Innovation robot handsheking

 

 Innovation-

: Better hands may help robots grasp meaning

 

 is our regular column in which we highlight emerging technologies and predict where they may lead

Two recent studies show that roboticists are applying some fresh thinking to the building and operation of robot hands, and a third suggests why the work is so important – it could be vital for domestic robots learning how to be useful around the home.

Silicon Valley start-up Willow Garage put its PR2 robot on the market earlier this year. It sports two gripper-equipped arms and has demonstrated its ability to use them to fetch a cold beer or fold a towel. But it relies on sophisticated sensors and extensive pre-programming to know how best to grasp an object and how hard to squeeze to maintain a firm grip without causing damage.

Siddhartha Srinivasa at Intel Labs in Pittsburgh, Pennsylvania, and colleagues think they have found a way to do it without the pre-programming. At last week's International Conference on Intelligent Robots and Systems in Taipei, Taiwan, his team discussed using Amazon's Mechanical Turk to help robots pick things up.

AMT is an online service that uses a human workforce to carry out tasks that are simple for people but difficult for computers and robots to complete unaided. It's already been used to help lost robots find their bearings, but Srinivasa's team used it to ask people to annotate images of objects the robot comes across. By adding outlines of objects, and grouping objects, the humans helped the robot to pick up a range of objects including a milk carton and a box of Pop-Tarts.

Too smart?

This may be applying more intelligence to the task than is really needed. Eric Brown at the University of Chicago and colleagues have just developed a "dumb" robot gripper that can pick up a range of unfamiliar and even delicate objects with no prior knowledge of them.

"Our gripper is simpler because it does not need tactile sensing, a computer, or precise vision, other than the need to know the general location of the object," says Brown, who developed the device with a team from robot manufacturer iRobot, based in Bedford, Massachusetts, and the Pentagon's Defense Advanced Research Projects Agency.

The "hand" is simply a rubber bag filled with small glass spheres that flows around the object to be picked up. A pump then creates a vacuum in the bag, jamming the spheres in place and hardening the robot gripper around the object. Because the gripper moulds itself around the object before hardening, its force is distributed evenly across a large surface area, meaning it can pick up even delicate objects like raw eggs without crushing them, says Brown.

Ikea test

Robert Platt of the Computer Science and Artificial Intelligence Laboratory at the Massachusetts Institute of Technology says the robot gripper would be great for picking up difficult objects like small nuts and bolts on an assembly line, for example, or turning a door knob in your home – but he doubts whether a dextrous hand alone will lead to more intelligent robots. "It wouldn't be able to put together a piece of Ikea furniture," he says.

The members of a Europe-wide project called Paco-Plus might disagree. Co-ordinator of the project Tamim Asfour at the Karlsruhe Institute of Technology in Germany says picking up and handling objects is central to robot learning. The strategy should sound familiar – such manual exploration is used by babies to learn about their bodies, the objects around them and their environment.

"Our robots are able to explore the environment and learn about objects and the actions they can apply to them," he says. One of the project's robots – Armar III – has now gleaned enough knowledge from its kitchen-like environment to learn how to open a dishwasher door, pick up a cup, turn it upside down and put it into the machine.

Read previous Innovation columns: How to delete corporate logos from view, Online army turns the tide on automation, What's the right path for indoor satnav?, TV networks to become social networks, CERN collides with a patent reality, Sunrise boulevards could bring clean power, Hand-held controls move out of sight, Mobile malware develops a money bug, Reinventing urban wind power.