Thrown Rocks Are No Threat to These Drones That Have Mastered Dodgeball

The time when a nasty drone with a precisely thrown rock is knocked out of the air may soon be over. Researchers at the Zurich University upgraded a drone with a special camera that can quickly identify approaching obstacles, which caused craft to avoid them with response times just as fast as 3.5 milliseconds.

Even drones piloted by a skilled pilot can take advantage of an obstacle avoidance system, which allows the craft to watch for obstacles in its flight path and automatically prevent a crash if a pilot sees no danger or responds quickly enough to avoid. Such systems are absolutely critical for drones built to fly autonomously, whether it be exploration in hazardous areas after a natural disaster, or for companies like UPS that deliver and have been products to consumers develop flying drones as an alternative to four-wheel vans. Those big brown UPS trucks are impervious to seagulls bending over of course, but a drone full of packages would easily be whipped through the air such a collision.

The cameras and image processing technologies that autonomous drones currently use to detect obstacles allow for response times inside anywhere from 20 to 40 milliseconds, according to the UZH researchers. That is fast, but if you count the speed of the drone yourself (some can fly over well 150 miles per hour) many obstacles, such as flying bird, another drone or even a static object, 20 milliseconds is not enough to avoid a collision.

To improve response times, the researchers equipped a drone with a so-called event camera. Your typical video camera relies on a sensor that exposes all pixel elements simultaneously before sending them for processing; Objects can only be detected by algorithms after an entire frame – and often multiple frames – has been analyzed. That is what contributes to the delays in response times. For comparison, an event camera has pixels that are individually ‘smart’ and only pass on data to be processed when a change in what they see is detected. If they don’t see anything newThey remain silent, allowing algorithms and processors to focus their efforts on pixels that have detected an approaching obstacle, speeding up the whole process.

The UZH researchers developed their own custom algorithms that can recognize threats detected by the event camera after taking images for a very short time, taking into account the speed and direction of the drone itself. The results are reduced reaction times down to just 3.5 milliseconds, making a drone much more effective at avoiding a fast-moving obstacle.

It is an improvement, but not perfect yet. From their tests, the researchers found that a drone equipped with the new vision system could dodge a large ball three meters away at a speed of about 22 miles per hour more than 90 percent of the time. Knowing the size of the incoming object in advance helped improve results, but in the field, a pair of event cameras on the drone, which generated stereoscopic images, could allow them to calculate the size of a threat themselves. It cannot yet be used to dodge bullets, however autonomous drones can fly 10 thanks to current improvements in response times times faster than they can now, which means you may not have to wait that long for your drone-delivered pizza to arrive.


Leave a Comment