Help Exporting Character Rig

[archaeology]

HAnim Export from Maya

Well, I have been able to export out HAnim characters from Maya, but only if the Joint Orient attributes of the joints are frozen first.

To see the result visit the following URL:
http://onaslant.ndsu.edu/x3d\_content/hanim/

However I have an issue that I still need to work out. As I said above, the export only works if the Joints have their "Joint Orient" attributes frozen before a mesh is bound to the skeleton. This is because the HAnim specification requires that the "rotation" field of the HAnimJoint node be set to "0 0 1 0" by default. Any other value is considered a a deformation movement.

I thought that this was a good solution for setting up character rigs and exporting them to HAnim, but as it turns out the animators who tested the export stated that the act of freezing the Joint Orient attribute makes it impossible for them to animate their characters because the attribute is frozen, the joints no longer rotate as you would expect a real-life person's joints to rotate.

Solution?

Well, I have a solution, but unfortunately, I do not have the math skills or 3D knowledge to make it work. I need somekind of formula to convert the "Rotate" value of an non-frozen joint to a "Rotate" value of a frozen joint so that if a joint of a skeleton is rotated, the identical joint (with its Joint Orient attribute frozen) on an identical skeleton moves its children to the same 3D world position as the children of the original skeleton.

Below is the API I code that I wrote to do this. It can also be viewed online at the following location:
http://onaslant.ndsu.edu/x3d\_content/hanim/

This code is obviously incorrect. Any suggestions on how to properly write this code is greatly appreciated. Any formula should have the frozen joint's Rotate attribute set to 0 0 0 when the original joint is set to 0 0 0.

void webX3DExporter::updateHARot(MNodeMessage::AttributeMessage msg, MPlug &plug, MPlug &otherPlug, void *clientData)
{
if ( msg & MNodeMessage::kAttributeSet )
{
if(plug.partialName(false, false, false, false, false, true) == "rotate" ||
plug.partialName(false, false, false, false, false, true) == "rotateX" ||
plug.partialName(false, false, false, false, false, true) == "rotateY" ||
plug.partialName(false, false, false, false, false, true) == "rotateZ" ||
plug.partialName(false, false, false, false, false, true) == "jointOrient" ||
plug.partialName(false, false, false, false, false, true) == "jointOrientX" ||
plug.partialName(false, false, false, false, false, true) == "jointOrientY" ||
plug.partialName(false, false, false, false, false, true) == "jointOrientZ" ||
plug.partialName(false, false, false, false, false, true) == "rotateAxis" ||
plug.partialName(false, false, false, false, false, true) == "rotateAxisX" ||
plug.partialName(false, false, false, false, false, true) == "rotateAxisY" ||
plug.partialName(false, false, false, false, false, true) == "rotateAxisZ")
{
double add0=0;
double add1=0;
double add2=0;

		double get0=0;
		double get1=0;
		double get2=0;

		MObject nodeObj = plug.node();

		MFnDependencyNode jNode(nodeObj);

		MPlug rPlug = jNode.findPlug("rotate");
		MPlug xr = rPlug.child(0);
		MPlug yr = rPlug.child(1);
		MPlug zr = rPlug.child(2);

		MPlug haPlug = jNode.findPlug("hanimRotation");
		MPlug xha = haPlug.child(0);
		MPlug yha = haPlug.child(1);
		MPlug zha = haPlug.child(2);

		MPlug joPlug = jNode.findPlug("jointOrient");
		MPlug xjo = joPlug.child(0);
		MPlug yjo = joPlug.child(1);
		MPlug zjo = joPlug.child(2);

		MPlug raPlug = jNode.findPlug("rotateAxis");
		MPlug xra = raPlug.child(0);
		MPlug yra = raPlug.child(1);
		MPlug zra = raPlug.child(2);

		xjo.getValue(get0);
		yjo.getValue(get1);
		zjo.getValue(get2);

		MEulerRotation joEuler(get0, get1, get2, MEulerRotation::kXYZ);
		MMatrix joMat = joEuler.asMatrix();//.inverse();

/*
add0 = add0+get0;
add1 = add1+get1;
add2 = add2+get2;
*/
xra.getValue(get0);
yra.getValue(get1);
zra.getValue(get2);

		MEulerRotation raEuler(get0, get1, get2, MEulerRotation::kXYZ);
		MMatrix raMat = raEuler.asMatrix();//.inverse();

/*
add0 = add0+get0;
add1 = add1+get1;
add2 = add2+get2;
*/
xr.getValue(get0);
yr.getValue(get1);
zr.getValue(get2);

		MEulerRotation rEuler(get0, get1, get2, MEulerRotation::kXYZ);
		MMatrix rMat = rEuler.asMatrix();

		MMatrix rMat1 = rMat.operator *(raMat);
		MMatrix rMat2 = rMat1.operator *(joMat);

// MMatrix rMat1 = joMat.operator *(rMat);
// MMatrix rMat2 = raMat.operator *(rMat1);
MEulerRotation nEuler;
nEuler.operator =(rMat2);

/*
add0 = add0+get0;
add1 = add1+get1;
add2 = add2+get2;
*/
add0 = nEuler.x - raEuler.x - joEuler.x;
add1 = nEuler.y - raEuler.y - joEuler.y;
add2 = nEuler.z - raEuler.z - joEuler.z;

// add0 = nEuler.x;
// add1 = nEuler.y;
// add2 = nEuler.z;

		xha.setValue(add0);
		yha.setValue(add1);
		zha.setValue(add2);
	}
}

}

[joojaa]

how do you interpolate the animation in hanim?

[archaeology]

QUOTE(Joojaa @ 12/30/05, 10:42 AM)
how do you interpolate the animation in hanim?
[snapback]223652[/snapback]

I've setup custom X3D Interpolator nodes in Maya that are counterparts to X3D interpolator nodes. For instance, in X3D there is an OrientationInterpolator node. The RawKee plug-in therefore has an x3dOrientationInterpolator node.

With the setup I have going now where the Joint Orient attributes must be frozen, I make a connection from the Joint node's "Rotate" attribute to the x3dOrientationInterpolator's "orientation" attribute.

When exporting the plug-in runs through the timeline based on attribute settings in the x3dOrientationInterpolator node and then stores the appropraite interpolation information in the x3dOrientationInterpolator node that is written to disk later on in the export process.

To get around the Joint Orient problem, I would like to calculate a new rotate value, that would rotate a frozen joint. I would store this value in a custom attribute on the Maya joint node as "hanimRotation". Then instead of connecting the Rotate attribute of the joint to the "orientation" attribute of the interpolator node, I would connect the "hanimRotation" attribute to the "orientation" attribute of the interpolator node.

I would monitor the Rotate Axis, Joint Orient, and Rotate attributes of the joint node and based on the input from these attributes update the hanimRotation attribute with a new value every time one of the monitored attributes has changed.

Aaron

[joojaa]

yes but se thing is that rotation in euler form of interpolation cant be just rotated and expect hthe animation to stay the same, bevcause the gimbal lock chages place.

[archaeology]

QUOTE(Joojaa @ 12/30/05, 11:21 AM)
yes but se thing is that rotation in euler form of interpolation cant be just rotated and expect hthe animation to stay the same, bevcause the gimbal lock chages place.
[snapback]223656[/snapback]

Hmmm... gimbal lock... I was afraid that might be involved. Well, I should clarify what the custom interpolator node does just to be sure we're on the same page. You may already understand, so if I'm just repeating something you already know, I apologize.
The custom interpolator node stores the value of the orientation attribute when the plug-in runs through the time line. The value is stored as a euler angle until it's written to disk, just before the values are written out, the plugin converts the euler angle to a quaternion values and writes them out as xyzw. So the actual interpolation of the playback in the X3D browser is of xyzw values.

[archaeology]

QUOTE(Joojaa @ 12/30/05, 11:21 AM)
yes but se thing is that rotation in euler form of interpolation cant be just rotated and expect hthe animation to stay the same, bevcause the gimbal lock chages place.
[snapback]223656[/snapback]

Sorry, my mistake. It doesn't convert it to a Quaternion, but instead converts it to an AxisAngle.

[rob-bateman]

if you get an MDagPath for a bone, you can use inclusiveMatrix() and inclusiveMatrixInverse() (of it's parent transform) to extract a local space matrix. Simply decompose the matrix to get rotation, scale & translation. That will ignore any jointOrients, rotateOrients. You might ge some probems with scale/rotate pivots being different, though this only affects transforms, not joints.

[joojaa]

Yes sure, but i still can se many caveats at this. eppecialy is the interpolstion happens in a parentted linkage, wich might lead to wierd interpolation problems. Quarteronions could hnadle it quite well but im not sure of how axis survives this angle, tough it should.

[rob-bateman]

QUOTE(Joojaa @ 01/03/06, 06:03 AM)
Yes sure, but i still can se many caveats at this. eppecialy is the interpolstion happens in  a parentted linkage, wich might lead to wierd interpolation problems. Quarteronions could hnadle it quite well but im not sure of how axis survives this angle, tough it should.
[snapback]223814[/snapback]

numerically axis angle and quaternions are the same, the only thing that differs is that quaternions save you a number of sin/cos calls when you interpolate (and are thus a bit more desirable for run-time usage)....

[archaeology]

QUOTE(Rob Bateman @ 01/03/06, 05:33 AM)
if you get an MDagPath for a bone, you can use inclusiveMatrix() and inclusiveMatrixInverse() (of it's parent transform) to extract a local space matrix. Simply decompose the matrix to get rotation, scale & translation. That will ignore any jointOrients, rotateOrients. You might ge some probems with scale/rotate pivots being different, though this only affects transforms, not joints.
[snapback]223810[/snapback]

Thanks for the input guys.

I'll give it a try.

Aaron

[joojaa]

QUOTE(Rob Bateman @ 01/03/06, 06:46 PM)
numerically axis angle and quaternions are the same, the only thing that differs is that quaternions save you a number of sin/cos calls when you interpolate (and are thus a bit more desirable for run-time usage)....
[snapback]223823[/snapback]

Yeah thats what i seemed to remeber. However thet is asuming they actualy interpolate form axis angele to axis angle, if they just interpolate channels of axis angle vaues to axis angle its a different thing alltogether.

Now why the hell would anybody do that, Well simply put you would hanve a easy to understand ease/tangent in controll that neither axis angle nor quat direct interpolation gives intuitively. But then youd be notmuch better than euler angles to begin with. And taht wouldnt be possible to work out.

Wich brings us back a full orbitaround to whare we started. Now if someone could just develop some GOOD way of adjusting euler curves in thisway (tough granted you can retime the interpolation, but ist still lacking something essential in terms of information)

[archaeology]

QUOTE(Rob Bateman @ 01/03/06, 05:33 AM)
if you get an MDagPath for a bone, you can use inclusiveMatrix() and inclusiveMatrixInverse() (of it's parent transform) to extract a local space matrix. Simply decompose the matrix to get rotation, scale & translation. That will ignore any jointOrients, rotateOrients. You might ge some probems with scale/rotate pivots being different, though this only affects transforms, not joints.
[snapback]223810[/snapback]

The following code almost works. The resulting euler angle rotates around the axis of the joint that has non-zeroed Joint Orient and Rotate Axis attributes. However, the angle is offset by the Joint Orient and Rotate Axis values. So if Joint Orient attribute equals -90 50 -90, the Rotate Axis attributes equals 10 -10 20, and the Rotate attribute equals 0 0 0, the custom "hanimRotation" attribute ends up equaling -54.494 67.731 -62.727. However, the hanimRotation attribute needs to equal 0 0 0 whenever the Rotate attribute equals 0 0 0 and rotate from there. Basically, I need to remove the offset. Any ideas on how to do that? I appreciate the help you guys have given.

Thanks,

Aaron

CODE
if ( msg & MNodeMessage::kAttributeSet )
{
 if(plug.partialName(false, false, false, false, false, true) == "rotate" ||
 --snip code--
 plug.partialName(false, false, false, false, false, true) == "rotateAxisZ")
 {
     MObject nodeObj = plug.node();

     MDagPath adp;
     MDagPath::getAPathTo(nodeObj, adp);
     MFnDagNode jNode(adp);

     MObject nodeObjP = jNode.parent(0);
     MFnDagNode pjNode(nodeObjP);
     MDagPath adpp = pjNode.dagPath();

     MMatrix jMatrix = adp.inclusiveMatrix();
     MMatrix pjiMatrix = adpp.inclusiveMatrixInverse();
     MMatrix localMatrix = jMatrix.operator *(pjiMatrix);
     MMatrix defMat;
     MEulerRotation fRot;
//      fRot.operator =(jMatrix);
     fRot.operator =(localMatrix);
     fRot.decompose(fRot.asMatrix(), MEulerRotation::kXYZ);

     MPlug haPlug = jNode.findPlug("hanimRotation");
     MPlug xha = haPlug.child(0);
     MPlug yha = haPlug.child(1);
     MPlug zha = haPlug.child(2);

     xha.setValue(fRot.x);
     yha.setValue(fRot.y);
     zha.setValue(fRot.z);
 }
}

[rob-bateman]

CODE
MVector trans;
double Scale[3];

// get path to transform
MDagPath path;
fnX.getPath(path);

// get path to parent transform
MFnDagNode fnP(fnX.parent(0));
MDagPath pp;
fnP.getPath(pp);

// get local space matrix
MMatrix mat = path.inclusiveMatrix() * pp.inclusiveMatrixInverse();
MMatrix matrot;

// extract translation from matrix
trans.x = mat[3][0];
trans.y = mat[3][1];
trans.z = mat[3][2];

// get scale from matrix
Scale[0] = sqrt( mat[0][0]mat[0][0] + mat[0][1]mat[0][1] + mat[0][2]*mat[0][2] );
Scale[1] = sqrt( mat[1][0]mat[1][0] + mat[1][1]mat[1][1] + mat[1][2]*mat[1][2] );
Scale[2] = sqrt( mat[2][0]mat[2][0] + mat[2][1]mat[2][1] + mat[2][2]*mat[2][2] );

// create 3x3 rotation matrix
matrot[0][0]=mat[0][0] / Scale[0];
matrot[0][1]=mat[0][1] / Scale[0];
matrot[0][2]=mat[0][2] / Scale[0];
matrot[1][0]=mat[1][0] / Scale[1];
matrot[1][1]=mat[1][1] / Scale[1];
matrot[1][2]=mat[1][2] / Scale[1];
matrot[2][0]=mat[2][0] / Scale[2];
matrot[2][1]=mat[2][1] / Scale[2];
matrot[2][2]=mat[2][2] / Scale[2];

// convert rotation to quaternion
MQuaternion R = matrot;

// convert to Euler angles (probably a very bad idea indeed;) )
MEulerRotation EulerR;
EulerR.decompose(matrot,MEulerRotation::kXYZ);

R and EulerR can be used as rotation values that ignore the orients - so it's never going to be 0,0,0 unless the jointOrient, rotateAxis and rotation are all 0,0,0 (since it's a concatonation of all three). It looks as though the values you are getting are the correct ones tbh. I'm not entirely sure why youd want zero'd values anyway? I can't see how it can affect any rendering engine for a game anyway.... ?

Anyhow, if you were to remove the rotations completely, then you will have to ensure that no instancing is supported in your engine since each instance would have to have it's normals & vertices transformed by the rotation values, relative to it's parent transform. In theory, you could remove the rotation by then simply using positional information for each joint (and a default rotation of 0,0,0).

During your export process, you might be able to just placing a locator node under each bone, create a parentConstraint to keep that locator relative to the bone, and use the above code to extract it's matrix (remembering to use the locators transforms, not the joint transforms!).

I still think that is far more hassle than it's worth, since either way the code for rendering and animating that will be identical whether you zero the rotations or not..... It just seems a lot of work for no benefit imho.

[archaeology]

Rob,

I appreciate the advice, its just that I'm locked into supporting the HAnim ISO Spec for X3D as this code is for the general X3D exporter for Maya. As such the HAnim ISO specification states that HAnimJoint nodes need to have a default setting of "0 0 1 0" when exported. If it were up to me, I'd just freeze the Joint Orient attribute before animating and binding the skeleton to the mesh, that way I could skip this entire formula issue all together. But the testers are complaing about having to do that.

I've refined my code a bit based on the code you posted. For the first joint in the skeleton, the code below does what I want it to do. However, the process starts to break down for subsequent joint nodes in the skeleton. The code works for the second joint in the skeleton chain if the first joint has not yet been rotated. However, if the first joint is rotated then the rotations are off for the second joint. The third joint... well, not even close. Although oddly, if the first joint hasn't been rotated, the third (Joint Orient frozen) joint's rotate attribute has the same value first non-frozen joint's Joint Orient attribute. Does that make sense? The sentence is a little hard to follow, sorry.

Screenshots:
Simple Skeleton
hanimRotation attributes of joints in the non-frozen skeleton connected to rotate attributes of joint nodes in frozen skeleton

No Rotation any joints
http://onaslant.ndsu.edu/x3d\_content/hanim/frozen01.jpg

Rotation of joint1 on nf-skeleton
http://onaslant.ndsu.edu/x3d\_content/hanim/frozen02.jpg

No Rotation of any joints highlighting the second joint in each skeleton
http://onaslant.ndsu.edu/x3d\_content/hanim/frozen03.jpg

Rotation of joint2 on nf-skeleton
http://onaslant.ndsu.edu/x3d\_content/hanim/frozen04.jpg

Rotation of joint1 and joint2 of nf-skeleton
http://onaslant.ndsu.edu/x3d\_content/hanim/frozen06.jpg

Shows how the rotate attribute of the third joint of the f-skeleton has the value as the Joint Orient attribute of the first joint of the nf-skeleton
http://onaslant.ndsu.edu/x3d\_content/hanim/frozen05.jpg

So I guess I have a partial solution for the first frozen joint anyway... it rotates from the non-frozen joint's axis, but has a 0 0 0 rotate value when the non-frozen joint has a 0 0 0 value. Of course, this doesn't take into account Rotate Axis, but I've given up on that aspect of the problem. Besides, the testers say they can live with having the Rotate Axis value always set to 0 0 0.

Anyway, here's my new code. Any advice on how to get the entire skeleton to behave like I'd want?

CODE
     MObject nodeObj = plug.node();
     MDagPath adp;
     MDagPath::getAPathTo(nodeObj, adp);
     MFnIkJoint aj(nodeObj);
     MFnDagNode pt(aj.parent(0));

     MQuaternion ao;
     aj.getOrientation(ao);
     ao.normalizeIt();
     ao.invertIt();

     MMatrix tp = adp.inclusiveMatrix();
     MMatrix ps = pt.dagPath().inclusiveMatrixInverse();
     MMatrix lm = tp * ps;
     MQuaternion lq;
     lq.operator =(lm);
     MQuaternion adjq = ao * lq;

     if(pt.typeName().operator ==("joint"))
     {
   MFnIkJoint pj(aj.parent(0));
   MQuaternion jq;
   pj.getOrientation(jq);
   jq.normalizeIt();
   jq.invertIt();

   adjq = jq * adjq;
     }

     MEulerRotation fRot;
     fRot.operator =(adjq);

     MPlug haPlug = aj.findPlug("hanimRotation");
     MPlug xha = haPlug.child(0);
     MPlug yha = haPlug.child(1);
     MPlug zha = haPlug.child(2);

     xha.setValue(fRot.x);
     yha.setValue(fRot.y);
     zha.setValue(fRot.z);

[archaeology]

Well, no matter how hard I tried, I couldn't get the my code to work the way I wanted it to... and I almost gave up. However, about two weeks ago I realized I was going about it all the wrong way. Instead of writing code that calculated the h-anim orientation of a node, I duplicated the rig, set the intermediateObject attribute of each joint in the duplicate rig to true, froze the transform attributes of each joint in the duplicate rig, and then used point and orient constraints to constrain each duplicate rig joint to the corresponding joint of the original rig. In the end a relatively easy answere I should have thought of to begin with. Of course, this makes it slightly complicated for the content author to use, so I spent the last week writing a character setup interface for it. The nice thing is, not only does this technique take into account the JointOrient attributes, but the RotateAxis attributes as well.

If anyone is interested in trying out this character export plug-in, visit:
http://onaslant.ndsu.edu/x3d\_content/rawkee/

Thanks for all the help,

Aaron