Self driving stuff is really hard, IMHO the US should invest in controlled autonomous vehicle infrastructure and reduce the amount of variables necessary for the vast majority of miles driven. Make it so the vehicles can go through cities at 80mph.

I also think that US should build more train tracks. Much easier to automate.

I've had this same thought many times, and I'm sure others here have as well.

Has anyone come across a serious study/proposal along these lines?

It's an ongoing project (but aimed mostly at making driving easier for humans).

Full self driving is inevitably going to be a combination of road technologies paired with on-vehicle technologies.

Humans, the current most sophisticated driving intelligence, crash all the time.

AI, will need to be smarter than humans to outcompete in a world made for us, but if we bake some simple additional signals into the roads and other vehicles, the whole thing might work.

I am no longer optimistic about the next 5 years.

I'm curious whether this foretells any potential future change in strategy for e.g. Cruise/Waymo, which are similarly ultra-expensive.

Eh, likely the same brand of boardroom “vision” that led to both companies being so far behind starting to develop EVs.