Identification of Motion Models for Living Beings

Abstract

Recent trends in computer graphics animation have shown the emergence of two alternative paths towards the quest for realistic motion control. On the one hand, physics-based modeling provides a comprehensive framework to design dynamic motions that can react and adapt to varying environments. On the other hand, advances in motion capture technology have proved to be a practical fast track towards realism. At the beginning of this paper, it will be recalled that none of these approaches is a panacea. In order to mix their respective advantages, the paradigm of motion identification is investigated and illustrated on a study case. Motion identification bridges the gap between dynamic models and motion capture by automatically finding the best match between the possible dynamic behaviors of a specified class of semi-physical models and real-world trajectories. This paper builds upon a vast theoretical tool-box coming from classic Identification theory and discusses how this framework can be applied within the context of animating synthetic living beings. A major point is to show that motion identification requires a radically new way of thinking about dynamic motion models.