I'm sure I'm just naive because I don't work in this space, but I keep wondering why they don't do machine learning from a satellite perspective. Yes, the cars need to notice and react to things in front of them. However, a lot of problem areas and odd navigation can be identified from space. I'm surprised they don't do some massive slow exposure shots from a satellite and see how traffic flows across the entire US. I want a car that can react in realtime to new problems, but it has a massive amount of storage so it can probably be told ahead of time what roads, interchanges, freeway onramps, etc.. are problems, and what should guide its decision making.