Verbs and Adverbs: Multidimensional Motion Interpolation

reating believable animated human figures proves difficult, even with the most sophisticated software available. Once an acceptable animation segment has been created, either by an ani-mator or through motion capture, the results remain difficult to reuse. The additional work to modify the animation may take almost as much time as creating the original motion. Furthermore, the exact style or structure needed for a particular motion may not be known until run-time. Interactive applications, such as a 3D video game or a virtual environment require a real-time con-trollable animation system. Research into controllable human figure animation divides into three major groupings: procedural, simulated , and interpolated. Procedural animation uses code fragments to derive the degrees of freedom (DOF) values at a particular time. The procedures can be as sophisticated as needed to provide different motion styles or react to different conditions of the simulated environment. Dynamically simulated figure animation uses controllers together with a simulated human to generate motion. The degree to which this method succeeds relies on how accurately we can understand and model human motion. Unfortunately, both these methods can alienate classically trained animators and use motion-capture technology ineffectively. This proves important since animators and motion-capture systems each produce compelling results. To leverage their qualities, a system must use what these resources provide. The third major grouping, interpolated animation, uses sets of example motion with an interpolation scheme to construct new motions. The primary problems to solve with this approach are to provide a set of meaningful, high-level control knobs to the animator or runtime system, maintain the aesthetic of the source motions in the interpolated motions, and extrapolate motion. Plus, it is difficult to acquire the examples— each is precious. Additionally, motion interpolation must be efficient for use in a run-time environment rather than earlier in a production pipeline. For these reasons, we chose an interpolation scheme using radial basis functions. This article describes a system for real-time interpolated animation that addresses some of these problems. Through creating parameterized motions—which we call " verbs " parameterized by " adverbs " —a single authored verb produces a continuous range of subtle variations of a given motion at real-time rates. As a result, simulated figures alter their actions based on their momentary mood or in response to changes in their goals or environmental stimuli. For example, we demonstrate a " walk " verb that can show emotions such as happiness …