I've been wondering that too.
The point of driverless cars is supposed to be a way to get us to that utopian transportation vision where we can go anywhere automatically by telling our transportation device where we want to go. This has been "possible" for decades but for one problem: all proposed systems required new tracks/roads be built that were separated from the current road system. That's prohibitively expensive. So in walks Google, and a few others, and says "We have all this technology, let's create something that interoperates with existing traffic on existing roads."
And they do some demos, and everyone thinks they've solved the problem.
Only they haven't. Google's cars, for example, have to drive on a "virtual track". There are holes in the track. Some of them are holes in the map, others are temporary detours and or obstacles that means the cars are unable to navigate them because it doesn't have enough information. To make driverless cars "work" as well as they appear to do at all across the whole country, Google is going to have to keep a constant, updated by the minute, map of the entire US road system, not just the official roads, but the private roads, the position of every driveway, etc.
So the DMV's comments aren't actually entirely out of order. Forget emergencies, you will have to take over every few hundred miles, assuming Google can update its databases to some decent compromise between up-to-the-second and "good enough", simply because the cars are going to have problems continuing.
Me? I'd prefer we look at our transportation system again and ask if this is really what we want and need. And if we're going to continue legally mandating suburban development and banning urban development, perhaps we need to look into improving PRT technologies and making them work.