Skip to main content
eScholarship
Open Access Publications from the University of California

Automating Expressive Locomotion Generation

  • Author(s): Kim, Yejin
  • Neff, Michael
  • et al.
Abstract

This paper introduces a system for expressive locomotion generation that takes as input a set of sample locomotion clips and a motion path. Significantly, the system only requires a single sample of straight-path locomotion for each style modeled and can produce output locomotion for an arbitrary path with arbitrary motion transition points. For efficient locomotion generation, we represent each sample with a loop sequence which encapsulates its key style and utilize these sequences throughout the synthesis process. Several techniques are applied to automate the synthesis: foot-plant detection from unlabeled samples, estimation of an adaptive blending length for a natural style change, and a post-processing step for enhancing the physical realism of the output animation. Compared to previous approaches, the system requires significantly less data and manual labor, while supporting a large range of styles.

Main Content
Current View