By means of the soccer matches between robots, researchers and students alike are seeking answers to the different issues of robotics and artificial intelligence. The problems arising during a soccer game are essentially the same as those faced in the case of robots applied in many different aspects of life. The wide-spread popularity of these matches therefore contribute to improvements in the aforementioned fields, as well.
The RoboCup soccer simulation is a game of emulating real, physical robots in a virtual gaming space, where the competing robots are controlled entirely by software solutions, therefore providing a higher abstraction level for the researchers in the field of artificial intelligence.
The challenge of performing tasks in the simulated environment is further increased by the need to design a real time and multi-agent system, the stochastic operation, the lack of information about the environment, and the exceptionally large state space. On the other hand, it provides a valuable opportunity to investigate problems arising in real life.
This thesis paper offers an insight into the simulation environment and the requirements for soccer simulation players. I present several algorithms for solving fundamental problems concerning the capabilities of the robotic soccer players. I also introduce my training module for testing the implemented capabilities.
Furthermore, I examine the machine learning methods used in implementing more advanced features, then I explain an algorithm for intercepting the ball, which I have elaborated using the Q-learning method of reinforcement learning.