WALTHAM, Mass. -- Evolution, until now the unchallenged domain of living organisms, may soon become possible for robots as well. So say computer scientists at Brandeis University, where a simple computer-based form of evolution -- nature's own design strategy -- has succeeded in designing LEGO structures without any assistance from humans. It's the first successful leap from today's computer-aided design into the futuristic realm of fully automated design, and the first baby step toward the artificial intelligence community's longstanding dream of evolutionary robotics, where robots might eventually adopt some form of evolution.
Brandeis's Pablo Funes and Jordan Pollack, the researchers who masterminded this achievement, say this trial-and-error approach to designing LEGO bridges, tables, and cranes lays the groundwork for robots capable of reworking their own hardware without any human guidance at all. Their work appears in the most recently published issue of the journal "Artificial Life," and the university has filed for patent protection on the work.
"The necessity of allowing robots' brains and bodies to evolve together has been around since the dawn of evolutionary robotics," says Pollack, associate professor of computer science in Brandeis's Volen National Center for Complex Systems. "Our view is that in nature there is never a brain without a body, and that small changes to both must be made in a co-evolutionary fashion."
To make this rudimentary evolutionary hardware, Pollack and Funes, a graduate researcher, constructed a computer program capable of plucking out structurally sound LEGO designs from a sea of possibilities. They then provided the program with simple optimization goals, such as spanning a distance and carrying a weight -- laying the groundwork for the fully automated design of LEGO-based structures such as a two-meter bridge, a crane capable of lifting a one-kilogram weight, and a table that could support the same weight.
After the computer analyzed hundreds of its own designs and spit out the best ones, Pollack and Funes broke out their LEGO bricks and built the structures. Almost without exception, they found, the computer had engineered structures that were structurally sound.
"We're not saying these structures are engineering marvels, but we've shown that even a simple evolutionary program, paired with the right physics, can design complex structures without any engineering expertise from humans. A number of other researchers have tried to evolve more impressive simulated structures, but ours are the first to translate into reality," Pollack says.
The computer's design success was largely a function of its skill in evaluating the integrity of LEGO structures, which Pollack says can be determined through the patent-pending algorithms he and Funes developed to analyze torque in networks of the sticky blocks. The program is no speed demon; it took a day and a half to design the two-meter bridge. But the cantilevered design it eventually came up with is, at least in Pollack's estimation, superior for a self-supporting LEGO bridge of that size.
"What I find most fascinating about these results," Pollack says, "is that a very simple algorithm 'discovered' sophisticated structures that took humans many centuries to design -- a cantilever for the bridge, and an inverted triangle for the crane."
The work was partially sponsored by the Office of Naval Research and the National Science Foundation. With a new research grant from the Defense Advanced Research Projects Administration (DARPA), members of Pollack's laboratory are now developing similar systems that feature limited motion, which is expected to allow fully automatic design of robot bodies and brains.
NOTE: LEGO is a registered trademark of the LEGO Group.
The above post is reprinted from materials provided by Brandeis University. Note: Content may be edited for style and length.
Cite This Page: