
BEAUTY may be only skin deep, but for humanoid robots a fleshy covering is about more than mere aesthetics, it could be essential to making them socially acceptable. A touch-sensitive coating could prevent such machines from accidentally injuring anybody within their reach.
In May, a team at the (IIT) in Genoa will dispatch to labs across Europe the first pieces of touch-sensing skin designed for their nascent humanoid robot, the . The skin IIT and its partners have developed contains flexible pressure sensors that aim to put robots in touch with the world.
“Skin has been one of the big missing technologies for humanoid robots,” says roboticist Giorgio Metta at IIT. One goal of making robots in a humanoid form is to let them interact closely with people. But that will only be possible if a robot is fully aware of what its powerful motorised limbs are in contact with.
Advertisement
“Skin has been one of the big missing technologies in the designing of humanoid robots”
Roboticists are trying a great variety of ways to make a sensing skin. Early examples, such as the CB2 robot, built at Osaka University in Japan, placed a few hundred sensors in silicone skin. But now “many, many sensing methods are emerging”, says Richard Walker of , London. Until a lot of robots are using them, it is going to be hard to say which are best suited for particular applications.
What’s more, there are many criteria the skin has to meet, says Metta: it must be resilient, able to cover a large surface area and be able to detect even light touches anywhere on that surface. “Many of these factors conflict with each other,” he says.
The iCub is a humanoid robot the size of a child of three-and-a-half years old. Funded by the European Commission, it was designed to investigate cognition and how awareness of our limbs, muscles, tendons and tactile environment fuels the development of intelligence. The iCub’s technical specifications are open-source and some 15 labs across Europe have already “cloned” their own, so IIT’s skin design could find plenty of robots to enwrap.
The skin is made up of triangular, flexible printed circuit boards which act as sensors, and it covers much of iCub‘s body. Each bendy triangle is 3 centimetres to a side and contains 12 capacitive copper contacts (pictured). A layer of silicone rubber acts as a spacer between those boards and an outer layer of Lycra that carries a metal contact above each copper contact. The Lycra layer and flexible circuits constitute the two sides of the skin’s pressure-sensing capacitors. This arrangement allows 12 “tactile pixels” – or taxels – to be sensed per triangle. This taxel resolution is enough to recognise patterns such as a hand grasping the robot’s arm. The skin can detect a touch as light as 1 gram across each taxel, says Metta. It is also peppered with semiconductor-based temperature sensors. This version of the skin will be released in May.
Later, IIT plans to add a layer of a piezoelectric polymer called PVDF to the skin. While the capacitance sensors measure absolute pressure, the voltage produced by PVDF as a result of its deformation when touched can be used to measure the rate of change of pressure. So if the robot runs its fingertip along a surface, the vibrations generated by friction give it clues about what that surface is made of. Such sensitivity might help it establish the level of grip needed to pick up, say, a slippery porcelain plate.
Philip Taysom, CEO of British company of Richmond, North Yorkshire, is not a fan of sensing skins based on capacitors, which he says can lose sensitivity with repeated use. Peratech’s answer is a stretchy, elastic material it calls quantum tunnelling composite (QTC). This comprises a polymer such as silicone rubber that is heavily loaded with spiky nickel nanoparticles. A voltage is applied across the skin, and when it is pressed, the distance between the nanoparticles within the polymer diminishes, which results in electrons flowing, or “tunnelling”, from one nanoparticle spike to the next in the area being touched. Crucially, the material’s electrical resistance drops dramatically and in proportion to the force applied, so the touch can be interpreted.
At the Massachusetts Institute of Technology’s Media Lab, is developing a QTC-based sensing skin for a commercial robot-maker which he declines to name. Instead of a tight, conforming skin, Whiton uses a looser covering, more akin to clothing. “We cover ourselves with textiles when we interact with people, so clothing may be a better metaphor as a humanoid’s pressure-sensitive surface covering,” he says.
Natural gestures, like tapping a humanoid on the back to get its attention, or leading it by the arm, can be easily interpreted because QTC boasts high sensitivity, he says. But novel skin capabilities could be on the way, too. For example, QTC can also act as an electronic nose. Careful choice of the material’s base polymer, says Taysom, means telltale resistance changes can be induced by reactions between volatile chemicals in the air – so it can become an e-nose as well as a touch sensor, able to detect, for example, a gas leak in your home. “This shows we can probably build into robots a lot of things that our skin can’t do. It’s another reason not to stick rigidly to the human skin metaphor,” says Whiton.
That’s not to say our skin isn’t a great influence. Shadow Robot will soon start testing a novel human-like touch-sensing fingertip from , a start-up based in California. Its fingertip comprises a rubbery fluid-filled sac that squishes just like a real fingertip, and is equipped with internal sensors that measure vibration, temperature and pressure.
Whichever of the emerging technologies prevail, sensing robot skins should help us get along with our future humanoid assistants, says Whiton. “Right now, robots are about as friendly as photocopiers. The interactions skins encourage will make them much friendlier.”
Parlez-vous robot?
If you’ve ever tried to direct a lost tourist to their intended destination, you’ll know how difficult directing someone that doesn’t speak your language can be. Directing robots presents a related challenge.
Typically, robots respond well to precise instruction sets but they are flummoxed if their instructions are given in the fuzzy, everyday language so beloved by humans. Now a team at the University of Washington in Seattle have developed translation software which could enable robots to understand a set of natural-language directions. The technology could make it easier to control robots in situations like search and rescue, where it can be preferable to send a robot rather than a human.
and her colleagues used the principles of machine translation – commonly used online to translate text of one language into another – to develop a navigation program for robots. Machine translation tools are designed to learn from previous efforts, improving their accuracy through experience.
The team first sent a small mobile robot to explore and map portions of two buildings on campus. The researchers then generated random paths through the maps and asked human volunteers to annotate the routes with natural commands, such as “turn right”, or “take the second left” that would have led to successful completion of each path.
Matuszek used these maps to train the navigation program, which learned to associate the various human commands with specific types of route-finding behaviour.
The navigation program was then run on a virtual robot, which was given natural-language directions for a variety of previously unknown routes through the maps. The virtual robot was able to successfully complete 10 of the 14 direction sets on its first attempt. The results were presented at the International Conference on Human-Robot Interaction in Osaka, Japan, in March.
“I’m glad to see work that is getting back to the original dreams of the field, like having a robot that you can talk with naturally,” says , a machine-translation researcher at the University of Texas in Austin.
He says that previous attempts to give robots instructions have favoured explicit rules for sentence structures, semantics and syntax. He says this “traditional” approach is labour intensive and requires strictly defined commands, making it cumbersome in emergency situations where many robots are deployed.
MacGregor Campbell