Trigger Volumes

[carli2]

But why does newton provide these two functions? NewtonCollisionIsTriggerVolume and NewtonCollisionSetAsTriggerVolume

[Julio Jerez]

because the are legacy from core 200. The kinematic body is a more powerfull concept

latee when the I add teh option DoNoGenerate contact to a collsion shape, then a full truuger will be a kenematc body in wich teh shape has that option set.
this will make lighwight. if teh option is set then it will generate contacts.

a dynamics body can also have a compond shape in which some subshapes has the do not generate contact option set to on. thoso will act as trigger that are part of the shape and you get the callback.

This is a more more peweful and flexible system, and it does no required special code pathway in the collision system.

Please try using kinematic body for trigger, you will see it is better.

[carli2]

In the moment I use the following code:

In the ConvexCast prefilter I kick the triggers out to get a view of the solid world:

Code: Select all

function myprefilter(body: PNewtonBody; collision: PNewtonCollision; userData: Pointer): cardinal; cdecl;
begin
 result:=integer((TWorldObject(userData).newton<>body) and (NewtonCollisionIsTriggerVolume(NewtonBodyGetCollision(body))=0));
end;

A collision in my game is marked as trigger and then newton is informed about the state:

Code: Select all

if trigger then NewtonCollisionSetAsTriggerVolume(col, 1);

How should I use your system now? How do I have to change the prefilter code to detect a triggered body?
I think I have to move the trigger code from collision creation to the object creation.

Is there some entry in the wiki that explains the difference bezween dynamic and kinematik?

[Julio Jerez]

You create a kinematic body

Code: Select all

Int mytriggercode = some special value
NewtonWorld* const world = scene->GetNewton();
NewtonCollision* const collision = CreateConvexCollision (world, &aligment[0][0], size, _BOX_PRIMITIVE, mytriggercode);
NewtonCollision* triggerBody = NewtonCreateKinematicBody(world, collision, &matrix[0][0]);
      NewtonBodySetUserData(m_tornadoBody, this);
      NewtonBodySetTransformCallback(m_tornadoBody, DemoEntity::TransformCallback);

In a convex cast you can check for the collision id of the collision shape of the hit body, if the ID match mytriggercode, then it is a trigger.

for body to body you iterate over the contact joint of the triggerBody, each joint will be a boy in teh trigger shape.
or you can do the same in a callback.

Sync to svn, I just place 25 players, in te hscene, the players are kenamti bodies, and they are all collected in the callback of the calling player.
you cn amove aroiudn and you see hwo i collide, by you can also make no colladibel by rejectin that contact.
you can do anything you want, and it is cost effective.

if you still has problem I will make a trigger demo.

The triggers are all there, it is just that they are no called triggers, because they are much more versitile than just triggers.

[carli2]

Yes, that works, thanks

(I abused the UserID of the collision to reflect the trigger flag and create kinematic bodies for triggers. the RPG works now)

[Julio Jerez]

(I abused the UserID of the collision to reflect the trigger flag and create kinematic bodies for triggers. the RPG works now)

That's is not abused, tha is the porpuse of that field is for the client application to do what si want to do with it.
one perfect ligitimal use if to make materials.
That oen fo teh big diffrentce with core 200, copre 200 shape were shared and teh coudl no have diffrent ID unless they where duplicated.
now two identical boxes, for example, can have different IDs and still share the data internally.

what Game is that you are working?

[carli2]

http://goldenwipf.de

It's a kind of universal game engine, so the engine may run untrusted code (from a server in a network game) and the physics are completely accessible via the scripting API.
It's a very big and interesting project, we will release in December.

[pHySiQuE]

Has anything been done that would cause NewtonCollisionClosestPoint() to stop working as an intersection test? It's supposed to return 0 if the bodies intersect, right?

[Julio Jerez]

I did not understand that question.
But I just re enable the closest point demo which use that function , and it seems to work

[pHySiQuE]

I am using the function NewtonCollisionClosestPoint() as an intersection test. If the function returns 0, the collisions are intersecting. If the function returns 1, they are not intersecting, and a closest point is returned.

This code demonstrates the behavior has changed, and returns 1 when collisions are intersecting. (It should return 0.) This means I can't use it as an intersection test:

Code: Select all

NewtonWorld* newtonworld = NewtonCreate();
NewtonCollision* newtoncollision = NewtonCreateBox(newtonworld,10,10,10,0,NULL);
dFloat mat0[16] = {1.0,0.0,0.0,0.0, 0.0,1.0,0.0,0.0, 0.0,0.0,1.0,0.0, 0.1,0.0,0.0,1.0};
dFloat mat1[16] = {1.0,0.0,0.0,0.0, 0.0,1.0,0.0,0.0, 0.0,0.0,1.0,0.0, 0.0,0.0,0.0,1.0};
dFloat contactA[3];
dFloat contactB[3];
dFloat normalAB[3];
if (NewtonCollisionClosestPoint(newtonworld, newtoncollision, mat0, newtoncollision, mat1, contactA, contactB, normalAB, 0)==1)
{
   printf("1");
}
else
{
   printf("0");
}

[Julio Jerez]

This si the code closet of closet point funtion

Code: Select all

bool dgCollisionConvex::CalculateClosestPoints (dgCollisionParamProxy& proxy) const
{
   _ASSERTE (this == proxy.m_referenceCollision->m_childShape);
   dgMinkHull minkHull (proxy);
   minkHull.CalculateClosestPoints ();

   dgContactPoint* const contactOut = proxy.m_contacts;
   dgCollisionInstance* const collConicConvexInstance = proxy.m_referenceCollision;
   dgVector p (ConvexConicSupporVertex(minkHull.m_p, minkHull.m_normal));

   const dgVector& scale = collConicConvexInstance->m_scale;
   const dgVector& invScale = collConicConvexInstance->m_invScale;
   const dgMatrix& matrix = collConicConvexInstance->GetGlobalMatrix();
   dgVector normal (matrix.RotateVector(invScale.CompProduct4(minkHull.m_normal)));
   normal = normal.Scale(dgRsqrt(normal % normal));

   contactOut[0].m_normal = normal;
   contactOut[0].m_point = matrix.TransformVector(scale.CompProduct4(p));

   contactOut[1].m_normal = normal.Scale (-1.0f);
   contactOut[1].m_point = matrix.TransformVector(scale.CompProduct4(minkHull.m_q));
   return true;
}

as you can see it does not return false ever. that funtion is called by thsi funtions

Code: Select all

dgInt32 dgWorld::ClosestPoint (dgCollisionParamProxy& proxy) const   
{
   dgCollisionInstance* const collision1 = proxy.m_referenceCollision;
   dgCollisionInstance* const collision2 = proxy.m_floatingCollision;
   if (!(collision1->GetConvexVertexCount() && collision2->GetConvexVertexCount())) {
      return 0;
   }

...
}

as you can see it can return 0 or false, by this is for the cases of shapes that can not be casted. stuff like heighfield, NULL collsion scene collsion
this is different that preivisue version wherer closest point returned 0, in new core 300 it always returns the closest point even if the shape are penetrations.

This is more useful beacsue it also give you teh same funtionality, to can detect penetration usin closet point by doingt this

Code: Select all

      bool penetration = false;
      if (NewtonCollisionClosestPoint(world, collisionA, &matrixA[0][0], m_castingVisualEntity->m_castingShape, &matrixB[0][0], &m_castingVisualEntity->m_contact0[0], &m_castingVisualEntity->m_contact1[0], &m_castingVisualEntity->m_normal[0], 0))
         penetration = (((m_castingVisualEntity->m_contact1 - m_castingVisualEntity->m_contact0) % m_castingVisualEntity->m_normal) >= 0) ? true : false;

[Julio Jerez]

thsi is your same example modified

Code: Select all

NewtonCollision* newtoncollision = NewtonCreateBox(newtonworld,10,10,10,0,NULL);
dFloat mat0[16] = {1.0,0.0,0.0,0.0, 0.0,1.0,0.0,0.0, 0.0,0.0,1.0,0.0, 0.1,0.0,0.0,1.0};
dFloat mat1[16] = {1.0,0.0,0.0,0.0, 0.0,1.0,0.0,0.0, 0.0,0.0,1.0,0.0, 0.0,0.0,0.0,1.0};
dFloat contactA[3];
dFloat contactB[3];
dFloat normalAB[3];
if (NewtonCollisionClosestPoint(newtonworld, newtoncollision, mat0, newtoncollision, mat1, contactA, contactB, normalAB, 0)==1)
{
   // shape has close point
   // cheack for penetration
      dFloat diff[3];
     diff[0] = contactB[0] - contactA[0];
     diff[1] = contactB[1] - contactA[1];
     diff[2] = contactB[2] - contactA[2];
      bool penetration = dotproduct (diff, noprmalAB) < 0 ? true : false;
}
else
{
  // shape do no have closest pooint
   printf("0");
}

[pHySiQuE]

I had to flip the sign of the diff vector, but yeah it works. Thanks!