NavFormer: Robust IGRF Forecasting for Autonomous Navigators

A new study introduces NavFormer, an innovative method for forecasting the International Geomagnetic Reference Field (IGRF) in contexts where coordinate systems are in motion. The research focuses on managing variations in triad magnetometer components, which change with sensor attitude even when the IGRF total intensity target remains invariant.

Solution Details

NavFormer addresses this challenge by forecasting the invariant target with rotation invariant scalar features. A Canonical SPD module is employed to stabilize the spectrum of window level second moments of the triads, avoiding sign discontinuities. This module builds a canonical frame from a Gram matrix per window and applies state dependent spectral scaling in the original coordinates.

Experimental Results

The results obtained across five flights demonstrate that NavFormer achieves lower error than strong baselines in various training scenarios, including standard training, few-shot learning, and zero-shot transfer. The project code is publicly available for the scientific community.