No replies to this topic

[batterj2]

Apologies if this is in the wrong section....

I'm trying to export poses from Layout into our (custom) engine and running into several difficulties, most notably knowing which rotations to extract, the order in which to combine them and convert from a LHR environment to a RHR environment (OpenGLES world). The idea is that we lerp between fixed poses to simulate animations (since there is only going to be a few).

The current approach has us writing a custom script to export a human-readable "pose file" which, for each selected bone in a hierarchy at t=0, contains:

• Bone name
• Parent bone name
• Length of the bone
• (If root bone, offset position)
• Rotation of the bone

The script for this is below:

	for (i = 1; i <= selection.size(); i++)
	{
		sceneItem = selection[i];

		name[i] = sceneItem.name;
		if(i == selection.size())
			parent[i] = string("");
		else
			parent[i] = sceneItem.parent.name;		
		length[i] = sceneItem.restlength;
		xPos[i] = sceneItem.getPosition(time).x;
		yPos[i] = sceneItem.getPosition(time).y;
		zPos[i] = sceneItem.getPosition(time).z;
		yRot[i] = sceneItem.getRotation(time).h;
		xRot[i] = sceneItem.getRotation(time).p;
		zRot[i] = sceneItem.getRotation(time).b;
		if(time == 0)
		{
			yRot[i] += sceneItem.getPivotRotation(time).h;
			xRot[i] += sceneItem.getPivotRotation(time).p;
			zRot[i] += sceneItem.getPivotRotation(time).b;
		}
	}

As I understand it, each given rotation is an offset rotation on a rotated coordinates axis (rather than a global rotation <- is that what world rotation is?)

When loading the data we determine the position of each bone by:

• creating a separate rotation matrix for heading, pitch and bank (pitch is negated)
• multiplying these rotations in the order: ((bank * pitch) * heading) to create a rotation matrix m
• if its not the root bone we multiply m by its parent rotation matrix m_p
• we then store m as the parent rotation matrix for any children bones
• we then create a 3D vector v of x,y,z where: x = y= 0 and z = (length of the bone) and rotate this vector by m
• using v we create a translation matrix m_t and then inverse this to create an inverse bind pose matrix m_i and store this for each bone

To determine if this works I've written some code to draw debug lines demonstrating the bones and for the most part it looks right up to about the 4th child bone (i.e. great-grandchild) where it looks like some data is missing to a correct rotation. In my case where arms are supposed to be pointing forwards they are in fact pointing upwards. If I turn IK off in Layout and repeat the process I appear to get a correct T-pose of our character.

So...

• does what I have said so far make sense?
• Is this a correct approach?
• Can anybody explain to me the differences and relations between rotations, world rotations and pivot rotations?

Additionally, does anybody where I can find useful information on similar techniques or an explanation of how Lightwave works? It seems any form of help is being systematically being erased from the WWW.

Many thanks for any help received!