An Architecture for General Purpose Physical System Simulation--Integrating Geometry, Dynamics, and Control

Abstract

Simulation of physical systems has long been of interest to scientists and engineers, and significant efforts have been directed toward the development of general purpose computer aided design and analysis systems. To date, however, success has been largely limited to the production of tools suited only for particular aspects of design: computer aided design systems have primarily emphasized specification of geometry; simulation systems from the mechanical engineering field have concentrated mainly on formulation and integration of an unchanging set of equations describing object behavior; and work by computer graphics and animation researchers has been aimed at producing good-looking animations without much regard for whether the generated motions were physically realistic. Flexible systems integrating all aspects of design and analysis have not yet been built. This thesis addresses the issues involved in developing fundamentally more powerful simulation systems. A system architecture for general purpose physical system simulation is proposed, and a prototype implementation, the newton system, is described. The architecture is based on a rich model-based object representation and provides a level of automatic analysis that encourages the kind of experimentation necessary for successful design. In particular, it is argued that by using geometric modeling techniques to include a full description of object geometry, previously difficult-to-incorporate simulation system features, such as collision detection and resolution, can be handled much more naturally. The Newton architecture incorporates a uniform exceptional event handling mechanism that allows the system to respond to a variety of simulation events that necessitate modification of object behavior descriptions. Combined with a general method for the formulation of object motion equations, the event handling mechanism supports automatic handling of collisions, changing contact relationships, control program state changes, and other events that can cause discontinuities in object motions.