Scan-wise refinement strategies for validating local geometric assumptions

Investigate scan-wise refinement strategies that further improve the validity of local geometric assumptions, such as planarity, in radar–inertial odometry pipelines that enforce explicit local geometry constraints during scan-to-submap registration and mapping.

Background

The paper shows that applying planarity constraints directly to an unfiltered radar map can degrade performance because the local planarity assumption is often violated by noisy or unreliable returns. In contrast, preserving only low-uncertainty points and combining them with a geometric residual significantly improves odometry.

From these observations, the authors conclude that a refinement stage during local map construction is necessary to enable effective enforcement of planarity constraints, and they explicitly note that exploring other scan-wise refinement strategies to further improve the validity of these local geometric assumptions remains to be done.