No More Zombies. High-Fidelity Character Autonomy for Virtual Small-Unit Training.

摘要

Virtual practice environments can offer situated, realistic learning experiences if properly implemented. 'Serious Games' delivered within these environments offer visually compelling experiences, but often suffer from a lack of realistic interactions with virtual characters such as teammates, adversaries, and other non-combatants. Artificially intelligent human behavior models-- intelligent agents--provide a variety of features not present in legacy computergenerated forces (CGF) systems; including goal-directed dynamic decision making, non-determinism, interactivity and transparency. Using a cognitive architecture, intelligent agents exhibiting these features can be brought to bear for virtual training environments to support both kinetic and non-kinetic small-unit training exercises. In partnership with the U.S. Army Research Lab's SFC Paul Ray Smith, Simulation & Training Technology Center (STTC), we have developed a suite of intelligent agents for virtual environments that can realistically engage human players in small-unit training scenarios. The centerpiece of this work is a knowledge-rich OPFOR sniper behavior capable of detecting and selecting friendly targets of opportunity, communicating with other insurgent support entities such as lookouts, and finding the appropriate escape path to avoid detection and capture. In addition to a sniper entity, we also developed agents to play supporting roles, including autonomous fire team agents and noncombatant townsfolk. In this paper, we describe in detail the challenges encountered during this effort and how intelligent agents can be exploited to improve training within systems such as the Dismounted Soldier Training System (DSTS). In addition, we outline the reusable integration architecture we developed to connect our agents to EDGE, a massively-multiplayer online virtual environment developed at STTC, and describe the design choices made to ensure that the architecture can be reused to connect both these agent.