Agent-based soldier behavior in dynamic 3D virtual environments

Abstract

Human behavior in virtual environments is commonly implemented as a finite state machine. This programming approach can be effective and challenging against human players, but its ability to realistically simulate the behavior of cooperative groups of soldiers is limited. This thesis covers the development of an agent-based system to control the behavior of infantry in 3D virtual environments. The system design divides the cognitive process into four modules: perception, mental model, goal, decision, and action resolution. Each module attempts to simulate both strengths and weaknesses of human perception and cognition, including instinctive reactions, perceptual error, memory degradation, context-dependent decision-making, and inference. Additionally, the soldiers are influenced by the actions and decisions of the agents around them, enabling cooperation. The resultant agent system was incorporated into a game-like interface and compared to a similar commercial game with standard AI. Overall, 72% of the test subjects thought that the agent behaviors were Mostly Realistic or Totally Realistic, and 81% found them to be equal to or better than those in the commercial game.