This topic is also here: http://www.gamedev.net/topic/633955-ragdoll-physics-rendering-issues/ but I realized after posting, it might make more sense in this forum? Feel free to remove the dupe.

Hi folks,
I have reached my wits end in trying to get my model to align with its ragdoll, and I was hoping you could help me a with a couple questions. I am using BEPU physics for my simulation, where my Player primarily moves around via a CharacterController object until he dies, and then the ragdoll is activated.


I am currently trying to simply get the player model to stay locked with the the "body" entity as an initial working case before attaching each limb. The problem I have is pictured below:


As the ragdoll moves, the model does not stay locked with the body entity, and the more it rotates, the more the model separates. I presume this has to do with one (or both) of these 2 things:


1. The world transform of the model represents the center of the CharacterController capsule (the pink thing pictured). I use an offset during drawing that adjusts the draw position, but this may be causing wierdness when rotations are added from the body entity. Should I essentially just pre-offset my CharController's position, and then pass that Transform to the ragdoll class and draw functions?

2. I am attempting to calculate the offset of the model to the ragdoll, which I then factor in when passing back the bone transforms, so that apples are compared to apples when the body does start to rotate/translate. I may be doing this wrong? This is my psuedocode for calculating that offset:
Matrix offset = Matrix.Invert(bindPose[bindPoseIndex]) * Entities["body"].WorldTransform
and my code for reapplying it when the bone transforms are being calculated:
Matrix.Invert(offset) * body.WorldTransform * actor.WorldTransform (actor is the char controller)


I am pretty sure that once I can figure this out, the rest of the ragdoll will be gravy, but I have been racking my brain for about 30 total hours on this issue to no avail, and I need some help thinking outside the box.

Thanks!

After thinking through what I said above, I tried a simple thing and it worked to solve the problem mentioned above. Here is what I did:

1. Stopped using the offset world position for my the ragdoll and draw transforms. This made it so my model always at least drew where it was supposed to in bind pose while ragdolls were enabled. Again, the offset was an artifact of using a physics entity that had it position set to the middle, and then having to reverse that when drawing.
2. Used an offset for the entire skeleton (in the bone-entity mapping phase). Prior, I had been only getting offsets for each bone, starting with the hip (aka bones[1]). Since the hip is centered in the middle of the body, it's transforms would be offset weird from the model's center, which is actually between it's feet. I believe this was also worsened by the fact that the physics entity's orientation UP is different than the models.

Now, I move onto binding each bodypart... I suspect I will have issues doing this as well; if so you'll be certain to see me back here

Ok, I definitely need help, as my mathbrain is simply non-functional here. Now that the root node is properly following my physics entities "body", I am struggling to get the transforms for the rest of the skeleton. I am getting crazy stretching, presumably because I'm not working in the right "space" when calculating those transforms.

This is what I'm doing, starting with how I calculate the offsets of physics entity-to-bone:
[source lang="csharp"] offsets.Add("Armature", Matrix.Invert(bindPose[0]) * Entities["body"].WorldTransform); //Calculate root offset first
for (int i = 4; i < input.Model.Bones.Count; i++)
{
int poseIndex = i - 3;
string bone = input.Model.Bones.Name;
if ((bone == "neck"))
offsets.Add(bone, Matrix.Invert(bindPose[poseIndex]) * Entities["head"].WorldTransform )... // get each "physics-tied" bone's relative offset[/source]
Now I calculate the boneTransforms:

[source lang="csharp"]Matrix[] boneTransforms = new Matrix[bindPose.Count];
Matrix finalOffset = offsets[0] * body.WorldTransform;
boneTransforms[0] = finalOffset * Matrix.Invert(actorWorld); //establish root

for (int i = 1; i < bindPose.Count; i++) //get offsets for other bones
{
int boneIndex = i + 3;
string bone = input.Model.Bones[boneIndex].Name;
if (bone == "neck")
{
offset = Matrix.Invert(offsets[bone]) * Entities["head"].WorldTransform;
boneTransforms = offsets[bone] * Matrix.Invert(actorWorld);
}
else //offsets for non-physics bones
{
boneTransforms = bindPose;
}
}[/source]
And finally, converting the bone transforms into world space:

[source lang="csharp"]Matrix[] worldTransforms = new Matrix[modelData.BindPose.Count];
Matrix[] skinTransforms = new Matrix[modelData.BindPose.Count];

worldTransforms[0] = boneTransforms[0];
skinTransforms[0] = inverseBindPose[0] * worldTransforms[0];

// Child bones.
for (int bone = 1; bone < worldTransforms.Length; bone++)
{
int parentBone = skeletonHierarchy[bone];
worldTransforms[bone] = boneTransforms[bone] * worldTransforms[parentBone];
skinTransforms[bone] = inverseBindPose[bone] * worldTransforms[bone];
}

return skinTransforms; //this gets passed into my shader (identical setup to what the XNA skinning example uses)[/source]

Again, this all works just fine for my root node, but the various bodyparts (in this case the head) ends up stretched and about 15 ft from where it should be. Can anybody help me figure out where I am going wrong in calculating the skintransforms?!