Re: Realtime Map Merger Improvements

Posted by derektan1995 on Jan 05, 2022; 1:21pm
URL: http://official-rtab-map-forum.206.s1.nabble.com/Realtime-Map-Merger-Improvements-tp8200p8799.html

Hi Mathieu,

Thanks for the detailed explanation. I'm currently trying out the second approach, but along the way, I chanced upon a question not directly related to this.

It seems from your suggested approach of map merging that we do not consider the optimized poses and constraints from individual robots. This is because we only extract sensor and pose data from the mapData topic (last node). This is understandably so because the intention of this map_merger node is to collect the robots' mapData incrementally (last node), feed it into the rtabmap class object, and to introduce new constraints considering mapData from all robots.  

I am confused as to how this rtabmap class could derive new constraints. In the case of ICP links, robots typically use their lidar scans to perform ICP against the map scans. However, on the central server collecting mapData messages, we are no longer able to subscribe to any lidar topics. I'm also unsure if there is a mechanism to use the lidar scans from the mapData (last node) to derive ICP constraints in this case.

As a result, I noticed that my merged map looks skewed, possibly due to the lack of constraints. On the contrary, my individual mapData Rviz maps that are overlayed together provides a more accurate reconstructed scene. Any insights to this?

Thanks as always,
Derek