MIT Scientists Create a Tiny Drone That Dodges Obstacles at High Speeds

Monday, 12 February 2018 - 7:02PM
Monday, 12 February 2018 - 7:02PM
MIT Scientists Create a Tiny Drone That Dodges Obstacles at High Speeds
< >
Wikimedia Commons

As anyone who's ever driven a motorbike or played a Sonic the Hedgehog game can tell you, sometimes it's incredibly difficult to dodge objects while you're moving at high speeds.

This is especially true of automated drones, whose navigation systems typically rely on plotting out a prescribed route before even beginning a journey, and trying to constantly keep track of where they're going while in the air.

This is fine when there's nothing in the way, but a drone has to be capable of spotting and reacting to potential hazards in the air, and that takes brain power. Constantly analyzing data from the drone's surroundings requires a lot of processor power, and as a result, there's a limit to how fast a drone can travel while still staying fully aware of its surroundings. Move too fast, and drones, particularly small ones, struggle to keep focused on where they're going.

Now, MIT scientists at the Computer Science and Artificial Intelligence Laboratory (CSAIL) believe they may have found an alternative strategy for drone navigation that allows small, zippy flying robots to get around more safely even while traveling at faster speeds. Apparently, the solution is actually to allow for a little less precision when mapping out flight routes.

The new system, named NanoMap, involves the bots making flight decisions on the fly, instead of coming up with a prescribed journey plan and then adapting it constantly as the journey progresses.

Traditional SLAM (Simultaneous Localization And Mapping) technology involves taking measurements of everything surrounding a drone at all times, and working to constantly fuse all of this data together into a solid map that means that the drone is completely aware of its environment in great detail.

The NanoMap system instead takes this data, and uses it to throw together a quick snapshot of the environment. The drone has a fuzzy idea of everything around it, and so long as it's not going to collide with anything, that's all it needs. Thus, a tiny robot brain can react quickly to a potential hazard when necessary, but there's no need to constantly keep an eye out to see if someone on the other side of the room is standing around.

According to Peter Florence, who's been working on the project:

Opening quote
"Because we're not taking hundreds of measurements and fusing them together, it's really fast. And when we want to plan our way around the world, we just search back through the views we already have. We have drones that can fly around now, but there's nothing that's as good as a hawk, for example, that can just blaze through a forest at top speeds. When you're going that fast, you just can't expect to have perfect geometric measurements of the world. You need something else."
Closing quote

This technology isn't entirely perfect - the drones don't get a particularly accurate view of their surroundings because they're constantly just keeping vaguely abreast of any potential hazards around that might distract them. This means that the technology isn't suited for drones that need to be more meticulous in their travel itinerary.

That said, it's understandable that, just like people, drones can't quite manage being both fast and accurate when doing a job.

Self-driving cars, for example, are better off with more traditional equipment. In cars, there's space to include enough processing power to allow for accurate measurements, and the machines rarely move so fast that they're going to tax their CPUs by trying to fully compute their surroundings.

For teeny tiny drones that need to move at speed, this system is a perfect way of saving computer memory while also maintaining a fast travel speed. These machines aren't as good at directions, but they'll get where they're going sooner rather than later, and sometimes, that's more important.

Science News