Tensor-field based localization project

• Funding agency: Swedish Research Council (VR) & Security Link
• Project leader: Isaac Skog
• Project partners: FOI, Swedish Defense Research Agency
• Duration: 2022 - 2027

News and results

• I together with Gustaf Hendeby (LiU) and Manon Kok (TU Delft), will give a tutorial on Indoor Localization using Magnetic Fields at Int. Conf. on Indoor Positioning and Indoor Navigation (IPIN), Sept. 2023.
• A new high-sensitive magnetic-field sensor array has been constructed. Public data sets will be published during the autumn.
• MSc. Chuan Huang started as a Ph.D. student in the project.

Project description

Vector fields, such as the earth's magnetic field and gravity field, are highly informative sources for localization. These exemplified vector fields are omnipresent, stable over-time, and variations in the fields, if measured accurately, provide a fingerprint highly correlated to the measurement location. Hence, they constitute a viable and robust source for localization in GPS-denied environments.

Thanks to the last decade's tremendous sensor technology development, high-performing and affordable accelerometer and magnetometer vector-sensor arrays can be constructed and integrated into portable devices and platforms. Similarly, as a camera can take an image of the surrounding environment, these sensor arrays can take an image-like measurement of a vector-field - a type of measurement that, in the literature, is referred to as tensor-field observation as it constitutes an assembly of spatially separated vector-field observations. Just as in computer-vision based localization systems, these tensor-field observations can be used for odometry via feature tracking, absolute localization via feature matching, and simultaneous localization and mapping (SLAM) via on-the-fly constructions of feature maps. However, unlike visual imaging, magnetic-field and gravity-field sensing is not impaired by smoke, dust, fog, etc. Further, in contrast to the hard-to-model structure of visual images, the imaged vector field must comply with, easy to model, physical laws. Based on these attractive properties, the purpose of the project is to explore how tensor-field observations, combined with inertial navigation1, can enable accurate, reliable, and cost-efficient "anywhere and anytime" self-localization. More precisely, the project aims to research and develop:

• Sensor-fusion and model-learning methods for tensor-field based odometry aided inertial navigation. This will reduce the inertial navigation error growth rate from cubic to linear in time, enabling significantly longer stand-alone operation times before the location error gets too large.
• Sensor-fusion and map-representation methods tailored for joint tensor-field based odometry and SLAM. Compared to existing vector-field SLAM solutions, this will drastically extend the length of the allowable exploration phases where new areas are mapped, as well as increase the accuracy and reliability through more informative field measurement.

Possible use cases of the developed self-localization solutions include pedestrian indoor navigation, underwater navigation, autonomous vehicle navigation in mines, aerospace navigation, etc.

Tutorials

The tutorial "Indoor Localization using Magnetic Fields" has been given at the IEEE International Conference on Indoor Positioning and Indoor Navigation (IPIN) 2023 by Gustaf Hendeby, Manon Kok, and Isaac Skog. The slides from the tutorial can be downloaded here:

1. Part 1: Introduction to Magnetic-Field Based Localization
2. Part 2: Magnetic-Field Based Odometry
3. Part 3: Magnetic field SLAM

If you have any questions regarding the tutorial material, feel free to send me an e-mail.

Publications

1. C Huang, G. Hendeby, and I. Skog A Tightly-Integrated Magnetic-Field aided Inertial Navigation System, IEEE Int. Conf. Information Fusion (FUSION), Jul. 2022.
2. I. Skog, G. Hendeby, and F. Trulsson, Magnetic-field Based Odometry – An Optical Flow Inspired Approach, IEEE Int. Conf. Indoor Positioning Indoor Navigation, Nov. 2021.
3. I. Skog, G. Hendeby, and F. Gustafsson, Magnetic odometry-a model-based approach using a sensor array, IEEE Int. Conf. n Information Fusion (FUSION), 2018.