This paper discusses what agents' autonomous learning is. Although the viewpoints of the agent have been given by a designer in conventional learning models, the agent ought to be able to replace their own viewpoints by itself in a real sense that the agent has autonomy. In this paper. it is assumed that an autonomous agent should be equipped with the following three characteristics; (a)self-generation of goals and evaluation criteria, (b)sense of values, (c) boundary setting between self and others. Especially, the discussion in this paper was focused on acquiring process of the second characteristic, (b)sense of values, by proposing Bi-Referential Model, which is based on developmental psychology. We discuss how the segregating processes of sense of values are dependent on the competition of shared resources, and how the Bi-Referential Model can reenact such segregating processes. And then, we compare our approach with some related works and explain the. major differences from those.