 |
|
Simulated human figures are used not only in games and for special effects, but also for ergonometric simulations, and even in synthetic training environments where personable interaction is the basis for comunication and evaluation.
Steve-Agent is an articulated humanoid figure which serves as an expressive front end for artificially intelligent, interactive training/evaluation software. The figure is assembled using the VRML hierarchical modeling format, loaded into an Iris-Performer based browser with a C-API, and animated interactively with natural looking movements using a collection of high speed transformation motors.
This project deals mainly with upper body actions, such as looking, pointing, grasping, and various facial expressions. While it does not directly address the broader problem domains of articulated locomotion and collision detection, the methods employed here have some relevant application to development in those areas.
This discussion will begin with a review of important background methods and terminology used in modeling and manipulating hierarchical transformations, and an explanation of conventions used in controlling 3D rotation. A set of animation primitives, called transformation motors, are introduced, and their behaviors are discussed in the context of nested coordinate systems and computational efficiency.
A simple solution is provided for handling extended joints, such as the elbow/wrist linkage, which typically pose a serious challenge for high performance inverse kinematic techniques. The figure dynamics are constructed from these primitives, and the agent is given a high level command set for executing dextrous tasks.