A robot on the move must be able to sense whether it is travelling in a straight line or spinning on the spot. But telling the difference is difficult with just a single camera for an eye. Yiannis Aloimonos, a computer scientist at the University of Maryland in College Park, told New Scientist that the best way to understand the problem is to imagine seeing the world through a cardboard tube.

Turning your head from side to side gives an image that is hard to distinguish from the view you get if you move sideways. His colleague Cornelia Ferm¿ller points out that we make the distinction with the help of cues from inertia sensors built into our ears, but robots do not have this faculty.

In 1998, however, Aloimonos and Ferm¿ller mathematically proved that being able to see in all directions allows precise motion sensing without the need for any other sensors. Now they have used this proof to develop software that processes images from a set of cameras arranged on the surface of a sphere to give omni-directional vision.

The solution is equivalent to imagining that you have an eye on the back your head, says Aloimonos. When you rotate your head, the images from the front move in the opposite direction to those from behind.

Whereas when you move sideways, the two sets of images move in the same direction, so it is easy to work out the direction of travel. The new device, nicknamed the "Argus eye" after a Greek god with eyes all over his body, consists of nine off-the-shelf digital cameras attached to a frame the size of a beach ball, although the team hopes to shrink the device in future. The images are passed to a computer for processing.

Aloimonos and Ferm¿ller say the software can easily identify the direction of motion in 3D when the Argus eye is moved. They will present the device at a robotics conference in Las Vegas this week.

This article will appear in the November 1 issue of New Scientist

Related Links
Center for Automation Research
SpaceDaily
Search SpaceDaily
Subscribe To SpaceDaily Express