#55 GIMPACT collision trimesh box bug

[Mathieu Roger]

The attached code file :
build a given trimesh (a complex one because its from a real example)
build a box
* check for intersections

-->these two does not intersect but collision is detected

Appearing on 0.8 and 0.9

[Mathieu Roger]

[Remi Ricard]

Logged In: YES
user_id=145206
Originator: NO

Hi Mathieu,

I was looking at the code and it is hard to see if there should be a collision or not.

The z values for the trimesh goes up to at least 76.5136 and the position of the box is z=42.

Can you modify your code to make it display the trimesh and the box using drawstuff (see the demo).

Remi

[Mathieu Roger]

screenshot

[Mathieu Roger]

Logged In: YES
user_id=1245769
Originator: YES

File Added: objects.jpg

[Mathieu Roger]

[Mathieu Roger]

Logged In: YES
user_id=1245769
Originator: YES

Hi Remi,

here you are (a modified version of demo_trimesh.cpp). I cannot find a way to print collision points. Hope it helps. As you can see, ship and box are quite far away.

Regards

Mathieu
File Added: demo_trimesh.cpp

[Martijn Buijs]

Does one of the geoms have identity rotation? It may be affected by this bug:

http://sourceforge.net/tracker/?func=detail&aid=2214773&group_id=24884&atid=382799

[Mathieu Roger]

Yes the trimesh has identity orientation as you may see in the code previously attached (test-ode.txt, that is the test case, expressed in CppUnit to reproduce the issue).

But I have just tried to set another orientation :

dQuaternion orientation ;
dQFromAxisAndAngle(orientation,1,1,1,0.2) ;
dGeomSetQuaternion(TriMesh,orientation) ;

which changes nothing at all and I still have the bug with that orientation. Maybe I am wrong in thinking that orientation is not axis aligned.

By the way, I am now under version 0.11.1 (on a debian box) and I still reproduce the issue.

It is easy to reproduce with the automatic test case I have provided, but as shown by svn history, no automatic test has been added for 3 years. I am sorry to be so rude, I love ODE and I am very grateful to all it's contributors, but I deeply believe that the first step to correct this bug is to automate it.
If you need I could transform the attached code to the UnitTest format already present in ODE, which will take a few minutes, but I have no ability to correct that issue.