April 23, 2013
Algorithm Shows Evolution Of Soft Robots
Lee Rannals for redOrbit.com — Your Universe Online
A group of researchers from Cornell University's Creative Machines Lab say they have developed an algorithm that can be used to witness virtual creatures evolving.
The team wrote in their paper titled "Unshackling Evolution: Evolving Soft Robots With Multiple Materials and a Powerful Generative Encoding" in the journal Proceedings of the Genetic and Evolutionary Computation Conference about their algorithm. They said the equation could be used to witness virtual creatures evolving their squishy, muscle-like features in order to teach themselves to walk.
"We show that combining a powerful generative encoding based on principles of developmental biology with soft, biologically-inspired materials produces a diverse array of interesting morphologies and behaviors," the authors wrote in their paper.
The algorithm incorporates concepts from developmental biology and how nature builds complex animals, including everything from jellyfish to jaguars. The scientists describe how they challenged human engineers to design robots made of these soft hard materials, but they said the human efforts paled in comparison to the designs resulting from evolution.
"Our results suggest that investigating soft robotics and modern generative encodings may offer a path towards eventually producing the next generation of impressive, computationally evolved creatures to fill artificial worlds and showcase the power of evolutionary algorithms," the researchers said.
[ Watch the Video: Evolving Soft Robots with Multiple Materials ]
A video showing the algorithm in action displays a creature evolving into a galloping, soft robot over 1,000 generations. Although 1,000 generations is relatively short compared to natural evolution standards, the team says it is enough elapsed time to demonstrate the power of evolution to create counterintuitive designs.
"Evolution did all the heavy lifting: there is no human in the loop after we start the Darwinian process. It is definitely evidence for evolutionï»¿ doing impressive things, not for intelligent design," Jeff Clune, assistant professor at the University of Wyoming and former visiting scientist at Cornell, said in a comment on YouTube about the video.
The team said they are also pursuing methods to minimize the need for expensive simulations and to evolve specific material properties instead of having a predefined palette of materials.
"These avenues are expected to allow increased complexity and diversity in future studies," the authors wrote. "In order to compare different approaches, the field would benefit from general, accepted definitions and quantitative measures of complexity, impressiveness, and naturalness. Such metrics will enable more quantitative analyses in future studies like this one."
Scientists at the Harvard-Smithsonian Center for Astrophysics (CfA) and their colleagues at the Heidelberg Institute for Theoretical Studies (HITS) said earlier this month they have developed software that builds a universe from scratch. This software generates a full-fledged simulation of the Universe, taking you through time from the Big Bang, and 14 billion years later into the future.