[jekdoce]

The thought provoking contrast presented in the article is camera derived sparse maps with lane lines, road edges, signs etc. vs lidar derived dense detailed models of the whole environment. Any thoughts around why the former is not enough? If it is enough it would certainly be preferable, since smaller in size and likely easier to maintain.

[tonygrue]

My guess is that the more detailed, dense map allows the car to precisely localize itself by using its view of the world to match against the map, yielding a location more precise than GPS (and for when gps doesn’t work); eg ICP point cloud matching/alignment/registration. More sparse maps would give you a less confident match, and you’d have to convert what the car is seeing to the sparse form.

[skgoa]

The former is enough, but having the latter makes it much easier.