Hi everyone,
I’ve been building a cloud-based Ground Control Station called Fukushima that runs entirely in the browser. It supports both ArduPilot and PX4 via MAVLink and includes onboard AI detection models for real-time object recognition during flight.
What it does:
- Full MAVLink control: ARM, Takeoff, Land, RTL, flight mode switching, PID parameter read/write
- Mission planning: Waypoint, Survey Grid, and Orbit modes
- Offline maps for 20+ cities worldwide (works in RF-denied environments)
- Fleet management: monitor multiple aircraft on a single dashboard
- Flight data recording with CSV export and map replay
- AI detection models running on companion computers (Raspberry Pi), exported as ONNX for edge deployment
AI Models available (plan-dependent):
- Basic object detection (people, vehicles)
- Collision avoidance
- Weapon detection
- Fire / smoke detection
- License plate recognition
- Aircraft detection
- Flag recognition (31 nations)
- Vehicle shape classification
- Camouflage pattern detection
- Boat detection (coming soon)
- Crowd detection (coming soon)
Tech stack:
- Backend: Python / Flask
- Frontend: HTML, CSS, JavaScript, Leaflet.js
- AI: ONNX Runtime on edge devices
- Communication: MAVLink over WebSocket
- No installation required — everything runs in the browser
Screenshots:
[Dashboard]
(screenshot-dashboard.png)
[Landing Page]
(screenshot-landing.png)
[Feature Overview]
(screenshot-features.png)
Links:
- Website: https://fukushima-gk.com
We offer plans from hobby ($10/mo) to defense ($5,000/mo). All plans include a free trial.
Would love to hear feedback from the ArduPilot community. Happy to answer any questions about the architecture or AI integration.
Thanks!