Science’s Dream Team

If they held a science fair at Disneyland, it might turn out something like this.

In one conference room, five toy truck-style robots engage in a surprisingly lively soccer match. In another office, would-be soldiers survey an expansive battleground using an Imax-style variant of virtual reality. And in yet another room, a mutant copy machine translates Japanese documents into English.

That was the scene last week at an open house to celebrate the 25th anniversary of USC’s Information Sciences Institute. Its ebullient executive director, Herbert Schorr, sported a custom-made sweatshirt that said “USC ISI 25 Years” on the front and “Grand Pooh-Bah” on the back.

It’s hard to imagine getting much work done at ISI, which occupies six floors of a Marina del Rey high-rise with panoramic vistas of the waterfront, the Santa Monica Mountains, downtown Los Angeles and a steady stream of jets taking off from nearby LAX.

But for a quarter-century, researchers at the institute have been turning out practical technical innovations ranging from the military’s first portable computer to the Internet domain name system that has made “.com” a household word.

Four refugees from Rand Corp. founded the Information Sciences Institute in 1972 so that they could focus their work on the nascent field of computer science. The group, led by Keith Uncapher, went searching for a university partner to gain access to academic resources. After being turned down by UCLA, USC’s School of Engineering jumped at the chance to operate the center.

Since then, ISI has focused on larger, more applications-oriented research projects than are typically tackled by university researchers. The institute has developed expertise in areas including artificial intelligence and advanced computer networks, and it won $52 million in research contracts--80% from the Defense Advanced Research Projects Agency.

“We have to guess what areas of computer science are up-and-coming and figure out what problems need to be solved,” Schorr said.

Of course, it takes a special kind of scientific insight to recognize the need for soccer-playing robots. The players, about the size of small dogs, are equipped with cameras so that they can find the soccer ball. When one of them has the ball in its sights, the motor kicks in and the robot drives over to the ball and heads toward the goal--hopefully the one belonging to its opponent.

A team of eight researchers has invested more than a year on the robo-soccer project, and ISI’s self-proclaimed “Dream Team” placed third in the international RoboCup ’97 competition in Japan in August. The robots are an ideal vehicle for developing technology for computer vision, artificial intelligence and autonomous agents, said Wei-Min Shen, a senior research scientist at ISI.

At the Open House on Wednesday, the robots beeped and spun around as they methodically scanned the field for the ball, which they can identify because it is red. But one of them mistook Shen’s brown leather shoes for the ball and proceeded to follow him around--to the delight of the roomful of spectators--until he took his shoes off.

Downstairs, virtual military commanders wore special 3-D glasses and stood in front of a slanted 4-by-6-foot screen to survey a digital version of Ft. Knox in Tennessee. Sensors connected to this “ImmersaDesk” kept track of where the commander was looking and adjusted the display accordingly. When standing right in front of the screen, the battlefield takes up one’s entire field of vision.

This experiment in immersive virtual reality--which eliminates the need for a bulky headset--is made possible by the Globus computational network, another ISI initiative. Globus links the supercomputing power of 3,000 data processors at 15 government and educational institutions into a single computational power grid that may someday be the equivalent of today’s electric power grids.

“We want to have an impact on the way people do science and use computers,” said Carl Kesselman, a project leader at ISI.

Meanwhile, Kevin Knight, another ISI project leader, demonstrated his prototype for an automatic language translation machine. The desktop device looks like a photocopier, except that a document that is placed face-down in Japanese characters will generate a new document in English instead of an exact copy.

First, the machine uses optical character recognition to identify the individual Japanese characters, and then splits them into words. A linguistics program identifies the noun phrases, verb complexes and other parts of speech, and diagrams each sentence as in a junior high school grammar exercise.

Next, the software translates each phrase into a conceptual “interlingua” phrase, then translates that phrase into English. Finally, the software uses statistical methods to select a final version of the sentence that sounds most like standard written English.

During a demonstration, a first cut at translating an unknown Japanese sentence yielded the English phrase, “Don’t have to frightening me.” Using a more advanced--and time-consuming--method, the same sentence was improved to, “I need not be frightening.”

Since Knight doesn’t speak Japanese, he didn’t know what the actual translation was and therefore couldn’t judge the accuracy of his creation. But he admitted, “It’s very unreliable and we’ve still got a long way to go.” Eventually, he hopes his machine will be able to translate between any two of 10 languages.

*

Karen Kaplan covers technology, telecommunications and aerospace. She can be reached at karen.kaplan@latimes.com