I'd love to see the tech in Microsoft's Kinect combined with this concept - it's already in the mass market, just need something to translate the movements (transmit video/audio) to a 'robot' on the other end.

These robots really need an LCD screen to show the face of the remote user though. I think we naturally attribute traits to the robot itself, not who's controlling it, and that's the weird part. Have the user's face appear and it's just "Oh, it's Bob. Hi Bob!"; (conceptual) problem solved.