There are indeed already multiple experiments on roads ongoing with autonomous trucks. The quote you place from @Turveyd is correct: GPS is not accurate enough for autonomous driving. It needs to be complemented by other sensors such as video cameras and radar. Plus very clever software that can process these video streams and detect all that needs to be detected.
The biggest problem no-one is talking about is how to program driverless cars to deal with a situation where no matter what the car does there's a high chance that someone is going to die. Even if all cars are networked together, you're never going to network the deer that jump onto the highway or the kid that runs into the street, and you're going to have dozens of instances every year where a car is put into an impossible situation. Optimally of course you want the car to steer into the place it will likely kill the fewest people. But what about when the "fewest people" are the occupants of the car in question? Will people really be OK with the fact that their car might affirmatively kill them in order to potentially save the lives of several others? Is everyone as altruistic as the fighter pilots that fly their damaged planes all the way to the ground to avoid hitting a populated area? Probably not. In which case, do you let every car program itself to prioritize the safety of it's own occupants over all else? The first time a car purposely steers into a crowd of moms with strollers and kills 20 babies because it otherwise would have collided with another car is probably the last time that algorithm is allowed to operate. There is no good answer, and since unlike human brains we can specify in advance how the machine will react in every situation we'll want to do that. Autonomous cars are full of promise now because they haven't had a major negative incident. When one does happen, as it inevitably will, these questions will start getting asked and the current irrational exuberance may fade a bit.
@Sig I agree with you that autonomous driving will not be the Holy Grail and that accidents will continue to occur. Whether those accidents will be of a different type than current accidents remains to be seen. I'm not sure whether the autonomous car will have an algorithm to calculate the "least possible loss" in case of an unavoidable accident.