Data-driven Interactive Balancing Behaviors

Animations driven by motion capture data are typically limited by their kinematic nature. In this paper, we describe a new data-driven approach for producing interactive dynamic balancing behaviors for an animated character. The result is a character that can interactively respond to single or multiple pushes in various directions and of varying magnitudes. A database of captured responses to pushes is used to create a model that supports hip, arm, and stepping strategies for balance recovery. An interactive push is modelled as a force impulse, which is then used to compute a momentum-based motion index in order to select the most appropriate recovery motion from the database. The selected motion is then adapted in order to provide a response that is specifically tailored to the given force impulse while preserving the realism and style of the original motion. Based on a relatively small motion database, our system is effective in generating various interactive balancing behaviors, for single and multiple pushes.