The real map of the world is just too complex and changing for anything other than a true AI to be able to operate within.

What i see is that all AVs will be interconnected, and constantly communicating with each other.

Funny, I think of it exactly the other way. Mapping the real world is within the realm of possibilities - I mean this article is 5 years old by now. Besides, sensory input will always trump map knowledge. I talked to someone from TomTom a while ago. IIRC, when a dozen of their users drive over a new road they'll push that update to other users.

Permanently connected and highly reliable wireless connections that risk taking depends on? I'll believe it when I see it. When you have such a train of V2V connected cars, it only takes one malfunctioning / package-not-arriving car to screw it up for everyone behind it. 4G LTE (or even 5G) might be fast, but whenever I am at a busy train station it chokes the fuck out. My Bluetooth gets choppy when I move my head too quickly, and that's at a distance of less than a foot from my phone. Wireless tech is worse the denser your urban area is, while for AV's the opposite needs to be true. It exists on production lines because those are static, isolated environments, while (urban) roads are much more dynamic and prone to errors and interference.

Segregating AV traffic from human-driven vehicles is key, though. So reconsidering road widths, composition, merging, etc, you wind up with something much more like bicycle paths, than the streets and highways we use today.

I think so too, at least for the near future, unless Google seriously gets their shit together. There's also a handful of public road tests, which could prove hopeful, but they are quite far removed from the ideal high-frequency, high-capacity bus replacement that I'm looking for.