We don’t think we are going out on a limb when we say that autonomous drones are going to be big. But, like the archetypal protagonist at the start of any hero’s quest, they need to learn a few things first.
One such challenge is how to fly quickly while managing to dodge obstacles. The reason this is hard for drones is because their cameras can only process images up to a certain speed since they have to do it frame by frame. Anything faster than 30 mph causes a bit of a headache.
That is where a new research project from the Massachusetts Institute of Technology comes into play. It builds on a technology called the Dynamic Vision Sensor (DVS), invented by researchers in Zurich. The DVS continuously visualizes scenes by looking for changes in brightness. This is similar to the way that the human brain senses visual information and means that it can work at short, microsecond intervals much faster than regular cameras.
The problem is the amount of processing time it requires due to the huge mass of data. By the time an obstacle has been detected and a control outputted, a drone would have already crashed.
MIT’s work represents a step forward. It is an algorithm that is able to isolate very specific changes in brightness, which has the effect of reducing complex scenes to their most essential elements.
“The DVS has had a lot of empirical successes,” lead author Prince Singh, a graduate student in MIT’s Department of Aeronautics and Astronautics, told Digital Trends. “However, there [hasn’t previously been] a concrete algorithm that can process the sensor’s ambiguous data to, for instance, control a dynamic system such as a drone. My work addresses the control of any system that has a linearized representation by making sense of the DVS’s ambiguous data. This work unifies the empirical successes observed, and most importantly, one doesn’t need knowledge of the problem’s geometry, as was the case for works until now.”
Singh is currently presenting the work at the Institute of Electrical and Electronics Engineers American Control Conference in Seattle. Next up, the plan is to publish a paper on the work. Hopefully, it won’t be too long after that before drones — and other autonomous robots — can take advantage of this high-speed breakthrough.
