Getting a Grip: Building the Ultimate Robotic Hand
From: Wired - 12/2007 - Vol. 15, No. 12
By: Gregory Mone

Enabling robots to handle physical objects means imbuing them with the
"hand-eye" coordination needed to recognize targets, guide their appendages
toward them, and finely manipulate the objects. Such robots must be designed
to learn from the errors they make, and this is the goal of a number of
roboticists building machines that are motivated to explore, fail, and learn
through tactile manipulation, much like a human infant does. The latest robot
developed by the Stanford AI Robot project, Stair 2.0, sports a more advanced
hand than its Stair 1.0 predecessor along with algorithms that allow the
machine to learn without human intervention, recording unsuccessful attempts
to manipulate objects so it will not repeat those same actions. The
University of Massachusetts at Amherst's UMan robot features an algorithm
that helps the machine determine how to operate its hand to manipulate
objects it does not recognize through experimentation, stippling the device's
mental perception of the object with a series of points. The machine measures
changes in the distances between those points as it gets a feel for the
target and deduces how to manipulate it. Meanwhile, the University of Genoa's
Laboratory for Integrated Advanced Robotics has created a humanoid,
five-fingered robot that is programmed to learn to manipulate objects via
study and mimicry of humans performing the same actions, using mirror neurons
as a template for the device's cognitive architecture. Areas with a
demonstrated need for such machines include elder care. 

Read the entire article at:
http://www.wired.com/science/discoveries/magazine/15-12/mf_robothand

Links:
STAIR, the STanford AI Robot
http://ai.stanford.edu/~asaxena/stairmanipulation/

Learning to Grasp Novel Objects
http://ai.stanford.edu/~asaxena/learninggrasp/

Mobile Manipulation
http://www-robotics.cs.umass.edu/~oli/research/mobmanip/

Laboratory for Integrated Advanced Robotics
http://www.lira.dist.unige.it/

BabyBot Hand
http://www.lira.dist.unige.it/babybot/hand.htm