Humans Can't Resist Robot Eyes - Reflexive Cueing With Pseudo-Social Stimuli

Abstract

Joint attention is a key mechanism for humans to coordinate their social behavior. Whether and how this mechanism can benefit the interaction with pseudo-social partners such as robots is not well understood. To investigate the potential use of robot eyes as pseudo-social cues that ease attentional shifts we conducted an online study using a modified spatial cueing paradigm. The cue was either a non-social (arrow), a pseudo-social (two versions of an abstract robot eye), or a social stimulus (photographed human eyes) that was presented either paired (e.g. two eyes) or single (e.g. one eye). The latter was varied to separate two assumed triggers of joint attention: the social nature of the stimulus, and the additional spatial information that is conveyed only by paired stimuli. Results support the assumption that pseudo-social stimuli, in our case abstract robot eyes, have the potential to facilitate human-robot interaction as they trigger reflexive cueing. To our surprise, actual social cues did not evoke reflexive shifts in attention. We suspect that the robot eyes elicited the desired effects because they were human-like enough while at the same time being much easier to perceive than human eyes, due to a design with strong contrasts and clean lines. Moreover, results indicate that for reflexive cueing it does not seem to make a difference if the stimulus is presented single or paired. This might be a first indicator that joint attention depends rather on the stimulus' social nature or familiarity than its spatial expressiveness. Overall, the study suggests that using paired abstract robot eyes might be a good design practice for fostering a positive perception of a robot and to facilitate joint attention as a precursor for coordinated behavior.

Keywords: anthropomorphism; collaborative robot (cobot); gaze-cueing; human-robot interaction (HRI); joint attention; robot design.

FIGURE 1

Set-up and sequence of events on a given valid trial. Each trial began with the presentation of a fixation cross in the centre of the “room.” After 900 ms, a display appeared with a “gaze” facing to the front. 1,000 ms later the gaze averted to one of the eight target positions. After a stimulus onset asynchrony (SOA) of 420 ms the target appeared in one of the eight positions. The target disappeared upon participants’ reaction or a time-out of 2000 ms.

FIGURE 2

(A) The four stimulus types, labelled respectively. Only the paired version is presented here. The top row includes the social (left) and non-social (right) stimuli. The abstract robot eyes are presented in the second row. (B) Image of the collaborative robot Sawyer used in the questionnaire. Here, presented featuring the single pixel eye design.

FIGURE 3

Mean reaction times for valid and invalid trials for the different stimulus type conditions arrow(s), robot pixel, robot crosshair and human. Error bars represent the standard error of the means.

FIGURE 4

Mean gaze cueing effect of the different stimulus type conditions arrow(s), robot pixel, robot crosshair and human. Error bars represent the standard error of the means.

FIGURE 5

Mean ratings for the three single items addressing task-related attributions (perceived accuracy, surveillance and usefulness) for the different stimulus type conditions arrow(s), robot pixel, robot crosshair and human. Error bars represent the standard error of the means.