Studies on Learning and Combination of Tactics in Gaming Simulation

Abstract

This paper introduces the design of a gaming-simulation in which several software agents try to find tactical rules with playing the game that aims at the second place. The agent has two neural network models for predicting behavior of other agents, including itself, and for achieving the target, respectively. In addition, the agent has a classifier model for tactical decision-making. This three-model structure enables us to monitor behavior of agents easily.