Skip to content

AI Drone – Automated Delivery

Welcome to the documentation of our project on automated object delivery using an FPV drone with AI assistance.

About the project

In this project we develop, implement, and evaluate a practical drone-AI application. A flight-ready FPV drone is extended with the following capabilities:

  • Autopilot functionality with ArduPilot
  • Position Hold and Altitude Hold based on LiDAR and Optical Flow
  • Delivery mechanism for the automated release of payloads
  • AI-assisted object detection via the Raspberry Pi AI Camera Module

Project status (2026-08-24)

The full autonomous delivery pipeline runs end-to-end in the ArduCopter 4.6.3 simulator; the MTF-01P sensor and the Raspberry Pi companion link are verified on the real aircraft. On 2026-08-21 a manual test flight ended in a crash whose cause is now fully understood and fixed in software and parameters. The flight controller hardware is recovered: the "destroyed" barometer was only unreachable behind a hung I2C bus — bent GPS-connector pins were straightened and, on 2026-08-24, stock 4.6.3 detected the barometer again. Real-flight milestones are unblocked, pending a parameter reload after the firmware flash. The crash & barometer recovery story and the full incident analysis are required reading and among the most instructive artefacts of this project (see also all documented Problems).

Quick start

Project structure

The documentation follows the task structure defined in the project brief:

Task Topic Documentation
1 Get familiar with hardware and software Hardware, Software
2 Extension for automated delivery Cross-cutting
3 Autopilot integration Autopilot
4 Position & Altitude Hold Sensors
5 Delivery mechanism Delivery System
5a Finding the target (AI detector) Landing Pad Detection
6 Frame extension Frame Extension
7 Documentation & presentation This site + poster + live demo

Team

This documentation and project was done by Lars Reul, Yichi Zhang, Amir Ebrahimi und Daniele Amore.