A Basic Study on Indicator of Transfer Learning for Reinforcement Learning

Reinforcement learning requires a lot of learning time for the agent to learn.Transfer learning is a method to reduce this learning time, but it has the problem that the user does not know which knowledge is effective in which environment until it is learned.Therefore, it is necessary for the user to consider the relationship between the source and destination when transferring knowledge.Therefore, this study proposes Indicator of adaptation criteria that can determine this relationship in advance.In simulations, we demonstrate the usefulness of the proposed method by using some example problems.