An Architecture for High-Level Human Task Animation Control

Abstract

There are many problems in developing a computer-generated animation of tasks performed by a human. Sequencing actions, which may be easy to do when only a single actor is involved, becomes much more difficult as more actors and objects begin to interact. Similarly, the actual planning of a task is difficult if one considers the capabilities which are being modeled. Resource allocation also must be considered along with alternate plans of action. These types of problems imply that a knowledge base and a planner need to be involved in creating the task animation. To produce the actual motion, techniques must be used to manipulate the geometry of the scene. Three basic techniques exist for accomplishing this, dynamics, kinematics, and constraints, with no single technique superior in all respects.

Although a few systems exist which attempt to address these problems, none succeed in a general fashion. Either work is concentrated in the animation production issues ignoring difficulties in sequencing and actor capabilities, or in the task planning stage, ignoring issues involving actual scene geometry. Those which seem to achieve success in both areas use a specialized architecture designed around the particular task to be animated.

An architecture is described which addresses issues in human task performance at two levels simultaneously. The architecture demonstrates a link between the task planning stage and the motion generation stage based upon an extensible set of virtual “processors” and communication pathways between them. The types of problems which can be encountered among these processors, including nomenclature conflicts and timing synchronization, are described.