More importantly, even if this kind of stuff doesn't enable full self-driving, it can still be used to make basic driving assistance much better - think of even the most basic cars having lane assist on most road, for example.

I don't think it's a significant expense, either. I'm not talking about ripping up roads and replacing all signs in one fell swoop. But e.g. when they redid the highway near me recently, it got little reflectors to mark the lanes. What if, say, every Nth of those had some kind of RFID thing in it? Or when they replace signs, why not put up a new one that broadcasts what it is? None of this is particularly expensive.

As to why not trains and buses - because they do not actually offer the same features as cars, and many people want those features. If you want to argue against cars on principle, it's a different conversation entirely. I do think that more quality public transport is desperately needed, but, speaking as someone who used it nearly exclusively for the first 25 years of my life (and in places that are designed around it, unlike US), it does not replace a car.

My point the road metadata isn't for collision avoidance, but for better lane and speed assist.