Towards Building an Interactive, Scenario-based Training Simulator Brian Magerko John Laird University of Michigan 1101 Beal Ave. Ann Arbor, MI 48109 734-936-3874, 734-647-1761 magerko@umich.edu , laird@umich.edu Keywords: training, automated direction, synthetic characters, user modeling, interactive drama ABSTRACT: This paper presents problems with non-interactive training simulators, and presents the design for an interactive system that will address these problems. The focus of this system is an omniscient Director agent that monitors the actions of the Trainee and other characters involved in a training scenario. Once the Director notices when the training goals of the scenario are not being met, it then dynamically modifies the state of the world to make reaching the training goals more likely. The pedagogical effectiveness of the training simulation should be increased through this observation and action combination. 1. Introduction Realism is an important value to uphold when creating military training simulations. The environment should be immersive and three-dimensional, complete with realistic opponents and teammates. To further provide a realistic simulation in this environment, the Trainee should have as much of a free range of action as possible; his actions should not be constrained by what the designers envision the Trainee would possibly do. We define the Trainee’s ability to act as he desires in the world as user flexibility. If the only goal of building a training simulation system were to present an absolutely realistic world, then we would worry about maximizing this user flexibility and nothing more. However, given that a training simulation is a pedagogical tool, there will be specific training goals for a given scenario. In order to achieve those goals, the Trainer needs the representational flexibility to formalize the goals for the training scenario, to script out in some abstract manner what should happen, and have a way to dynamically modify the scenario to ensure the training goals are achieved. We define the ability of the Trainer to specify the exact scenario he wishes as writer flexibility. This paper proposes a system that attempts to find a balance between user and writer flexibility. Finding the balance between user flexibility and writer flexibility lies in the monitoring of the Trainee’s actions and the intelligent modification of the environment. As the User executes actions in the world, those actions may begin to lead him down a path that diverges with the training goals that the Trainer has specified. Monitoring the Trainee and recognizing this possible divergence from the training scenario is a key element of improving the efficiency and efficacy of immersive training. Once a divergence is recognized, the Trainee must be encouraged in some unobtrusive, realistic manner to follow a path more consistent with the pre-defined scenario. How the world should change, as well as which global actions to execute to create this change, is the other side to our problem. Once something has gone awry, how do we reconcile the current state of the world with the desired one in both a realistic and non-obtrusive fashion? Inspiration for building the system we are proposing in this paper has come mainly from work done on interactive drama systems [1], [2], [3], and [4]. Laurel’s definition has come to aptly characterize this yet-to-be-realized art form [5]: An “interactive drama,” then, is a first-person experience within a fantasy world, in which the User may create, enact, and observe a character whose choices and actions affect the course of events just as