Ok, this is my point: There is a center of mass. Right now, with a typical FK/IK rig, we control this center of mass only implicitly; whenever we move a body part or the root, the position of the COM changes, which effectively changes the weight distribution in the whole character. That way it's really easy to mess up the weight in animation, unless we have a lot of experience. But even then it's merely guessing.
I don't consider this a given, inherent fact of animation; it's a problem that asks for solutions. So I was looking for alternatives, and I started this thread to find out if there are other people like me. However, I am aware of the fact that it's quite a hard problem. I don't expect to find a perfect solution, but ways to deal with it in a more effective and appropriate way than what is possible/available right now. To me, enhancing the rig was the most obvious starting point.
It seems that I didn't express myself very well in previous posts, so to make things a bit clearer:
When I said "pointConstraining the COM" I didn't mean to fix it forever and ever to some place in space. I meant pointConstraining to some control object, so you can change or animate the position, whatever you want. You could use it for the animation itself, or only to get the rig into a certain pose; your choice.
It's not the optimal solution for every situation. So "weight-based rig" should probably be replaced by "rig featuring FK, IK and weight-oriented behaviour" or some such.
I only referred to situations where you WANT the COM to stay (or to be) in a certain location. Be it above the left foot while walking, above the center of both feet in a standing position, a bit behind the feet when preparing for a back-flip, or constrained to a path-curve when jumping/falling.
However, if I try to animate some movement, and I don't know where the COM is in a certain pose, it's safe to say I have not yet understood the movement.
If you want unnatural movement, and weight doesn't matter anyway, why use a weight-oriented rig? However, with a weight-oriented rig you could control how unnatural the movement is!
failure: I really don't understand your sphere-example. The center of the sphere is it's COM anyway, so what's the point? If you animate the center of the sphere, you animate the COM. The whole point of this COM feature is that the weight/COM _IS_ obscured in a biped. This is the difficulty in animating a biped the right way, and I tried to think of ways to reduce that difficulty. Reduce, not magically take it away. I'm not talking about having the COM animated automatically, but about having the possibility to explicitly control and animate it. (As opposed to only implicitly with FK/IK!)
And: A fixed COM looks way more natural than a fixed root joint!
As for the simulation issues: Yes, it would be a step towards simulation, BUT only as far as it remains usable. After all, animation IS simulation; not computed but made by animators. Animators simulate life. So animation tools should simplify that process.
You're right, weight is a function in time, but a pose is only a snapshot at a given time. If you look at some pictures for rotoscoping, you can still identify the COM in every picture. So you can interpolate this as a function in time, the poses being edit points of a NURBS curve.
Joojaa: The way you place things CAN matter, depending on the situation. Ok, you can get the same result in this way or another. But if you are in a situation where it is important that the COM is here or there, being able to explicitly control the COM is obviously an advantage. You said: It's not good "when there are 2 things happening when you move this or that." But consider this: When you move a limb IK/FK style, there are more things happening anyway, because this implicitly changes the COM! It only depends on how you look at it. The COM is invisible/fictious, so you don't see when it moves. But that doesn't change the fact that it moves!
As for the up/down movement: This is what I meant when I said pointConstraining in xz only. For example when the foot is planted on the floor, the y position of the COM would be implicitly determined by the contact point and the state of the rest of the body. I could go into more detail if you want to hear it.