Levinson Group Only U.S. Lab Awarded Highly Advanced iCub Robot
The “adoption” process is almost complete, “parenting” classes have been taken and now all that is left is to wait for the new addition to the “family” to arrive from Europe. Adoption, parenting, and family are all analogies in the case of the new addition because it is built of metal and plastic but made to act, respond, and learn like a human child.
Sometime this fall the Beckman Institute will become home to a unique new student from Europe – the only one of its kind in the United States – in the form of a highly-advanced humanoid robot. If all goes as planned, in late October an “iCub” robot from a European “robot consortium” will go through the final phase of its adoption process and join Artificial Intelligence group member Stephen Levinson’s Language Acquisition and Robotics Laboratory.
The iCub will learn about language as a child does: through interactions with the world around it, by learning about its own body movements and motor skills, and by absorbing knowledge from its “parent-teachers”, aka the members of Levinson’s lab.
The iCub, as described by the RobotCub project that created it, is a “full-fledged humanoid robot ‘child’ with sophisticated motor skills and several sources of sensory information.” Levinson’s lab was one of seven research entities awarded a free iCub – estimated to cost about €200,000 – in an open call competition put out by the RobotCub project. The robot is so special in its capabilities that the Department of Defense wanted one to share among its research groups, but its proposal wasn’t accepted. The proposal from the Language Acquisition and Robotics Laboratory was accepted, thanks to Levinson’s area of research.
“People all over the world are clamoring to get these robots,” Levinson said. “The fact that we’re getting one is amazing.”
Not only will Levinson’s lab have the only iCub in the United States and North America, it will also be the only research group in the world using one to study language acquisition. That fact, as much as anything, is why the lab’s proposal was selected to receive one of the humanoid robots.
“That is the thing that really attracted them,” Levinson said. “We were very specific about language acquisition.”
When it arrives, this new addition to Levinson’s robot family will mark a huge leap up the robotics evolutionary ladder for the lab. The lab’s current lineup of three robots, known as Trilobots, feature wheels, sensors, and platforms for electronic equipment, but have limitations on their sensing and learning capabilities. The concept behind the iCub is to use its human-like motor skills, physical and sensing capabilities, and highly advanced open source software programming to recreate the learning processes of young children.
“The idea is that the richer the sensory motor periphery of the robot, the more elaborate learning that can take place,” Levinson said. “We simply want to experiment on that basis.”
Levinson said the new robot can reach, grasp, feel, and look with enhanced sensing abilities that most other robots, including his lab’s trio of Trilobots, just don’t possess. The new robot will not only be able to pick up a ball and learn its meaning, but also learn to distinguish properties such as whether the ball is hard or soft and rough or smooth, as a child does.
“The visual sense, the enhanced proprioception of the hands and arms, and even the legs, all of this provides a much richer input from which we hope we can learn a great deal more,” Levinson said. “Therefore there is much more information coming in and that allows it to build a more elaborate mental model of the world.”
– Stephen Levinson
The RobotCub project is associated with the second European Network for the Advancement of Artificial Cognitive systems, Interaction and Robotics, or EUCogII for short, and funded by the European Commission. EUCogII is a European network for researchers in artificial cognitive systems and related areas that funds educational outreach programs.
The mission of RobotCub, which bills itself as an international project on humanoid cognitive systems, from its beginning in 2004 was to create a robot that functions as closely as possible to a human child in order to advance our understanding of how it is humans learn. The RobotCub project’s stated goal is to develop a cognitive humanoid robot that is an “open-systems research platform for enactive cognition” toward promoting and enhancing research.
Toward that goal, robots have been adopted into labs across Europe to advance research in areas of artificial cognitive systems, cognitive neuroscience, and developmental psychology. The RobotCub project put out an open call in 2007. Levinson met David Vernon, then head of EUCogII, at a conference and after hearing about his work on robotics and language acquisition, Vernon suggested he send in a proposal.
Giorgio Metta, an Assistant Professor at the University of Genoa and member of RobotCub, said the seven winners of the open call prize were given an iCub free of charge for 99 years to pursue research.
“Stephen’s proposal was chosen by a panel of six experts as worth receiving one of the iCubs,” Metta said. “We will be delivering the iCub before the end of 2009, as soon as we can in fact.”
Learning how to operate and program an iCub to get the most out of it is not easy. So graduate students Lydia Majure and Logan Niehaus of Levinson’s lab got the enviable assignment of spending part of their summer on the Italian Riviera – learning the ways of a humanoid robot by day and getting their fill of squid and octopus at night.
The robot consortium, consisting of several European universities and laboratories including the Italian Institute of Technology and the Laboratory for Integrated Advanced Robotics (LIRA) at the University of Genoa, sponsors the school, which trains attendees on operating and programming the iCub robot. The robot consortium provided the robots, the training, and even the hotel and meals in the fishing town of Sestri Levante on the Italian Riviera where the school was held.
“Because of where we were it was seafood every night,” Niehaus said. “I had never had squid or octopus before and we were having that close to every other day.”
Majure loved the fare.
“I was thrilled. I love consuming cephalopods, especially when they’re fried,” she said. “Logan was complaining by the end about wanting some chicken, though. To each his own I suppose.”
But even the delicious sea creatures took a back seat for Majure to her first encounter with the robot.
“The most exciting part of the school was actually seeing the robot for the first time,” Majure said.
The iCub robot is a metallic and plastic version of a human toddler, with anthropomorphically correct metallic body parts covered by a plastic “skin”. It can sit and crawl, has 53 degrees of freedom (compared to 20 degrees for the lab’s current robots) in its articulated hands for manipulating objects and gesturing. It has approximately the same dimensions as a two-and-a-half year-old child (around three feet tall and weighing approximately 30 lbs.) and its eyes, ears, and hands have sensors that provide sight, sound, and touch.
– Lydia Majure
The training school consisted of working on project development for the robot’s software, and included writing some software code that could be useful later on when the robot arrives and the Levinson lab starts doing experiments.
“They want to make a system that is neurologically and psychologically feasible,” Niehaus said of the RobotCub project. “Over the past four years they have slowly built this up and revamped it. We put a small piece in that is part of a larger project.
“We learned how to program for this robot system, which actually turned out to be extremely simple. We can focus on research and not writing software, at least not worrying about the nuts and bolts of programming this robot.”
Majure said the training they got in Italy will allow the lab to start taking advantage of the robot’s capabilities almost as soon as it is delivered.
“It was really valuable for learning how to use the robot and the associated software interfaces,” she said. “That experience probably saved us a ton of time in learning the basics of the platform. Now, when the robot arrives this fall, I’m hoping that we can get straight to work on our research.”
Majure said that the robot’s motor skills and sensory capabilities will allow the lab to explore new lines of research.
“There are many new capabilities that this robot has which will increase our ability to research cognition,” she said. “For example, due to being able to move its head and track objects, we can build multisensory models of its immediate surroundings. We can investigate motor babbling (learning kinematics by random exploration of motion range) and learning by object manipulation.”
– Stephen Levinson
Levinson said funding for the trip came from Illinois graduate Bruce Wonnacott and his venture capital fund. Two more students from the lab will go to Italy in October. Levinson was hoping to have the robot delivered for his Director’s Seminar talk discussing the new addition on Sept. 16, or by the Beckman Institute 20th Anniversary Symposium Oct. 5-7.
“What Giorgio has told me repeatedly is that they can’t be rushed,” Levinson said. “They’ve got to do the assembly carefully and they’ve got to test it.”
Once the iCub is delivered and gets used to its new surroundings at Beckman, experiments will begin.
“This is important and should be emphasized: we are not going to program the robot to do anything. We’re going to program it to learn to do things,” Levinson said. “What we’re interested in doing is to understand how the brain implements the mind and how the mind controls activities and how those activities can be made purposeful. We are not trying to do any specific thing.”
One thing the lab will try to do is to program it to learn how to walk, something none of the labs around the world have tried to do. Currently an iCub can stand with a framework attached to prevent it from falling; programming one to learn how to walk was a task the robot consortium asked of Levinson’s lab.
“We made a commitment to EU cognition that we were going to try, if possible, to get the robot to learn how to walk,” Levinson said. “I know the robot hasn’t learned to walk but I don’t think anybody has even written a program to make it walk. That is one thing that we have not done before but we will try to do.
“It is going to do many things, all of which merge together with the common goal of language acquisition.”
Even the task of programming an iCub to learn how to walk has ramifications for language acquisition in Levinson’s lab.
“We’re going to try and do a complete sensory motor integration so that even things like walking will have a great deal to do with language,” Levinson said. “Walking is a form of spatial reasoning and spatial reasoning is an important underlying principle of natural language.
“Walking is a goal in the sense that people walk so it would be nice if the robot walked. But what we’re really exploring is the adaptive control mechanism that would allow stable walking and ultimately allow us to develop a sense of spatial reasoning from that. It is this basic underlying theory that we are trying to develop.”
Levinson has been engaged in research, either in industry at Bell Labs, or at the University of Illinois, for more than 30 years.
“I can honestly say that this is the most exciting event in my entire research career,” he said. “Unqualified, I can say that it is. And I’ve been at it for awhile.”
This article is part of the Fall 2009 Synergy Issue, a publication of the Communications Office of the Beckman Institute.