CONTACT: Stanford University News Service (650) 723-2558
"Seeing-eye" robot demonstrated
STANFORD -- A group of children invented a new game on Tuesday afternoon, June 13: ring around the robot. The place was the steps at the entrance to the Main Quad. The occasion was the first demonstration of a "visual robot" that has been developed over the last four months by graduate student Illah Nourbakhsh and visiting scholar David Andre with the sponsorship of Professors Carlo Tomasi and Michael Genesereth in computer science.
The visual robot, which is roughly the size and shape of the Star Wars robot R2D2, has one ability. It can avoid obstacles: steps, walls, people, cookie boxes, tripod legs. It was put through its paces by a couple of dozen adults and a half-dozen kids. Even the adults found it impossible to resist the temptation to block the robot's path and watch it veer away. Except for running into a woman with a black dress, it performed flawlessly.
What is significant about the robot is not its ability to avoid obstacles per se, but the way it does so. Most robots navigate using active sensors, like sonar that emit sound waves and detect obstacles by recording the echoes. The new Stanford robot works in an entirely different way. Its only sensors are three video cameras. Each of the cameras is focused at a different distance: near, mid-range and far, far being about 50 inches. The robot's microprocessor scans the image to determine the amount of contrast between adjacent pixels. When the contrast level is high, the image is in focus. In this fashion, the robot can tell whether it is about to run into something. In the case of the black dress, apparently there wasn't enough contrast in the dress so that the robot could tell when it was in focus.
"This system is more robust, operates with less knowledge and is much simpler than other obstacle avoidance systems," said Tomasi. Compared to sonar, the visual system can pinpoint the direction of obstacles much more precisely, he said.
"It is a good idea that works even better than we thought it could. It's such a simple idea, I don't know why someone didn't think of it a long time ago," added Genesereth, as he watched a group of children box in the robot. It works so well, in fact, that Stanford is patenting the technique.
The computer scientists' next step is to enhance the vision system so that the robot can navigate by visual landmarks. These activities are part of the department's "bookstore project" to design a robot that can go from the department to the library, check out a book and return.
This is an archived release.
This release is not available in any other form.
Images mentioned in this release are not available online.