Unnatural Selection: ‘Mother Robot’ Builds Evolving Baby Bots

It's always been something of a solace for humankind that robots can't evolve like we can, but we may not retain that advantage for long. Researchers have demonstrated a sort of rudimentary artificial evolution: a robot that creates generations of smaller "baby" robots, evaluates them and selects the best to "live" on — just like Mother Nature.

Related: Scientists Aspire to Nature's Genius With 'Biomimetic' Research

The study was led by Fumiya Iida at the University of Cambridge. A "mother robot" assembled smaller robots by gluing together pieces, some of which were equipped with motors, in various configurations. It then observed how quickly the resulting "children" could move, and then kept the designs of the fastest ones and disassembled the rest. It then included this information when creating a new "generation" — either mutating the successful designs or integrating them into less successful ones.

"Natural selection is basically reproduction, assessment, reproduction, assessment and so on," said Iida in the Cambridge news release. "That’s essentially what this robot is doing — we can actually watch the improvement and diversification of the species."

It works, too. After a few generations, the children were running twice as fast as the first batch. That's evolution at work.

This type of research is usually done virtually, with an artificial intelligence evaluating thousands or millions of possibilities in a computer simulation. But doing it in real life, with real robots, is the only way to show that it actually works, as Iida's lab has done.

Related: MIT's Cheetah-Bot Makes Running Jumps With Lifelike Ease

Future work may combine simulation and real-life testing, increase the complexity of the child robots, or use nature-inspired techniques to improve the efficiency of their gaits.

You can read the full paper in the open-access journal PLOS One.