Advertisement

Science Watch : Researchers Building Robotic Prototypes, Lego Block by Lego Block

Share

Computer scientists at Brandeis University say they have taken a significant step toward creating robots that will evolve into ever more sophisticated machines, capable of repairing and modifying their own hardware.

What they have come up with isn’t exactly the “2001” film star, Hal, or even R2D2 of “Star Wars.” But they have produced software that allows a computer to design structures, such as bridges and cranes, without human intervention.

“It’s a kind of science fiction,” conceded Jordan Pollack, associate professor of computer science at the university’s Volen National Center for Complex Systems in Waltham, Mass. “But it works.”

Advertisement

They pulled off their achievement with the help of toys designed for children, the ubiquitous Lego blocks that have spread from the kid’s room to prestigious research laboratories around the world. The small plastic blocks, manufactured by international toy giant Lego Group, can be attached to each other in a variety of ways, making it possible to build robotic prototypes in minutes and with minimal cost.

Lego blocks were chosen because they lend themselves to modular construction techniques, according to graduate researcher Pablo Funes, who has done most of the work on the project.

“Legos are modular in the sense that they can be combined in different positions,” Funes said. “Any brick can attach to any other brick.”

The Brandeis scientists created algorithms, or computer programs, that established the basic physics needed to build strong structures. Then the computer was given a single Lego brick, along with a goal, called a “fitness function”--in this case to design a bridge two meters in length.

That single brick “mutated” into other bricks, and the computer tried many ways to assemble the bricks into a bridge. Each step along the way, it compared the structure with the physics model contained in the software.

“That program basically calculates whether or not the structure will collapse,” Pollack said.

Advertisement

A day and a half later, the computer produced a cantilevered design for a bridge.

The scientists then broke out their set of Legos and built a real bridge exactly like the computer had designed.

It not only worked, Pollack said, but it was a good design. In fact, it took humans centuries to develop similar structures--bridges that are based on counterbalances.

The computer also designed a crane, based on an inverted triangle, and a table.

The computer came up with the designs through a trial-and-error process, adding one Lego at a time and checking against the physics model to see if that strengthened or weakened the structure. It had an amazing number of chances to get it wrong.

Two plastic bricks with eight studs--by which the bricks are connected--can be put together 24 ways, according to Lego Group. Three bricks can be assembled 1,060 ways. And just six bricks can be put together in 102,981,500 different ways. The computer used more than 200 bricks to design its bridge, so the possibilities are virtually infinite.

The scientists say they gave the computer no commands other than to start the task. But they are walking a fine line. Did the computer succeed because they told it too much about what it was supposed to do, or did it succeed because it was able to “think” through the challenge, aided by fundamental principles of physics?

Pollack thinks the computer did its own work because human values--such as symmetry--were left out of the equations. And he said they took pains to keep human input at a minimum.

Advertisement

“It doesn’t know about the triangle, and it doesn’t know about the counterbalance any more than a bird knows about a wing,” he said. “It’s simply exploiting those structures because the physics makes them better.”

The scientists see the experiment as a “baby step” on the road to computer-driven machines--or robots--that evolve both mechanically and “intellectually” as they work through increasingly complex assignments.

“It’s called evolutionary computation,” Funes said, and it is an attempt to copy the natural selection process that feeds evolution in the world of plants and animals.

Like survival of the fittest, the Brandeis computer attaches one Lego brick at a time, and discards the weaker for the stronger combinations.

It doesn’t mean Hal is lurking around the corner and planning a coup, because the challenging field of robotics is progressing at a very slow pace. The Brandeis computer might be able to design a pretty good bridge, but it can’t pick up a single Lego brick.

“These are designed for an 11-year-old’s hand,” Pollack said. “No one has a clue how to build a robot with that kind of manual dexterity.”

Advertisement

So the human body, as well as the mind, will undoubtedly remain in control for a long time. But someday . . .

*

Lee Dye can be reached at leedye@ptialaska.net.

Advertisement