A team from the University of Zurich has trained an artificial intelligence system to fly a drone in a virtual environment full of obstacles according to an article in Forbes. The research by the university’s Robotics and Perception Group under lead researcher Davide Scaramuzza in partnership with Intel included flying a drone around obstacles at 40 kph (25 mph), three times as fast as the previous best piloting software, says the article.
A paper describing the project, Learning high-speed flight in the wild, was published this month in the journal Science Robotics.
“Our approach is a stepping stone toward the development of autonomous systems that can navigate at high speeds through previously unseen environments with only on-board sensing and computation,” the paper concludes.
Scaramuzza’s group previously demonstrated the first autonomous drone able to beat human racing pilots around a course using sophisticated path planning with help from external sensors. The new work takes a somewhat different approach and is self-contained: the machine-learning AI effectively has an apprenticeship with an all-knowing master in the virtual world, and learns to imitate the master’s technique.
For the immediate future, this approach is likely to be applied mainly to small drones, enabling them to carry out aggressive manoeuvres faster than any human pilot. The paper indicates that even faster speeds could be achieved with better sensors and more accurate modelling; it is not clear what the absolute limits might be, add Forbes.
Watch a two-minute video showing how ‘Learning high-speed flight in the wild’ was achieved here.
For more information visit:
www.forbes.com
http://rpg.ifi.uzh.ch/
