A Scalable Approach to Control Diverse Behaviors for Physically Simulated Characters

摘要

Human characters with a broad range of natural looking and physicallyrealistic behaviors will enable the construction of compelling interactiveexperiences. In this paper, we develop a technique for learning controllersfor a large set of heterogeneous behaviors. By dividing a reference libraryof motion into clusters of like motions, we are able to construct experts,learned controllers that can reproduce a simulated version of the motions inthat cluster. These experts are then combined via a second learning phase,into a general controller with the capability to reproduce any motion in thereference library. We demonstrate the power of this approach by learningthe motions produced by a motion graph constructed from eight hoursof motion capture data and containing a diverse set of behaviors such asdancing (ballroom and breakdancing), Karate moves, gesturing, walking,and running.