In a new finding, researchers claim that they have developed a robot that doesn’t need explicit instructions to do a task like other robots require and can learn and complete a task by trial and error method like humans do. The newly designed robot can learn simple tasks like opening a cap of bottle and remember how it is done for the next time.
Researchers at the University of Berkeley named the robot as Berkeley Robot for the Elimination of Tedious Tasks (BRETT). The BRETT uses neural-based deep learning algorithms to learn tasks like humans do. This enables BRETT to better adapt to surrounding situations and complete a task with less prior knowledge when compared to other robots. Also, once learned it remembers the technique to complete the task so it takes less time to complete the same task in future.
To make BRETT complete the task researchers at Berkeley added a reward function to its learning process. Researchers assigned several tasks including stacking LEGO bricks, screwing on the cap of a bottle and fitting wheels onto a toy airplane. Every time robot succeeded or came closer to complete the task it was given a higher reward point. This way robot learned from trial and error method.
Time taken by BRETT was dependent on the amount of information to be learned while completing a task. Scientists found that it took 10 minutes to learn a task when exact coordinates of the object were given or it was told where to start and where to finish at. However, it took more than three hours to learn a task where it needed to figure out coordinates or position of the object itself.
Scientists at Berkely are optimistic that the BRETT is just a start and it will take another 10 years to make robots learn several complex task autonomously without human intervention. Also, they believe that one-day robots will become the robots depicted in movies like iRobot.