Correlation Analysis for Predictive Models of Robot User’s Impression: A Study on Visual Medium and Mechanical Noise

Abstract

Service robots are increasingly common. Studies show that the appearance and behavior of robots influence users’ impressions. Developers need to know whether users’ impressions of robots are in accordance with the purpose they designed or not. To assess such impressions, it is necessary for participants to meet robots in a real-world setting. Such studies are costly and take a lot of time. On the other hand, experiments are now extremely easy to perform over the internet. However, there is no evidence that impressions obtained from a video recording or a movie of a simulated avatar are comparable to an impression obtained in a real-world setting. In this study, we hypothesized that there are trade-offs between ease of collecting impressions and the real-world applicability of said impressions. We tried to use impressions obtained in “Recorded” and “Avatar” settings to predict the impressions in a “Real” setting by correlation analysis. In addition, we also performed muted real-world setting “Soundproof” to evaluate the influence of motor noise of robot motion for user’s impressions. In the experiment, two kinds of humanoid robots performed five kinds of motions. Participants gave their impressions in quantitative form under four different conditions: real-world, video recording, video avatar and muted real-world. Our study takes into account the effects of motor noise in addition to the medium on which the robot is seen. Our results show correlations between the “Soundproof”, “Recorded”, and “Avatar” settings. We found that motor noise affects participants’ impressions of the robot and that some trade-off relationships exist between the different conditions.