This story is over 5 years old.

This Robot Mom Builds Kids and Judges Which Is Best

It’s survival of the fittest, robo-style.
August 12, 2015, 12:15pm
A cubic robo kid. Image: The University of Cambridge.

Researchers at Cambridge University and ETH Zurich have built a robot mom capable of independently building a set of kids. But the robot mom not only assembles her automaton family from plastic cubes equipped with motors, she also evaluates which of her cubic kids is best, and bases her next generation on these winning traits. So it's survival of the fittest, but robo-style.

Taking inspiration from natural evolution, Professor Fumiya Iida from Cambridge University's Department of Engineering told me that the aim was to build creative robots that could eventually be used to accelerate and optimize design and innovation processes. The potential of building hundreds of different robots that perform better each time also exists.

"Robots are usually built for repetitive tasks, and for mass production instead of mass customization. But we are hoping that a robot can be an inventor," Iida told me. "That's why we've called this the Inventor Project—because the robot can actually invent something that we can't invent."

In a study published in the journal Plos One, the researchers describe building a mother robot (a robotic arm) that independently produced five batches of cubic kids. Apart from the initial command from the researchers to build robots capable of movement, the robot mom does all this sans human intervention.

According to Iida, each robotic kid has a "genome" (a reconfigurable program code) made up of a combination of one and five different "genes" (more code), that stores information about the robo-kid's build, motor commands, and shape. Robo-evolution, said Iida, follows the same principles as in the natural world—by "mutation" (when genes are modified, or a new genome is made by merging genes from different individuals).

For this robot selection experiment, the robot mom evaluated which robot kid was best by testing how long it took for each robot kid to travel from point A to B. The traits of the most accomplished movers were then passed down to the next generation. The researchers discovered that both design diversification and performance improved through the generations as the robot mom fine-tuned previous designs, as well as invented new shapes for her kids.

The replication prowess of the robo-mom is pretty impressive. Iida said that their robot could build a robot an hour, and could be operated for 24 hours straight; so the potential of building hundreds of different robots that performed better each time were great.

"Robots in factories are limited in terms of adaptability and dealing with uncertain environments. Once they step out of the factories they are useless," said Iida. "They are not designed to be creative, so the fundamental question about autonomy and adaptability is missing in robotic technologies."

Next up, Iida and his team aim to collaborate with civil engineers in order to apply their robotic technologies to the housing construction sector. Iida explained that builders usually had to learn to adapt quickly given that properties were constructed in different types of environments. He said that civil engineers in the construction sector had expressed an interest in their bio-inspired robotics, believing that they could help automate housing construction, making it cheaper and faster.

Iida also envisions that in the future this tech could be applied to everything from cars to mobile phones.

"Robots have been used to build cars, but the car designs are always the same," said Iida. He believes that their tech could pave the way for greater design customization, which would simultaneously diversify a product, while fitting individual user desire.

"This would probably be a more advanced form of mass-customization," he said, "where it's not just different from machine to machine, but the machines could also be adapted to different uses."