انجمن کامپیوتر ایران

This paper provides technical solutions for two problems selected from Soccer Simulation 3D domain. The agent movement equation problem and the localization problem are selected from this domain and investigated using several machine learning methods, including neural networks, evolutionary learning and statistical learning. Results show the remarkable advantage of reinforcing ordinary multi-layer perceptron neural networks with evolutionary algorithms. Also they confirm the superior performance of support vector machines for regression tasks when the underlying system is neither dynamic nor chaotic.