Intelligent Decision System Based on SOM-LVQ Neural Network for Soccer Robot

To make real-time and optimal decisions, the decision system of the soccer robot must be able to learn priori-knowledge given by humans and its own match experience. These two learning issues should be addressed respectively in two learning phases of the system, i.e., the off-line learning and the on-line learning. The SOM-LVQ model is thus proposed in this paper. Compared with conventional decision models and typical BP neural network model, the SOM-LVQ provides a knowledge structure based on which the on-line learning is much easier to fulfill. An intelligent decision system based on the SOM-LVQ model was built. In matches with different opponents, this intelligent system showed outstanding adaptivity and achieved much higher right decision ratio than others.