This Robot Keeps Trying Until He Gets it Right
Robot at Berkeley learns via Trial and Error.
BRETT (Berkeley Robot for the Elimination of Tedious Tasks) is a machine that takes a mundane task such as opening a bottle and learns to do it well by attempting multiple times. Using neural network based deep learning, it’s algorithm helps it learn from its failure and therefore find success. This is mirrors curiously close to the developmental process of the human brain evolving the idea of a program into an assistant that wouldn’t need a new set of code to perform a new task, but rather enough time to discover how to do it effectively.
BRETT is very much still an infant however. On average it can take him up to 3 hours to learn even the simplest tasks but the technology is promising and a very small grain on the milestone that will be self-teaching A.I.
Feedback or more from the writer: angeldamion.com