Scientists teach robot to learn new skills via trial and error Scientists at University of California, Berkeley have taught robots to learn. New technique, called �deep learning�, is a system of algorithms that enable robots to learn motor tasks through trial and error. It is very similar concept to how humans learn behaviours and get to know how things work around them in early childhood. It is already called a major milestone in the field of artificial intelligence.Scientists posing with robot they nicknamed BRETT. It is a major milestone in the field of artificial intelligence as this is robot that can learn by itself, bringing idea about robots helping around the house closer. (Photo courtesy of UC Berkeley Robot Learning Lab)The concept of robot learning new things by itself has been puzzling scientists for quite some time. But now they can demonstrate their achievements. They showed what robot can do by making it complete various tasks � putting a clothes hanger on a rack, assembling a toy plane, screwing a cap on a water bottle, and more. Everything was done without pre-programmed details about surroundings of the robot.Professor Pieter Abbeel of UC Berkeley Department of Electrical Engineering and Computer Sciences said it is a new way to empower robot and make it learn without changes to software. ��The key is that when a robot is faced with something new, we won�t have to reprogram it. The exact same software, which encodes how the robot can learn, was used to allow the robot to learn all the different tasks we gave it�, he said.This is very important step in robotics. Robots are usually tested and demonstrated in controlled environments. Objects are where robot expects them to be and there are not many obstacles to overcome. However, if robots are going to assist us in our homes, they have to be prepared to operate in environment that is constantly changing. You cannot reprogram your robot then you placed a chair in a different place, it has to learn to walk around it by itself.This learning ability did require tremendous amount of programming anyway. Usually, robots have to be programmed to handle the vast range of possible scenarios. Now scientists are trying �deep learning� method, which is loosely inspired by the neural circuitry of the human brain when it perceives and interacts with the world. Humans are not born pre-programmed with possible scenarios, so a robot needs to be able to learn too.Deep learning programs create �neural nets� in which layers of artificial neurons process overlapping raw sensory data, whether it is sound waves or image pixels. It helps robot to categorize new object and patterns, learn how to behave around them. Deep learning is already used by programs, such as Siri on iPhones, Google speech-to-text program or Google Street View, but learning to accomplish motor tasks has proved to be far more challenging.BRETT demonstrates its abilities to learn without pre-programmed knowledge about its surroundingsLike in many cases, practise here makes perfect. Scientists were testing their software with a Willow Garage Personal Robot 2 (PR2), which they nicknamed BRETT (Berkeley Robot for the Elimination of Tedious Tasks). The algorithm controlling BRETT learning included a reward function that provided a score based upon how well the robot was doing with the task. Robot was presented with different tasks, such as placing blocks into matching openings or stacking Lego blocks � almost like a child playing with educational toys.BRETT observes the scene, position of his hands. Then tries to move � movements that bring the robot closer to completing the task will score higher than those that do not. Robot will learn actions with highest score and will keep this technique for the future encounters with similar tasks. BRETT could master a typical assignment in about 10 minutes, but when he is not given the location for the objects in the scene and needs to learn vision and control together, the learning process takes about three hours.With more data robots will soon be able to learn much more complex things. It is a tremendous leap forward in robotics and artificial intelligence, bringing idea of house robots a little step closer. Even though learning processes are still not perfect, this shows that future might be just as we imagined it � with robot butlers.Source: UC Berkeley