Parameter Spaces, Spacetime Control and Motion Graphs for Automating the Animation of Videogame Characters

Candidate: Lorne McIntosh
Type: Master of Science (MSc), School of Interactive Arts and Technology
Date: September 12, 2011
Senior Supervisor: Steve DiPaola
Thesis: Download Thesis Document

Abstract

Character animations are a crucial part of many interactive productions, from training simulations to videogames. As these productions have become more sophisticated, the growing number of character animations required has made standard animation techniques like key-framing and motion-capture increasingly expensive and time-consuming. Procedurally generating animations appears to offer a solution. This thesis extends and combines work from several areas of procedural animation to create an end-to-end system for the automatic generation of character animations for interactive productions. Specifically, our architecture pairs Spacetime Control, used to automatically generate new physically-valid clips of character animation, with a data-driven playback technique, used to automatically generate continuous streams of character motion from these clips in real-time. Our approach exploits the natural parameterization present in videogames and character motion to organize and automate the procedural generation of large quantities of character animation. It also supports rapid-prototyping, easily handles animation design changes, and may potentially be operated from start to finish by a single user. We demonstrate this architecture with a working implementation and show results from an example scenario starring a humanoid character capable of dozens of generated motions including standing, walking, running, turning and stepping.