AI Seminar

Animation by Example

Michael Gleicher

Motion for computer animation is notoriously difficult to create. In order to achieve the expressiveness, subtlety and realism of quality motion, practitioners have relied on either capturing the movements of real performers, or labor and skill intensive manual specification methods. Such methods create specific, short clips of motion. These clips may provide the desired quality, but lack the flexibility required when all movements cannot be pre-planned. In contrast to clip-based methods, motion synthesis approaches can flexibly create motions on the fly, but (to date) have not provided sufficient quality.

In this talk, I will survey our efforts to create high-quality motion for animation in a flexible manner. I will begin by reviewing some of our previous efforts in motion editing, the problem of adapting motions to meet new needs. I will discuss how the successes and failures of these approaches have lead us to a number of new directions. I will describe several of our recent results, including preserving the fine details of motions during editing, creating high-level control abstractions for motion, and synthesizing new motions by assembling pieces of existing motions. Combined, these developments promise to allow flexible creation of high-quality motion based on an initial set of example motions.

Sponsored by

CSE Graphics