inflation with kangaroo - precision problem with box collision

Hello,

I've got a problem about inflation and collision with kangaroo (I am not using the latest version 2.0 but the older). In the definition, a geodesic sphere mesh is inflated between two boxes. The inflation works well but it seems to fail when "colliding" the edges and corners of the boxes (see the .jpg below), the mesh penetrate into them. I have tried to use brep, meshes less or more subdivised, ajust kangaroo settings, change its solver... but without result.
Is there someone who have the same problem ? Do you have any solution ?

thanks, Ludovic

Attachments:

inflationfail2.jpg, 172 KB

Replies to This Discussion

Permalink Reply by Daniel Piker on May 4, 2015 at 6:52am

Hi Ludovic,

In Kangaroo 0.099, the collision is only preventing the vertices of the sphere from going inside the boxes. This does not totally prevent any overlap of the volumes, because other types of intersection can occur, such as when the collision happens between the edges of one mesh and another.

In 2.02 this is improved because the interaction can be 2-way, so the vertices of the box can also not go inside the sphere, and will push on its faces if we try to make them, which would give less intersection at the corner like you see above.

However, edge-edge collisions are still not included, so there is still a degree of approximation that occurs (most noticeably when the shapes have sharp edges and corners). Refining the mesh will reduce this, though of course denser meshes get more intensive to compute.

I do have some ideas about how to implement a mesh-mesh collider that also deals with such edge-edge collisions, but it could take a while.

For an idea of the complexity involved in real-time collision detection, take a look at the contents of this book on the subject:

http://realtimecollisiondetection.net/books/rtcd/toc/

(and that's just for rigid objects - deformables add even more problems)

Permalink Reply by ng5 Alex on May 4, 2015 at 7:05am