Clarification on Interactive Mode Functioning and Map Updates

[andrycapod]

• Operating System:
  • Docker container Ubuntu 22.04
• Installation type:
  • source
• ROS Version
  • ROS 2 Humble
• Version or commit hash:
  • 2.6.10
• Laser unit:
  • Sick TiM 571 (simulated)

Description
I'm experimenting with the interactive mode in slam_toolbox using the online_sync mode. Here's what I did:

1. Launched online_sync mode from slam_toolbox.
2. Enabled interactive mode using the toggle from the provided RViz panel.
3. Added the Interactive Marker display in RViz.
4. Manipulated the pose graph using the interactive markers (e.g., moving nodes).
5. Pressed "Save Changes".
6. Verified that the manualLoopClosureCallback in loop_closure_assistant.cpp was being called correctly.

Question
I am unsure how these interactive modifications are supposed to reflect on the map:

1. Are the changes applied immediately and visible on the 2D map in RViz?
   (e.g., will the occupancy grid change in real time after modifying the graph?)

2. Or is the only way to see the result of the graph changes to save the map and reload it?

At the moment, I can manipulate nodes and apply changes, but I don't visually observe any difference in the map in RViz. Could you please clarify how this is intended to work?

Any insight would be appreciated, thanks!

[SteveMacenski]

1. No, the pose graph is updated so the next time the occupancy grid is generated from it (usually on the time in your params file) it will be reflected.

2. No, just wait for the next occupancy grid update. The slam system maintains a pose-graph as its internal data structure. The occupancy grid is an output of the SLAM and isn't used or consumed internally. Its only for external application use

[andrycapod]

Thank you very much for the clarification!
Just to confirm I understood correctly: the updated pose graph will be reflected in the occupancy grid the next time it is regenerated, based on the map_update_interval. That makes sense.

What I find interesting—and as shown in the attached video—is that when I enable interactive mode and move the first node of the pose graph, then hit "Save Changes", the occupancy grid does immediately reflect the updated pose, which is exactly what I would expect.

However, this doesn't seem to happen for other nodes. When I move and adjust another node (with help from the scan alignment), then press "Save Changes", not only is the occupancy grid not updated, but the node actually snaps back to its original pose, as if the change were discarded. This can also be seen clearly in the video.

Is this expected behavior?
Could this be related to something else (e.g., the use of laser odometry), or am I misunderstanding how the interactive editing is supposed to work?

For context, I'm using the default parameters from the online_sync mode.

Thanks again for your time and support!

Here the link to the video

[SteveMacenski]

Honestly speaking, its been many years since I developed or used that feature. It wouldn't entirely surprise me if that particular feature wasn't completely flushed out. Many of these 'advanced' features I was working on for a new SLAM workflow which I dropped when I left that job early in my career. I productionized a few of them, but not all of them were needed at the time so after making a POC they were shelved until needed.

We move the node with the data, but perhaps there's other metadata that also needs to be updated. I don't recall off hand what all I do or do not update along with that node update. If you'd like to use that feature, maybe that's something you can dig into and see what's going on. Maybe we need to update the neighboring node's constraints distances (which would surprise me if I didn't; but I don't recall) or even ask the system to reprocess a manually adjusted node.

This library in my eyes is in a maintenance-only status for my efforts. I haven't been developing anything new here and if I were to create a new SLAM and localization system, it would be a fresh re-write. If you have a vested interest here in using this and want to help maintain, make big contributions on these features, or develop it, please let me know and I'm happy to add you as a comaintainer / developer :-)

[SteveMacenski]

Oh it looks like I even recorded this in #70

[andrycapod]

Thanks a lot for the help!

I'll try to take a closer look and see what might be going on. In case I manage to fix something or come up with useful new features, I'd be happy to open a PR.

I know this might not be the best place to ask, but just to confirm — as it seems from what you said, you're not planning to develop any new features for slam_toolbox anymore (unless they come as PRs), both for Humble and the other branches. Is that correct?

[SteveMacenski]

At the moment, yes. This is in a stable place that I've shown to scale to very large maps successfully and used in a variety of applications. Sure there are points of improvement and new features to add, but they are not being personally prioritized by myself beyond supporting users to contribute back their changes to the project. We receive alot of interest and PRs in this repository (understandably), so with my limited time and resources, I think its best that I prioritize getting those contributions and improvements for their deployed applications into the project rather than myself developing the improvements.