Exploiting Sparse Semantic HD Maps for Self-Driving Vehicle Localization

August 2020

tl;dr: Use lane graph and traffic sign to build sparse HD map on highways to perform cm-level localization.

Overall impression

The sparse HD map uses lane graphs (LLD) and traffic sign (TSR) and only takes 0.03%, orders of magnitude smaller storage space than traditional HD maps.

The proposed method aims to tackle localization on highways where loop closure cannot be used.

Sparse HD map does not require detailed knowledge about the appearance of the world (dense geometry and texture) –> this leads to better generalization.

It is interesting to see that

Key ideas

Technical details