Added Agile Properties The University of Zurich allows drones to dodge dynamic impacts on objects such as basketballs

For drones, avoiding obstacles is a vital technology. But at present, commercial drone systems are largely stuck in avoiding static obstacles such as buildings, and dynamic obstacles such as birds are not caught. But the engineering team at the University of Zurich has recently developed a new system that can quickly evade the blue ball thrown at it.

Added Agile Properties The University of Zurich allows drones to dodge dynamic impacts on objects such as basketballs

The current barrier avoidance system for drones typically takes 20 to 40 milliseconds to deal with changes in their surroundings, the researchers said, more than enough to avoid buildings or other static obstacles, but there is nothing they can do about birds or other drones.

To this end, the team modified existing drones and reduced processing time to 3.5 seconds with an enhanced algorithm. The key to the transformation is the Event Camera, which does not analyze each pixel in the scene when an object is detected moving, but only for changes in light intensity. This means that pixels that are not moved remain “silent” in a state, reducing processing load and speeding up reaction times.

Based on the active camera, the team also made algorithmic adjustments to the drone to not only observe the “activity” of all pixels, but also correct the movement of the drone itself in real time. In the first round of tests, only for active cameras, team members threw objects at the camera and adjusted the algorithm to the size and distance of the object, eventually increasing the accuracy to between 81% and 97%.

Added Agile Properties The University of Zurich allows drones to dodge dynamic impacts on objects such as basketballs

The team then equipped the camera with the drone and repeated the test. In the end, the drone managed to evade 90 percent of the motion obstacles, including a ball thrown from 3 m (9.8 ft), which hit the drone at 10 m (32.8 ft) per second.

When the drone’s algorithm knows the size of an object, an active camera is fine. However, without knowing the size, two cameras are required to measure the incoming obstacle and react appropriately. Davide Falanga, lead author of the study, said: “Our ultimate goal is that one day drones will be as flexible as human-operated drone pilots. “