Selforganization of character behavior by mixing of learned movement primitives.

Abstract

The real-time synthesis of natural looking human movements is a challenging task in computer graph-ics and robotics. While dynamical systems provide the possibility to parameterize behavior in a flexi-ble and adaptive way, the reconstruction of details of human movements with such systems is a chal-lenge due to the large number of involved degrees of freedom. We present an approach for the synthe-sis of realistic human full-body movements in real-time that is based on the learning of motion primi-tives, or synergies, from motion capture data apply-ing a novel blind source separation algorithm. By application of kernel methods we map such com-ponents onto low-dimensional dynamical systems that can be iterated in real-time, and which are in-tegrated in a stable overall system architecture. We demonstrate how this model can be integrated with other key elements of computer animation systems, such as style morphing, synchronization with ex-ternal rhythms or navigation. The performance of our approach was validated by self-organization of complex behaviors like dancing. 1