Communications of the ACM
Drones Learn to Navigate Autonomously By Imitating Cars and Bicycles
Researchers in Switzerland have developed the DroNet algorithm, which enables drones to fly by themselves through the streets of a city and in indoor environments after being trained on traffic rules and examples from cyclists and car drivers.
Credit: Zsolt Vaszary
Researchers at the University of Zurich (UZH) and the National Center of Competence in Research Robotics in Switzerland have developed the DroNet algorithm, which enables drones to fly by themselves through the streets of a city and in indoor environments.
They say the deep-learning algorithm was trained on traffic rules and examples from cyclists and car drivers.
DroNet generates two outputs for each single camera input image, including a steering angle to keep the drone navigating while evading obstacles, and a collision probability so the drone can identify and promptly respond to dangerous situations.
The researchers have demonstrated that their DroNet-equipped drones learned not only to navigate through city streets, but also to do so in completely different environments, where they were never trained.
"With this algorithm we have taken a step forward towards integrating autonomously navigating drones into our everyday life," says UZH professor Davide Scaramuzza.
From University of Zurich
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found