NewtonCollisionClosestPoint() and compound shapes

[Bird]

Hi Julio,

When painting I use NewtonCollisionClosestPoint() to help to set the user defined gap between instances. But it wouldn't work with Compound shapes like those created with NewtonMeshApproximateConvexDecomposition() because of there's a dgAssert(0) on line 435 of dgNarrowPhaseCollision.cpp.

Any chance you could fix that?

-Bird

[Julio Jerez]

Oh yes I did no implemnt
bool dgCollisionConvex::CalculateClosestPoints (dgCollisionParamProxy& proxy) const
for compound yet, teh code is ter commnete out,

I will make funtional tomorrow, it is no that diffcult.

[Bird]

I will make funtional tomorrow, it is no that diffcult.

Great! Thanks

-Bird

[Bird]

Oh yes I did no implemnt
bool dgCollisionConvex::CalculateClosestPoints (dgCollisionParamProxy& proxy) const
for compound yet, teh code is ter commnete out,

I will make funtional tomorrow, it is no that diffcult.

Hi Julio,

Do you think you will be able implement this? The actual line is 465 of dgNarrowPhaseCollision.cpp, not 435 like I said previously in error.

-Bird

[Julio Jerez]

yes, I will, thsi is actaully part of the SDK API, I just did no have tiem to write iot.

I did not do last weekend, I will try this next one, it is actaully very simple to do it.

[Bird]

I did not do last weekend, I will try this next one, it is actaully very simple to do it.

Hi Julio,

Sorry to keep bugging you about but it's kind of a big problem for my project.

And did you ever get the chance to try to handle this problem? http://newtondynamics.com/forum/viewtopic.php?f=9&t=8579

I'd love to be able to do completely intersection free painting if possible.

-Bird

[Bird]

Hi Julio,

Any chance you can take a look at this? I need to be able to paint with Compound Shapes and this is a showstopper for me. Thanks -Bird

[Julio Jerez]

Is this the same a the problem with the long skinny box?

Oh I see the closest point on compound is no implemented in the engine core.
OK let us do this, I will do the long skinny box bug, and add the compound closest point for compound tomorrow.

[Bird]

OK let us do this, I will do the long skinny box bug, and add the compound closest point for compound tomorrow.

Excellent. Thanks!

-Bird

[Bird]

OK let us do this, I will do the long skinny box bug, and add the compound closest point for compound tomorrow.

Any luck?

-Bird

[Julio Jerez]

Ok I am doing this now.
but I am a lite confused here.
the function that is commented out is this

Code: Select all

dgInt32 dgWorld::ClosestPoint(const dgCollisionInstance* const collisionSrcA, const dgMatrix& matrixA,
                       const dgCollisionInstance* const collisionSrcB, const dgMatrix& matrixB,
                       dgTriplex& contactA, dgTriplex& contactB, dgTriplex& normalAB,dgInt32 threadIndex)
{
   dgKinematicBody collideBodyA;
   dgKinematicBody collideBodyB;
   dgContactPoint contacts[DG_MAX_CONTATCS];

   dgCollisionInstance collisionA(*collisionSrcA, collisionSrcA->GetChildShape());
   dgCollisionInstance collisionB(*collisionSrcB, collisionSrcB->GetChildShape());

   collideBodyA.m_matrix = matrixA;
   collideBodyA.m_collision = &collisionA;
   collisionA.SetGlobalMatrix(collisionA.GetLocalMatrix() * matrixA);

   collideBodyB.m_matrix = matrixB;
   collideBodyB.m_collision = &collisionB;
   collisionB.SetGlobalMatrix (collisionB.GetLocalMatrix() * matrixB);


   dgContactMaterial material;
   material.m_penetration = dgFloat32 (0.0f);

   dgContact contactJoint (this, &material);
   contactJoint.SetBodies (&collideBodyA, &collideBodyB);

   dgCollisionParamProxy proxy(&contactJoint, contacts, threadIndex, false, false);

   if (collisionA.IsType (dgCollision::dgCollisionCompound_RTTI)) {
      dgAssert (0);
/*
      return ClosestCompoundPoint (&collideBodyA, &collideBodyB, contactA, contactB, normalAB, threadIndex);
   } else if (collisionB.IsType (dgCollision::dgCollisionCompound_RTTI)) {
      dgInt32 flag = ClosestCompoundPoint (&collideBodyB, &collideBodyA, contactB, contactA, normalAB, threadIndex);
      normalAB.m_x *= dgFloat32 (-1.0f);
      normalAB.m_y *= dgFloat32 (-1.0f);
      normalAB.m_z *= dgFloat32 (-1.0f);
      return flag;
*/

can you verified that this is where you are getting the assert? I do no remember the convex cast using that function.

[AntonSynytsia]

Just in case it wasn't mentioned, closest distance doesn't work with tree and compound collisions. Although it does work with all convex shapes, i.e. cylinders, cones, spheres, and convex hull itself.

[Bird]

can you verified that this is where you are getting the assert? I do no remember the convex cast using that function

This is a separate issue from convex cast. When I'm painting I use NewtonCollisionClosestPoint() to allow the user to set a gap between instances so when the user is painting with compound shapes that;s when I'm hitting the assert.

Thanks taking the time to look into this!

On another note, have you ever looked at DLib? ( I just discovered it) http://dlib.net/ Looks extremely well done with lots of goodies that you might find useful for Newton.

-Bird

[Julio Jerez]

Ok now you have compound/convex and compound/compound

@ AntonSynytsia
the closest distance does not support closest distance to collision trees, or heightfield, do you nee that?.
It does support compound collision as they are consideredd convex.

@Bird
Sync again, and tell me how it goes.
tomorrow I will do the long shape placement. Is that working fine?

[AntonSynytsia]

@Juleo
No, I think I won't need that for tree collisions. I was just testing.