Engineers Program Robot to Create Its Own “Children”

Mother-bot displays laws of natural selection, improving the design of each generation of offspring.

Engineers at the University of Cambridge have designed a robot capable of developing miniature “baby” robots, improving upon them with each successive generation.

If modern advancements in robotics and AI haven’t cemented public fears of a Hollywood-like robot apocalypse yet, it wouldn’t be surprising if this announcement did.

Researchers based the experiment around natural selection and Darwin’s theory of evolution. The “mother” robotic arm was programmed to build miniature robots capable of “walking” along a surface.

Once unleashed, she was given total autonomy to design ten generations of robots using plastic, motorized cubes. Five hundred autonomously constructed “children” were produced over the course of five separate experiments. Each baby-bot took around 10 minutes to design.

The mother-bot’s AI allowed it to use each previous generation as an experiment in learning what worked and what didn’t. Each of these children had a unique “genome” made up of between one and five “genes,” which decided the programming concerning the child’s shape, construction and motor commands.

These genes were modified throughout subsequent generations, resembling the real-life process of mutation.

To determine the fitness of each robo-child, the mother-bot tested how far it could travel from its starting position to another point in a given amount of time. Successful babies remained unchanged in the next generation, while the failures were modified.

Researchers found the fastest of the children at the end of the process were more than twice as quick as those of the first generation. This was possible as the mother-bot designed new shapes and gait patterns throughout the experiment.

“One of the big questions in biology is how intelligence came about — we’re using robotics to explore this mystery,” said lead researcher Fumiya Iida of Cambridge’s Department of Engineering, who collaborated with researchers at ETH Zurich.

“We think of robots as performing repetitive tasks and they’re typically designed for mass production instead of mass customization, but we want to see robots that are capable of innovation and creativity,” Iida explained.

These experiments contribute to the growing field of evolutionary robotics, investigating the creation of autonomous robots without human intervention. Until today, most research in this field has been done through simulation.

Breaking out into the real world, the idea of replicating robots is both an exciting and frightening concept. In industrial settings, manufacturers in need of a greater automated force may simply need to buy cheap parts in bulk and let their one robot design a fleet of fresh workers.

But what’s to stop robots from building and redesigning almost infinitely? Can they design and redesign AI software to the point where it rivals or even exceeds human intelligence? At the moment, that doesn’t seem possible.

“It’s still a long way to go before we’ll have robots that look, act and think like us,” said Iida. “But what we do have are a lot of enabling technologies that will help us import some aspects of biology to the engineering world.”

If and when they do look, act and think like us, let’s remember to tell them about Isaac Asimov’s “Three Laws of Robotics.”

The experiment’s results were published in the open access journal PLOS One.