Positioning in Spatially Inhomogeneous Static Magnetic Fields Using Low-Cost Sensors

QC 20260505

Abstract

Accurate positioning plays a crucial role in modern society, enabling efficient and safe transportation, supporting emergency response, and underpinning autonomous systems. Indoor positioning has attracted significant research attention because global navigation satellite system signals are unreliable or unavailable indoors, while the growing demand for navigation and location-based services in complex indoor environments continues to increase. One promising approach is magnetic-field-based positioning, which exploits spatial variations of indoor magnetic fields.

However, current magnetic-field positioning systems still have some limitations. Many existing solutions rely on pre-collected magnetic field maps of the environment, which impose high overhead and maintenance costs. In contrast, solutions that do not require collecting maps prior to positioning often depend on low-drift odometry, such as pedestrian dead reckoning or visual odometry, to maintain accuracy, or assume 2D motion to simplify the positioning problem. These dependencies limit the usability and deployment flexibility of magnetic-field-based positioning systems in indoor environments.

In this thesis, we investigate how a magnetometer array and other low-cost sensors can be utilized to overcome these limitations. This research addresses four problems: 1) how an array of low-cost magnetometers and an inertial measurement unit (IMU) can be accurately and efficiently jointly calibrated to minimize measurement errors; 2) how an inertial magnetic odometry system can be constructed; 3) how consistency of the uncertainty reported by an inertial magnetic odometry system can be improved; and 4) how an inertial magnetic simultaneous localization and mapping (IM-SLAM) system can be built to provide long-term, accurate positioning.

The first problem is addressed in Paper A, where we proposed an efficient and accurate method to jointly calibrate a magnetometer and an IMU. The proposed method has similar calibration accuracy to that of the state-of-the-art, while being one order of magnitude faster in calibration. Thus, the proposed method is a reliable and effective vii choice for jointly calibrating magnetometer-IMU pairs. The second problem is addressed in Paper B, where we developed a magnetic-field-aided inertial navigation system with a horizontal error of less than 2 meters after more than 3 minutes of navigation. This system has the potential to solve one of the key challenges of existing magnetic-field simultaneous localization and mapping solutions — the very limited allowable length of the exploration phase during which unvisited areas are mapped. The third problem is addressed in Paper C, where we proposed an observability-constrained magnetic-field-aided inertial navigation system to improve the consistency of yaw uncertainty, thereby enhancing the long-term reliability of inertial magnetic odometry systems. Finally, we address the fourth problem by developing loosely coupled and tightly coupled IM-SLAM systems. These two system architectures serve as practical reference designs for practitioners. More importantly, the tightly coupled IM-SLAM system achieves a final positioning error of less than 2 meters over an approximately 200-meter trajectory.

The work in this thesis demonstrates the feasibility of building inertial-magnetic positioning systems with moderate accuracy. A potential application of the proposed systems is for the positioning of emergency response officers, e.g., in mining and fire rescue, where low visibility poses a significant challenge to rescue operations.

Link to DiVA