In this paper, we suggest an r-learning (robot-learning) system for children who are unfamiliar with educational tools. Our system is characterized with the following features. Firstly, the system provides children with personalized educational instructions considering their learning ability. In our system, learning contents for them are selected according to their learning history extracted from human-robot interaction data. Secondly, our system helps them learn educational contents easily and joyfully through various behavioral interactions with a robot. Third, r-learning scenarios are modeled with Petri-net to handle exception during learning and robot contents can be composed by use of predefined robot behaviors. Finally, we implement these system S/W using OPRoS which is a software platform for robotic services.