To people in the household really help, you must first learn robot better cope with unknown situations. The problem: While researchers are now increasingly intelligent machines before programmer to even the most complex tasks to solve, but in an unstructured environment is hardly helps. At the University of Massachusetts Amherst has now developed a robot that also can use items, which he never before encountered. The UMass Mobile manipulator “, shortly UMan, pushes new objects on a table around, to see how they move. Is it appropriate parts covered, he experimented with them – and then manipulate them according to new tasks. “It’s similar to when a baby imagine that a new toy discovered and all existing parts touch to find out how things move,” said Dov Katz, master’s student at the university and principal author of the study. He led the investigation along with his computer science Professor Oliver Brock. “One of the challenges in robotics is always an intelligent robot can work, even if it’s the shape of an object not previously known,” says Andrew Ng, computer scientists at Stanford University and robot-known researchers. “I think this work is an important step in this direction.” If someone is currently one wanted to teach machines, a pair of scissors to use, he should be much to write software – code that defines what exactly is a pair of scissors and how the two blades together. “Katz and Brock are now a completely new approach, in which the robot itself with the scissors play and find out how the two blades are connected.” UMan uses a normal webcam, to move from the top of the table to look. The analysis of the differences between adjacent pixels it detects where the edges of an object is. Then he prods the subject and changes based on the resulting need for movement, if necessary, the image that he’s been made. Then he plays more with the object and watching how its individual parts behave in relation to one another. The object is to the front and rear pushed to the width and length, and also at an angle of 45 degrees on both, if necessary, to the movement to fully understand. Wherever no longer possible movement, identified the robot then a joint. UMan then uses all this information to find out how the object is best manipulated. The robot can also determine if there are multiple joints, and how these relate to each other. Katz openly admitted that his team of the work of researcher Paul Fitzpatrick has inspired. The scientist LIRA-Lab at the University of Genoa allows the robot also an object “tap” to distinguish it from its visual background. “At the Amherst study I like that they come from virtually the same action headshot more information,” he says. This is the robotic equivalent of a human “Herumfummeln” on a subject to find out how to use it. Until now UMan can not lift objects – instead, they are directly on the table surface manipulated. The machine has already successfully learned how he scissors and various wooden toys around. UMan is somewhat smaller than a human being built – his single arm is about one meter long. It can be divided into seven degrees of freedom to move, which makes it similar flexibility as a human arm do, says Katz. The arm is in turn a hand with three fingers, which can rotate. The researchers expect that his experience gained UMan soon to take advantage of new objects. In a computer simulation has been a learning process tested with the machine at the sight of similar objects derived handle this. “If he has learned as a pair of scissors works, it recognizes the sight of a Takers that a similar structure.” In the simulation UMan joints could already see where it an object in a direction expressed – currently it takes six. Katz plans, however, go even further: UMan will soon contact solely on its image recognition. The developed algorithms to be next year in hardware. “This work provides me a step forward – to a men schenahnlicheren process, as objects of robots manipulated, experienced and recorded,” says Josh Smith, an Intel Grief sensors working. The UMan-philosophical approach was also interesting because it is perception and sensors combined with physical manipulation.