My observation is based on experience in working with 3D models and animation.  Tracking can mess up easily if your character is not skinned/rigged properly or if the software is not interpreting the data correctly. 

A simple example:

Natal is tracking the arm and reads a rotation of 90 degrees, if the software reads the coordinates but turns counter-clockwise instead of clockwise then Natal has given the proper cordinates yet the software interpreted it wrong and now although the arm is in the right place it's twisted horribly.  That's a very basic example but it illustrates that it can easily be the software and the not the hardware that's at fault.

Also, you're ignoring the Ricochet demo where you could clearly see the avatar on screen was tracking properly to her movements.

Putting ragdoll-style contraints on the bones of the rig aren't going to solve the issues they were seeing in the avatar tracking. It may provide something of a safety net, or even allow them to come up with several solutions to a particular image, and choose the "best" one.

In the end, the problem was the lighting on stage, as I said.

It worked much better on the show floor, where the lighting was set up much better. They also didn't show any demos that involved multiple people at the show, that I recall, except in their silly "what if" video. The number of problems with solving that are downright staggering, and they aren't going to solve it without turning your living room into a fully-lit lab.

For it to be useful for games, I believe a tracking marker will eventually be required. Sony's wand demos were pretty good -- far superior to the simplistic 2D stuff MS was showing, and the PSEye, frankly, has done for years. Combine that with Natal's nice camera and the software they are working on.. then you might have something feasible. Even a wristband with a reflective patch might do the trick.

The media's craze, over what basically amounted to a premade video and some very simple show floor demos, still kind of boggles me.  The "what if?" imagination that MS has stirred up with Natal is really something.  But it really is just that, at this point -- a dreamy image.  I wish them luck though -- if they succeed it'll really be spectacular.

As I understand it, Natal doesn't need lighting because it's an active sensor, not a passive one. It doesn't rely on light in the environment bouncing off an object and into the camera, it sends out its own pulses of infrared light and measures how long the light takes to return to construct its 3D image.

Now, Natal does also have a conventional video camera which would rely on environmental light, but that seems to have little to do with the interface. It mostly just seems to be for scanning people and objects and sending those pictures to the software. You could still play Burnout in a dark room, you just couldn't videoconference with your cousin.

I would disagree about the lighting on the stage.  The camera emits its own infrared light to detect depth.

I would have to say the problem lies in their skeletal system algorithm for a rotating body.  Something got wonky when Kudo turned around.  It will most assuredly be fixed before release.

That's quite a leap, and only those types of people call others that name.

On topic, someone please present Johnny Lee's credentials other than he recently earned his PHD at Carnegie Mellon.  Other than being an Internet sensation with his Wii-mote projects, why are you debating what he says?

Why did the Natal demo have +200ms of annoying lag?
(while this head tracking experiment with the wii-remote has almost no visible lag)