Behavioral software agents for real-time games

Abstract

This paper describes the real time autonomous software agents include intensive simulations such as interactive computer entertainment software. The layered artificial intelligence (AI) architecture presented is based on the subsumption architecture, a proven control system for mobile robots that consists of parallel levels of independent behaviors. Higher-level behaviors modules can subsume lower-level ones through the limited number of inputs and outputs. This AI system can be highly optimized and extended various ways. A C++ program is presented demonstrating simple three-layer AI agents. Experiments with this program show that relatively little memory and processor time required per agent; furthermore, a modern computer game would be able to support thousands of simultaneous agents with this behavior system.