![[2602.13690] Physics Aware Neural Networks: Denoising for Magnetic Navigation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.13690] Physics Aware Neural Networks: Denoising for Magnetic Navigation
Summary
This paper presents a novel framework for denoising magnetic navigation data using physics-aware neural networks, addressing challenges in airborne systems affected by magnetic noise.
Why It Matters
As GPS technology can be unreliable in certain environments, enhancing magnetic navigation systems is crucial for various applications, including aviation and autonomous vehicles. This research introduces a method that improves data accuracy and reliability, which could significantly impact navigation technology.
Key Takeaways
- Introduces a physics-based framework for denoising magnetic navigation data.
- Utilizes constraints from Maxwell's equations to enhance predictive accuracy.
- Demonstrates improved performance over classical deep learning methods.
- Employs synthetic datasets to address data scarcity in training.
- Highlights the importance of continuous-time dynamics in modeling magnetic time series.
Computer Science > Machine Learning arXiv:2602.13690 (cs) [Submitted on 14 Feb 2026] Title:Physics Aware Neural Networks: Denoising for Magnetic Navigation Authors:Aritra Das (1), Yashas Shende (1), Muskaan Chugh (1), Reva Laxmi Chauhan (1), Arghya Pathak (1), Debayan Gupta (1) ((1) Ashoka University) View a PDF of the paper titled Physics Aware Neural Networks: Denoising for Magnetic Navigation, by Aritra Das (1) and 5 other authors View PDF HTML (experimental) Abstract:Magnetic-anomaly navigation, leveraging small-scale variations in the Earth's magnetic field, is a promising alternative when GPS is unavailable or compromised. Airborne systems face a key challenge in extracting geomagnetic field data: the aircraft itself induces magnetic noise. Although the classical Tolles-Lawson model addresses this, it inadequately handles stochastically corrupted magnetic data required for navigation. To address stochastic noise, we propose a framework based on two physics-based constraints: divergence-free vector field and E(3)-equivariance. These ensure the learned magnetic field obeys Maxwell's equations and that outputs transform correctly with sensor position/orientation. The divergence-free constraint is implemented by training a neural network to output a vector potential $A$, with the magnetic field defined as its curl. For E(3)-equivariance, we use tensor products of geometric tensors representable via spherical harmonics with known rotational transformations. Enforcing phys...
