Open-source drone autonomy, from the parts list up

What it does Langostino is a reference build for a real, flying autonomous drone. A Raspberry Pi runs ROS2 Humble on top of an INAV flight controller, with Python nodes for the flight-controller adapter, LiDAR, and the AI layer that does the higher-level decision-making. The repo covers the whole journey: a bill of materials, an assembly guide, 3D-printable mounts, custom INAV firmware builds, and an offline map-tile server for the ground station.

The interesting bit The stack is deliberately split: INAV keeps the reflexes (stabilization, failsafes) while the Pi-side AI layer does the deciding. It also comes out of a Bittensor-adjacent community — the org is literally named swarm-subnet — with a stated roadmap toward decentralized training, which is an unusual pedigree for a drone project.

Key highlights

• Full hardware-to-software reference: BOM, assembly guide, printable STL parts, and custom INAV firmware in inav-custom-firmware/
• ROS2 Humble architecture with Python nodes for the FC adapter, LiDAR, and AI
• Ground-station maps run fully offline via the bundled map proxy
• Deep-dive article series covering hardware, wiring, and software architecture
• MIT-licensed, with a Discord, contributing guide, and good-first-issues

Caveats

• The README never says what the AI layer actually does or how it was trained — architecture details live in external Substack articles
• Gym PyBullet Drones and reinforcement learning are name-dropped, but the training pipeline isn’t explained in the repo
• The Bittensor connection is a badge and a roadmap link, not an explanation

Verdict If you want to understand a complete drone stack and own a screwdriver, this is a rare end-to-end reference. If you wanted a drop-in autopilot or a sim-only weekend project, keep browsing.