This post is in continuation to https://discourse.ros.org/t/georeferencing-pointcloud-map/12454
It is mentioned in ROS REP-105(link) that we should switch between map frames when doing large-scale mapping with 3D-pointclouds in PCL, since it uses FLOAT32 and leads to errors with large values. Given the Japan plane coordinate system, seems that it is easier to create precise large-scale 3D pointcloud maps using autoware.ai in Japan. While it seems the same is not possible with other global coordinate systems/projections like ECEF/UTM etc., since they lead to a loss in precision due to floating-point approximation for large values. Is it so? Is there an implemented architecture as mentioned in ROS REP-105 which provides a general approach for the same globally?