Decision-Making of Football Agents with Support Vector Machine

Robocup has attracted much attention for Artificial and Computational Intelligence researchers. Robocup involves various aspects of problems, i.e., cooperation with team mates, dynamic problems, imperfect information, uncertainties caused by noise, and so on. Therefore, it is quite difficult to design football agents. In this paper, Support Vector Machines, one of the most famous machine learning algorithms, are used to decide if the agents carry out basic skills, such as shoot and through balls, which are given in advance. That is, firstly, data, i.e., the position and directions of balls and players, is collected by playing given skills naively. Then, labels indicating the success/fault of the skills are added to the data. Secondly, SVM learns the data. Finally, the SVM decides if the skill should be carried out. Several experiments on game plays with stronger team binaries at Japan Open elucidate the effectiveness of the proposed method.