On the Applicability of General LiDAR Registration to V2X Data Alignment

Abstract

Vehicle-to-Everything (V2X) cooperative perception in autonomous driving is attracting increasing attention as a means to overcome the limitations of individual vehicle sensors and expand the perception range. Achieving this objective requires precise alignment of heterogeneous sensor information into a unified coordinate system. However, unlike conventional registration problems, V2X environments present unique challenges such as low overlap, sensor asymmetry, and limited communication bandwidth. In this study, we systematically evaluate representative alignment approaches under the V2X scenario using the DAIR-V2X dataset. The experimental results reveal that while each method shows certain advantages, they also exhibit significant limitations in terms of robustness, accuracy, and communication e ciency under realistic conditions. These findings highlight that although registration techniques are well established in general robotics, they are not directly transferable to V2X environments, motivating further research into robust and communication-aware solutions for reliable V2X data alignment.