fixing adding tables to the collision world

[nlyubova]

@vrabaud : fyi

[hersh]

There is a new bug here. Consider what happens when j == 0. triangles[0] = 0, triangles[1] = 0, and triangles[2] = 1. So the first triangle is always a line. Then because of the loop ending just before vertices.size() - 2, the last (needed) triangle is missing.

I think the correct code should look like you have it, with lines 132 and 133 changed to read:

triangles[i3++] = j+1;
triangles[i3] = j+2;

[sachinchitta]

This PR seems to have a lot of changes that are not relevant to the fix being made. @nlyubova could you please update this PR and remove all your debug comments.

[nlyubova]

I've just cleaned the last commit, sorry for the debug comments.

[sachinchitta]

Could you address hersh's comment too.

[dornhege]

I believe that hersh's comment is the correct way.

If I understand this correctly, tables are added as planar meshes. This is a problem for the point cloud updaters, where a 3D convex hull is produced that degenerates. This leads to arbitrarily large portions of the octree being removed. So, I'd recommend to add a fixed table height.

[nlyubova]

When I try the Hersh's way, the table is not displayed, so I do not know a better way to fix it

[davetcoleman]

@nlyubova do you have any interest in moving this change to the new repo? https://github.com/ros-planning/moveit

otherwise I will close it for now

[v4hn]

It seems everyone involved in this request agrees that there is a problem with adding tables to the collision world.
I'd like to keep this open as a reference until this is either fixed or at least ported.

[davetcoleman]

i closed it because no good solution seemed to have ever been found, but i'll close PRs with a little more caution