Why I can't relacate in localization mode

[derwee]

Rtab-map building 3D map is great.
I set in localization mode with the pre-built map. But starting point still begins with the graph of map, and It don't relocate to new position.
refer to
http://official-rtab-map-forum.206.s1nabble.com/Map-misalignement-during-localization-td474.html
It tells that I need to adjust robot position in order to get green background in the loop closure view. But We can't find a loop closure. Is there something wrong with my setting and operation? I look forward to hear from you and appreciate.

[matlabbe]

Hello derwee,

it seems that the loop closure hypothesis is not high enough to trigger a loop closure. To see the loop closure hypotheses, you can show the "Likelihood" and "A Posteriori PDF" in Window menu -> Show view...

Example:


In "A posteriori PDF", the corresponding hypothesis of the old image should be near the threshold to trigger a loop closure (green background when accepted). You can set the plots in graphics view mode (see right-click menu) to show images related to each dot in the graph.

cheers

[derwee]

Hi Matlabbe,

Thanks for your great help.

In "A posteriori PDF", the corresponding hypothesis of the old image should be near the threshold to trigger a loop closure (green background when accepted). You can set the plots in graphics view mode (see right-click menu) to show images related to each dot in the graph.

May I know which parameter in setting file can directly be adjust above thredhold you pointed? thanks.

[matlabbe]

Hello,

In Preferences dialog under RTAB-Map Settings panel, the parameter is called "T_loop : Loop closure threshold." In the settings file or in ROS, the parameter is "Rtabmap/LoopThr" (default 0.11).

Note that 0.11 is already quite the lower that you can set while avoiding most of the false positives. I rather recommend to see in the likelihood view if there are images that have a much higher score than the others. If no, maybe there are not enough hypotheses (not enough images in memory) to find an appropriate likelihood. Generally, having a minimum of 20 images/nodes in the graph is required.

cheers