RTABMAP localization metrics

[robertsenps]

Hi all,

I'm trying to benchmark the rtabmap localization using 3d lidar. Are there any available metrics that I can use to quantize the match rate between lidar pointcloud and generated 3d map? So I can know the accuracy of the robot from the acual position.

Thanks

[matlabbe]

Normally, having a ground truth is the best way to do that.

If you don't have a ground truth, you could look at the Statistics -> Loop/OdomCorrectionNorm statistic


The higher the jumps, the higher the correction or localization error.

[robertsenps]

Hi Matthieu,

Thanks for the reply. Is the statistic available as a topic? And regarding ground truth, what is your opinion about what should be used for a ground truth when we run rtabmap in the real robot?

[matlabbe]

Hi,

The statistics are published in /rtabmap/info topic, in statKeys and statValues fields (http://docs.ros.org/en/noetic/api/rtabmap_ros/html/msg/Info.html). They can also appear in rtabmap_ros/Info rviz display.

It depends on what kind of robot. For an outdoor robot, maybe an RTK GPS. For indoor robot, if you have OptiTrack or VICON systems, that could make a very accurate ground truth. For both approaches, if the ground truth is published though TF, say /world -> /base_link_track, you can set ground_truth_frame_id and ground_truth_base_frame_id respectively to save the ground truth in rtabmap database. RTAB-Map can then compute RMSE directly (that will appear in Statistics) and you can compare both paths (slam vs ground truth) in rtabmapviz or rtabmap-databaseViewer.

cheers,
Mathieu

[Masoumeh]

Hi

would you please explain about "ground_truth_frame_id and ground_truth_base_frame_id"?what is the difference?
I have also rtk gps, that i want to use it as groundtruth, which of these parameter should be set? and set to what?

thanks