Enabling real-time physics simulation in future interactive entertainment

Publisher Website
Google Scholar
Abstract
Interactive entertainment has long been one of the driving factors behind architectural innovation, pushing the boundaries of computing to achieve ever more realistic virtual experiences. Future entertainment applications will feature robust physics modeling to enable on-the-fly content creation. However, application designers must provide at least 30 graphical frames per second to provide the illusion of visual continuity. This constraint directly impacts the physics engine, which must deliver the results of physical interactions in the virtual world at a fraction of this frame rate. With more sophisticated applications combining massive numbers of complex entities, the cost of robust physics simulation will easily exceed the capability of today's most power machines.This work explores the characteristics of real-time physics simulation, and proposes a suite of future-thinking benchmarks stressing different situations that represent the demands of future interactive entertainment. With this suite, we then explore techniques to help meet these demands, including parallel execution, a fast estimation approach that self-regulates error, and a value prediction technique that is allowed to get "close enough" to the real value. We demonstrate that parallel execution together with the proposed fast estimation approach can satisfy the demands of nearly all of the PhysicsBench suite.

This publication has 8 references indexed in Scilit: