My (admittedly limited) knowledge of the cars is that they 'cheat' a lot, in that they don't just look at sensors to make decisions rather they look at a map of data to determine ie approach this corner at maximum 40mph, pot holes on left side of road at this point, always stop at this intersection as it's tricky to see, etc. So the parallel between improving maps data with phones will be improving the car detail map data with cars.
