Robots cannot usually learn by recording the outcomes of the previous actions; they cannot do this even when it comes to simple tasks. They can work in controlled environments such as medical labs, but when they are put in new situations the robots are not able to perform the task which they are required. Scientists at the University of California have developed an algorithm which allows robots to learn motor tasks by using trial and error. This is precisely the method which people use to learn new tasks. This discovery represents a major milestone in the domain of artificial intelligence. The project will be presented at the International Conference on Robotics and Automation (ICRA) which will take place on May 28 in Seattle.
The algorithm employed by the researchers is called deep learning. It has the neutral circuitry of the human brain at its base. As a consequence it enables mechanical beings to think similarly to the way humans do. This can one day allow scientists develop household robots which can help with doing chores which individuals would rather skip such as repairing household appliances, replacing light bulbs and folding laundry.
Professor Pieter Abbeel from Berkeley’s Department of Electrical Engineering and Computer Sciences (University of California) was the lead author of the study. According to him what they did was a new approach to enabling robots to learn. He also added:
“The key is that when a robot is faced with something new, we won’t have to reprogramme 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.”
The robot was named BRETT, which stands for Berkeley Robot for the Elimination of Tedious Tasks. In order to test its learning abilities the researchers required it to do different task without pre-programming details about the environment. The tasks which BRETT successfully performed included screwing the cap on a water bottle, placing a clothes hanger on a rack and put together a toy plane.
Households are different than closed medical facilities and assembly lines. The objects inside the house are scattered and they can change their place over time. As a consequence household robots would have to adapt to changing conditions. The neutral nets which deep learning uses employ artificial neurons which analyze the incoming data both in auditory and visual formats. It puts data together in order for pattern recognition to be possible. This type of artificial intelligence is very similar to how the human brain functions.
Image Source: topix.com