Host: Japan Society for Fuzzy Theory and Intelligent Informatics (SOFT)
Car racing game is a game of computer programs in IEEE CEC 2007 Car Racing Competition, where two car agents compete with each other for passing through waypoints in a two-dimensional real-number plane. Previously, we applied a fuzzy Q-learning to selecting an action of a car agent in a single trial. In order to pass through the current waypoint in fewer steps, we introduced a Step Reward. When the car agent can much easily pass through the current waypoint than the opponent in a competitive game, however, it is more important to pass through the current waypoint in a good direction for the next waypoint than pass through the current waypoint in few steps. In this paper, therefore, we propose a Direction Reward for the current waypoint in the front direction, where the car agent can easily pass through the next waypoint. We have results of the experiments for various situation.
