Theoretical Background

[nismo]

Hello ,
I am a beginner in the visual SLAM area and I stumbled upon RTAB-Map. After trying it with ROS and a stereo camera I was impressed with results.But, I am having trouble understanding the principles and the theoretical background that it uses so that I can describe it to others. Is it a particle filter - graph based SLAM mixture? because I see some particle filtering used in the code but I cannot really figure it out... would you mind giving me an overview of the system (reminder : I use stereo).
Thank you very much in advance!

[matlabbe]

Hi,

The paper "Online Global Loop Closure Detection for Large-Scale Multi-Session Graph-Based SLAM" gives a description of the SLAM back-end used by RTAB-Map. Like the title says, it is a Graph-based SLAM approach, not a particle filter approach. The particle filter (along with the Kalman filter) you are referring is an experimental way to filter visual odometry noise (it is not used by default).

There is no official documentation yet about visual odometry used in RTAB-Map (you may find other posts about that though).

cheers,
Mathieu

[nismo]

Alright thank you very much. One more question. Is there any way to get the poses.txt file and the high-res point clouds other than through rtabmapviz?? A service I could call or something?

[matlabbe]

No, it is not possible. We would need to add a service on rtabmap node to generate a point cloud using similar parameters than in Export dialog of rtabmapviz. I added an issue.

For the poses, you can already get them by subscribing to /rtabmap/mapGraph topic. If mapping is paused and mapGraph is not published, you can force a republishing by calling service /rtabmap/publish_map (with global=true, optimized=true, graphOnly=true).

cheers,
Mathieu

[nismo]

Thank you very much Mathieu! Keep up the good work!

Best regards!