A Kansei-evaluation Study on the Uncanny Valley of Real-world Robots

Abstract

Roboticists believe that highly, but not perfectly, realistic human-looking robots elicit negative feelings in humans. This is the so-called ‘uncanny valley’ response. Most research on the uncanny valley has used the morphing technique and the morphing rates as the objective scale. However, those findings might come from unnatural and/or unexistable appearance of morphed faces. The present study investigated the uncanny valley by measuring participants’ impressions of some existing robots whose degree of human likeness was evaluated by Scheffé’s paired comparison method, a kansei-engineering approach. The survey demonstrated the uncanny valley like Mori’s (1970) hypothetical graph. Furthermore, as the X-axis, ‘experience’, one of the two independent dimensions in mind perception (Gray, Gray, & Wegner, 2007), showed the pattern of the uncanny valley, but ‘agency’ did not. These results suggest that robots become unnerving when people ascribe to them experience, rather than agency.