The Berkeley Robot for the Elimination of Tedious Tasks – also known as BRETT – is able to learn from trial and error in much the same way as human beings. The mechanized being is able to perfect motor skills by experimenting and recording what works and what does not, and adjusting its behavior accordingly.
Typically, robots are unable to learn even simple tasks by recording the results of previous actions. They can excel in tasks such as painting car doors on an assembly line or running medical tests. New situations, however, often result in robots being incapable of completing the work they were ordered to perform.
Algorithms in "deep learning" – which is based on the neural circuitry of the human brain – now allow mechanical beings to think more like humans, in what researchers are hailing as a "major milestone in the field of artificial intelligence." This could one day lead to the development of household robots capable of folding laundry, replacing light bulbs, repairing household appliances and more.
University of California Berkeley researchers demonstrated their new system by having BRETT perform a series of tasks, which included assembling a model aircraft and stacking Legos, without receiving detailed information about its surroundings. A score was recorded for each completed step in the process, providing BRETT a means of recording success rates.
"What we're reporting on here is a new approach to empowering a robot to learn. 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," said Pieter Abbeel from UC Berkeley.
Unlike assembly lines and closed medical facilities, objects around the home are often scattered, and move around the house over time. This means that household robots would need to be able to adapt to changing conditions. One method for creating this flexibility is to program a robot with a wide range of possible scenarios, providing instructions on how to carry out tasks in each case.
"For all our versatility, humans are not born with a repertoire of behaviors that can be deployed like a Swiss army knife, and we do not need to be programmed. Instead, we learn new skills over the course of our life from experience and from other humans," said Sergey Levine, a postdoctoral researcher on the project.
Deep learning employs neural nets, consisting of artificial neurons that process incoming data in visual and auditory formats, compiling information in an effort to recognize patterns. This artificial intelligence is similar to the way human brains process information, and may be required for household robots, if they are ever to become commonplace.
The researchers will present the results of their development of BRETT at the International Conference on Robotics and Automation (ICRA) in Seattle on May 28.