Putting sensors or transponders in roads and intersections to control autonomous vehicles is an antiquated (what, twenty years old?) idea. The infrastructure cost would be prohibitive...

Yes, but on the other hand the current approach of self training and largely ignoring edge cases hasn't yet turned out to work well enough. Emphasis on "yet", OK, but I wouldn't be quite so dismissive. Sticking the equivalent of 5 cent RFID tags on curbs every so many metres wouldn't cost a trillion dollars.

<CYNICISM>
Besides, there is probably no need to figure out which general approach will work best. In the end the decision on whether to let current generation cars loose will be primarily political, and perhaps insurance-related, not based primarily on the technical or third-party safety concerns. Get used to looking both ways twice when crossing the street till the cars at least manage the concept of (say) object permanence : )
</CYNICISM>

Jim