Commercial drones are pretty smart. They can make their way from A to B without your help, as long as you tell them where B is first. But if there’s anything but clear sky along the way, then you may either have to step in to help or let the thing crash into a surprise tree or building. Unless you’re running Andrew Barry’s smart drone, that is.
Barry, a PhD student at MIT’s Computer Science and Artificial Intelligence Lab, came up with a smarter, faster way for his drone to scan the world around it. This lets the drone fly at high speed and still react fast enough to avoid obstacles.
The normal way for a flying vehicle to assess its surroundings is to use two cameras to make a 3-D image. It then searches for through the image to see what’s there, first checking a meter away, then two meters and so on, until it has a model of the world ahead. This is dead slow. CSAIL says that drones navigating like this can manage only 5-6 mph.
Barry’s drone does the same, but it only pays attention to what’s 10 meters in front. Nothing closer, nothing further. It’s like driving by only looking at what’s in front of the hood of your car. This method produces decisions that are 20 times faster than before.
“You don’t have to know about anything that’s closer or further than that,” Barry says. “As you fly, you push that 10-meter horizon forward, and, as long as your first 10 meters are clear, you can build a full map of the world around you.”
This would be suicide on the roads, but the sky is much emptier, so Barry’s method works just fine. It computes the video at 120 frames per second, pulling out depth data at a speedy 8.3 milliseconds per frame. This allows the drone to hurtle along at 30 mph, veering around obstacles with no human help. It’s like a self-driving version of the Speeder Bikes from the Star Wars: Return of the Jedi movie.
Right now, this is just a one-off, $1,700 project hand-built from off-the-shelf parts. But in the future, this is the kind of tech that could make parcel delivery drones practical.