Abstract
A pursuit problem is a multi-agents' benchmark problem, where four hunters pursue and capture the prey in a cell environment. We have proposed a staged view that reflects a human's view where we can see an object easily in the neighborhood but more difficult in the longer distance and easily in the center direction but more difficult in the righter and lefter directions in both cell environment and real number environment. With the staged view, the information that hunters can obtain becomes unclear depending on distance and direction, and we classify the hunter's view as "neighborhood", "short range", "middle range", and "long range". We showed that hunters can capture the prey even with the staged view. In this paper, we propose a staged view that reflects scopes of human's view more than previous one, that is, scopes where we can watch with both eyes (neighborhood), we can see with both eyes (short range), we can see with only one eye (middle range), and we can not see (long range). We apply a fuzzy Q-learning to the pursuit problem with the proposal staged view. A result shows that hunters can learn effectively and capture the prey even with the proposal staged view.
