That also said, Google has a lot of experience driving cars in daylight, in good weather, in a very limited area, and mostly on freeways. That is very different from turning those vehicles loose in Boston, at night, in a snowstorm, without any human even present. Five to eight years to purchase (or rent by the hour) an autonomous robo-taxi is absolute fantasy territory. Personally, I think 20 years is probably very optimistic. (Yes, there are a lot of advances being made but there's a huge gap between where we are and where we need to get to replace taxis.)
Regarding vehicles though, I don't agree. Yes, they won't work in the snow or inclement weather yet. But there are numerous places around the world where the weather is sufficiently stable that autonomous vehicles could operate the majority of the time. The value proposition is simply too high for anyone to throw their hands up and say, "Well, we can't do this." and move on to another project.
It's going to happen slowly (see: Tesla Autopilot slowly gaining more features with each point release), and then all of a sudden.
[1] http://theaviationist.com/2015/04/16/salty-dog-502-aar-omega...
I'm sure we're going to get there. The question is whether it's one decade or five to have no human present under a wide range of conditions.
Google is covering tens of thousands of miles in autonomous mode every month [1], and Tesla is refining their algorithms and mapping data with every mile Tesla Model S owners drive (whether they've paid for Autopilot or not, the vehicle just needs to have the hardware).
Hockey stick and all that jazz.
[1] https://www.google.com/selfdrivingcar/reports/