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. For a car agent's action selection, in this paper, we propose a Q-learning method in which a car agent can pass a present waypoint with the smaller number of steps. If the number of steps to pass the present waypoint is less than that of the similar past state, we give additional rewards (called step rewards) to all car agent's actions from the first step to the present waypoint except several last actions. As a result of the simulation, the number of steps of the car agent with the step rewards became a little less than that without it. Because the car agent runs so fast that it cannot efficiently pass the next waypoint.
