I think the robo-car is the self cleaning kitchen of the 2010's. When your ML algorithm makes an error, it won't just be a drop in some abstract accuracy measurement -- someone will die. This is a hell of a lot of liability to take on. Google's self-driving car is a promo stunt. You would need a lot of infrastructure for it to be safe -- and with that infrastructure will come loss of the "freedom" you describe -- e.g. It'll be a kind of train.

Not like you have that freedom now. Yeah, maybe in some rural area compared to hoofing it for 30 miles :-), but we're not talking about that -- you'll still use cars there. It's not true in suburbia. That's why we have malls and big box stores with huge parking lots built next to highways. You are unnaturally constrained to the highways and roads that have been built -- the freedom is illusory.

Speed. Well, you're attempting to compare Caltrain to the alternatives. In no way does Caltrain compare to Metro north or NJTransit -- the reason people choose to drive is not because cars are intrinsically faster in those areas, but because the mass transit in the bay area sucks. It's like saying in 1997 that "people have chosen rationally" not to use a phone as a web browser. Yes, they have, very rationally in fact, but it says nothing about the intrinsic ability of a phone-like device to be a web browser.

Efficiency is a red herring. Even the post you linked to is inconclusive in that regard. If you step out of the box and consider that the same improvements to cars could be made to buses or trains, then apply the economy of scale a massive investment in transit would mean -- it comes out a little differently. In the same post, there is a comparison to transit in other countries -- the comparison gets a little weak at that point.