Computational Generation of Terrestrial Animal Locomotion

Abstract

While the animation of humans is a well established topic in computer graphics, the animation of animals remains far less explored. Part of this is because in contrast to the wealth of easily obtainable motion data for humans, similar motion data for animals or fictional creatures is substantially more difficult or impossible to obtain. In this thesis I develop techniques for animating animals which do not rely on any motion capture data, but instead automatically synthesize motions by exploiting principles from physics and biomechanics. I address three main components to the problem of animal motion synthesis. First, I present an optimization technique which is capable of solving for a cyclic gait for an animal without requiring any motion data, and which can additionally solve for the most efficient type of gait for the animal to perform as well as refine the shape of the animal itself if desired. Second, I address the problem of increasing the realism of the synthesized gaits by learning from a database of ground-truth animal gaits. Finally, I move beyond gait synthesis to the synthesis of kinematic controllers which can be interactively guided by a user to create animations in real-time.