Each person has their private physical and/or psychological area where they do not want to share with others during social interactions. This area gives them comfort about interactions and its size usually depends on various factors such as culture, personal traits, and acquaintanceship. This issue may also arise in case of human-robot interaction, especially when the robot is required to generate a socially competent interaction strategy toward people they are interacting with. Here, we propose a new robot exploration strategy to socially interact with people by considering the social relationship between the robot and each person. To that end, two definitions of interaction area are made: (1) Acceptable area allowed to be shared with other people and robots, and (2) Private area where a human does not want to be interfered by others. Based on these definitions, the robot can optimize the path to maximize the frequency/degree of visiting the acceptable area of each person and to minimize the frequency/degree of trespassing into the private area of them at the same time in an iterative way. In this paper, the social force model (SFM) of each person, based on the potential field concept, is designed by a fuzzy inference system and its parameter is optimized by the reinforcement learning model during interactions. We have shown that the proposed model can generate a suitable SFM of each person, which was quite similar to a ground truth model, allowing to plan a path to simultaneously optimize the two factors of interaction area, respectively.
Pakpoom Patompak, Sungmoon Jeong, Itthisek Nilkhamhang, Nak Young Chong, Learning Social Relations for Culture Aware Interaction, URAI2017, International Conference on Ubiquitous Robots and Ambient Intelligence, Maison Glad Jeju, Jeju, Korea from June 28-July 1, 2017.
Outstanding paper award