A Comprehensive Approach to Validating the Uncanny Valley using the Anthropomorphic RoBOT (ABOT) Database

Abstract

The uncanny valley hypothesis posits that people’s emotional responses to robots are increasingly positive as robots’ resemblance to humans increases. However, when robots closely, but imperfectly resemble humans, people’s responses turn negative, only to revert back once their appearance more closely resembles humans. These sharp emotional transitions (i.e., peaks and valleys in emotional response) from positive to negative, and then back to positive, are collectively referred to as the uncanny valley. In this project, we attempted to validate the uncanny valley with the largest set of real-world robots currently available in open source format (the ABOT Database). Participants saw static images of 251 robots which varied in their degree of human-likeness, and rated them on uncanniness. We found significant empirical support not only for the hypothesized uncanny valley but an additional valley. This unanticipated valley emerged when the robots’ appearance had low to moderate human-likeness. Unique combinations of appearance dimensions of human-like robots may be responsible for the presence of an additional valley for robots that only moderately resemble humans. These findings of uncanny valleys in the existing robots may have important implications for robot design.