Opponent zigzag movement model capture and prediction in robotic soccer

Robotic soccer research as a standard problem to foster Artificial Intelligent research in robotics and related fields still needs improvement in game play modeling. Action anticipation decision could be based on opponent game play model. Artificial Neural Network has been used to capture game play model, but still has problems of slow learning convergence and inaccuracy. Artificial Neural Network algorithm need to be a small, fast, and robust application for small devices of robot soccer. In this paper, experiments of model capturing simulated zigzag soccer movement has been done using Radial Basis Function Neural Network, Inverse Function Neural Network, pseudo-inverse Singular Value Decomposition, existing and modified Simurosot robot soccer prediction algorithm. Model capture results show, despite higher error in zigzag movement model capture, the Inverse Function Neural Network and the modified Simurosot prediction algorithm yield lower error in movement prediction.