Globally Consistent 3D LiDAR Mapping With GPU-Accelerated GICP Matching Cost Factors

This paper presents a real-time 3D LiDAR mapping framework based on global matching cost minimization. The proposed method constructs factor graph that directly minimizes costs between frames over the entire map, unlike pose graph-based approaches minimize errors in space. For minimization, we use voxel data association-based GICP is able to fully leverage GPU parallel processing. combination of and computation enables constraint relative with small overlap creation densely connected graph. process managed voxel-based metric can quickly be evaluated GPU. We incorporate an external loop detection order help factors avoid convergence local solution. experimental result KITTI dataset shows approach improves estimation accuracy long trajectories.