Currently, all costmap publisher / subscription when using raw messages (e.g. costmap messages containing the raw costmap from 0-255 rather than the occupancy grid message of 0-100 for internal use by the task servers such as behavior and smoother) send the full costmap each time. This is unideal, even though we compose all our nodes into the same process, which does help greatly with compute time.

It would be better if we mirrored the Update methods from the OccupancyGrid publication to minimize the number of times a full, raw, costmap is sent over the network and only publish the updated regions. This would decrease compute resources and latency for the behavior and smoother server (or other servers for client applications using the raw costmaps).

This is a great first issue for a new contributor! Just make sure to look over the code and make sure that all instances of costmap pub/sub are using these classes to make sure that everyone can properly handle the updates (which I believe they all should be, minus some unit tests)