Costmap not getting cleared properly

[stevemacenski]

Costmap not getting cleared properly

[SteveMacenski]

I'm not sure what you're referencing. I see the white points (I presume those are the voxel grid points not your sensor readings, right?) marking the costmap in that table in the center. After the acceleration period the points and costmap points disappear.

I think you may be referencing those odd triangular regions after other areas decay that later go away? I can say that I've never seen that before. All of this data looks too perfect, is this a simulation? I notice things updating incredibly slowly, this could be a function of computational load on your computer if you're using gazebo- the depth camera plugin itself really takes a toll. The costmaps update based on changes so it could be its taking a bit of time for all the information to get through the pipe from the last iteration.

Can you confirm this behavior in 'real life'?

[SteveMacenski]

I think I understand better, those white points are sensor outputs not actually the voxel grid, right? Its very likely there are still points marked in the grid, try visualizing the actual /voxel_grid.

[Kapcirt]

You are right, this is a simulation.
For the other comments, the whites points are the voxel grid and not the sensor input and it's really slow because I'm updating the global costmap at a frequency of 5Hz and plublishing it at 2 Hz but now that you said it, it is indeed much slower.

I'll try to test it in real environment to see how it is behaving.

Thanks for your help, I'll keep you updated if I find anything new

[SteveMacenski]

If this is a simulation issue, certainly we should work to resolve it since many people rely on it. PRs are welcome

[SteveMacenski]

I'm going to close this unless there's an update here saying you have issues in real-life. If you're finding that it works in real life fine and you've debuged your simulation environment, we should file a new ticket around simulator testing.

[Kapcirt]

I think I found what was wrong, I just needed to have an update frequency lower then the publish frequency. In my previous configuration they were equal and I guess (not sure though) that there was a conflict between the update and publishing of the costmap.

https://youtu.be/5Jl_rIEvC10

PS : When trying to use the mapping mode, this error popped up
what(): Time is out of dual 32-bit range
From what I read, it is a problem that occurs in simulation because the ros::time type is uint32_t and it cannot express negative value which can happen in simulation ( https://answers.ros.org/question/209452/exception-thrown-while-processing-service-call-time-is-out-of-dual-32-bit-range/ ). Just to inform you in case you wanted to make it simulation friendly.

Thanks for the help

[SteveMacenski]

Awesome! Glad you got it working. Having the publish higher than the update may result in undefined behaviors- nice catch.

Mapping mode is mostly a toy just to show that it is possible to generate 3D maps of size with low overhead, but I should look into that. The fixes seem minimally simple.

[Kapcirt]

Hey, I still got some issues (it is only working when the obstacle_laser_layer and the stvl_layer and enabled together), I was wondering what is your global costmap resolution ?
Since this parameters is getting set by the static map, I can't really modify it and I was thinking that it might be important

[Kapcirt]

Here is another video https://www.youtube.com/watch?v=y4ctBYCbiiI&feature=youtu.be from the top view we can see that the clearing is half done at some spot.
Is it possible that it has something to do with the updateBounds that doesn't include points ?

[SteveMacenski]

The global costmap is the same resolution as the static map, if the static map is included as a layer.

It looks like you're in the odom frame, not map frame, which may skew what the results you're looking at are. Your resolution doesn't appear to be consistent, but its difficult for me to tell without the voxels visualized at their true size (unless they are and you've set the resolution incorrectly).

[SteveMacenski]

Hi! I'm compiling some photographs of people's robot that have tried or are using the STVL for my upcoming ROSCon 2018 talk on this layer and its use. I'd love if you took a screen shot of your simulations or if you had hardware as an example! thanks!

https://roscon.ros.org/2018/

[Kapcirt]

Hi! Sorry for the late answer, here are the picture of the simulation and the hardware on which I tested the STVL.
Good luck for your talk at the ROSCon !

[SteveMacenski]

Awesome! Thanks for the note!