[Georgia Tech][GVU][Animation lab][Search]

Adding Variability to Cyclic Animations

[Eye Graphic] [Eye Graphic] [Eye Graphic]

Creating appealing humanoid animations is a difficult and time-consuming task. Cyclic motions such as running or walking are particularly hard to animate because limitations in time and technology often result in only a small number of distinct cycles being produced, whether the method is keyframing, motion capture, or simulation. These few cycles are then repeated to create an animation of the desired length. Unfortunately, the repetitiveness of the resulting motion is often noticeable and no matter how high the quality of the individual cycle, the resulting stream of animation appears unnatural.

For example, if a running motion is created through motion capture, it should look realistic, because it was obtained from a real person. But the limited field of view of most motion capture equipment means that only a few strides of running over ground can be obtained. If an extended sequence of running is required, the repeating cycle will quickly become obvious. A run could be videotaped and rotoscoped with an animator adding in variations of the motion, but that requires an animator to edit the motion, and controlling the support of the foot becomes problematic. Simulations of running can also appear repetitive: the control system attempts to maintain a steady-state running pattern and therefore it reduces the variability of the running motion. Even with a complicated simulation such as a full-body dynamic model with many controlled degrees of freedom, a good control system produces cyclic motion with little variability between strides.

This work seeks to fix this problem by determining how to introduce natural-looking variability into cyclic animations of human motion. We assume that variability is noticeable and we hypothesize that a motion that possesses it looks more natural. From a biomechanical perspective, a fundamental attribute of biological systems is that motion varies from performance to performance. We leverage some of the biomechani-cal research on variability and incorporate it into our work. In this framework, introduc-ing variability into animations equates to modifying the trajectories of the joint degrees of freedom from their nominal values, and can be viewed as adding a type of noise to the animation. Our experiments are confined to running motions, in part because ani-mating a convincing run is extremely difficult, and in part because running is a more constrained activity than walking. Through human subject testing, we seek to answer the question of whether adding noise produces a more natural-looking animation, and if so, how much and what type of noise should be added.

We make a distinction between variability in a motion and the style of a motion. Style is a measure conveying the difference between the motion of two subjects or the overall emotional expressiveness of a motion, while variability is a mea-sure conveying the changes between repetitions of a task. A clear example of a stylistic difference in motion would be a dejected walk as compared with a buoyant walk; an example of variation in a walk would be the arm occasionally brushing against the torso. Stylistic differences are very interesting from an animation perspective, but our concern in this work is with variability.

The slides for the talk given at the 1999 Eurographics Workshop on Animation and Computer Simulation are here.

See which level of noise added to a simulation of a male running you think looks most natural by downloading the following animations and viewing them:

Our results for fifty subjects are as follows:


Publications

Project Members


Copyright 1999

Questions or comments? Email bobbyb@cc.gatech.edu.