One of the many ways researchers are working to improve the performance of autonomous drones is by having them compete against humans in the business of racing them. It takes some skill to pilot a tiny quadcopter through confined spaces at blistering speeds, and by developing algorithms that outperform these feats, we may usher in a generation of drones with incredible capabilities. Through a novel algorithm that can plot a flight path with great efficiency, scientists at the University of Zurich are now claiming to have done just that.
In the space of a few years, drone racing went from an underground hobby for aircraft enthusiasts to a professional sport, and among the bodies facilitating these high levels of competition is the Drone Racing League. For the 2019 season, organizers included for the first time a dedicated competition for developers of autonomous drones, who could pit their self-piloted aircraft against one another for significant cash prizes.
A drone developed at Delft University claimed first prize in the inaugural event, proving 12 percent faster than the next fastest autonomous drone in the field. But in a special bonus round, it was no match for professional human pilot Gabriel "Gab707" Kocher, trailing its counterpart by five seconds.
Now, in the space of less than two years, the University of Zurich researchers claim to have closed this gap, albeit in a very different setting and with a few caveats. They say that previous algorithms for autonomous drones would rely on simplifications of either the quadcopter system or the flight path itself. Their novel algorithm improves on these by more accurately considering the limits of the drone, and calculating "time-optimal trajectories" that accelerate and decelerate at just the right rates through different segments of a course.
“The novelty of the algorithm is that it is the first to generate time-optimal trajectories that fully consider the drones’ limitations,” says study author Davide Scaramuzza.
The team proved the worth of their new algorithm by using it to navigate a quadcopter through a race circuit. External cameras were used to capture the motion of the drone and give it real-time information on its location, which then informs the algorithm going forward. Control of the quadcopter was then turned over to two professional drone racing pilots who were given time to practice on the course beforehand.
All the laps completed by the algorithm were faster than the human pilots, and the performance was more consistent, as once it determined the optimal path through the course it was able to reliably repeat it. The scientists say this is the first time an autonomous quadcopter has outperformed human pilots in a drone race, but it will be a while before those competing in the Drone Racing League find themselves losing to a computer.
This is because not only does the algorithm rely on the external cameras to gauge its position in the course, it takes about an hour of computing for it to calculate the time-optimal trajectory. These are two factors the researchers are looking to address before the algorithm finds its way into commercial use: reducing the computational demands of the algorithm and enabling it to rely on onboard cameras instead.
But the algorithm still represents a significant step forward for the technology, and could prove useful for drones built for all sorts of applications. Whether they're completing search and rescue operations, inspecting buildings or delivering cargo, the goal is to have them doing so with great speed, efficiency and reliability.
The research was published in the journal Science Advances, while you can see the drones in action in the video below.
Source: University of Zurich
