First, the team tested the system without a drone by throwing objects of various shapes and sizes in the direction of the camera. The algorithm detected the objects with a success rate of between 81 and 97 percent and took only 3.5 milliseconds to detect them.
They then installed the cameras in a drone and threw objects at it in flight, both indoors and outdoors. In more than 90 percent of cases, the drone was able to evade - even when a ball was approaching it from a distance of only 3 meters at 10 meters per second.
According to Scaramuzza, this shows that event cameras can increase the navigation speed of drones tenfold, which expands their range of applications: "One day, drones will be used for a wide variety of purposes, such as delivering goods, transporting people, taking aerial photographs and, of course, for search and rescue operations. But the ability of robots to reliably detect obstacles approaching them also plays a decisive role in other areas, such as the automotive industry, mining and remote inspection with robots".
In the future, the scientists plan to develop the system...
...with an even more agile quadrocopter. "Our goal is an autonomous drone that navigates just as well as a human drone pilot," says Davide Falanga, PhD student and first author of the study. "If autonomous drones navigate as reliably as human pilots, we can also use them for missions that are out of the line of sight or out of range of the remote control system."