There's a great video from Pieter Abbeel, professor at Berkeley, where they show a robot doing all types of household tasks. Accompanying it, he notes that everything shown in the video is controlled by a human. The tech is all there (costs, non-withstanding...), they just need a brain to direct the motors. linky Now, that "just" turns out to be pretty complex, even if you hard-wire the highest level of logic "Do dishes" => "Collect dishes" => "find dishes" ; "grab dishes", you still have many black-boxes in there that you all need to perform with a high degree of accuracy. You don't want to put the TV remote in the dishwasher, and determining how to pick up a dish--even what orientation a dish is in--is not a straightforward problem. Then you start wondering if it is robust against the cat walking in front of a sensor... Each of these is generally tackle-able, but I can't help but wonder if version 1.0 will only support circular plates with only minimal scraps of food. Still even if your robots don't have any sense of objects or time, and are only grabbing and moving items with pre-assigned tracking tags, that's enough to make entire jobs obsolete.
Contextualization of machine vision has been the bugbear of artificial intelligence for 20+ years. The advances we've seen over the past 10 or so have all been a function of the steady advances in computing power and speed prompted by the Internet and smart phones. We're past the need for a quantum leap; at this point it's the incremental crawl to asymptotic perfection. Which is kind of a drag. There will be no "holy shit" moment. It'll be "huh. I just bought an Asimo that does windows at Best Buy - future, yo" moment.
Yeah, it's sort of funny how people gave up on "programming consciousness" and just figured out that if they can abstract away actions, they can script the highest level and get a robot / program that is still perfectly useful. Makes me think of machine learning as "the working-man's AI".