MIT robot plane deletes the pilot

When the robots come for you, at least they won't scratch the walls. MIT research into autonomous flight has delivered a robotic plane that can thread its way, at speed, through enclosed and indoor conditions, without requiring preconfigured flight plans or GPS navigation. The plane has significantly longer flight time than autonomous helicopters, though introduced a fair few problems of its own.

Unlike helicopters, which can hover, rotate on the spot, easily travel in three-dimensions and go sideways, planes must keep moving and have reduced flexibility in where they can redirect themselves. MIT's solution was a custom-designed aircraft with shorter, chunkier wings that combine tight turning, the possibility of relatively low speeds without stalling, and reasonable cargo capabilities for the AI smarts and camera equipment.

Inside, along with the cameras which allow for the bird's-eye view in the video below, the plane has a laser rangefinder, accelerometers and gyroscopes to track whereabouts in the room it is, what speed it's traveling, acceleration, and more. At any one moment in time, the plane is figuring out 15 different values, MIT says, and part of what makes the new system special is a duo of algorithms – one fast and rough, the other slow and accurate – to first filter out the most relevant data and then crunch only that.

It's not all magic, at least not yet. The plane must be preloaded with a high-res digital map of the area, unlike helicopters which are capable of building their own maps. That's next on the MIT team's agenda, however, boosting the algorithms and building in more visual information to the other sensors.

Of course, an alternative approach might be pairing the best of helicopters and planes, using both in sequence. An autonomous helicopter could enter an unknown area and quickly gather reconnaissance data of the environment, taking advantage of the flexibility of movement to be more comprehensive, and then be followed by an autonomous plane using that fresh data but bringing greater flight time to the table.

[via Gizmodo]