An Unsupervised Learning Method for an Attacker Agent in Robot Soccer Competitions Based on the Kohonen Neural Network

RoboCup competition as a great test-bed, has turned to a worldwide popular domains in recent years. The main object of such competitions is to deal with complex behavior of systems which consist of multiple autonomous agents. The rich experience of human soccer player can be used as a valuable reference for a robot soccer player. However, because of the differences between real and simulated soccer world, it needs to fit such precious transcendental knowledge to use in the simulated soccer game. On the other hand, Reinforcement Learning (RL) as a common method in this domain because of its trial-and-error nature does not have great performance in using transcendental knowledge. Thus, this method is limited to complex multi-agent learning problems. Among various frameworks of intelligences, in general, Artificial Neural Networks (ANN) and specially Kohonen neural networks with its feed-forward architecture and its ability in discovering any relationships of interest that may exist in the input data may be considered as a powerful tool in clustering. This paper puts forward an unsupervised learning method based on Kohonen network to create a powerful Tactics layer in decision-making section for an attacker agent. The approach presented in this paper is based on the combination of expert’s knowledge and data obtained from the simulated world. This system is applied to the attacker agents of ULA 2006 soccer team. Simulation results revealed that the chosen approach is superior with respect to the other intelligent techniques.