Autonomous Airplane Dodges Obtacles

//

There are drones that can hunt hurricanes, land on aircraft carriers and combine helicopter and airplane flight. And now they can even dodge obstacles indoors, without access to global positioning system satellite signals.

It's a massive computing challenge, but an important one for designing autonomous fliers that can operate when they can't "hear" a remote operator or satellite. GPS signals can occasionally be blocked by bad weather or in combat situations by jamming. A plane that can navigate with no outside help at all has an advantage over remotely controlled drones.

PHOTOS: Five Scariest Bioweapons

MIT's Robust Robotics Group has built a plane that can fly around a room — specifically, weaving around pillars in a parking garage, where no GPS signal can penetrate. They were able to write an algorithm that allowed the airplane to determine its own acceleration, speed, orientation and position.

While there are some air vehicles that can navigate around obstacles and even coordinate with each other in the air, they tend to be helicopters. Helicopters can control their flight more precisely at low speeds, but they can't stay up long because they use a lot of power just to hover. A fixed-wing aircraft can stay up longer on a given amount of fuel (or charge), but it can't manuever as easily as a helicopter.

To deal with that, the MIT researchers gave the plane relatively short and stubby wings that allow it greater maneuverability and slower airspeeds.

The MIT team cheated a bit, giving the aircraft a "map" of it's surroundings that they knew was accurate. But that isn't so different from having map data uploaded to any robot's memory. Even with that, the plane still had to figure out where it was and where it was pointed. That required an on-board laser rangefinder, accelerometers and gyroscopes. A dozen values had to be calculated in a fraction of a second.

BLOG: Flapping, Winged Robot Perches On Hand

The next step is a plane that can figure out a map of its environment as it goes along. That might require adding visual information to the rangefinder — a big challenge, but they have proved it is possible, at least in principle.

via MIT

Credit: MIT