As a rule, robots have to learn through explicit instruction, whether it's through new programming, watching videos or holding their hands. UC Berkeley's BRETT (Berkeley Robot for the Elimination of Tedious Tasks) isn't nearly that dependent, however. The machine uses neural network-based deep learning algorithms to master tasks through trial and error, much like humans do. Ask it to assemble a toy and it'll keep trying until it understands what works. In theory, you'd rarely need to give the robot new code -- you'd just make requests and give the automaton enough time to figure things out.
As you might suspect, though, this brain-like 'bot isn't ready for the real world yet. It takes 10 minutes to learn a task when you tell it exactly where it needs to start and stop, and 3 hours if it has to learn those positions itself. BRETT isn't drawing from a wealth of experience, as you do, so it doesn't make those logical leaps that help you grasp a concept quickly. With that in mind, the researchers are optimistic that the technology will improve dramatically over the next several years as robots get better at handling lots of data. Eventually, artificial intelligence could be good enough that robots would be ready for anything their designs allow.