True Autonomy Requires a Different Navigation Architecture
By Ken Devine
As autonomous systems continue to emerge and become more capable, we are reaching an inflection point in their development. Undeniably, our reliance on them is increasing, whether it’s robots working in our factories and homes, autonomous vehicles driving us around, or drone platforms strengthening our national security. Yet many of the systems we describe as autonomous still depend on two external connections: they need GPS to know where they are and communications to receive information or direction from somewhere else.
Those dependencies become increasingly consequential as autonomous systems move deeper into contested, infrastructure-poor, or GPS-denied environments. In aerospace, for example, the International Air Transport Association (IATA) recorded more than 580,000 instances of GPS signal loss between August 2021 and June 2024. In defense, the Center for Strategic and International Studies (CSIS) identified communications as a central vulnerability for unmanned systems. It is increasingly clear that autonomy depends on software capable of navigating, planning and executing at the edge when communications and centralized command and control are unavailable.
For OEMs, the strategic question is therefore how to design systems that continue to understand their position and environment after those external dependencies degrade. That requires moving toward navigation architectures built around multiple independent sources, including GPS, inertial, visual, radar, celestial, magnetic, and other modalities.
AI that’s used to make real-time decisions must factor in navigation availability and accuracy as part of mission planning, simulating and evaluating potential routes partly based on how well it believes the vehicle can maintain its navigation and positioning integrity. This relationship between intelligence and navigation becomes critical in GPS-denied environments. When GPS becomes unreliable, autonomous systems need independent signals from multiple sources to maintain position and a software architecture capable of managing these inputs so on-board AI can make the most informed decisions.
Magnetic navigation is one of those sources. MagNav uses quantum magnetometers and AI to sense Earth's crustal magnetic field and produce a clean signal that autonomous systems can use as an unspoofable source of navigational truth. SandboxAQ’s AQNav software compares those signals against known magnetic anomaly maps to deliver positional clarity without external hardware or transmissions. By interfacing with existing inertial, visual, and other navigational systems, that information can help an autonomous system maintain an independent understanding of its location when GPS becomes unreliable.
We have flight-tested AQNav extensively for nearly 3.5 years, with more than 480 hours flown over tens of thousands of miles of land, water, and varying terrain, on board nine aircraft types ranging from drones and prop planes to business jets and large military transports. The tests reinforced our view that MagNav, as part of a broader navigation architecture, will play a significant role in enabling the future of autonomous flight, and that the software orchestration layer is equally important, if not more so, as the types of sensors used.
Autonomy Lessons Learned on the Road
The aerospace industry is not the only sector to arrive at the same crossroads – an incredible amount of groundwork and precedent exists from autonomous vehicle (AV) testing. Waymo recently passed 200 million miles of fully autonomous operation and concluded that multimodal sensing is indispensable to autonomy at scale. Its vehicles combine cameras, lidar, and radar because each provides different information and has different strengths and failure characteristics. This is a similar approach to aviation navigation, where an integrated, system-of-systems approach ensures accuracy, redundancy, and resilience.
Similar to MagNav, Waymo also uses high definition maps as a prior, allowing onboard compute to focus more resources on responding to real-time changes in its environment – e.g., a car stopping short or an animal darting into the street – rather than continuously recalculating its position along its relatively static route. Waymo's experience with autonomous driving has demonstrated the importance of closed-loop simulation for evaluating interactions and scenarios that cannot practically be reproduced at sufficient scale in the physical world.
Aerospace autonomy needs the same architectural principle, but adapted for environments where infrastructure may be unavailable and interference may be intentional. Unlike slow-moving cars with a limited operational radius, autonomous aircraft travel much farther and faster and therefore need real-time navigational signals to ensure mission success, especially in GPS-denied environments. But given the smaller form-factor of unmanned aerial systems, it is critical that these enhanced navigation capabilities do not add weight or further tax already constrained compute resources needed for guidance, optics, and other critical functions.
Building Navigation into the Autonomous Stack
The practical opportunity is to make resilient navigation part of the software architecture of autonomy. AQNav is designed to provide magnetic positioning information that can be integrated with an existing navigation stack, complementing GPS, inertial, visual and other available modalities rather than requiring an OEM to redesign the platform around a new navigation system.
Over time, I expect evolving autonomous platforms will continuously evaluate which navigation sources are available, how trustworthy each source is, and how their planned trajectory will affect future positioning uncertainty. The combination of independent physical sensing, edge software, and intelligent planning will be fundamental for developing truly sovereign autonomous systems, but will require better sensors, better maps, and significantly more sophisticated integrations. The result, however, will be autonomous systems that can continue operating even when individual dependencies disappear.
As we ask machines to make increasingly sophisticated decisions in the physical world, maintaining an independent and resilient understanding of their position is critical for the success and safety of the mission – whatever it may be. Advanced AI and increasing compute performance means that these systems will continue to become smarter and more capable, but at the end of the day they still need an accurate source of positioning information to make reliable operational decisions on the behalf of human operators. Earth's magnetic field gives us another persistent physical reference for doing that, and combining it intelligently with other modalities can help OEMs build autonomous systems capable of operating well beyond the GPS tether.
