Hello,

> This is useful discussion topic. Some people have already been exploring
> it with the OccupancyGridUpdate message
> (http://ros.org/doc/groovy/api/map_msgs/html/msg/OccupancyGridUpdate.html)
> in particular Stéphane Magnenat. There was a REP partially related to
> this topic
> (https://github.com/stephanemagnenat/rep/blob/91dc02f8f2a82c5ef4820d2fc9f0f6d1fd1d0c20/rep-0129.rst)
> but I'm not sure what the final status of it was.

Thanks David for pointing out my previous work on this topic, and sorry 
for the delay in answering. My goal was both to address 2D and 3D maps. 
For 2D maps, I envisioned replacing a part of the map, and for 3D, to 
use deltas on the point clouds.

However, during the writing of these REPs, I talked to many people and 
came to the realisation that the core problem was that the 
stream-oriented data processing model of ROS (topic) was not well suited 
for maps, which are fundamentally a data-oriented structure. Hence, 
instead of fitting a square block into a circular hole, my colleagues 
and I started working on defining a data-oriented solution to the 
problem. The result is a work-in-progress document on a database-style 
map manager API [1]. This work is mostly conceptual at the model level 
for now (that is, ROS-agnostic), but we expect a first "satisfactory" 
version (I would not say stable) in the coming weeks, from which we 
could start a first experimental ROS implementation. We welcome comments 
on this document, but do not forget that it is a work-in-progress draft 
(as seen by the many TODOs currently in the document) that might not 
even be self-consistent.

kind regards,

Stéphane

[1] 
https://github.com/ethz-asl/map_manager_api/blob/master/map_manager_api.rst

-- 
Dr Stéphane Magnenat
http://stephane.magnenat.net