A game play agent learns to play as if it is controlled by human player

Abstract

The authors are developing the learning agents for the platform game in the FSS2012 Game Competition. The agent is expected to run with human-like behavior or as if it is controlled by the human player. Before the implementation of learning perspective, we are investigating how human player controls the game play character. Now we are implementing hard-coded agent based on our investigation. In the next step, we should derive some kind of generalized learning scheme to be the potential ability for the agent. Finally, meta-controller for rule selection is to be developed. All these functionality may be completed before the competition.