Sometimes, what seem like mundane tasks to us are extremely difficult for a computer; sure, if a computer is programmed to do a single specific task, it can probably do that task perfectly, but asking it to do anything it isn’t programmed to do will near always result in failure.
Cue learning algorithms.
Researchers at the University of California-Berkley have enabled robots to mimic human learning; that is, learning through trial and error. This means that robots can, in a way, program themselves to do menial tasks like assembling toys or hanging clothes.
Pieter Abbeel: “What we’re reporting on here is a new approach to empowering a robot to learn.”
Co-researcher Trevor Darrell added that “The challenge of putting robots into real-life settings, like homes or offices, is that those environments are constantly changing. The robot must be able to perceive and adapt to its surroundings.”
A new branch of artificial intelligence called deep leaning creates “neural nets” that serve as artificial neurons that allow machines to process raw data from their sensors, such is images or sounds, then categorizes these data into patterns.
Researches used a Willow Garage Personal Robot 2 which they named BRETT in an experiment to learn how long it would take BRETT to learn simple task like matching block shapes.
On average, it took BRETT about 10 minutes to master each task if it was given the coordinates. Without being given coordinates, it would take BRETT at least three hours to learn.
The findings will be published on May 28 at the International Conference on Robotics and Automation (ICRA) in Seattle.
Will Hawking be there? I doubt it.