An autonomous drone guidance system built on a specialized AI model, autopilot technology, and powerful onboard computing. Real-time cognitive functions run on the drone itself — detection and interception do not depend on a continuous operator link.
Four subsystems form a single onboard guidance chain.
Detects and tracks targets in thermal video; trained on real battlefield footage.
ArduPilot flight control integrated over MAVLink telemetry and CRSF control links.
Raspberry Pi 5 with AI HAT runs inference and guidance algorithms on the drone in real time.
Digital and analog thermal cameras supply the video feeds for day and night operation. Once a target is detected, the model keeps it tracked and the autopilot steers the drone until the target is reached — automatically.
The onboard chain runs from camera feeds through AI inference to the flight controller, with the operator link used for supervision and launch — not required during terminal guidance.
| Feed | OSD rate |
|---|---|
| Digital thermal | 59–60 FPS |
| Analog thermal | 24–25 FPS |
Readings from demonstration-footage on-screen displays; rendering frame rate of each feed, not a controlled benchmark.
The onboard cycle repeats continuously until the target is reached. The loop repeats every frame; no operator input is required.
Thermal video feeds the onboard pipeline.
The AI model finds the target in frame.
Lock is maintained frame after frame.
MAVLink commands flow to ArduPilot.
The drone arrives at the target.
Division of labor: the AI decides where to fly; ArduPilot handles how to fly there.
Because track and guide run onboard at display frame rate, the loop keeps closing even when the link to the operator is delayed or lost — the drone finishes the approach on its own.
Flight — the drone detects the target and reaches it. Automatically. Tell us your mission profile and we'll walk you through the demonstration footage.
Request a demo