Adaptive Choreography for User’s Preferences on Personal Robots Takamitsu Matsubara, Shizuko Matsuzoe, and Masatsugu Kidode Abstract—We propose an approach to efficiently adapt the non-verbal information of personal robots to a user’s prefer- ences. As non-verbal information focuses on gestures except meaningful gestures, in our method the adaptation problem is formalized as choreography; the action (as one unit of the non- verbal information) is assigned to each state as preferred by the user. Since the state is defined based on the classification of language focusing on the shallow discourse structure proposed by [1], our method is constructed with independence of tasks and situations unlike previous methods. Furthermore, using a preferences database obtained from multiple users, we produce a User Preference Model (UPM) that can represent various user preferences by a small number of parameters. A new user is asked to assign actions on a few states to adapt the UPM for the user preference. After the adaptation, the UPM can be used to predict actions on other states as preferred by the user, that accomplishes adaptive choreography. We implemented the proposed method with a personal robot, and validated it through experiments on multiple tasks with human subjects. As a result, we confirmed that the user’s impressions of the robot were greatly improved by our method for multiple tasks. I. I NTRODUCTION In human-human communication, non-verbal information such as facial expression, direction of eye gaze, or gestures, is recognized as a crucial factor for improving the impression made on the partner [2]. Desirable non-verbal information depends not only on the task and situation, but also on the preferences of the partner. Therefore, for smooth and comfortable communication, understanding of the partner’s preferences through interactions is important. Inspired by such considerations in human-human communication, for human-robot communication, many studies have been made on capturing the user’s preferences with robots that could express non-verbal information. Ogawa et al. [3] demonstrated the effect of rhythmic entrainment of a robot’s non-verbal cues, such as nodding and gestures, in response to a user’s vocal communication for improving impressions on the partner. Yonezawa et al. [4] used ambient gaze tracking to give the appearance of mutual attention and mutual gaze by a stuffed-toy robot and showed that these behaviors improved the feelings of the partner for the robot. Shiwa et al. [5] presented experimental results showing that a slow response time by the humanoid robot “RobovieII” was forgiven when the robot made use of “conversational fillers” such as the Japanese expression “etto” (resembling T. Matsubara and M. Kidode are with Graduate School of Information Science, Nara Institute of Science and Technology, Japan, {takam-m, kidode} @is.naist.jp S. Matsuzoe is with Graduate School of Systems and Infor- mation Engineering, University of Tsukuba, Japan, {s1130184} @u.tsukuba.ac.jp I prefer that you shake your head widely… Oh! That's good! Fig. 1. A rough sketch of the choreography. “well...” or “uh...” in English). Moreover, Mitsunaga et al. [6] developed an adaptation mechanism based on policy gradient reinforcement learning for robot proxemics and gaze behaviors using initial default parameters for human-robot proxemics based on social spatial zones [7]. Previous studies have demonstrated the effectiveness and importance of user-adapted non-verbal information by the robot to cause good impressions with the user; however, these studies have focused on specific elements of non- verbal information such as nodding, direction of eye gaze and proxemics assuming particular tasks and situations. Focusing on each task and situation, the adaptation problem has also been constrained by introducing several assumptions that might only be satisfied within the task. Thus, these approaches are constructed with Task dependence. Although task dependence allows us to make the complex tasks of user adaptation in non-verbal communication much simpler and more feasible, the adaptation problem needs to be stated for each task and situation. To alleviate this issue, in this paper, we explore a Task- independent approach to adapting the non-verbal information of personal robots to a user’s preferences, inspired by the recommender systems [8]. In our approach, the problem of adaptation of the non-verbal information to user preferences is formalized as the choreography; the action (as one unit of the non-verbal information) is assigned to each state as preferred by the user. The main features of the proposed method are: 1) Task independence 2) Quick adaptation The details of these features are as follows. 1) Task independence indicates that our method can be commonly applied to various tasks and situations. To achieve this property, we utilize the SWBD-DAMSL (Switchboard Discourse Annotation and Markup System of Labeling) tag [1] for the state definition. Since the SWBD-DAMSL tag