| A major component of a RoboCup simulated soccer team is the online coach agent. The ability of the coach agent to oversee everything happening in the soccer field provides for the introduction of high-level, strategic planning to the whole team.; RoboCup Team Formation Recognition is the problem of recognizing the team formation used by the opposing team. With this information, the coach can plan for a counter-formation to gain strategic advantages over the opponents.; This thesis examines a neural network-based approach for solving the RoboCup Team Formation Recognition problem. Systematic experiments are used to evaluate several single hidden-layer neural networks of different sizes in search of an optimal number of hidden nodes to employ. Once this number is found, the 5-Fold Cross-Validation tests are used on neural networks against a heuristic-based approach, called the Geometric Method.; The results presented in this thesis is a major step towards building a competitive online coach agent for use in a RoboCup simulated soccer team. By incorporating an accurate team formation recognition module, the online coach can better model the opponents and thus gain strategic advantages over the opposing team,... |
