Abstract. In this paper, we present a Human Robot Interaction (HRI) system which can determine people's interests and intentions concerning exhibits in a museum, then proactively approach people that may want guidance or commentary about the exhibits. To do that, we first conducted observational experiments in a museum with participants. From these experiments, we have found, mainly three kinds of walking trajectory patterns that characterize global behavior, and visual attentional information that indicates the local behavior of the people. These behaviors ultimately indicate whether certain people are interested in the exhibits and could benefit from the robot system providing additional details about the exhibits. Based on our findings, we then designed and implemented a network enabled guide robot system for the museum. Finally, we demonstrated the viability of our proposed system by experimenting with a set of Desktop Robots as guide robots. Our experiments revealed that the proposed HRI system is effective for the network enabled Desktop Robots to proactively provide guidance.