To advance artificial intelligence, robots are simply asking for help to humans. That is the idea advanced by Willow Garage, a company specializing in robotics based in Palo Alto, California. According to the company, asking for help is the second nature of human beings and this behavior could help solve one of the thorniest problems in artificial intelligence.
Until now, these systems designed by different companies can recognize simple shapes, ranging from pens to mugs. But when the parameters were modified (or new lighting angle different example), the reliability rate is literally broke his face. Therefore, it was hardly possible to develop machines capable of moving into a new environment or manipulate a whole bunch of objects. This is exactly the type of problem has met Willow Garage in developing his Personal Robot 2.
But where artificial intelligence is more difficult to learn and recognize new forms, the human mind seems to encounter no difficulties. Thus, Alex Sorokin, a scientist at the University of Illinois has said that it would be a good idea to transfer that advantage to machines. Garage working with Willow, the system put in place based on the “Mechanical Turk” from Amazon, an area for managing small work units distributed to many people.
Specifically, the robot takes a picture of the object not recognized and sends it to the Mechanical Turk. Here, people can use the software Sorokin to surround the item and attach a name beside it. An interesting system, since each processed image is paid between 3 and 15 cents. In preliminary tests, the robot moved the offices of Willow Garage, regularly photographing his environment. A few minutes later, catches were annotated returned to the machine. Result: an accuracy rate of 80%.
If the score is particularly high, Alex Sorokin believes it is probably possible to go beyond 90% if other people verify the responses received. The potential of such a project is very important. For a cleaner example, just to let her understand the new environment to allow then full autonomy in a house. At the slightest concern or unexpected event, such as the presence of a new cabinet, just capture an image and send its contents to be identified.
“This is a fantastic idea,” said John Leonard, a specialist in robot at MIT (Massachusetts Institute of Technology). Potentially, this could allow robots to operate alone for long periods without human intervention.