Abstract
Reinforcement learning is one of effective mechanisms which enable an agent to acquire its behavior. But when an environmental change occurs, the agent has to learn again from scratch. In this paper, the agent system is constructed by using the probabilistic network in order to cope with this problem. In concrete terms, an agent acquires its action through reinforcement learning, and its result is replaced to a probabilistic network which is a more flexible expression. Through the navigation problem, the availability of the proposal technique is shown.
