Finding and recognizing objects by touch in your pocket, in the dark or among items on a cluttered table top are distinctly human skills -- ones that have been far beyond the ability of even the most dexterous robotic arms.
Rodney Brooks, a well-known roboticist, likes to demonstrate the difficulty of the challenge for modern robots by reaching into his pocket to find a particular coin.
Now a group of roboticists in the Department of Biomedical Engineering at the Georgia Institute of Technology in Atlanta, led by one of Mr. Brooks' former students, has developed a robot arm that moves and finds objects by touch.
In a paper published this month in the International Journal of Robotics Research, the Georgia Tech group described a robot arm that was able to reach into a cluttered environment and use "touch," along with computer vision, to complete exacting tasks.
This ability is vital if robots are to leave the world of factory automation and begin to undertake tasks in human environments, like patient and elder care or rescue missions during emergencies.
"These environments tend to have clutter," said Charles Kemp, the director of the Healthcare Robotics Lab at Georgia Tech and Mr. Brooks' former student. "In a home, you can have lots of objects on a shelf, and the robot can't see beyond that first row of objects."
The development is part of a wide range of advances in the last two years that foretell a world in which robots will move freely in human environments, to be able to work near them and with them.
For the safety of workers, industrial robots are either kept in metal or glass cages, or protected from humans by "light curtains," which cause the robots to stop if a human approaches.
That has begun to change with a new generation of robots from companies like Rethink Robotics in Boston, and Universal Robots in Denmark, that make robot arms that can operate safely in proximity to human workers.
Robots, guided by machine vision, have also been limited by their inability to reach into spaces, the way living creatures can, to pick out an object. They are, in fact, programmed to avoid contact.
"We're flipping that on its head," Mr. Kemp said. "Let's say contact with the arm is fine, as long as the forces are low."
The Georgia Tech researchers have produced a robot arm that can reach and then use software to control its sense of touch, making it possible to find specific objects in a collection or area.
Mr. Kemp said the researchers were able to achieve success, both with a robot and with digital simulations, after a relatively small series of attempts, and using a simple set of primitive robot behaviors.
The algorithms used gave the arm qualities that seemed to mimic human behavior. For example, the robot was able to bend, compress and slide objects. Also, given parameters designed to limit how hard it could press on an object, the arm was able to pivot around objects automatically.
The arm was designed to essentially have "springs" at its joints, making it "compliant," a term roboticists use to define components that are more flexible and less precise than conventional robotic mechanisms. Compliance has become increasingly important as a new generation of safer robots has emerged.
The robot also has an artificial "skin" that can sense pressure or touch.
"If we look in biology, it's not just compliance at the joints that all of these organisms have," he said. "They also have good sensing of forces across their entire body."
The researchers built their software for a simulated "cluttered" world and for an operating robot. The robot's arms were designed by Meka Robotics, a San Francisco company that makes a variety of robot components. The software is based on the Willow Garage Robot Operating System, or ROS, which is intended to be shared freely.
The Georgia researchers have made their software open source as well, and shared instructions to make and adapt low-cost robot skin, in the hopes that other robot makers will improve and advance what they have done.
The research was financed by the "Maximum Mobility and Manipulation" program at the Pentagon's Defense Advanced Research Projects Agency.
The researchers chose not to describe the new robotic behaviors as "emergent" -- a term used for a complex system created from simple component functions -- but Mr. Kemp said it was a reasonable description.
"To me, it does seem like sort of emergent intelligence, with the robot moving through this complex environment," he said.