In this paper, an action selection mechanism (ASM) is proposed for each agent given its role in a multiagent system for robot soccer. The robot soccer game has very dynamic and uncertain characteristics. Furthermore, there exist competitions between agents. The soccer-playing robot should take an appropriate action according to its given role such as striker or sweeper. A computational mechanism for such action selection is essential in this regard. Initially, a relatively simple ASM is designed for a situation with no opponents. Then, the mechanism incorporates some additional action selection schemes considering the opponents' behaviors. Particularly, a multilayer perceptron (MLP) is used to learn how a human being selects an action given a situation. The effectiveness of the proposed ASM is demonstrated through a real robot soccer game of S-MIROSOT (single-robot micro-robot world cup soccer tournament).
展开▼
