Implementation of a re-meshing based implicit surface reconstruction #36
Conversation
|
Test PASSed. |
|
Test PASSed. |
|
Test PASSed. |
|
Test PASSed. |
…s, incl new launchfile and node
|
Test PASSed. |
|
Test PASSed. |
|
Test PASSed. |
|
Test PASSed. |
| @@ -21,6 +21,9 @@ class MeshModel { | |||
| static bool create(const std::string &off_pathm, MeshModel::Ptr *meshmodel_ptr, | |||
| bool verbose = false); | |||
|
|
|||
| MeshModel(Polyhedron &p); | |||
There was a problem hiding this comment.
Why can't we stick to the ::create syntax?
| @@ -35,7 +36,7 @@ class IsoSurfaceManifold : public CollisionManifoldInterface { | |||
| const double body_radius_; | |||
| bool is_constructed_; | |||
| const cgal::PolyhedronPtr original_manifold_; | |||
| cgal::Polyhedron collision_manifold_; | |||
| std::unique_ptr<cgal::MeshModel> collision_manifold_; | |||
There was a problem hiding this comment.
Why does this have to be a unique_ptr? Seems rather exotic.
| CGAL::Bounded_side side = m_is_closed ? m_side_of_ptr->operator()(p) : CGAL::ON_UNBOUNDED_SIDE; | ||
| if (side == CGAL::ON_BOUNDARY) return m_offset_distance; | ||
|
|
||
| cgal::Point closest_point = m_tree_ptr->closest_point(p); |
There was a problem hiding this comment.
Following the previous design of cad-percept, we tried to use MeshModel as our 'universal' cgal interface. Operations like distance queries are already part of that. Why not use it?
| bool success = offset_constructor_->execute(*original_manifold_, body_radius_, &collision_poly); | ||
|
|
||
| // non-explicit constructor of MeshModel takes care of this conversion. | ||
| *collision_manifold_ = collision_poly; |
There was a problem hiding this comment.
Wouldn't it make more sense to just create a new mesh-model and reset the pointer?
| @@ -16,12 +20,14 @@ void IsoSurfaceManifold::setBodyAttitude(const Eigen::Quaterniond& attitude) { | |||
| } | |||
|
|
|||
| double IsoSurfaceManifold::signedDistance(const Eigen::Vector3d& position) { | |||
| throw cad_percept::NotImplementedException(); | |||
| cgal::Point position_cgal = cgal::eigenVectorToCgalPoint(position); | |||
There was a problem hiding this comment.
don't we now have our nice implicit conversions for this? ;)
| << std::chrono::duration_cast<std::chrono::nanoseconds>(t_per_call).count() << " [ns]" | ||
| << std::endl; | ||
|
|
||
| EXPECT_TRUE(true); |
There was a problem hiding this comment.
can't we at least test some expectation here? like time > 0?
Needs to wait until #35 is merged.Ready to review.Contains yet another offset surface method.
Short summary of this method:
We ignore the meshing topology of the original mesh, and only use the original mesh to define an (implicit) meshing domain
This implicit meshing domain is evaluated by a function which returns the distance of a point to the offset surface, see
cad-percept/cpt_collision_manifolds/include/cpt_collision_manifolds/offset_surface/meshdomain_strategy.h
Line 27 in e9864a1
A meshing process is started to create a mesh that encloses this meshing domain. See lines following from here:
cad-percept/cpt_collision_manifolds/src/offset_surface/meshdomain_strategy.cc
Line 40 in e9864a1
Advantages are: It's quite generic and we have some control over how the resulting mesh is discretized (params in the meshing process). Runtime somewhat independent of original mesh resolution.
Disadvantages are: Really slow. Not feature preserving (e.g. sharp boundaries). Both are because we ignore the original mesh topology.
For now good as a intermediate test method to test impact of mesh quality on algorithms further down the line.
9sec solution (facet_size 0.03):
100sec solution (facet_size 0.01):
100sec solution (facet_size 0.01) for a larger offset distance:
45sec solution (facet_size 0.02) for a even larger offset distance to test self-intersections:
