Robots emerge from simulation

By Eric Smalley, Technology Research News

Living things have at least two advantages over machines -- we reproduce and we are honed by millions of years of evolution. Researchers at Brandeis University have developed a system for designing robots that makes it clear those advantages are destined to fade into history.

The researchers’ Genetically Organized Lifelike Electro Mechanics (Golem) project combines a genetic algorithm that allows populations of virtual robots to evolve toward a desired set of characteristics, with a rapid prototyping machine that builds the robots' body parts automatically.

The long-range goal of the project is to reduce human involvement in the process of building machines like robots to defining their tasks and supplying the raw materials to build them, said Jordan B. Pollack, associate professor of computer science at Brandeis.

The general idea was "evolving [robot] bodies and brains together in simulation and then trying to figure out how to transfer them to reality," he said. The solution was to make the robots’ body parts emerge from a vat of liquid plastic via stereo lithography, a rapid prototyping technique that uses computer-controlled lasers to set the light-sensitive liquid.

Postdoctoral researcher Hod Lipson developed the "universal robot set" starting point for the genetic algorithm by adding motion to a truss simulator, Pollack said. Truss simulators model the components of structures like bridges.

The genetic algorithm started with a population of 200 virtual machines and selected those that were best at moving. The algorithm replicated the selected individuals, modified the copies and then added them into the population in place of other individuals, sometimes randomly and sometimes in place of the least fit. The algorithm repeated the process 300 to 600
times.

The robots the system built are little more than toys. They consist of plastic tubes and joints and electric motors. The motors have to be inserted by hand. The robots are less then a foot long and their bodies are combinations of simple geometric shapes. (See photo)

The machines' only task is to move in one direction on a horizontal surface. But the simple devices demonstrate a process of combining automated designed with automated manufacturing.

Pollack and Lipson "have opened a new way of designing systems whereby one doesn't have to look at past experiences," said Pradeep Khosla, Dowd Professor of engineering and robotics at Carnegie Mellon University. "The structures that come out of [the simulations] are not totally intuitive."

For example, one of the robots is roughly pyramid shaped with a rod in the center. The rod presses downward at angle, shuffling the machine forward.

The next phase of the Golem project is to increase the complexity of both the robots’ components and their tasks, Pollack said. The researchers plan to use robotic arms to assemble the next generation of the robots from existing parts, he said. Pollack is also working on projects aimed at giving robots the ability to learn and adapt once they're on the job.

"You don't expect a transfer from simulation to reality to work. You expect it to be a noisy sort of process that gets you most of the way there but doesn't really give you a working robot," Pollack said.

The automated design and manufacturing system could be used to produce useful machines in seven years and could be used to make toys in two years, Pollack said. Pollack's work on evolutionary computing is funded by the Office of Naval Research. The Golem project is funded by DARPA. The researchers publish their work in the August 31 issue of the journal Nature.

Timeline:   7 years
Funding:   Government
TRN Categories:   Artificial Life and Evolutionary Computing; Robotics
Story Type:   News
Related Elements:   Photo, Technical paper “Automatic design and manufacture of robotic lifeforms” in the August 31 Nature




Advertisements:



September 20, 2000

Page One

Hue-ing to quantum computing

Robots emerge from simulation

Software sorts Web data

Processor design tunes memory on the fly

Superconducting transistor debuts




News:

Research News Roundup
Research Watch blog

Features:
View from the High Ground Q&A
How It Works

RSS Feeds:
News  | Blog  | Books 



Ad links:
Buy an ad link

Advertisements:







Ad links: Clear History

Buy an ad link

 
Home     Archive     Resources    Feeds     Offline Publications     Glossary
TRN Finder     Research Dir.    Events Dir.      Researchers     Bookshelf
   Contribute      Under Development     T-shirts etc.     Classifieds
Forum    Comments    Feedback     About TRN


© Copyright Technology Research News, LLC 2000-2006. All rights reserved.