Developing intelligent agents for training systems that learn their strategies from expert players

摘要

Computer-based training systems have become a mainstay in military andprivate institutions for training people how to perform certain complex tasks. Asthese tasks expand in difficulty, intelligent agents will appear as virtual teammatesor tutors assisting a trainee in performing and learning the task. For developingthese agents, we must obtain the strategies from expert players and emulate theirbehavior within the agent. Past researchers have shown the challenges in acquiringthis information from expert human players and translating it into the agent. Asolution for this problem involves using computer systems that assist in the humanexpert knowledge elicitation process. In this thesis, we present an approach fordeveloping an agent for the game Revised Space Fortress, a game representative ofthe complex tasks found in training systems. Using machine learning techniques,the agent learns the strategy for the game by observing how a human expert plays.We highlight the challenges encountered while designing and training the agent inthis real-time game environment, and our solutions toward handling theseproblems. Afterward, we discuss our experiment that examines whether traineesexperience a difference in performance when training with a human or virtualpartner, and how expert agents that express distinctive behaviors affect thelearning of a human trainee. We show from our results that a partner agent thatlearns its strategy from an expert player serves the same benefit as a trainingpartner compared to a programmed expert-level agent and a human partner ofequal intelligence to the trainee.