1 Media Technology, The Faculty of Engineering and Science, Aalborg University, VBN2 Mobility and Tracking Technologies, The Faculty of Engineering and Science, Aalborg University, VBN3 Aalborg U Robotics, The Faculty of Humanities, Aalborg University, VBN4 Department of Architecture, Design and Media Technology, The Faculty of Engineering and Science, Aalborg University, VBN5 The Faculty of Engineering and Science, Aalborg University, VBN
In previous research we showed how a cross‐cultural multimodal corpus of human face to face interactions can serve as empirical foundation for implementing multicultural agents (e.g. Rehm et al. 2009). To this end, the corpus has been annotated in relation to a number of communicative phenomena like gestures, postures, proxemics, communication management, etc. Based on these annotations, statistical models have been derived and implemented in embodied conversational agents in order to simulate culturespecific communicative behavior in these agents. Although this was a successful approach, there have always been some discomforts about it due to the following reasons: - Subjectivity of the annotation - Cultural bias of the annotation and the implementation/design of the agents - Applicability of results to agents (they might represent a culture of their own) Meanwhile, we have moved to physical agents, i.e. robots. Here the difference between humans and robots is more apparent, due e.g. to limited expressive channels or reduced degrees of freedom. For instance the Nao platform which we are using has less joints in arms and legs then a human, making movements look different, independent on how careful movements have been designed. Thus, we suggest a new methodological approach for modeling the behavior of robots and for making them culturally aware which could also be of benefit for the use in multicultural agents. Instead of collecting data from human interactions, potential users of the system have to create the behavior themselves, allowing us to collect a cross‐cultural database of behavioral parameters for the robotic system under development. This reduces any subjective or cultural bias that was previously introduced. Instead, the cross‐cultural database of behavioral parameters allows data‐mining the statistical models that will drive the generation of communicative behavior in the robot based on informants that are immersed in the target culture.
Lecture Notes in Computer Science: 15th International Conference, Hci International 2013, Las Vegas, Nv, Usa, July 21-26, 2013, Proceedings, Part Iii, 2013, p. 431-440
Culture Aware Robots; Human robot interaction; Affective Computing
Main Research Area:
Lecture Notes in Computer Science (lncs)
15th International Conference on Human-Computer Interaction (HCI International 2013)Human-Computer Interaction