Abstract
We applied a reinforcement learning with fuzzy state to a pursuit game
in a real number environment, where hunters were not able to capture the prey
for a difficult capture condition because they used only observed information for the
prey. In this research, when the prey is nearer than a certain distance from a hunter,
the hunter uses the observed information for not only the prey but also the other
hunters who are nearer than a certain distance from the prey. This means that a
hunter switches a simple Q-table to a complex one. All combinations of observed
information for the other hunters severely increase the number of states. Therefore,
we translate the observed information into relative relations for hunters position
with the prey, where the number of states does not increase too much. And we
simulated the method for various distance of switching Q-table.
