Racing games have always been one of the most popular game genres among all types of game consoles. Players of racing games have to control a racing car and drive it to the goal as quickly as possible. A lot of skill is required to manoeuvre the car, which involves movements such as turning, accelerating, decelerating and overtaking.
A successful arcade game should not disappoint players by setting a very hard target. On the other hand, too easy a task may make the player feel bored and childish. Striking a balance between the two poles involves a lot of art. In racing games, the performance of the computer-controlled cars(CCCs) is one of the dominating factors. However, it is important to know what kind of mechanism is suitable to govern those computer cars, so that they can be comparably competitive with human players. In this paper, a Fuzzy-Neuro System (Yager and Zadeh, 1994) has been used, instead of the typical knowledge base driven AI controller (Funge, 2004) to simulate what a human driver does. By using the neural network (Bekey and Goldberg, 1993; Fu,1994; and Zalzala 1996), a new feature can be achieved, that cannot be found in other AI mechanisms (Rolston, 1988; and Luger, 2002)—i.e., the ability to learn. This enables CCCs computer cars in a racing game learn to make faster turns and improve performance with each lap for any racetrack provided.
This paper aims to model cars as self-contained objects with standard car control inputs and car behavior outputs, which will be used equally by all players for car racing games. Therefore, the play conditions of the game are equal and fair both for the player and the computer.
How can an AI system (Schwab, 2004) be arranged to have CCCs race around user-defined circuits so that the CCC's performance improves with each lap and rapidly becomes competitive with and even superior to the human controlled car performance? |
