Tech Report CS-04-04
Evaluating and Tuning Motion Graphs for Character Animation
Paul S. A. Reitsma, Nancy S. Pollard
Realistic and directable humanlike characters are a continuing goal in animation. Motion graph data structures hold much promise for achieving this goal. However, the quality of the results obtained from a motion graph depends on properties of the graph that may be difficult for a user to control. This paper presents an approach to evaluating and tuning a motion graph data structure for a user's particular application. One key to evaluating a motion graph is to first embed it into the task domain. The goal of embedding the graph into the task domain is to capture all possible paths that might result from "playing back" the motion graph within the given environment. This paper describes an algorithm for accomplishing this embedding that is computationally efficient and also has the advantage of preserving the flexibility of the original motion graph. We then define an evaluation criterion on this embedding and describe a technique for tuning the motion graph based on this criterion. These techniques can assist the user in selecting an appropriate set of motion clips, adapting them for a given task, and answering the question of when sufficient motion is available for an application.