Georeferencing GLIM maps

[VRichardJP]

I would like to georeference GLIM maps. There is already a gnss_global extension module which deals with GNSS data, but it does not fully satisfy my needs. In particular:

• The module uses Pose messages instead of NavSatFix, so GNSS sensor data must first be processed (e.g. with navsat_transform_node?). This is a bit inconvenient.
• The module computes the transformation T_world_utm only once, after the mapping device has travelled a few meter. So travelling longer distance won't improve the transformation (while new pose constraints are still added to the graph.)
• When the map size is not big (e.g. <1km) or when the path contains many loops, using poor GNSS data may be detrimental to the map. In other words, there are cases where I would rather not add any GNSS-based constraint to the graph (yet I would still like to georefence the map).

To address these issues, I am thinking about the following:

• Add support for NavSatFix messages. NavSatFix could be converted to UTM or MGRS with geographiclib. NavSatFix messages could be filtered based on fix status or position covariance.
• Use the first valid NavSatFix message to compute a GNSS origin and compute all submap_coords relatively to this origin point (thus avoiding errors on large floating point values).
• Add a enable_gnss_factor-like option. If enabled, use current logic to add pose factors to the graph (but use the GNSS origin instead of T_world_utm)
• Add a on_update_submaps callback, which receives submaps optimized by the main global mapping module and computes T_world_utm (using exactly the same logic than the current T_world_utm initialization). This T_world_utm transform is what is needed to georeference the map. There is no save callback yet, but I think it could be nice to dump the transformation data on save so that exporting a georeferenced map can be automated. Otherwise, the user could use tool like pdal to apply the transformation on the exported map manually.