For each month this technology's market deployment is delayed, there's a 9/11 of American deaths in car accidents. Even if Google went ahead with maximally-unsafe public beta tests and killed a hundred people, that would still probably cause an enormous net savings of lives.
If 3000 people kill themselves (and/or their passengers) each month, that's "ok", that's accidents, they happen, it's noone's fault.
If driverless cars kill 100 people each month, then IT'S SOMEONE'S FAULT. Someone is liable. Someone can be sued. Someone in that case being the company that makes these driverless cars.
This means that personal cars will have a human somewhat in control to absorb the blame for a very long time.
I think the first driverless vehicles we'll allow on the roads will be unmanned trucks. They don't need a human onboard since they only transport goods, not people, and there's a huge efficiency gain if you can remove truck drivers, since they need to eat and sleep.
And then when people have gotten used to driving alongside fully automatic vehicles, public opinion might shift to also allow people to be transported by similar systems.
Rail and water-transport are a lot mot cheaper than trucks (either manned or un-manned). The only thing that keeps them (the trucks) running is the cheap price of gasoline in the States.
The road to driverless vehicle public acceptance is paved with extremely limited trials and an absolute focus on safety.
There are obviously a lot of legitimate safety questions about robot cars. Are they better or worse at avoiding pedestrian collisions than a good driver? An average driver? How does the average severity of an accident vary between human drivers and robots? It will take some expensive testing and stats/actuarial work to sort out whether they would actually save lives at this point or anytime soon.
There are some important security questions too. How many exploits will be found per year? Will they be used to commit anonymous acts of violence? Do the cars connect to a network of any kind and pass viruses? If a car has been off or out of range during an infection, does it have to be quarantined from other cars and forced to update? "Pulling over to restart driving-service." How do you verify that an infected car has actually updated? Lots of completely untested, life-critical systems.
When other vendors enter the market, what kind of standards will they have to meet? How do we verify the security of their car software, will we require they be open source? If a vendor goes out of business, who is responsible for maintaining their upgrade distribution network?
There's a lot of important policies to decide on, it's going to take awhile.