A lot of it is software. The hardware is mostly solved except for cost and there's no reason to think it won't go through the same cost reduction curves that are common throughout the industry. So, IF you can get to viable technology for specific use cases, there's not much reason to think you can't get to workable economics.

> So, IF you can get to viable technology for specific use cases, there's not much reason to think you can't get to workable economics.

This seems like a good analysis, but are there particularly good reasons to believe we will be able to produce a car that drives through an urban environment, at any cost?

Some day. At this point, IMO it would be foolish to say never or not for a century. We could find ourselves in a place where the current path that's mostly deep learning gets close but not close enough. But we seem to be moving forward pretty well even if the last 5 to 10% takes a couple decades longer than many people think.

But I'm still inclined to think that generalized door-to-door autonomy without a competent human present is still a long way out.

Waymo cars already drive in Phoenix, Arizona (i.e. urban environment) without human drivers.

https://www.wired.com/story/waymo-google-arizona-phoenix-dri...