Drones are agile things, but they are not known for their quick reactions. If you want to knock one from the sky, a well thrown ball or even a spear should do the trick. However, not for much longer, as researchers at the University of Zurich have created a drone that can autonomously dodge the objects thrown at it, even at close range.
You can see the quadcopter showing these skills in the video above (although nobody tried it with a key). Well, some of those releases are quite easy, but the drone continues to react completely autonomously. And although we have seen quadrocopters that can handle static objects like trees, avoiding the moving elements in the air is much more complicated.
"We really wanted to expand the limits and see what these robots are capable of," researches Davide Falanga on customs at the University of Zurich The edge.
Giving drones and auto-dodge would be useful for many use cases. It would make the drones safer, causing them to dodge flying birds or nearby humans. It would also be useful for military and police deployments. If you have a drone monitoring a protest, for example, being able to dodge thrown objects is a very useful skill.
Falanga says that dodging dynamic objects is beyond the knowledge or even the most commercial advanced drones on the market today. He says that Skydio's R1 drone probably has the best autonomous characteristics, but "he still has trouble avoiding moving objects."
As Falanga and his colleagues, Suseong Kim and Davide Scaramuzza, unpack their research work, there are many reasons for this limitation. Technical factors, such as the responsiveness of unmanned aircraft engines and the latency of their sensors, create bottlenecks. What is easy for a human being (well, most of the time) is incredibly complicated for electronics.
However, the unmanned aircraft of the University of Zurich has a great advantage over commercial quadcopters: an advanced sensor known as an event camera. While traditional cameras record a set of frames every second and pass them to the software for processing, the event cameras only send data when the pixels in their field of view change in intensity. This means that they use less data and have lower latency. In other words: a faster response time.
However, event cameras are still rare. They cost thousands of dollars and are not usually seen outside of a research laboratory. Falanga says that they will eventually reach the mainstream, but it will take years of development to reduce them at a reasonable cost. "I absolutely believe that in the long run I think we will see more and more use of these cameras," he says.
Until then, drones will remain vulnerable to anyone with a good eye and a strong launch arm.