problem in simulation

[polarbear]

Hi,
I use robot model Husky in Gazebo to do a simulation in stabmap(rviz mode).
But the result looks wrong and i don`t know hou to deal with it.
Like the screenshot.The model,proj_map,cloud_map and voxel_cloud are not at the suitable position.
I feel it may because some wrong with the TF setting, could you give me some suggestions?
Thanks!

[matlabbe]

Hi,

Can you show TF in RVIZ? Also:

$ rosrun tf view_frames
$ evince frames.pdf

Maybe it is a problem with the orientation of the camera (TF).

cheers,
Mathieu

[matlabbe]

Can you show how rtabmap is started? (in particular what is the value of frame_id)

[polarbear]

Thanks,I change the ‘frame_id’ to base_link.The car and cloud_map are at the same plane now.
But the proj_map is still at the wrong plane like old screenshot.
So i try to find the work parameters in the setting of Occupancy Grid to deal with it like the 'frame_id' problem.But fail.

[polarbear]

Hi,
The proj_map problem is solved now by change Global options frame to /map.
And a new problem comes.
In rivz,i control the car to explore the environment.Sometimes it suddenly teleport to another place.But in gazebo,the car is still at where it was.Next,the mapping map looks error,two map overlap.

[matlabbe]

Hi,

Can you show a screenshot with TF? So we can see if it is the odometry, a bad loop closure, or rtabmap detected a new map should be created (high odometry covariance 9999). For the second, if in your simulation you have a lot of repeated textures, loop closure detection may find more false positives. In this case, you can decrease "RGBD/OptimizeMaxError" (maybe 0.05 or 0.1 depending on how precise is your odometry) so that these wrong loop closures are rejected.

cheers,
Mathieu

[polarbear]

frames.pdf
Hi,
I`m sorry for reply too late.My computer system is broken in a update and i fix it took a lot of time.
Here is my frames and the screenshot when mistake happen.

[polarbear]

It`s another situation that wrong loop closures are rejected like you say.Decreasing "RGBD/OptimizeMaxError" is worked.

But i don`t know why sometimes the upper floor`s situation will happen.All that pointclounds created after teleport are even not at the same plane.

And I found i will get the higer quality of map  in rtabmap window than in the rviz,is it my illusion？

[matlabbe]

Hi,

That one seems indeed a wrong loop closure, which can be solved be decreasing "RGBD/OptimizeMaxError" as you did. If point clouds are not on the same plane after the wrong loop closure, it is because the graph is optimized in 6DoF. You can force 3DoF optimization with Reg/Force3DoF=true (EDIT Optimizer/Slam2D=true for version < 0.11.12).

Between RVIZ and rtabmapviz, it depends on the MapCloud plugin parameters and Preferences->3D Rendering parameters respectively.

cheers,
Mathieu