Edge AI: why the intelligence has to live on the platform

There's a simple design question at the heart of every autonomous system: where does the thinking happen? For a lot of consumer technology, the answer is “in the cloud.” For defence autonomy, we think the answer has to be “on the platform.”

The link cannot be assumed

In the field, communications are contested. Links get jammed, degraded or simply drop. A system that depends on a distant server to make decisions stops being useful the moment that connection fails — which is often exactly when it matters most.

Running the intelligence on the device itself removes that dependency. The system keeps sensing, deciding and acting whether or not it can talk to anyone.

Latency and data both argue for the edge

There are two more reasons. First, latency: round-tripping every decision to a server and back is too slow for something moving through the real world. Second, data: what a system sees can stay on the system, rather than being shipped elsewhere.

Connectivity should be a bonus, never a crutch. A capable system earns its keep with the link down.

What that demands of us

Edge AI is harder to build. It means designing models and software that run in real time on small, low-power hardware, within tight limits on size, weight and power. That constraint shapes a lot of our engineering — and we think it's the right constraint to design around.