Proxy Simulations for Efficient Dynamics

Abstract

Proxy simulations reduce the cost of simulation in large virtual worlds, such as those used in training simulations or computer games. A proxy takes the place of an accurate simulation for objects that are out of view, while the accurate model continues to manage visible objects. A proxy must ensure that objects enter the view at reasonable times throughout the simulation and in states that reffect their time spent out of view. The quality of a proxy simulation is measured by how well it maintains reasonable behaviors, where the deffinition of reasonable depends on the environment and its application. We present two examples of proxy simulations based on discrete event models: one for city traffic simulation and another for multi-agent path planning and motion. For these examples, we demonstrate dynamics computation speedups of over two orders of magnitude as the environments grow in size and complexity.