'Baby' robot learns like a human

A robot that learns to interact with the world in a similar way to a human baby could provide researchers with fresh insights into biological intelligence.

Created by roboticists from Italy, France and Switzerland, "Babybot" automatically experiments with objects nearby and learns how best to make use of them. This gives the robot an ability to develop motor skills in the same way as a human infant.

The robot consists of a one-armed torso with a pair of cameras for eyes and a grasping hand. It has an in-built desire to physically experiment with objects on the table in front of it and an ability to assess different forms of interaction and learn from mistakes. If the robot fails to grasp an object securely, for example, it remembers and tries a differently strategy next time. One unbidden skill developed by Babybot was the ability to roll a bottle across its table.

One video (avi format) shows the robot experimenting with a rubber duck, while another clip shows Babybot examining a ball.

Cluster computing

"We started with knowledge from developmental psychologists and neuroscientists," explains Giorgio Metta at Genoa University in Italy, a member of the research team. "What we're doing is the same as what neuroscientists do but from an engineering perspective."

Babybot's "brain" is actually a cluster of 20 computers running several neural networks. This is software that mimics a biological neural system and learns in a similar way - by establishing and altering the strength of links between artificial neurons. By adjusting the neural network software and observing the robot's learning behaviour, the roboticists can test different neuroscience models.

"The idea is fantastic," says Steve Grand, founder of UK robotics research company Cyberlife Research, who has also worked on simple learning robots. "It's the only way you can research the development of intelligence or artificial intelligence."

However, Grand believes fundamental differences between the human brain and computers used to control learning robots like Babybot may mean that such machines can never become as intelligent as us.

The one and only

"Babybot is based on artificial neural networks, which are not in the slightest the same as biological ones," Grand says. "Since we only have one example of intelligence in the universe, the brain, it might help to keep closer to biology. For all we know it might be the only way of doing it."

Plans to build the successor to Babybot are already underway, again with input from neuroscientists. The first functioning version of a robot toddler, dubbed 'Robocub', should be completed in spring 2007.

"The goal is to build a humanoid 2-year-old child," explains Metta. This will have all of Babybot's abilities and the researchers hope it may eventually even learn how to walk. "It will definitely crawl," says Metta, "and is designed so that walking is mechanically possible."

Metta and colleagues hope that other researchers will use their Robocub designs. They also plan to make copies of Robocub available to other experts. "Everything about it will be open source, including the hardware, so anyone can use it in their own work," Metta says. [NewScientistTech]