Czech scientists work to give open-sourced iCub robot the “human touch”
Scientists from the Czech Technical University in Prague (ČVUT) have an unusual, constantly evolving colleague – a humanoid robot named iCub. While not the first such robot at the university’s Faculty of Electrical Engineering, iCub is the first equipped with all key senses – sight, sound and touch.
The pearly white iCub robot is about as tall as a four-year-old child. Its joints and muscles are controlled by 53 electric motors, two cameras act as its eyes, and two microphones serve as its ears.
But what sets iCub apart from other humanoid robots is its sense of touch, which is made possible by 4,000 sensors embedded in its electronic skin, says robotics researcher Lukáš Rustler.
“It can hear and see, which is basic, but at the same time, it has so many sensors on its body, it has the sense of touch.
“Many robots can move, and many robots are humanoid. This one is different in that it has ‘skin’, which is comprised of triangles, each with 12 sensors that act like a smartphone screen.
“It can also do yoga or catch a ball, for example. My colleagues are working to teach him to recognize people, so to understand where a person is and, for example, play board games with them.”
That electronic skin, along with the robot’s size, also makes iCub ideal for simulating processes by which babies acquire tactile knowledge of their surroundings – including learning about their own bodies.
Dr Matěj Hoffmann heads the humanoid robotics group at the Czech Technical University, which acquired iCub to advance research into human brain cognition through artificial intelligence.
“We’re interested in how a new-born brain learns to use its body, to understand its dimensions, what it can do, how to reach for objects, and the like. Primarily, we’re interested in the first and second years of life, how quickly a child learns to do a huge number of things from scratch.”
The iCub is an open-sourced, humanoid robot, with the hardware design, software and documentation all released under a General Public License. This means Czech Technical University researchers have the freedom to run, study, share, and modify the software for their own purposes.
Dr Hoffmann and his colleagues have used that freedom to create a neural network for their iCub. It draws on an algorithm directly inspired by the functions of the human brain, which they have studied in collaboration with developmental psychologists and neuroscientists from abroad.
“We’re studying how the human brain works, which is far more adept than any robot or computer, and will remain so for a very long time. We’ve traced how the brain receives tactile signals through the skin and processes them. Based on that, we created our own neural network, running throughout the ‘skin’ of the iCub.”
Human intelligence is inseparable from the physical body and its sensory and motor systems. That’s why computational models alone are not enough, says Dr. Hofmann, and the Czech researchers want to help humanoid robots learn to “calibrate” themselves to adapt to their surroundings.
“Humanizing” robots in this way can help them move more autonomously and safely in populated environments – and help them help humans in a variety of ways.