-
assigned issue to
GPU Mode - very roughly resoluted scans in the shape of a triangle
GPU
The GPU mode is really helpful, but it means that the scans in medium to long distance are no longer really usable. Algorithms cannot be tested realistically. Only when the GPU delivers similar results as the software mode, it is really usable. Is there already a solution for this problem?
Software
Comments (6)
-
-
reporter I'm also using Gazebo 7.14.0. Thanks for the hint, I had a closer look at the Gazebo issue tracker. This issue https://bitbucket.org/osrf/gazebo/pull-requests/2955/fix-gpuraysensor-vertical-rays/diff seems to be the one, but was alleady merged into the Gazebo7 Branch at 2018-06-18. So it should be included in Gazebo 7.14.0. But it seems to be still not fixed. Any idea?
-
Did you solved this problem? Since I meet the same problem during my simulation. I’m using Gazebo 7.16.0. My CPU will be overloaded without GPU support.
-
I have the same problem using Gazebo9.17. The quality of the point cloud produced by GPU is awful.
-
reporter Nothing ever changed, sadly. I don’t have the gazebo knowledge to fix it myself.
-
I just tried changing
resolution
in the sdf/urdf because it was mentioned in the gazebo PR, but no success. There is no solution for GPU accuracy that I am aware of.
https://osrf-migration.github.io/gazebo-gh-pages/#!/osrf/gazebo/pull-requests/2955/page/1 - Log in to comment
Confirmed in Gazebo 7.14.0 and 9.5.0 on my laptop with GTX 970M
The problem is probably with Gazebo and not this ROS node because the same exact same code is used for CPU/GPU. Is there anything like this on the Gazebo issue tracker?