Large-scale magnetic field mapping and magnetic map-based localization
Fast large-scale magnetic field mapping with Gaussian process regression
There are many magnetized materials in real environment and they make magnetic fluctuations. These fluctuations can be used as landmarks and localization methods based on the fluctuations has been proposed by some researchers; however, almost all methods use a linear magnetic map which stores magnetic data according to travel distance. Hence, these methods cannot estimate a 2D (or 3D) robot pose and its heading angle. Although the 2D pose must be estimated for the autonomous navigation, it is significant time consuming to build the 2D magnetic map. In this work, I proposed a very fast large-scale 2D and 3D magnetic field mapping method and realized the magnetic map-based localization in large-scale environments. The key technique of the method is use of the Gaussian process regression to estimate unmeasured magnetic fields.
Magnetic data collection with a robot that has a LiDAR-based localization module
I used a robot that has a LiDAR-based localization module to measure the magnetic field. The LiDAR-based localization gives us very accurate robot pose and magnetic data with accurate position can be easily collected. However, it is significant time consuming to measure all of the magnetic fields even though the robot is used. I overcome this problem by using the Gaussian process regression.
Dense magnetic field mapping with Gaussian process regression
The bottom left and middle figures are an occupancy grid map and the trajectories of the magnetic sensors estimated by the LiDAR-based localization. There are unmeasured magnetic fields; however, the Gaussian process regression enables us to build a dense magnetic map from the coarse magnetic field measurement result. Magnetic fluctuations can be seen from the mapping result.
Localization with the magnetic map
The Gaussian process regression also estimates uncertainty of the regression and it is valuable for the probabilistic localization. I implemented the particle filter-based localization with the magnetic map and realized 2D localization in large-scale environment. The bottom figure show the localized trajectory (left) and its estimation errors (right) according to the position and heading angle.
Extension to 3D
To extend the method, I developed a 3D magnetic field measurement robot that has a simple manipulator. The robot recognizes 3D objects and moves the manipulator with a collision avoidance strategy. 3D magnetic data can be quickly measured by the robot. I realized large-scale 3D magnetic field mapping (visualization). As far as I know, there are no works that realized this kind of large-scale 3D magnetic field visualization.
Publications
The large-scale magnetic field mapping method with Gaussian process regression is presented in [1]. This work also presents the particle filter-based localization with the magnetic map. I extended this work and presented the large-scale 3D magnetic field mapping method in [2].
[1] Naoki Akai and Koichi Ozaki. "Gaussian processes for magnetic map-based localization in large-scale indoor environments," In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 4459-4464, 2015. (ResearchGate)
[2] Naoki Akai and Koichi Ozaki. "3D magnetic field mapping in large-scale indoor environment using measurement robot and Gaussian processes," In Proceedings of the Indoor Positioning and Indoor Navigation (IPIN), pp. 1-7, 2017. (ResearchGate)