About pose of the rtabmap estimation

[plaid]

Where can I find the pose of the rtabmap estimation results in the topic?

[matlabbe]

Do you mean the odometry correction? rtabmap publish it under tf /map -> /odom. The mapToOdom transform can be also received in mapGraph topic.

[plaid]

I want to check the pose estimated by rtabmap. I want to compare that with the ground truth!!

[matlabbe]

There is a mechanism to feed a ground truth TF to rtabmap (with ground_truth_frame_id and ground_truth_base_frame_id parameters), it will compute RMSE live in Statistics->GT panel and show both trajectories in rtabmapviz.

You may also export the graph in TUM format after the mapping:
rtabmap-export --poses rtabmap.db

cheers,
Mathieu

[plaid]

I see. Thanks for your reply. I found a topic called rtabmap/localization_pose/pose/position, what is this?　I thought this topic was an estimated pose and was comparing this to ground truth.

[matlabbe]

No, localization_pose is published when a loop closure happens or on localization. It is the pose of base_link in map frame with the marginal covariance from graph optimization. It is not related to ground truth.

[plaid]

I see. So it's impossible to compare the output from a topic to ground truth?

http://arxiv.org/pdf/1811.09895.pdf

I would like to do a plot like the one in the paper you described in the gazebo simulation, can I do something similar with what you described above?

[matlabbe]

Well, you can see TF world->map->base_link (slam) compared to a world->tracker (ground truth) in real-time if you want. If you want a topic, just subscribe to TF topic and record those transforms.

However, it is more convenient to just export the graph after mapping in TUM RGB-D daset format and use TUM's python evaluation script to compare with a ground truth that you can record in same format.

plaid wrote

I would like to do a plot like the one in the paper you described in the gazebo simulation

Which simulation, and which paper?

[plaid]

I wrote it wrong. The paper is the linked research paper that I mentioned in my previous reply.

It means that I want to do a comparison (rtab-ground truth plot) like the one in that paper in my gazebo simulation.

matlabbe wrote

You may also export the graph in TUM format after the mapping:
rtabmap-export --poses rtabmap.db

Using the method you taught me, I got rpe.pdf and the average translation error was 0.001, which is almost identical to Gt. It's working great, is it possible that I'm doing something wrong and not doing it correctly?　The results were so great that it was as if I had made a mistake. So I would like your opinion.Thanks!

[matlabbe]

Hi,

can you share the rtabmap pose file and the ground truth file that you used? or copy/paste the values inside a "raw" xml tag.

like this...

The figures shown in that paper were using the RGB-D dataset evaluation tool (https://vision.in.tum.de/data/datasets/rgbd-dataset/tools) that I described in my previous post.

cheers,
Mathieu