It does not use what's often called "HD maps". It uses vision + NNs to guess the lanes (even if they aren't visible), with some help of regular map data, but vi

It does not use what's often called "HD maps". It uses vision + NNs to guess the lanes (even if they aren't visible), with some help of regular map data, but vision takes priority.