Scalable Magnetic Field SLAM in 3D Using Gaussian Process Maps (1804.01926v2)

Abstract: We present a method for scalable and fully 3D magnetic field simultaneous localisation and mapping (SLAM) using local anomalies in the magnetic field as a source of position information. These anomalies are due to the presence of ferromagnetic material in the structure of buildings and in objects such as furniture. We represent the magnetic field map using a Gaussian process model and take well-known physical properties of the magnetic field into account. We build local maps using three-dimensional hexagonal block tiling. To make our approach computationally tractable we use reduced-rank Gaussian process regression in combination with a Rao-Blackwellised particle filter. We show that it is possible to obtain accurate position and orientation estimates using measurements from a smartphone, and that our approach provides a scalable magnetic field SLAM algorithm in terms of both computational complexity and map storage.

• The paper introduces a novel magnetic field SLAM method that uses Gaussian process regression to produce precise 3D maps from smartphone magnetometer data.
• It integrates reduced-rank GP regression with a Rao–Blackwellised particle filter to efficiently segment and manage large indoor environments.
• Experimental results demonstrate that the approach effectively corrects odometry drift and scales with low computational complexity for extended trajectories.

Scalable Magnetic Field SLAM in 3D Using Gaussian Process Maps

The paper presents an innovative approach to 3D simultaneous localization and mapping (SLAM) using magnetic field anomalies as a source of position information. This technique holds particular promise for indoor navigation, where traditional GPS-based systems are ineffective. The authors, Manon Kok and Arno Solin, leverage local anomalies detected by magnetometers present in modern smartphones without necessitating additional infrastructure. The process hinges on constructing a comprehensive map of magnetic fields based on Gaussian process (GP) regression.

Methodology Summary

The featured SLAM system comprises several methodological advancements:

• Magnetic Field Mapping: The environment's magnetic field is captured via a Gaussian process model, which incorporates well-established physical properties derived from Maxwell's equations. The resultant magnetic field map accounts for both Earth's magnetic field and deviations caused by ferromagnetic materials in buildings.
• Computational Efficiency: A major challenge in executing continuous SLAM over large volumes is the computational load. This paper addresses the problem by implementing reduced-rank Gaussian process regression combined with a Rao–Blackwellised particle filter (RBPF). This approach segments the environment into hexagonal blocks, thereby enabling efficient processing and reducing the computational complexity that scales with the number of measurements.
• 3D SLAM Configuration: The authors assert that previous approaches primarily operated in 2D or 2.5D, restricting their applicability in varied environments. This research pioneers a fully 3D mapping approach applicable for real-world 3D environments, as demonstrated through experimental trajectory mapping on a smartphone.

Key Results

Highlighted outcomes and empirical evaluations of the presented SLAM method include:

• The proposed framework successfully maintains a high level of accuracy in mapping and localization. It effectively corrects positional drift from odometry through improved map localization as validated by comprehensive empirical data.
• The authors provide concrete examples of their method applied to capture trajectories within indoor and mixed environments, proving the method's scalability and robustness in handling extended 3D movements.
• The dimensions and regional representations of magnetic field maps are efficiently managed by approximating large map areas with small state dimensions, specifically using 256 states per hexagonal block covering approximately 260 cubic meters.

Implications and Future Directions

The authors' approach to scalable 3D SLAM utilizing magnetic field anomalies presents significant implications in the domain of indoor localization. It can potentially enhance personal navigation devices by utilizing pervasive smartphone sensors without augmenting additional infrastructure. As such, the method presents appealing applications in navigation for augmented reality systems and ambient intelligence environments.

Future research directions highlighted by the authors include:

• An exploration of inertial pedestrian dead reckoning (PDR) systems as replacements for current odometry solutions, aiming towards a fully GPS-independent localization framework.
• The potential application of a Rao–Blackwellised Particle Smoother to refine trajectory estimations further.
• Continued optimization to further reduce computational demands, which could extend the method's applicability to real-time navigation systems.

In conclusion, this paper provides a substantial contribution to 3D magnetic field SLAM methodologies. By integrating Gaussian processes with particle filters, the authors demonstrate a proficient method for handling large-scale, three-dimensional environments, paving the way for future development in ubiquitous localization technologies.