Return to Session D8 Next Abstract

Session D8: AI/Machine Learning: MagNav

Multi-Aircraft Generalizability of Platform Denoising Models for Magnetic Navigation
Kimberly M. Moore, Patrick Neary, Favour Nerrise, Alexander McNeil, Jonathan Ouellet, Eddie Rodriguez, Eric Shoup, Luca Ferrara, Elijah Williams, Kyle Palko, Craig Fitzpatrick, SandboxAQ
Location: Ballroom E
Date/Time: Wednesday, Jun. 5, 8:35 a.m.

Magnetic Navigation (MagNav) is a critical technology that is able to provide absolute positioning in GPS-denied environments. This technology works by comparing a measured field to a reference magnetic field map of the Earth, in order to generate a PNT solution. Importantly, the MagNav device must be calibrated (denoised), to remove platform noise present in the onboard sensor measurements. While multiple techniques for calibration are being explored in the community (e.g. Tolles-Lawson or Artificial Intelligence (AI)), an equally important question is the generalizability of such models. In this talk, we explore the broad question of generalizing calibration models across multiple tail numbers of the same airframe, and over different flights.
In collaboration with the United States Air Force, SandboxAQ has collected magnetic field data onboard multiple C-17 flight campaigns, including operational exercises, as well as training flights at Joint Base Charleston. In particular, the area surrounding Joint Base Charleston was recently remapped by the United States Geological Survey (USGS) as part of their EarthMRI initiative, providing new high-quality ground truth magnetic anomaly maps for comparison. Our unique dataset across multiple aircraft tails provides an outstanding opportunity for probing long-term technical questions of model generalizability. We performed extensive AI transfer learning analysis to investigate the generalizability and durability of individual calibration models, and discuss the results in terms of operational implications for magnetic navigation on new aircraft.
SandboxAQ is a quantum information science company that spun out of Alphabet in 2022, bringing novel solutions combining AI and quantum technologies to market. The multi-disciplinary team includes scientists specializing in geomagnetism, physics, and AI as well as hardware engineers with expertise in prototyping and small form factor sensors.



Return to Session D8 Next Abstract