Self-Driven Soft-Body Creatures

摘要

Virtual characters play an important role in computer generated environments, such as, video games, training simulations, and animated films. Traditional character animation control methods evolve around key-frame systems and rigid skeletons. In this paper, we investigate the creation and control of soft-body creatures. We develop creatures that learn their own motor controls and mimic animal behaviours to produce autonomous and coordinated actions. Building upon passive physics-based methods and data-driven approaches, we identify solutions for controlling selective mesh components in a coherent manner to achieve self-driven animations that possess plausible life-like characteristics. Active soft-body animations open the door to a whole new area of research and possibilities, such as, morphable topologies, with the ability to adapt and overcome a variety of problems and situations to accomplish specified goals. We focus on two and three-dimensional deformable creatures that use physics-based principles to achieve unconstrained self-driven motion as in the real-world. As we discuss, control principles from passive soft-body systems, such as, clothes and finite element methods, form the foundation for more esoteric solutions. This includes, controlling shape changes and locomotion, as movement is generated by internally changing forces causing deformations and motion. We also address computational limitations, since theoretical solutions using heuristic models that train learning algorithms can have issues generating plausible motions, not to mention long search times for even the simplest models due to the massively complex search spaces.