Thursday, May 10, 2007

Guessing Robots Find Their Way

By Princessa

Robots that use "guesswork" to navigate through unfamiliar surroundings are being developed by US researchers.

The mobile machines create maps of areas they have already explored and then use this information to predict what unknown environments will be like.

Trials in office buildings showed that the robots were able to find their way around, New Scientist reported.

Making robots that can navigate without prior knowledge of their surroundings was a huge challenge, the team said.

Most mobile robots do this using a technique called SLAM (simultaneous localisation and mapping), whereby they build up a map of their unknown environment, using various sensors, whilst keeping track of their current position at the same time.

But this technique is slow because a robot must explore a great deal of terrain to know its precise location. It is also prone to errors.

So the team from Purdue University, in Indiana, has developed a new approach.

The robots create a 2D map of the area they are exploring, but when they come to an unknown area, they check back through this information to see if it seems similar to any areas that have already explored.

They do this using an algorithm - a step-by-step problem solving procedure.

Professor George Lee who carried out the research, said: "The robot gets to a new area and thinks: 'Have I seen these sorts of things before?' Then it goes back and looks at its stored data.

"It might then think: 'Hey, this is very, very similar to something I've seen before, I don't need to explore that room or corner.' And this saves time for it to explore other areas."

He said it was similar to the human navigational process, where we build up a "mental map" of our surroundings by recognising familiar sights.

Some limitations

The scientists first tested the algorithm using virtual mazes and offices. Their computer models revealed that the robots could navigate successfully while exploring a third less of their environment than robots that simply used SLAM.

Then tests carried out using real robots inside a university office building showed that the new navigation technique was also faster and less prone to errors than SLAM.

However, the new method did have some limitations, Professor Lee said.

"Indoors, in places like office buildings, it works well; outdoors, where the scene isn't as repetitive, the result is not that good."

Self-navigating robots could have many applications, Professor Lee told the BBC News Website.

The US defence department is currently focusing on self-driving automobiles.

Professor Lee's research was funded by the National Science Foundation and was published in the journal IEEE Transactions on Robotics.

No comments: