Navigating without the sky

Satellite navigation is so reliable in daily life that it has become invisible. It is also, from a defence perspective, remarkably fragile — and it is usually the first thing an adversary removes.

The physics explain why. A satellite positioning signal is transmitted from roughly twenty thousand kilometres away by a modest transmitter. By the time it arrives it is extraordinarily weak — below the background noise. Overwhelming it locally takes very little power.

Jamming is the lesser problem

Jamming denies the signal, and a well-designed system notices immediately: the fix is gone, and it can fall back to other methods. Unpleasant, but honest.

Spoofing is worse. A spoofer transmits a plausible but false signal, and the receiver reports a confident, precise, wrong position. The system does not know anything is amiss. Every decision downstream is then made on a false premise, and the more the platform trusts its navigation, the further astray it goes.

A system that knows it is lost is manageable. A system confidently reporting the wrong position is dangerous.

What replaces it

There is no single substitute. What works is combining several independent methods with different failure modes, so that the loss of any one degrades the estimate rather than destroying it.

• Visual odometry. Estimating motion by tracking how the scene shifts across the camera between frames. Accurate over short intervals, and dependent on there being something to look at.
• Mapping and relocalisation. Building a map of the surroundings while determining position within it, and recognising places seen before to correct accumulated error.
• Inertial dead reckoning. Carrying the estimate forward using motion sensors. Excellent briefly, drifting steadily, and entirely independent of the outside world.
• Fused estimation. Weighting each source by how much it can currently be trusted — and, critically, tracking how confident the combined estimate deserves to be.

Managing error, not eliminating it

Navigation without external reference is fundamentally about error management. Every dead-reckoned estimate drifts; the question is how quickly, and how often you can correct it against something recognisable. This is why featureless environments are so difficult — open water, desert, snow, darkness and smoke all deprive vision of the landmarks it needs.

Repetitive environments create the opposite failure. A corridor that looks identical to four other corridors invites confident relocalisation to the wrong one, which produces the same dangerous condition as spoofing: precision without accuracy.

The feature that matters most

If we had to name a single property that distinguishes a serious navigation stack from a demonstration, it would not be accuracy. It would be knowing how accurate it currently is.

A system that reports 'my position estimate has degraded and I no longer trust it' has given the operator something actionable. A system that keeps reporting a confident figure as its error grows has given them a trap. We design for the former, and we treat honest uncertainty as a first-class output rather than an afterthought.

Why we treat denial as normal

A great many autonomous systems are built assuming satellite navigation is available, with a degraded fallback bolted on afterwards. We invert that: the system is designed to work without it, and treats a satellite fix as a useful corrective when it happens to be available and trustworthy.

This is more expensive to build. It is also the only version that survives contact with an environment where someone is actively trying to make your system lost.