HGeo-TopoMap: Boosting Topological Mapping with Hierarchical Geometric Priors

Researchers introduced HGeo-TopoMap, a vision-based topological mapping method for autonomous driving.

It combines inverse perspective mapping priors with learned spatial and geometric consistency to improve centerline instance detection and connectivity.

On OpenLane-V2, it outperformed baseline methods on centerline, lane-segment, and robustness benchmarks.

The approach could make autonomous vehicle mapping more accurate and reliable in challenging driving conditions.