Human Tracking Field Test — Autonomous Target Engagement in Real Conditions
February 2, 2026
The Reality of FPV Drone Pursuit
"You are walking in a field and suddenly hear a strange noise from the sky. You look up but see nothing. The noise grows louder. A small, almost invisible point quickly increases in size and approaches you. You try to run — maybe gain a few meters — but the insane noise is the last thing you hear before the explosion."
This account comes from someone who survived an FPV drone attack. It captures the psychological reality of being targeted by an autonomous aerial system.
Our Field Test
We recently conducted a critical milestone test: validating our Edge Autonomy module against a live human target. Our volunteer experienced sensations remarkably similar to the account above — with one important difference: no explosion.
What We Tested
Scenario: Open field, volunteer acting as target, drone operating in autonomous mode with pilot ready to override.
Process:
- Drone detects human as valid target
- System acquires and locks tracking
- Drone approaches while maintaining continuous track
- Pilot switches from autonomous to manual control before impact
Results
Target Acquisition: The detection model successfully identified the human target and acquired lock within acceptable latency, despite unpredictable movement patterns.
Tracking Performance: As the volunteer attempted evasive maneuvers — changing direction and speed — the tracking module maintained continuous lock and adjusted trajectory predictions in real-time.
Approach Phase: The drone closed distance while maintaining stable tracking. The pilot override system worked flawlessly, allowing seamless transition to manual control.
Cold Weather Operations: At -12°C, our system performed without issues:
- Raspberry Pi 5 and Hailo accelerator required no additional cooling
- Camera optics remained clear with no frost accumulation
- All subsystems operated within normal parameters
Why Human Targets Matter
Static objects and slow-moving vehicles are relatively easy tracking challenges. Humans present the hardest test case:
- Unpredictable movement patterns
- Rapid direction changes
- Variable speed
- Posture changes affecting detection
Successfully tracking a human in harsh conditions validates that our system can handle virtually any target type.
Technical Architecture
Our Edge Autonomy module processes video at 30+ FPS using:
- Raspberry Pi 5 — main processing unit
- Hailo-8 AI accelerator — 26 TOPS neural network inference
- Custom Betaflight integration — seamless flight controller communication
- HC4051N data stream switcher — instant transition between autonomous and manual modes
Total weight: under 100 grams. Hardware cost: under €200. Zero cloud dependency.
Safety and Control
The pilot override system is fundamental to our design philosophy. Autonomy assists the operator but never removes their ultimate authority. During this test, the experienced pilot demonstrated the system's fail-safe by smoothly transitioning control before any collision occurred.
Applications
This capability has direct applications in:
- Defense — terminal guidance in contested RF environments
- Security — autonomous patrol and threat tracking
- Search and rescue — locating and following survivors
- Wildlife research — tracking animals without constant manual piloting
Next Steps
We continue field testing across varied conditions:
- Low-light and high-contrast scenarios
- Multiple target types (vehicles, groups)
- Different environmental factors (precipitation, dust, terrain variation)
Contact Us
If you're developing FPV platforms or need reliable autonomous tracking capabilities, we'd like to discuss integration and partnership opportunities.
Test conducted: Winter 2025/2026
Temperature: -12°C