Live: 300+ Best Black Friday Deals Live: Black Friday TV Deals BF Deals Under $25 BF Deals Under $50 5 BF Splurges 8 BF Must-Haves 15 Weird Amazon BF Deals BF Cheat Sheet
Want CNET to notify you of price drops and the latest stories?
No, thank you

Natural selection: Mama robot builds self-evolving baby-bots

Researchers at the University of Cambridge have built a robot that can build other robots, test them, and improve on the results.

The mother-bot building a baby from motorised cubes. Screenshot by Michelle Starr/CNET

The theory of natural selection popularised by Charles Darwin has now been demonstrated in robots. A "mother" robot built by researchers at the University of Cambridge can design, build, test and improve upon her baby robots with each successive generation, creating an evolving series of ever-better machines.

The research was published on June 19 in the open access journal PLOS One.

The team, led by Dr Fumiya Iida of Cambridge's Department of Engineering, built the mother-bot and programmed her to build a robot capable of moving. After that, everything the mother did was autonomous. Using plastic cubes with motors inside, she built 10 generations of 10 "children," across a total of five separate experiments, totalling 500 autonomously constructed baby-bots.

"Natural selection is basically reproduction, assessment, reproduction, assessment and so on," lead Iida, who worked in collaboration with researchers at ETH Zurich, said in a statement. "That's essentially what this robot is doing -- we can actually watch the improvement and diversification of the species."

For each generation, the mother-bot deployed and tested the children, using the data collected to improve on the robots. By the final generation, the robots were able to perform tasks twice as quickly as robots in the first generation.

Here's how it worked. Each child robot had a unique "genome" made of one to five genes, containing all the information about its shape, configuration, construction and commands. The mother-bot tested the children by seeing how far they travelled from a starting point within a certain amount of time. The most successful children were left as they were, while the slower ones were mutated using traits from the successful robots.

While there were still differing speeds for the final variation, the researchers found that the fastest individuals in the last generation moved, on average, more than twice as fast as the individuals in the first generation.

The team also found that the mother was not only able to tweak the design of the children, she was able to introduce new shapes and gait patterns -- some, the team said, that a human designer would not have been able to build.

"One of the big questions in biology is how intelligence came about -- we're using robotics to explore this mystery," said Iida. "We think of robots as performing repetitive tasks, and they're typically designed for mass production instead of mass customisation, but we want to see robots that are capable of innovation and creativity."

Usually, evolutionary robotics is studied by running computer simulations. While this allows for testing of a wide range of evolutions, they often don't work as well in the real world as they do in the simulation. These "mother-bot" experiment removed that problem.

But, while the experiment was successful, it also has a problem: Time. The mother robot, for instance, takes about 10 minutes to design, build and test each baby robot. This could be cut down by introducing computer simulations into the process. If the mother robot could accurately know which traits would work best for the task, it would cut down on the time spent building and testing the less successful robots.

Watch the mother-bot at work making a baby in the video below. Awwww.