Turtle-inspired localization on robot

In nature, some animals exhibit impressive navigation capability using ambient magnetic field. Particularly, certain kinds of sea turtles can associate geomagnetism to spatial representation for positioning in transoceanic migration across the seemingly clueless sea. In robotics, the previous works on magnetic navigation can position a robot using ambient magnetic field, but they focused on an exploitation of extensively-explored magnetic map, i.e. the magnetic field of every inch of the region of interest is recorded for mapping. In this work, we propose an algorithm that is based on our analysis on how sea turtles navigate at sea under magnetic disruption as investigated and reported by biologists [1], [2]. We propose a direct likelihood method that generates pseudo training data to improve the estimation accuracy of the Gaussian mixture models. The experimental evaluation demonstrates that our localization algorithm exhibits stable and accurate positioning results. This work contrasts with the previous works which focused on magnetic localization using extensive data collection. On the contrary, we address whether magnetic localization is still feasible under scarce data samples and how to overcome this challenge.