NewtonCollisionCalculateAABB bug?

[Marc]

No, I started using it with 2.11 and because of the strange results I got and because you haven't released the version puplicly yet, I tried with another newton version and got different results. And then I tried with even another version and got different results again.

[Marc]

Ok, I just tried that: Using NewtonCollisionForEachPolygonDo on the collision and doing a simple min and max of all vertices, I get the correct bounding box:

original mesh manually calculated bb: RAW BB: -3.126976 -3.616533 -0.019121 3.500324 3.568864 10.969679
calculated with the supportvector code: NTC BB: -2.006744 -3.275425 -0.019121 3.480575 3.260570 10.969678
calculated with min/max/NewtonCollisionForEachPolygonDo: NTC BBITER: -3.126976 -3.616533 -0.019121 3.500324 3.568864 10.969678

As a sidenote: There is a small rounding accuracy with the last digit of the last value. That's not bothering me though.

[Julio Jerez]

so the Box using support vertex is wrong?

[Marc]

Yes, unless my code from above, which basically resembles your code from further above is wrong and causes this issue.

[Julio Jerez]

I see, that looks like a big Bug I will check it out out.


The rounding error is because the AABB using support vertex comes from a Dot product projection while the AABB calcualted fron checking all vertices comes from a direct compare.

[Julio Jerez]

you are righr mark in it totally bogus, I just tested It must happen in version 2.11
I am fixing it now, an dI will post Version 2.11 even if teh destruction is not finish yet.
There are few important Bug fixes that are important.

[Marc]

Ok. I'm looking forward to the new version

[Julio Jerez]

Ok I fixed it, it was a stupid bug I intrudeced in 2.10 when I changed the format to make it share code with the compound shape.
I added this test code to the SDK demos and it seems to work fine.

Code: Select all

   {
      NewtonCollisionInfoRecord collisionInfo;
      NewtonCollisionGetInfo (collision, &collisionInfo);
      if (collisionInfo.m_collisionType == SERIALIZE_ID_TREE) {
         int count;
         dVector p0(-1000, -1000, -1000);
         dVector p1(1000, 1000, 1000);
         const dFloat* vertexArray;
         int vertexStrideInBytes;
         int vertexCount;
         int indexList[256];
         int attributeList[256/3];
         count = NewtonTreeCollisionGetVertexListIndexListInAABB (collision, &p0[0], &p1[0],
                                                     &vertexArray, &vertexCount, &vertexStrideInBytes,
                                                                 indexList, sizeof (indexList)/sizeof (indexList[0]),
                                                                 attributeList);

         // check AABB accuracy

         vertexStrideInBytes /= sizeof (dFloat);
         dVector minP1 ( 1.0e10f,  1.0e10f,  1.0e10f, 0.0f);
         dVector maxP1 (-1.0e10f, -1.0e10f, -1.0e10f, 0.0f);
         for (int i = 0; i < vertexCount; i ++) {
            for (int j = 0; j < 3; j ++ ) {
               minP1[j] = vertexArray[i * vertexStrideInBytes + j] < minP1[j] ? vertexArray[i * vertexStrideInBytes + j] : minP1[j];
               maxP1[j] = vertexArray[i * vertexStrideInBytes + j] > maxP1[j] ? vertexArray[i * vertexStrideInBytes + j] : maxP1[j];
            }
         }

         dVector minP;
         dVector maxP;
         dMatrix matrix (GetIdentityMatrix());
         CalculateAABB (collision, matrix, minP, maxP);

         minP1 -= minP;
         maxP1 -= maxP;
      }
   }

It is late now I will post beta 2.11 tomorrow.
Thank you for the Bug, I would never find it

[Marc]

You are welcome! Thanks for fixing it!

[Marc]

Ok in the public 2.11 version the aabb now seems to be correct for me as well