If you extend some combination of Moore's law, an assessment of how sensors are improving (slowly), advances in automation since the DARPA Grand challenge, and look at other automation projects (Stanfords, many of the current manufactures advances ) - the point at which you can have a consumer car do a 100% urban drive from home to office (without prior mapping assessment - Google's routes are highly surveyed prior to being driven automatically ) in a first-world country is about 15 years out - say 2027. The premium will mostly be for the sensors, by then - and will probably start off being a $25,000 package in 2027.

Way before then, though, you will see many (and already do) of the advances making their way into cars - assisting you with lane drifting, automatically braking to avoid rear-ending the car in front of you, etc... We'll probably see "automated cruise-control" on freeways about 10 years prior to the full automation at a $40K premium - mostly putting it in the space of commercial vehicles (Long haul Trucking, Long Distance Busses) before Consumers have access to it - I expect to see a large number of these vehicles on freeways by 2017.

I'm interested in when an automated race car will be able to beat a human. I have friends who say it's 30 years out before an automated system will be able to make the kinds of adjustments that a human can, but 30 years is a long time out. Shelley is only 18 seconds back on Thunderhill - if we can knock just 1 second a year off that time, we could see parity in just a couple decades.