Construction of state space in RoboCup

Machine learning has been widely applied to deal with problems in complex environments such as RoboCnp which is an ideal platform for research on AI and robotic. However, there are some very challenging problems in completing the machine learning in such a complex environment. One of them is how to construct an appropriate state space, which should have two main features to well describe the main characters of the environment states and to be small enough to be processed by ANN, RL, or other methods. In this paper, a new method to construct an appropriate state space in the complex environment is proposed, which fit the above two requirements. The authors also have completed a sample state space to describe the middle field situation in RoboCup simulation game, which can be used to do the route decision in RoboCup