Drones

Drone plays dodgeball to demo fast new obstacle detection system

Drone plays dodgeball to demo ...
Researchers at the University of Zurich have developed a new system that lets drones dodge high-speed obstacles
Researchers at the University of Zurich have developed a new system that lets drones dodge high-speed obstacles
View 2 Images
Researchers at the University of Zurich have developed a new system that lets drones dodge high-speed obstacles
1/2
Researchers at the University of Zurich have developed a new system that lets drones dodge high-speed obstacles
An event camera image, clearly showing the incoming ball (and the person throwing it)
2/2
An event camera image, clearly showing the incoming ball (and the person throwing it)

Obstacle avoidance is a crucial piece of technology for drones, but commercially-available systems just aren’t fast enough for some situations. Now, engineers at the University of Zurich have developed a new system that gives drones such fast reflexes that they can play – and win at – dodgeball.

According to the researchers, most current obstacle avoidance systems take about 20 to 40 milliseconds to process changes in their surroundings. That’s fine for a drone gently approaching a building and finding its way inside, but it’s no match for fast-moving obstacles like birds or other drones. That makes navigation a problem in certain situations, like when there are a lot of drones together or in dynamic environments like disaster zones, or when a drone just needs to move fast.

So for the new study, the researchers kitted out a quadcopter drone with cameras specially designed to detect fast movement, as well as new algorithms that made them even faster. This cut the reaction time down to just 3.5 milliseconds.

The cameras in question are known as event cameras. Rather than analyzing every pixel in a scene to determine if something has moved, in event cameras individual pixels respond only when they detect a change in light intensity. That means pixels without movement stay “silent,” reducing the processing load and speeding up reaction times.

An event camera image, clearly showing the incoming ball (and the person throwing it)
An event camera image, clearly showing the incoming ball (and the person throwing it)

Even so, the Zurich team found that existing event cameras weren’t tuned for use with drones. To adapt, they developed their own algorithms that not only keep watch for all pixel “events” in their sight, but also correct for the drone’s own movement in real time.

The first round of tests involved just the cameras, as the team threw various objects at the cameras to test how well the algorithm could detect them. Depending on the size of the object and how far away it was thrown from, the system was between 81 and 97 percent accurate.

Next they put these cameras onto a drone, and repeated the test. The drone successfully dodged incoming objects more than 90 percent of the time, including a ball thrown from 3 m (9.8 ft) away that was traveling at 10 m (32.8 ft) per second. One camera was enough when the drone was programmed to know the size of the object, but for times when it didn’t know, two cameras were used to let it measure up the incoming obstacle and react appropriately.

Not only does this help the drone dodge flying, falling or thrown objects, but it could help drones safely move faster. The team says it could translate to an increase of up to 10 times faster.

“Our ultimate goal is to make one day autonomous drones navigate as good as human drone pilots,” says Davide Falanga, primary author of the study. “Currently, in all search and rescue applications where drones are involved, the human is actually in control. If we could have autonomous drones navigate as reliable as human pilots we would then be able to use them for missions that fall beyond line of sight or beyond the reach of the remote control.”

The research was published in the journal Science Robotics. The drone can be seen dominating at dodgeball in the video below.

Dynamic Obstacle Avoidance for Quadrotors with Event Cameras (Science Robotics 2020)

Source: University of Zurich

3 comments
nameless minion
I'm a cynic and a catastrophizer and I wonder if this avoidance capability might be used to create more effective military weapons or weapons for terrorists.
Bob809
nameless minion- Your cynical attitude means that a HK (Hunter Killer, from the Terminator movies for those not in the know) is on it's way to your home, with lessons learned from this baby steps HK. Just what I was thinking. Teach them to avoid ground/air to air missiles, cannon fire etc. :-)
nick101
Hmmm, didn't look like they were trying all that hard to hit the (undoubtedly expensive) drone.