The characteristic task of service robots that can interact with humans is to achieve human-robot collaboration. Mutual adaptation is considered to be an important characteristic of robots, required for carrying out such collaborative tasks. Here, we introduce the concept of mutual adaptation, propose a learning model, and describe an experimental task to explain the above concept. A waiter robot performs a collaborative task using a platform system, which is developed by a constructive approach. The interactive and manual modes of this system are compared by performing a preliminary experiment to evaluate the effectiveness of the robot’s autonomous function. The results indicate that the robot’s autonomous function works well when operated in the interactive mode under short time or slow speed conditions.