[robotics-worldwide] [software] Release of the Source Code of "DroNet: Learning to Fly by Driving"
We are excited to release open-source our deep-learning project for
autonomous drone navigation in city streets, called "DroNet".
Due to the danger that flying a drone can cause in urban environments,
collecting training data results impossible. For this reason, DroNet
learns how to fly by imitating the behavior of manned vehicles (cars and
bicycles), which are already integrated in urban environments. Designed
as a fast 8-layers residual network, DroNet produces two outputs for
each single input image: a steering angle (to keep the drone navigating
while avoiding obstacles) and a collision probability (to let the drone
recognize dangerous situations and promptly react to them).
The DroNet project is well documented and easy to use. The detailed
README.md explains how to run the algorithms, train it, and test it. In
addition, information on how to download and pre-process the datasets is
This project will also be presented at this year's IEEE International
Conference on Robotics and Automation, in Brisbane
Antonio Loquercio, Ana Isabel Maqueda, Carlos Roberto del Blanco, Davide
DroNet: Learning to Fly by Driving
IEEE Robotics and Automation Letters, January 22, 2018
The authors: Antonio Loquercio, Ana Maqueda, Carlos del Blanco, and