[gregcoombe]

I think that there is grey-area between data that is sufficiently static to represent in a map (buildings, lane lines) and data that is dynamic and must be handled by on-board perception systems (pedestrians, road debris). A good way to think of this is that the map is encoding priors about the environment for use by the perception system. It may be possible in the future to do all of this on-board (like humans do!) but the computational constraints would be quite high.

[gregcoombe]

To add to this: it's not just computational complexity, but it's also about safety. There is a human review process for every element of the map, which enforces a level of quality and safety that is extremely difficult to ensure from a fully on-the-fly system.