Self organized character animation based on learned synergies from full-body motion capture data

We present a learning-based method for the realtime synthesis of trajectories for character animation. The method is based on the approximation of complex movements by a small set of spatial movement primitives, or synergies. Applying kernel methods, such learned primitives are associated with nonlinear dynamical systems that, similar to central pattern generators in biological systems, produce highly complex trajectories in real-time by self-organization of behaviour. It is demonstrated that the novel method is suitable for the simulation of highly realistic human movements in interactive character animation. Different simple applications are discussed, including crowd-animation, navigation combined with the dynamic variation of movements styles, and the animation of a folk dance.