Remove free cells beyond walls

[derektan95]

Hi Rtabmap community,

I would like to repost this question from the Rtabmap Forum to seek some advice from a wider community.

I am using gazebo simulation with rtabmap, along with a 3D lidar for localization and mapping. I noticed some free cells that is often mapped beyond the walls in my simulation, as seen below. Is there a way to modify certain params such that there's no free cells behind the walls (I understand that this isn't realistic in the real world)?

This is an example of no free cells beyond obstacles:

Warm Regards, Derek Tan

[matlabbe]

Hi,

Is the second figure created with Cartographer?

What are the Grid/**** parameters that you used for rtabmap? To know in particular if that 2D occupancy grid map is created from OctoMap projection or from the 2D ray tracing approach. If you can share the database, that would be nice.

cheers, Mathieu

[derektan95]

Hi @matlabbe,

The second figure is taken from the internet as an example, but I have tried cartographer in the past to get a clean map like this earlier on.

As for the Grid parameters, I have tried to create both 2D occupancy grid map from octomap projection, and from the 2D ray tracing approach. Both have issues where free cells exist beyond walls. Nevertheless, I will upload the rtabmap database files for both Grid/3D=True and Grid/3D=False very soon.

Warm Regards, Derek Tan

[derektan95]

Hi @matlabbe,

Here are the database files you have requested for. This folder contains 2 db files where Grid/3D=True and Grid/3D=False. As you will see, there are white cells beyond the walls.

Db Files: https://drive.google.com/drive/folders/1oCluv92DReBZqwBDLn7GPfS57FBwLFaY?usp=sharing

Thanks again, Derek Tan

[matlabbe]

2D (rtabmap ray tracing):

3D (octomap ray tracing):

In both cases, the ray tracing seems overshooting a little, maybe because of angle resolution. For rtabmap ray tracing, this is used here for each obstacle.

[derektan95]

Hi Mathieu,

Thanks for the feedback! Any suggestions on how I can reduce this overshoot? Is increasing the angle resolution a good idea (i.e. lower magnitude)?

It seems like the problem is increasingly worsened for rays hitting walls that are further away from the robot.

Warm Regards, Derek

[matlabbe]

The farther it is, the more accurate the slope computation should be. Don't know if using double instead of float in that function could help for ray tracing on far obstacles. Other option would be to stop ray tracing free space 1 cell before the actual obstacle cell.

[derektan95]

Hi Mathieu,

I tried stopping ray tracing of free space 1 cell before the actual obstacle, at the proposed code location here. However, I realized this technique only works on my 2D occupancy grid map only if I set Grid/3D=false. This does not work if octomap is activated. May I know how I can implement this feature for octomap raytracing?

[matlabbe]

The ray tracibg for the 3d case is done by the OctoMap library. You may do a similar hack in their code.