There are mainly two kinds of methods to alleviate the onerous task of generating natural-looking human character animation: kinematic synthesis and dynamic simulation. In the former one, it is easy for animator to design how the avatar to act, but difficult to generate physically realistic performace of the action in the virtual environment. While the later one, which can figure out very accurate physical movements of all objects in a scene, does not allow the animator to specify the desired motion in an intuitive way.

Our work is to make the virtual character learn how to move from the motion capture data, and to use the knowledge to drive the virtual character in the physical simulation to obtain realistic animation.

Return to Conference Programme