The talk gives an overview of the topic of utilizing magnetic field distortions for robotic navigation as it was elaborated in the corresponding PhD thesis. The work comprehensively illuminates the various aspects that are relevant in this context, for example the characteristics of magnetic field environments, which are assessed and examined for their usability for robot navigation in various typical mobile robot deployment scenarios.
A strong focus of the work lies on the self-induced static and dynamic magnetic field distortions of complex kinematic robots, which could hinder the use of magnetic fields because of their interference with the ambient magnetic field. In addition to the examination of typical distortions in robots of different classes, solutions for compensation and concrete tools are developed both in hardware (distributed magnetometer sensor systems) and in software. In this context, machine learning approaches for learning static and dynamic system distortions are explored and contrasted with classical methods for calibrating magnetic field sensors.
In order to extend probabilistic state estimation methods towards the localization in magnetic fields, a measurement model based on Mises-Fisher distributions is developed in this thesis.
Finally, the approaches of this work are evaluated in practice inside and outside the laboratory in different environments and domains (e.g. office, subsea, desert, etc.) with different types of robot systems.