Your perspective is appreciated. Am I completely off base in characterizing this as "GIS from hell" vs. "The Bicentennial Man?" 'cuz google's approach seems to be a definable amount of data while Tesla's seems to be a whistling-in-the-dark systems approach.
If you don't have a premade map you have to generate the map while you figure out where in the map you are. This is called, creatively, simultaneous localization and mapping, and abbreviated SLAM because DARPA likes cute acronyms. Doing that requires the robot to explore a bit, essentially learning what Google tells its cars while trying to drive. I think you've mentioned playing with a Roomba before. You know how it just kind of zips around randomly at first? That's what it's doing. Now imagine it weighing N tons. Roomba isn't the smartest robot in the world, but all I'm aware of are closer to Roomba than something you want on the road.