Help wanted: Constructing a CollisionObject from multiple primitives

[tfachmann]

I am using the fcl library as transition to have a better representation of collision objects. As I don't have a mesh for all of my objects (yet) I have used a collection of spheres with a given radius to approximate the collisions.

While switching to fcl, I now want to be able to represent both: a complete mesh and my previous (naive) method as FCL Objects. Is there a functionality to construct an CollisionObject which is the union of Spheres?

I have found fcl::generateBVHModel which appears to provide the functionality I need, but it didn't work as expected (maybe I just didn't understand it). I have used it as such:

typedef fcl::BVHModel<fcl::OBBRSSf> MeshModel;
// t_transformations describes a relative transformation to the center of the sphere,
// where all have the same radius t_radius
fcl::CollisionObjectf toFCL(const double t_radius, const std::vector<Eigen::Matrix4f>& t_transformations)
{
  std::shared_ptr<MeshModel> geom = std::make_shared<MeshModel>();
  for(size_t i = 0; i < t_transformations.size() - 1; ++i) {
    auto sphere = fcl::Spheref(t_radius);
    fcl::Transform3f trafo(t_transformations[i]);
    fcl::generateBVHModel(*geom, sphere, trafo, 32, fcl::FinalizeModel::DONT);
  }
  auto sphere = fcl::Spheref(t_radius);
  fcl::Transform3f trafo(t_transformations[i]);
  fcl::generateBVHModel(*geom, sphere, trafo, 32, fcl::FinalizeModel::DO);
  geom->computeVolume()   // == 0   (?)
  return fcl::CollisionObjectf(geom, fcl::Transform3f::Identity());
}

Am I using the interface correctly? If not, is it even possible to construct an CollisionObject from multiple primitives? If I'm using it correctly, I can elaborate further on what goes wrong / is unexpected.

[ddengster]

hi, i'm also trying to get things to work with multigeometry swept collision. How you're constructing the models seems to be correct (though you might want to know that the fcl::generateBVHModel function converts the circles to triangles and can be inefficient).

Now, on my setup with even just 1 box being added to the fcl::BVHModel<fcl::OBBRSSf> , then combined with a fcl::SplineMotion and tested for swept collision, i get an infinite loop with the distanceRecurse function here: https://github.com/flexible-collision-library/fcl/blob/master/include/fcl/narrowphase/detail/conservative_advancement_func_matrix-inl.h#L496 and it eventually results in a stack overflow exception. Are you doing something similar to this and getting the same results I have?

If someone could help in this area it would be great!

[tfachmann]

Hi, thanks for your feedback regarding the construction of the BVHModel geometry. For my use-case, I assume it would be best to not triangulate the spheres by constructing a mesh, but better hold a vector of fcl:Spheres instead to then check the collision against each sphere.

I was only using the fcl::collide() and fcl::distance() interface, where I have not encountered a stack overflow exception. My error was more on a semantic layer which I had blamed to the construction of the mesh. Thus, your error is probably something else.