Abstract
Car racing game is a game of computer programs in IEEE 2007 Car racing Competition, where two car agents compete with each other for taking waypoints in a two-dimensional real-number plane.In a competition game, if our car agent is advantageous to the opponent's car agent, our car agent should act as the next way point is located in the forward direction of the car agent when the car agent pass through the current waypoint.In the previous research, the forward direction was defined as an angle range of 45 degrees in the right and left from the direction of a car agent and it learned such actions by applying fuzzy Q-learing with additional rewards when the next waypoint was located in its forward direction. In this research, wepropose two methods of narrowing the angle range and giving more rewards as the next waypoint is located in closer direction of the car agent. In addition, we examine the case of giving negative rewards if the waypoint is not located the forward direction of a car agent.We found that the combination of methods of giving more rewards and giving negative rewards was the best simulation result.
