Flying robots know how to avoid obstacles like birds

Intelligence is like a bird: Flying Rescue Robot (Flying Rescue Robot) will avoid obstacles.

Being able to orient themselves in damaged forests, tunnels or buildings, this machine can be of great value in search and rescue activities.

Small-sized flying machines are quite popular, and global positioning technology (GPS) provides guides. Currently, Ashutosh Saxena, assistant professor of computer science and his team is addressing a difficult problem: how to keep the 'car' from crashing into walls and branches. The controller does not always react fast enough, and the radio signal may not reach all the positions the robot is going to.

The test car is a quadrotor (4-propeller flying model), a commercially available aircraft, the size of a card table with four helicopter propellers. Saxena and his team have programmed quadrotors to avoid corridors and stairs using 3D cameras. But in nature, these cameras are not accurate enough at great distances to plan a route around obstacles. Therefore, Saxena is building on the methods that he developed earlier, to turn a flat-image video camera into a 3D model of the environment by using signals like: converging lines straight, the external dimensions of similar objects and objects are in front or behind each other - the same way people use to complement their stereoscopic vision unconsciously.

Two graduate students Ian Lenz and Mevlana Gemici trained robots with 3D images of obstacles such as twigs, poles, fences and buildings; Robot's computer learns the characteristics of all images that have in common, such as color, texture, shape and context - such as a tree branch, attached to a tree. Set rules to decide what is an obstacle burned inside the chip before the robot flies.

In flight the robot breaks down the current 3D image of the environment into small parts based on clear boundaries, deciding what objects are obstacles and calculating a path through these obstacles. As close as possible to the path it was ordered to follow, continuously adjusted when the vision changed.

It was tested in 53 independent flights in an obstacle-filled environment and succeeded in 51 cases, the remaining two failing due to the wind. The results are presented at the International Conference on Intelligent Systems and Robots in Portugal, October 7 to 12.

Saxena plans to improve the robot's ability to respond to environmental changes like wind, and allow it to detect and avoid moving objects, like birds, for experimental purposes, he Ask everyone to throw tennis balls at this flying car.

The project is supported by a grant from the Defense Advanced Research Projects Agency (Advanced Research Projects Agency).