Artificial intelligence has the potential to grow even smarter with the latest invention from the University of California, Berkley. There, a research team developed an AI algorithm that uses trial and error to learn from its previous mistakes. The robot carrying out the algorithm is named BRETT (Berkley Robot of the Elimination of Tedious Tasks), and it is a PR2 robot from Willow Garage. UC Berkley’s algorithm uses “deep reinforcement learning” to develop an awareness of the robot’s surroundings.
According to researcher Chelsea Finn,
“Moving about in an unstructured 3D environment is a whole different ballgame,” said Finn. “There are no labeled directions, no examples of how to solve the problem in advance. There are no examples of the correct solution like one would have in speech and vision recognition programs.”
Without any pre-programmed knowledge of its environment, BRETT can successfully assemble basic objects after fumbling about for a few minutes. As the robot attempts each successive build, the time it spends fumbling about becomes shorter due to new awareness of its surroundings.
The first task of assembling a toy airplane wheel actually takes a painstaking 12 minutes for the robot to complete, definitely a tedious task. Eventually BRETT applies the same algorithm learned from the toy airplane to Lego bricks. This time, the robot achieves the task almost right away—a success arising from many small failures.
After BRETT’s AI mastered that motion, the researchers applied the algorithm to all kinds of motions, like correctly twisting caps onto a pill bottle and thermos. Perhaps BRETT will, one day, be able to open that tightly sealed jar of pickles that has been sitting in my fridge.
Watch BRETT try, try, try again, below.