Rethinking Navigation Using Quantum Sensing and AI
Quantum sensing and AI are unlocking next-gen navigation — resilient, precise, and independent of GPS
Revolutionizing Navigation with Quantum Sensing and AI: A Deep Dive into MagNav
By Favour Nerrise, AI/ML Resident at SandboxAQ & PhD Candidate at Stanford University
Introduction
Hi everyone! I’m Favour Nerrise, an AI/ML resident at SandboxAQ, where I work on quantum sensing for navigation technologies like AQNav. Combining my PhD research in electrical engineering at Stanford with real-world applications at SandboxAQ has been an incredible journey. Today, I’m excited to share how we’re rethinking navigation using quantum sensing and AI — specifically for airborne applications.
Why Quantum Sensing?
Quantum sensing leverages the principles of quantum mechanics to surpass the limits of traditional sensors (like those in your smartphone). While GPS is widely used, it has vulnerabilities:
- Jamming/Spoofing Risks: GPS relies on satellite signals that can be blocked or manipulated.
- Drift Errors: Over time, GPS accuracy degrades — a critical flaw for aviation and defense.
Enter Magnetic Anomaly Navigation (MagNav):
- Uses Earth’s magnetic field as a passive, jam-resistant signal.
- Enables precise latitude, longitude, and altitude tracking without satellites.
- Ideal for clandestine or high-stakes missions where GPS is unreliable.
The Challenge: Noise vs. Signal
Aircraft generate electromagnetic noise (e.g., from engines or electronics) that interferes with MagNav sensors. Our goal? Separate the true magnetic anomaly signal from the noise. Traditional methods like Tolles-Lawson compensation help, but we turned to AI-driven solutions for greater accuracy.
Physics-Inspired Machine Learning
We tackled this using Liquid Time-Constant Networks (LTCs), a neural architecture inspired by biological systems (like the C. elegans worm!). Key advantages:
- Efficiency: Solves high-dimensional problems with minimal training.
- Closed-Form Solutions: Bakes in physical assumptions, reducing computational overhead.
- Real-Time Processing: Critical for airborne navigation.
Results: 64% Error Reduction
Testing on the Air Force-MIT MagNav Challenge dataset, our LTC model outperformed traditional methods:
- 58–64% lower error vs. Tolles-Lawson baselines.
- Successfully extracted weak magnetic signals from noisy aircraft data.
- Demonstrated potential for real-time, drift-resistant navigation.
Future Directions
- SciML Integration: Embedding deeper physics into models for better generalizability.
- Unsupervised Denoising: Leveraging large datasets to auto-detect noise patterns.
- Scalability: Optimizing for longer flights and diverse environments.
Closing Thoughts
Quantum sensing and AI are unlocking next-gen navigation — resilient, precise, and independent of GPS. At SandboxAQ, we’re pushing these boundaries further, with exciting applications ahead.
