Guessing robots navigate faster: research
Last Updated: Thursday, May 10, 2007 | 10:39 PM ET
CBC News
U.S. researchers are developing robots that use the equivalent of guesswork to find their way around unfamiliar places.
The robots, being tested by scientists at Purdue University in West Lafayette, Ind., create a map as they travel through strange territory, which they use to predict what to expect in new environments.
The technique, called prediction-based simultaneous localization and mapping (P-SLAM) could one day let robots easily navigate complex surroundings such as office buildings unaided, Purdue professor of electrical and computer engineering C.S. George Lee told New Scientist news service.
The new approach differs from the typically difficult and time-consuming method used by robots to create maps, which involves scanning terrain and measuring distances to create their maps.
Lee and his colleagues instead developed a technique that mimics the way in which humans create mental maps. Their algorithm uses patterns created for chunks of territory or cells that the robot has already explored and predicts what unexplored cells will be like based on stored data.
Predictions for new cells that have a high-confidence ranking can be left unexplored to save time, while cells with lower confidence scores would be examined and fully mapped.
Lee says tests at Purdue found that the robots saved time and made few mapping errors.
His research is published in the scientific journal IEEE Transactions on Robotics.
