Field Test Update: Direct Hit from 500+m

calendar February 20, 2026

Results that speak for themselves

Our latest round of field testing delivered the clearest validation yet of GPO Edge Autonomy — our AI-powered autohoming module for FPV drones.

In the test captured on video, the system autonomously detected and locked onto an unclassified target from a distance of over 500 meters. It maintained continuous tracking throughout the approach and guided the drone to a direct hit. No pilot intervention during the terminal phase. No cloud processing. No GPS. Just onboard AI running on edge hardware.

This represents a significant precision improvement over our previous test cycles, and it didn't happen by accident.

What we changed

Higher camera resolution

We've upgraded to a higher-resolution camera module. In autonomous terminal guidance, every pixel matters — better resolution means earlier detection, tighter bounding boxes, and more accurate tracking at distance. The reason we couldn't do this before wasn't the camera itself. It was processing speed.

Faster frame processing

Our system needs to process each frame within approximately 33 milliseconds to maintain 30 FPS — the minimum for reliable real-time tracking. With the Raspberry Pi 5 and Hailo-8 AI accelerator as our compute platform, we've optimized the pipeline to handle higher resolution frames within this budget. Faster processing unlocked the resolution upgrade.

Triple video stream architecture

This is the biggest architectural change in this update. We now split the camera output into three independent streams:

  1. Flash storage stream — Raw video is written directly to onboard flash memory. This provides complete, uncompressed mission footage for post-flight analysis, debugging, and model training. No frames are dropped, regardless of what the AI pipeline is doing.

  2. Analog video buffer — A dedicated stream feeds the pilot's analog video display (goggles). This ensures the operator always has real-time visual awareness, even if they're not actively controlling the drone. Manual override remains instant at any point during the mission.

  3. AI detection and tracking stream — The third stream feeds directly into our target detection and tracking pipeline. This is where the Hailo-8 accelerator does its work — running inference on every frame, maintaining the bounding box, handling occlusion recovery, and feeding guidance corrections to the Betaflight flight controller via MSP protocol.

All three streams run in parallel on the same hardware. The total module weight remains approximately 100 grams, and the hardware cost stays under €200 per unit.

Why this matters

FPV drones are effective, affordable, and fast — but they have a well-known vulnerability. In the final seconds of a mission, RF signal degrades. Electronic warfare can sever the link entirely. When the pilot loses control, the mission fails.

GPO Edge Autonomy exists to solve exactly this problem. Our module gives the drone the ability to complete its mission autonomously when signal is lost — or when the pilot simply activates autonomous mode for the terminal phase.

The latest test results confirm that this isn't theoretical. It's working. In the field. On real hardware that can be integrated into existing FPV platforms without redesigning the airframe.

What's next

We're continuing to iterate on detection accuracy, tracking robustness, and guidance precision. Every field test generates data that feeds back into our pipeline — better training data, refined algorithms, tighter integration.

If you're a drone manufacturer, defense integrator, or investor interested in scalable autonomous systems for FPV platforms, we'd like to hear from you.

Get in touch:

Share

Back to blog