A Robot With Beanbag Hands Learns the Gentle Touch

[one...two...tree 2,762]

The simple act of picking something up requires a plethora of decisions: Is the thing light or heavy? How much force do I need to grip it? If I grip too hard, will I crush it with the might of my mighty hands?

As we grow up we become fairly practiced at the art of picking up (objects, that is), so our brains will do most of this for us without a lot of conscious thought. But all those variables—plus adapting to a surprise on the fly—mean that picking things up with the proper force is one of the most difficult skills to teach a robot. That's why the design by Eric Brown's team is so clever.

In this week's Proceedings of the National Academy of Sciences, Brown and colleagues demonstrate (paper in press) their "universal gripper," a successful prototype of a robot hand. It's based on an idea that's been around for a while, and it looks like a beanbag on a robot arm, because, well, that's kind of what it is.

"Our gripper is simpler because it does not need tactile sensing," says [brown]. His team has created a robot hand that is simply a rubber bag filled to about 80 per cent of its volume with glass spheres, each 100 micrometres in diameter. [New Scientist]

A beanbag typically grips our bodies when we're being lazy, but the advanced version created by Brown's team (which includes scientists from DARPA and iRobot, the Roomba inventors) can squeeze those glass spheres around a host of different objects.

The gripper is placed over objects, deforming around them. The air in the bag containing the granules is then removed by suction, causing the granules to pack tightly together – known as jamming – and the hand to grip on to the object. No computation is required, although at present a researcher must decide when to evacuate the air, and the gripper is able to lift a diverse selection of objects, from a car's shock absorber to a raw egg. The hand is even capable of lifting and pouring a glass of water, or of picking up a pen and using it to draw. [Nature]

But will the simple universal gripper win out over robotic hand designs that feature human-like fingers?

The moldable gripper can function without needing sophisticated artificial intelligence software. Fingered robotic hands may be more suitable for manipulating objects but require "a central processor or brain for a multitude of decisions, many of which have to be made before the hand even touches the object, for example about how wide to spread the fingers apart," according to the researchers. [Scientific American]

http://blogs.discove...e-gentle-touch/