In the past couple of years drone racing has taken off. Pilots that were flying for fun on makeshift tracks in their spare time are suddenly travelling the world on professional contracts. Racing organizations like DRL are signing multimillion-dollar sponsorship deals with some of the biggest names in business.
The sport has captured imaginations because of the way it combines emerging technologies with time-tested entertainment: drones, RC hobbyists, racing, speed, virtual reality… all play a part in making drone racing a spectacle that people want to get involved with.
The sport of the future
Drone racing’s position as the ‘sport of the future’ also means it’s a great partner for other emerging technologies. One of those is the field of computer vision.
This week scientists in Switzerland have published research detailing their exploits in building an autonomous racing drone that could one day compete with the best human pilots.
Building a drone that can navigate around obstacles and make trajectory decisions based on its environment is no mean feat.
It’s made all the more difficult when you think about the restrictions most racing drones are designed with. They need to be small and light, not weighed down by multiple cameras and sensors.
And so the first challenge was devising a system that wouldn’t stop the drone getting off the ground. The second was developing a computer vision system capable of responding to its environment within a fraction of a second and making adjustments during flight.
Human drone racing pilots rely on incredible coordination and no small amount of muscle memory to handle tracks at speeds upwards of 100mph.
Researchers from the Robotics and Perception Group at the University of Zurich’s Department of Informatics and ETH Zurich’s Department of Neuroinformatics have developed a system that attempts to bring together that level of perception and real-time responsiveness, despite the constraints of working with a small racing drone.
The system relies not on sonar, lidar, radar or any type of clever sensor. Instead, it combines a convolutional neural network with a state-of-the-art path-planning and control system.
In English, that means that the drone uses cameras to see the world and react to it. The system is trained in pretty comical fashion: the researchers have to hold the drones and pretend to fly them through the course. Throughout the process, the drone collects thousands of images and the data is used to train the neural network on how to properly fly through the obstacle course.
Once it’s the drone’s turn to fly unaided, it is able to handle the track with relative ease. As University of Zurich roboticist Antonio Loquercio explains, “The drone receives an image from the camera and the neural network outputs, Hey drone, now you have to go two meters to the left.
So what does all of this mean? Well, it’s best if you watch the video below to see what the researchers have achieved. Put simply, this is a drone racing system that can fly unaided, learn how to fly around tracks and navigate obstacles and adapt on the fly to moving gates.
Should Drone Racing Pilots Be Worried?
No doubt this technology is incredible, and it will probably find its use among a range of commercial applications. But actual autonomous drone racing seems a step too far. Having said that, building this technology into tiny racing drones is an ideal way to highlight the progress of the project.
As it stands, nobody wants to pay to watch a bunch of robots fly autonomously around a track without making a mistake. The whole reason we enjoy sport is for the human drama, the competitiveness and the emotional rollercoaster that comes with it. None of those things would easily apply to a bunch of AI-powered racers.
For now, professional drone racing pilots need not fear. Your jobs are safe.