Application of Fuzzy Q-learning to Car Racing Game

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. If the current waypoint is taken, the next waypoint becomes the current. In each action, the car agent can get some information on itself, the other car agent, the current waypoint and the next waypoint. In this paper, we apply to it a fuzzy Q-learning that is an extended version of Q-learning, one of methods of reinforcement learning, to acquire the actions that car agent takes the current waypoint without the other car agent. We define the state space of fuzzy sets about the distance and angle to the current waypoint and the speed of car agent. We illustrate this method with several patterns of rewards.