This paper introduces the virtual steward robot which can interact with the inhabitant and assist/serve daily lives in the intelligent home environment. This virtual robot agent has learning and emotional interaction capabilities, and it is capable of performing human-friendly interactions. The learning system tries to reveal the empirical behavior patterns of the user"s daily life style, and appropriately controls each device with the acquired knowledge through reinforcement learning techniques and fuzzy logic. In order to achieve friendly interactions with the user, a virtual 3D avatar has been programmed to perform facial expressions in responding to the recognized emotional state of the user. For this task, we have developed an emotional model, and an affective relation between robot and the user.