AI-based algorithm uses "compressed evolution" to create functional robots
Evolution is a very slow process, due largely to the fact that nature doesn't "know" in advance which features of an animal will be beneficial. A new AI-based algorithm does know, however, allowing it to design purpose-specific robots within a matter of seconds.
The algorithm was developed by a team of American scientists led by Northwestern University's Asst. Prof. Sam Kriegman.
Two years ago, Kriegman made headlines when he revealed tiny bioengineered robots known as Xenobots, which were capable of reproduction. An early version of the algorithm was used in that study, to determine the optimum body shape for the Petri-dish-swimming li'l bots – they ultimately ended up looking a bit like Pac Man.
Whereas a supercomputer took months to come up with that design, the algorithm has now been refined to the point that it runs on an ordinary laptop computer, and it can deliver results within less than half a minute. It was trained utilizing the principles of natural evolution, hugely boosted by the fact that the programmers could look back through time to see what sort of biological traits worked and what sort didn't.
"Evolving robots previously required weeks of trial and error on a supercomputer, and of course before any animals could run, swim or fly around our world, there were billions upon billions of years of trial and error," said Kriegman. "This is because evolution has no foresight. It cannot see into the future to know if a specific mutation will be beneficial or catastrophic. We found a way to remove this blindfold, thereby compressing billions of years of evolution into an instant."
As a test of the technology, the researchers asked the algorithm to create a robot that could walk across a flat surface. Importantly, the system had no knowledge of human attempts at walking robots to draw upon.
Starting with a digital block of material about the size of a bar of soap, the algorithm came up with multiple successive designs, each one based upon the computer-modeled successes and failures of those that preceded it. After 26 seconds and nine such generations, the system arrived at a design that could walk half its body length per second utilizing three inline legs. The scientists proceeded to build a physical silicone model of the robot, which could indeed walk (albeit in a shuffling sort of way) as air was pumped in and out of its body.
It is hoped that once the technology is developed further, it could be used to optimize the design of robots intended for applications such as searching for trapped survivors at disaster sites, making repairs in sewage systems, or even performing medical procedures within the human body.
"When humans design robots, we tend to design them to look like familiar objects," Kriegman said. "But AI can create new possibilities and new paths forward that humans have never even considered. It could help us think and dream differently. And this might help us solve some of the most difficult problems we face."
The study is described in a paper that was recently published in the journal Proceedings of the National Academy of Sciences.
You can see the robot in action, in the video below.
Source: Northwestern University