This is called a Dynamic Shape Display. MIT built it, and it uses a Kinect sensor. In one of its applications, instead of a user in real life manipulating something on a TV, it puts the user on a TV, manipulating something in real life. See for yourself.
inFORM, the name of the display, has other uses, too — the latter portion of the video shows someone interacting with the display itself, for example, scrolling through a series of 3D representations of geometric formulae. One could see, potentially, how this might also become a type of game or game controller unto itself in the future.
The video shows how the device works at the end — the display itself is a grid of pins attached to actuators that are then manipulated by the computer to form the shapes seen on its surface. Whether that's the outline of a VW, or a hand picking up a torch, it all works the same.
The remote test here simulates a type of video conference, with the person on the screen remotely interacting with a ball where the conference is originating. You can imagine any number of other applications, too, particularly for scientists working with a hazardous object, or something in a hazardous environment.
There's more about inFORM, and those who built it at MIT, at the link.
inFORM [Tangible Media Group, Massachusetts Institute of Technology. Thanks Eric!]