Laboratories in Europe are adopting iCub robotic toddlers constructed from modular pieces and open source software to study learning and cognition
30 May 2008—Researchers across Europe are becoming parents to bouncing baby robots. By teaching them to walk, open doors, shake hands, and even talk, scientists hope to figure out how human children learn to do the things we adults take for granted.
The team behind the iCub robot believes that robots, like children, learn best from experience. Like a toddler who progressively learns about his own motor skills and how to interact with the world, the iCub—the size of a 3-year-old child, with sensor-equipped hands, eyes, and ears—has touch, sight, and hearing to explore its surroundings and develop its cognitive abilities.
The iCub is the baby of RobotCub, the European Unionfunded project that aims to advance research on the use of humanoid robots to understand human learning. Scientists in Europe and beyond believe humanoids can be essential tools in the study of human intelligence, which many of them argue is linked to the structure of the human body and the way it can interact with its surroundings. Their argument: because the physical body and its actions together play as much of a role in cognition as does the brain, mimicking human actions is essential to understanding the components of intelligence, like reasoning or memory.
”If you want to develop humanlike intelligence in a robot, you need to shape the robot as closely as possible to a human body,” says Giorgio Metta, a professor at the University of Genoa in Italy and a longtime collaborator on the RobotCub project.
Ultimately, the EU hopes the work of the iCub team will not only advance the use of robots in industry but also support efforts by researchers in the areas of developmental psychology and cognitive neuroscience to test new hypotheses on cognitive functions in the human brain.
The RobotCub project was first launched in 2004 to compete with research under way in Japan and the United States. What differentiates iCub from its competitors is its modular hardware and software and open-source development. The modular design allows researchers not only to modify or add new components more easily but also to work separately on various aspects of the robot; iCub’s head, for instance, was designed at the Instituto Superior Técnico, in Lisbon, while its legs were developed at the University of Salford, in the United Kingdom. The robot’s open sourcebased software is available to anyone who wishes to download it.
”In Europe, we’ve done our best to catch up to the Japanese, who are still very active in humanoid research,” Metta says. ”But we’ve pursued a different strategy—one based on a single open platform that can be developed by many different groups in many different places. It’s more effective to have a critical mass of researchers tackling an area as complex as this.”
The adoption program is the latest phase of the project. In it, six European research labs, including Imperial College London, Université Pierre et Marie Curie in Paris, and the Technical University of Munich, will each receive an iCub and help train it to learn about its surroundings. Metta expects developers to build more than 15 iCubs for research projects across Europe.
The Imperial College London project focuses on the development of cognitive architectures that use what’s called internal simulation—a robot’s ability to use models of itself and the world around it to anticipate the results of its actions—as the primary method of causal reasoning about everyday objects, explains Yiannis Demiris, senior lecturer at the department of electrical and electronic engineering at the college.
The robots can develop these models in two ways. One is through a process of exploration, in which the robot tries out different actions and learns their consequences. Or the robots can learn by imitation, in which they observe humans or other robots performing tasks and form new models from them, according to Demiris. ”Especially for the imitation part, we will be applying models of the mirror system, found in humans and monkeys, which is activated when an observer executes an action but also when the observer sees another agent executing the same action,” Demiris writes via e-mail. A separate project will provide baby with its first words. Under the newly funded ITALK (for Integration and Transfer of Action and Language Knowledge in robots), project iCub will get an important upgrade. The ITALK team will take the robot toddler’s ability to sort shapes into corresponding holes in boxes and stack wooden blocks to a new level by equipping the robot with the skills to name objects and even its own actions.
About the Author
John Blau writes about technology from Düsseldorf, Germany. He explained German resistance to carbon caps on European cars for IEEE Spectrum and for IEEE Spectrum Online he reported on Europe’s search for alternative peer-to-peer TV technology.