Drones are the most agile aircraft around, just look at some of the racing quadcopters, and you get a pretty good idea of just how fast they can take a turn or avoid obstacles. That said, they are terrible at avoiding moving objects, and can be knocked out of the sky with the slightest of hits. That may no longer be the case; however, as researchers from the University of Zurich have developed an autonomous drone that’s capable of dodging thrown objects, even at close range.
The problem drones have with dynamic avoidance is one of latency — the time it takes for onboard sensors to engage the drone’s motors when it encounters a moving object is bottlenecked, usually by the camera system it uses. Traditional cameras take in data using a set number of frames each second, which is then passed on to a computer that uses software to process the data and is then sent back to the drone telling it where to travel.
To get around that issue, the researchers used an advanced sensor known as an event camera, which only transmits data when a change of pixel intensity is encountered due to an object moving at speed. This means the process of identifying and course correction is reduced, thus reducing the bottlenecking.
The researchers designed their drone using an Insightness SEEM1 smart sensor, which grabs those intensified pixels, and hands them off to an Intel Up Board that uses the DBSCAN algorithm to process the information, and sends it a Luminier F4 AIO Flight controller to move the drone out of the path of the object. The researcher’s platform isn’t perfect by any means, it can’t dodge super-fast objects, such as missiles or planes in flight, but it’s an excellent first step in the right direction.