Though the demo was quite impresive, and definitely more advanced than any other I've seen before, the "shakyness" is still there, detaching the "augmented" and the "regular" reality.

What is it preventing a more accurate analysis of the video frames? Too low resolution? Changing lighting conditions? Lack of processing power? I can't see how these cannot be solved with current optical and computing equipment.

Is it, then, a matter of not having developed sophisticate enough image analysis (vision) algorithms?

The shakiness has always confused me a little. Can they not dampen the momentum vectors somewhat - so little back-and-forths in the actual calculations don't propagate up to the render?

I thought it was a cool demo. This passed through my mind while watching it:

It's strange to me how the 2d images in the 3d info popout immediately looked off. A 3d interface seems to warrant a 3d image.