Abstract
This chapter proposes an efficient modular artificial neural network (ANN) architecture for the intelligent decision making of a robot in a robot soccer systems with different team configurations. In conventional ANN based approaches, the decision making systems are trained separately for different team configurations and which leads to an increased computational overheads and learning time. The technique discussed in this chapter can alleviate this situation, by making use of a flexible modular ANN architecture capable of accommodating different team configurations without any repeated learning phase as the team configuration changes. The basic building block in this modular decision making system is an ANN developed for a MiroSot small league system. Decision space of the MiroSot small league system is simple and two dimensional so that its prediction accuracy is high. The simulation results indicates that the modular decision making system developed for higher team configuration maintains the same level of prediction accuracy as that of the smaller team configuration. In the modular approach decision variables are shared among the various ANN blocks so that it need not be trained again.