A Large-Scale Quantitative Analysis of Avatars in VR and AR

Area of Science:

• Human-Computer Interaction
• Virtual Reality
• Augmented Reality
• Computer Graphics
• Social Computing

Background:

• Avatars are crucial in virtual and augmented reality (VR/AR) environments.
• Limited understanding exists regarding avatar representation within research.
• A comprehensive analysis of avatar imagery in scientific publications is needed.

Purpose of the Study:

• To systematically analyze avatar representations across a large corpus of VR/AR research.
• To identify trends and biases in avatar design, including gender, ethnicity, body type, and visual style.
• To provide actionable insights for improving avatar diversity and reducing bias in future systems.

Main Methods:

• Collected and analyzed 14,440 avatar images from 4,659 publications.
• Hand-labeled each image for gender, ethnicity, body representation, and visual style.
• Linked image data with publication metadata (keywords, affiliation, year) for comprehensive analysis.

Main Results:

• Identified a dominance of realistic white male avatars in VR research.
• Observed lower-fidelity and more gender-diverse bodies in AR.
• Noted a post-2021 increase in stylized designs and diversity labels.
• Found a mismatch between female-associated keywords and avatar occurrences.

Conclusions:

• Current avatar research exhibits significant representational biases.
• Emerging trends indicate a move towards greater diversity and stylized designs.
• Recommendations include adopting balanced starter libraries, bias dashboards, and representation checklists to mitigate bias.