Animation of Humanlike Characters: Dynamic Motion Filtering with a Physically Plausible Contact Model

Abstract

Data-driven techniques for animation, where motion captured from a human performer is “played back” through an animated human character, are becoming increasingly popular in computer graphics. These techniques are appealing because the resulting motion often retains the natural feel of the original performance. Data-driven techniques have the disadvantage, however, that it is difficult to preserve this natural feel when the motion is edited, especially when it is edited by a software program adapting the motion for new uses (e.g. to animate a character traveling over uneven terrain). In fact, many current techniques for motion editing do not preserve basic laws of physics. We present a dynamic filtering technique for improving the physical realism of motion for animation. As a specific example, we show that this filter can improve the physical realism of automatically generated motion transitions. Our test case is the physically challenging transition contained in a double kick constructed by merging two single kicks.