A Methodological Outline and Utility Assessment of Sensor-based Biosignal Measurement in Human-Robot Interaction

Abstract

Sensor data taken during a human-robot interaction (HRI) have high potential for usage as new, objective measures of an interaction, either replacing or supplementing survey techniques that are currently most common in HRI research. Sensor data can be taken in large quantities quickly, naturally, and discreetly. They also have the potential to reflect a user’s biosignals—information about the user’s inner state (such as stress and attention) when interacting with the robot. We previously conducted three studies attempting to use sensor data as a measurement in HRI, with methodological differences in three different experimental environments. In this paper, we reanalyze and add new data to the previous findings under a consistent methodology, consolidate what correlations we find, and can conclude that sensor data is a useful metric in HRI across a wide range of experimental setups and subject pools. We fully describe the methodology we determined to be most effective, from selection of sensors to data analysis techniques to HRI experiment setup, as a basis for how this methodology can be used in other HRI studies. We describe necessary steps in the analysis of a large amount of sensor data (over 100,000 sets) and how sensor data can be compared with survey and behavioral data. Based on these correlations, we find that the most effective sensors are temperature sensors, tactile sensors, and face distance measurements. We also find that higher measurements across all of these sensors are more correlated with both survey and behavioral measurements reflecting positive thinking towards a robot (including non-technophobia, reciprocal behaviors, and positive ratings of the robot) during an interaction. Based on these results, we argue that robot sensor usage is an important and objective metric for HRI research.