduce the computational complexity of magnetic field-based SLAM.
• To improve position accuracy by compensating for dead-reckoning drift.

Main Methods:

• Representing the magnetic field map with a reduced-rank Gaussian process (GP) using Laplace basis functions.
• Deriving analytic expressions for the gradient of the learned magnetic field.
• Applying an extended Kalman filter (EKF) utilizing the magnetic field gradients for SLAM.

Main Results:

• The proposed EKF-based algorithm achieves a computational complexity of O(Nm2) per time step, significantly lower than the O(NpNm2) of RBPF methods.
• The algorithm effectively compensates for position drift in integrated odometry measurements.
• Performance in drift compensation is comparable to existing RBPF-based magnetic field SLAM techniques.

Conclusions:

• The EKF-based approach offers a computationally efficient alternative for magnetic field SLAM.
• This method enables accurate localization by leveraging magnetic field gradients.
• The algorithm demonstrates practical applicability in real-world scenarios, validated on diverse datasets.