Detecting cultural identity via robotic sensor data to understand differences during human-robot interaction

Abstract

Socially-assistive robots (SARs) have significant potential to help manage chronic diseases (e.g. dementia, depression, diabetes) in spaces where people live, averse to clinic-based care. However, the challenge is designing SARs so that they perform appropriate interactions with people who have different characteristics, such as age, gender, and cultural identity. Those characteristics impact how human behaviors are performed as well as user expectations of robot responses. Although cross-cultural studies with robots have been conducted to understand differing population characteristics, they have mainly focused on statistical comparisons of groups. In this study, we utilize deep learning (DL) and machine learning (ML) models to evaluate whether cultural differences show up in robotic sensor data during human-robot interaction (HRI). To do so, a SAR was distributed to user's homes for three weeks in the US and Korea (25 participants), while collecting data on the human activity and the surrounding environment through on-board sensor devices. DL models based on that data were able to predict the user’s cultural identity with roughly 95% accuracy. Such findings have potential implications for the design and development of culturally-adaptive SARs to provide services across diverse cultural locales and multi-cultural environments where users’ cultural background cannot be assumed a priori.