I occasionally give talks on the future of robotics and since about 2000, I've been predicting that the technology required for cars to drive themselves would be feasible around 2020 or so. My definition of feasible for this means robust enough, safe enough, and inexpensive enough that a manufacturer could sell autonomous vehicles to willing buyers.
However, I always added the following caveat: while the technology would be there, it might take decades before the public accepts autonomous vehicles. This acceptance would include social tolerance of the whole concept of driverless cars, liability laws, and vehicle licensing.
In some sense social tolerance might be the biggest issue as it is an important driver of the other issues. As an example of resistance to this sort of thing is that while airplanes could fly themselves, virtually nobody (including me) is willing to get on an airplane with no pilot, and most of us are uncomfortable (to say the least) with the concept of large, potentially explosive aircraft cruising around above our heads with no human guidance or backup.
On the other hand, cars are not aircraft, the cost of a pilot or two or three relative to the overall cost of operating an airline is relatively small, and autonomous cars seem more like robots and lots of people think that robots are cool. Furthermore, there are several very difficult tradeoffs regarding driving that society has and is increasingly being faced with.
For example, the fatal accident rate per mile driven for drivers over 75 goes way up and our society is aging rapidly. This leaves the unfortunate choice of either limiting the mobility of many older drivers by taking away their licenses or allowing them to continue to drive and risk them killing themselves and others. The autonomous car provides a third, and likely preferable, option.
Liability laws are a tough issue as well. I wrote a humorous post about an early accident involving an autonomous research vehicle, but more seriously, if a manufacturer, with deep pockets until bankrupt, is liable for every accident, it's a huge disincentive to produce robotic cars. While I used to think that this would be the biggest stumbling block, now I'm not so sure. With Toyota being scapegoated and incurring massive costs when some drivers couldn't remember which pedal was the gas and which the brake, and others committing fraud to ride on that wave, perhaps the impact won't be as bad as I thought relative to what reality currently is.
So that leaves licensing. Governments are generally slow to respond, so I figured it would take forever for the government innovations required for licensing driverless cars. But I was wrong. Nevada has just issued the worlds first license for an autonomous car! And California might do the same soon!
So at this point, I'm definitely encouraged, and I think there could be some sort of licensed autonomous vehicle available for purchase sometime early next decade.