Abstract

Intelligent Agents for Individual
and Team Training Applications

Azad M. Madni, Ph.D. and Carla C. Madni
Intelligent Systems Technology, Inc.

H. Barbara Sorensen, Ph.D.
Warfighter Training Research Division/AFRL

Sharon K. Garcia, Ph.D.
AFRL/HEPC, Brooks AFB

Abstract

Intelligent agent technology has advanced to the point where it can potentially provide a costeffective replacement for teammates, coaches, and opponents in simulation-based training (SBT). However, the introduction of intelligent agents into training applications is not without its share of challenges given the many roles that intelligent agents can potentially play and the unique requirements of each role. For example, intelligent agents can play the role of teammates, opponents, supervisors, subordinates, coach, and mentor. The required level of intelligence depends on the functions that the agent is required to perform in these various roles. For example, agents performing as “extras” in a scene require little or no intelligence (i.e., they don’t have to be intelligent), whereas an After-Action Review (AAR) agent requires the ability to diagnose deficiencies relative to, for example, the school solution. To perform some functions, agents require an understanding of the problem domain; for others they merely need to understand their own tasks. In some cases, agents need the ability to plan. In other cases they merely need the ability to react. A pedagogical agent might need a human persona to enhance learner motivation and learner-system interaction. It is important to realize that when intelligent agents are used to replace a human, it is not necessarily the case that the functions performed by the human are mapped to a single agent; rather, for ease of authoring and maintenance, it might be entirely advantageous to employ multiple, cooperating agents to perform the functions. The implementation of these various kinds of agents for simulation-based training (SBT) can take a variety of forms that vary in their level of sophistication. For example, a pedagogical agent typically requires some reasoning facilities to diagnose student deficiencies and create a remediation plan. Such an agent can be implemented using agent modeling languages or expert system shells. At the other end of the spectrum, agents that fulfill the role of “extras” can be conveniently modeled as time-driven, message-passing entities. Similarly, superior and subordinate agents that are not key actors in a scenario can be represented as event-driven, messagepassing entities to reduce implementation costs. Depending on the training context, teammate implementations can vary from simple condition-action, messaging entities to sophisticated cognitive agents based on cognitive architectures such as ACT-R or Soar. In this paper, we provide specific examples of the use of intelligent agents for individual and team training. In particular, we present a detailed discussion of an intelligent pedagogical agent along with an example of its use. We also present the key issues in the design of synthetic teammates and intelligent adversaries.

From:

Madni, A., Madni, C., Garcia, S., and Sorensen, H.B., Proceedings of AIAA Infotech@Aerospace, Hyatt Regency, Crystal City, Arlington, VA, September 26-29 2005.