First of all, I'm very happy to use RTABMap for about 6 months till now, especially the Lidar SLAM one. I'm curious about a node that your team created, that is pointcloud_assembler. I saw that it can assemble last-n pointclouds to create a dense pointcloud. But may I know what is your initial purpose or problem that led you to make pointcloud_assembler? is it just for making a denser pointcloud, or it is used for another thing too like loop closure etc? Could you give a detailed information about this?
to assemble 2D vertical scans in 3D, then send the 3D cloud to RTAB-Map,
to assemble sparse 2D scans to get denser scans to send to RTAB-Map,
to assemble velodyne scans for 1 second (accordingly to odom), then send the cloud to RTAB-Map (updating a 1 Hz),
to assemble scans of a rotating lidar for duration of half rotation before doing a map update (similar to hovermap).
You would use it when you need to record all scans from a lidar without increasing RTAB-Map update rate. If you have a lidar at 20 Hz, and your odometry is accurate enough during 1 second, then we can accumulate the scans during that time and update the map just at 1 Hz. We could update the map at 20 Hz, but the graph will grow too fast (for online loop closure detection and map optimization).