Synthesis of Character Behaviour by Dynamic Interaction of Synergies Learned from Motion Capture Data

Abstract

Character animation ideally combines the competing requirements of high realism and flexible automatic generation of behaviour. A method for real-time human character animation is presented, which self-organizes character behaviour with high degrees of realism by dynamic coupling of ’synergies’ that are learned from motion capture data. Based on a new algorithm for blind source separation that considers time delays, highly compact generative models of body movements are learned from motion capture data. The learned components are mapped onto stable solutions of dynamical systems applying kernel methods, resulting in a coupled network of dynamic pattern generators whose state can be updated in real-time. This new framework is applied for crowd animation and the automatic generation of interactive behaviour for multiple characters.