2D BEHAVIOR BASED SOCCER TEAM - A NEURAL NETWORK APPROACH
Felipe Lira Santana Silva1; João Paulo de Almeida Barbosa2; Guilherme Sousa Bastos1
1 Universidade Federal de Itajubá; 2 Instituto Tecnológico de Aeronáutica
Resumo
RoboCup is an internatinal robotics competition created to stimulate artificial intelligence research all over the world. In a RoboCup's 2D Soccer Simulation League, a server simulates a soccer game with 11 players and 1 coach for each side. This server saves after each game two log files, that can be analyzed with Data Mining techniques to extract information about the game. The software developed in this work can read more than 30 log files in less than 2 hours, so it is possible to change the defensive behavior as soon as the game ends. A feed forward artificial neural network is then trained to recognize opponents' shoot and pass probabilities, and its output is used as sensors to activate specific defensive behaviors to suppress opponents' scoring ability. Practical tests showed a 17\% decrease in opponents' scoring when using the new Expertinos team.
Palavras-chave: Artificial Neural Network; Simulation 2D; RoboCup