System · 01

Perception Core

The onboard neural network that detects, classifies and tracks targets in real time at 30 FPS — entirely on the drone, with no datalink and no cloud.

1Camera Input
→
2AI Detection
→
3Target Lock
→
4Operator Authorisation
→
5Terminal Guidance

Step 4 is a hard gate: the loop cannot advance to terminal guidance without an explicit operator decision.

The perception core is the part of AKINAK-AG that sees. An onboard neural network detects and classifies persons, vehicles, drones and wildlife in real time at 30 FPS, then holds lock through manoeuvres. Predictive algorithms handle occlusions and re-acquisition, so a target that briefly leaves the frame is picked up again rather than lost.

Fully onboard, millisecond response

The real-time processing pipeline runs entirely on the drone — millisecond response times, zero radio emissions, no cloud dependency. Nothing about detection or tracking needs a command link or a ground station, which is what lets the module keep working when the link is jammed or deliberately switched off.

Where it sits in the loop

Detection and tracking run continuously. They feed the terminal guidance stage, which can only start after explicit operator authorisation. The same perception core drives intruder detection in critical-infrastructure protection.

Drone detection means identifying an intruding UAS during perimeter patrol — a detection-and-alerting function, not interception.

Perception Core: questions

What does it detect?

The onboard neural network detects and classifies persons, vehicles, drones and wildlife in real time, and holds lock through manoeuvres, occlusions and re-acquisition. Drone detection here means identifying an intruding UAS during perimeter patrol — a detection-and-alerting function, not interception.

What compute does it run on?

Inference runs onboard at 30 FPS on a compact AI accelerator. The exact host and accelerator are not published — they are shared under NDA or with the evaluation kit.

Does detection depend on a ground station or the cloud?

No. The full processing pipeline runs on the drone, with millisecond response times, zero radio emissions and no cloud dependency.