Watch a Drone Autonomously Dodge Trees at 30 MPH

Hobby drones are getting more independent.

|
Nov 3 2015, 3:34pm

This team from MIT's Computer Science and Artificial Intelligence Laboratory or CSAIL—the lab that brought you soft robots pining for the 90s and devices that can see through walls—built a drone that's able to maneuver through obstacles and quickly calculate potential collisions so that the drone won't run into trees.

Sure, it's not the stuff of actual drone races, where you actually have a person manning the thing and you're watching it zip through at 100 miles per hour. High speed is relative. Unmanned, 30 miles per hour is actually really fast when you have to navigate through boughs and branches.

How collision detection works is pretty simple. If a drone were trying to calculate the distance between it and a solid object, it would take a photo. Doing this continuously for all the objects around it would be processor-intensive, however. The team said it would slow the drone down to around 6 miles per hour.

There's an easier way to do it. Instead of wasting computing on objects too far away, the drone can just continually process object distances within a 10 meter radius in front of it.

Indeed, while drone races are tremendously exciting to watch, it's not like NASCAR, where it's mostly a show of speed and maneuvering around an elliptical track. There are far more challenges since they usually take place in the woods, or some other obstacle-laden course where handling, not necessarily speed, is most important. But you have to admit: seeing how a drone recognizes and maneuvers through obstacles still makes for some Star Wars-level entertainment.

Stories