Data-Driven Approach for Human Locomotion Generation

Abstract

This paper introduces a data-driven approach for human locomotion generation that takes as input a set of example locomotion clips and a motion path specified by an animator. Significantly, the approach only requires a single example of straight-path locomotion for each style expressed and can produce a continuous output sequence on an arbitrary path. Our approach considers quantitative and qualitative aspects of motion and suggests several techniques to synthesize a convincing output animation: motion path generation, interactive editing, and physical enhancement for the output animation. Initiated with an example clip, this process produces motion that differs stylistically from any in the example set, yet preserves the high quality of the example motion. As shown in the experimental results, our approach provides efficient locomotion generation by editing motion capture clips, especially for a novice animator, at interactive speed.