Delivery drone

An autonomous drone system that delivers packages across a city.

When
Dec 2019 to Apr 2021
Tools
Python, PyTorch, TensorFlow, OpenCV
Top left, the drone camera: reading QR codes, then finding the landing pad. Simulation, 2.5× speed.

The problem

This project involved developing an autonomous drone system to deliver packages across a city. A custom AI combined multiple machine learning algorithms to address flight control, navigation, and delivery challenges.

The drone above a grid of parcels; its camera feed shows their QR codes
The camera reads the QR code on each parcel.
The drone carrying a parcel past buildings
Carrying the parcel across the city.
The drone camera feed with green crosshairs locked on a landing marker
The camera locks onto the landing pad.

How I built it

  • Flight control.

    A redesigned cascaded PID control system, with a custom regression model that predicts its gain factors.

  • Navigation.

    A custom deep learning navigation algorithm, built with PyTorch and TensorFlow, plans the route across the city.

  • Delivery.

    A Haar cascade trained in Python reads imagery from the onboard camera to scan parcels and find the landing pad.

Results

Patented

Indian patent “Unmanned Aerial Vehicle (UAV) and Navigation Method Thereof”, IN 202221010717, plus two IJRASET papers.

70%

better flight stability, measured by control precision.

54%

shorter flight time with the deep learning navigation algorithm.

30%

higher delivery accuracy from the onboard camera.