Exploring Interpersonal Distance with Virtual Agents on a Naked-Eye Stereoscopic Display

Abstract: This study investigates how virtual agents displayed on a naked-eye stereoscopic (3D) screen influence interpersonal distance, as well as discomfort and likeability, during a common corridor passing task. The increasing adoption of "on-screen agents"—virtual body agents displayed on devices such as smartphones and large public screens—offers an alternative way to physical body robots, reducing their high costs and potential physical risks. However, a question remains: can these virtual agents enable users to maintain appropriate interpersonal distances in real-world scenarios? In particular, traditional 2D displays often fall short in conveying the sense of physical presence that humans instinctively need to maintain a comfortable personal space. This deficiency highlights a major challenge in designing natural and effective human-agent interactions.This study directly addresses this gap by focusing on the maintenance of interpersonal distance in dynamic, pre-interaction scenarios within a simulated environment, specifically tailored for immersive technologies. We set out to answer two key research questions:•RQ1: How does interacting with an on-screen agent presented on a stereoscopic display affect the maintenance of interpersonal distance?•RQ2: How do the display conditions (3D vs. 2D) of the on-screen agent affect user comfort and likeability during dynamic interactions?To investigate these questions, we leveraged a Looking Glass naked-eye stereoscopic display, that offers multiple viewpoints without any wearable devices, providing a robust platform for simulating depth perception. We designed a controlled experiment where participants walked past a human-like 3D agent in a simulated corridor. The experiment employed a within-participants design across three different conditions: OFF condition (the display turned off), 2D condition (stereoscopic display function disabled), and 3D condition (stereoscopic display fully enabled). To consider potential order effects, all 15 Japanese participants (aged 21-28 years) experienced all three conditions in different orders. During the task, we precisely measured the shortest distance that participants reached while passing in front of the agent. After completing the task, participants rated their discomfort using the Japanese translation of the RoSAS scale and their liking for the agent using the Godspeed scale. All statistical analyses were conducted using Wilcoxon signed-rank nonparametric tests with Bonferroni-Holm corrections.Findings regarding interpersonal distance in RQ1 showed the following key findings:•There was a significant difference between 3D and 2D conditions (3D condition: Mean = 1046.67 (mm), SE = 62.96; 2D condition: Mean = 755.00 (mm), SE = 74.33; p < .01, Bonferroni-Holm corrected). •There was a significant difference between the OFF and 3D conditions (OFF condition: Mean = 403.67 (mm), SE = 26.07; 3D condition: Mean = 1046.67 (mm), SE = 62.96; p < .001, Bonferroni-Holm corrected), and between the OFF and 2D conditions (OFF condition: Mean = 403.67 (mm), SE = 26.07; 2D condition: Mean = 755.00 (mm), SE = 74.33; p < .01, Bonferroni-Holm corrected)These results strongly suggest that stereoscopic displays enhance the visual presence and realism of virtual agents, thereby working to bring human spatial behavior closer to real-world social norms regarding interpersonal distance. Furthermore, these highlights that the mere presence of the agent influences spatial behavior, regardless of the type of display.On the other hand, RQ2 (ratings of discomfort and likability) showed no significant differences between the 3D and 2D conditions (Discomfort; 3D condition: Mean = 3.36, SE = 0.16; 2D condition: Mean = 3.18, SE = 0.15; p = .27, n.s. , Bonferroni-Holm corrected, Likeability; 3D condition: Mean = 2.51, SE = 0.13; 2D condition: Mean = 2.48, SE = 0.12; p = .88, n.s., Bonferroni-Holm corrected). This result suggests that in a dynamic task such as walking, the display format alone may not be the main driver of emotional responses. It is possible that participants' attention was distributed throughout the environment, or other factors, such as the passive behavior of the agent, had a greater impact on perceived comfort and likability than nuances in display depth.This study shows that virtual agents displayed on a stereoscopic screen have a significant effect on interpersonal distance in dynamic situations, bringing interactions closer to natural human social behavior. This finding is an important step towards achieving more natural and comfortable interactions with on-screen virtual agents, and will contribute greatly to the development of augmented reality (AR), virtual reality (VR) and mixed reality (MR) simulations in particular.This study has limitations: small sample size (15 Japanese participants). Future work needs diverse groups and varied agent behaviors (e.g., dynamic gaze) for more natural interactions.

Keywords: On-screen agent, Interpersonal distance, Stereoscopic display

DOI: 10.54941/ahfe1006908

Cite this paper:

​