Overview
Atmaca is my LLM-powered autonomous-flight project where you control a UAV with plain English/Turkish commands. Everyday instructions like "take off and go north" or "climb 10 meters" are interpreted by a language model (Gemma 3) acting as the drone's brain and turned into flight.
How It Works
The system splits into a "brain" and a "body". The brain is a Gemma 3 model running on Ollama; it converts the user's text command into a structured JSON decision (e.g. "takeoff 10 m"). The body runs on Python + DroneKit; it receives the JSON over HTTP, translates it into MAVLink commands for the UAV (SITL or real) and feeds telemetry back.
Outputs & Status
The result is a working demo and an autonomous-flight test video (v3.5). I published it open-source (MIT) on GitHub — a pioneering R&D effort showing end-to-end natural-language UAV control.
Natural Language → Flight Flow
Command
Plain language: "take off, go north"
LLM Brain
Gemma 3 turns it into a JSON decision
Body
Python/DroneKit → MAVLink
UAV
Autonomous execution + telemetry
3D Model
Tech Stack
Videos
Thinking about an AI/LLM-powered smart UAV?
I can adapt an LLM-integrated autonomous control layer that turns natural-language commands into flight, for your scenario.
I'll build the same for you
Message the assistant — talk through whether I can build something like this, the approach and a ballpark, and get to a booking in minutes.