Virtual Pet

This project aims to design and generate virtual pets (VPs) based on their personality and appearance traits. To understand how VPs' appearance traits relate to their personality, we followed design guidelines from the Five Factor Model (FFM). We modeled several initial VPs that follow these design guidelines and then used these as datasets to generate more VPs. In this exhibition, we illustrate an AI approach for generating 3D artwork and provide a better understanding of users' perceptions of virtual pets' personalities.

The Context

The Metaverse has grown in popularity in recent years, and a lot of work has been done to explore the content in the virtual environment. For example, previous researchers explored the understanding of the player perception of narrative NPC roles in games, by proposing a methodological approach towards the visual design of non-player characters (NPCs) to fit specific narrative roles, because NPCs serve important functions for game narratives and influence player immersion. And the MBTI (Myers-Briggs personality inventory) test has been popular among the public for identifying individuals with personality disorders. We also found that previous researchers have applied machine-learning techniques to appearance generation. The popular personality test and the research on player roles in virtual environments motivated us to investigate the personality and appearance traits of virtual pets and generate them through machine learning algorithms with design guidelines.

Pet Design

To investigate the influence of pets' appearance traits on human perception of personality, we developed virtual pet characters inspired by real-life pets. Our methodology involved categorizing primary cat and dog breeds into six distinct clusters based on their appearance traits and creating a virtual pet character to represent each cluster. Using this approach, we aimed to make virtual pets with unique and recognizable physical characteristics while incorporating diverse appearance traits.

Study 1

The study aimed to understand users' perspectives on virtual pet characters, focusing on three primary goals: 1) comparing participants' perceived personalities of different styles of pets within the same cluster, 2) evaluating how our designed pet attracts people, and 3) understanding their perceptions toward keeping real pets and virtual pets.

The study design involved examining the influence of appearance (realistic and voxel visual style) and behavior (pets with or without animation) on participants' perceived personality and overall feeling (preference for offered pet images). Participants were randomly assigned to one of six pet clusters, and each participant evaluated all three conditions: real static pets (pet photos presented in a realistic style), static virtual pets (pet rendering pictures depicted in a voxel style), and animated virtual pets (animation clips showcasing rendering pets).

Results of Study 1

The quantitative and qualitative results showed that the style and appearance of virtual pets significantly impact participants' perceptions of their personalities. We also found that participants connected perceived characters with the appearances of virtual pets. We concluded that the design suggestions involve expected pet types, personality presentation factors, and interaction with pets, which benefit future virtual pet design.

Pet Genration

Building on the insights gleaned from our study 1 results, we now delve into the features that guided our generation process. These findings provide a strong foundation for our approach to generating virtual pet models. We have developed a unique 3D model generator that automatically creates virtual pets by hybridizing existing models and creating new ones based on input. The generation process is shown in right figure, which is mainly three steps. Divide and recombination, dyeing, and texturing.

Study 2

We conducted Study 2 through surveys and semi-structured interviews to gather users’ feedback on our generated characters. This study explored how participants perceive virtual pets' personalities by observing their appearances using quantitative and qualitative methods. Participants were presented with ten images, comprising three manually designed virtual pets (M1, M2, M3) and seven computer-generated ones (G1 to G7), refer to right figure, and requested to rate their likability on a scale of 1 to 5.

Results of Study 2

The statistical results showed that appearance features played a significant role in participants' likability ratings of virtual pets, specifically cobby cats with light-colored coats and decorations. Additionally, we identified significant correlations between certain personality and player-type traits and likability ratings of virtual pets.

The qualitative results show a strong correlation between virtual pets' physical features and personalities attributed by users. Appearance plays a significant role in conveying their personalities, as participants identified specific personality traits based on design elements like body shape, skin color, facial features, and expressions. Additionally, users' personalities can influence their preferred pet personalities, explaining why some prefer pets with corresponding personalities. Most participants found the voxel-style pets more relaxing and easier to create with intricate details than the realistic-style pets. Overall, appearance significantly shapes users' perceptions of virtual pets, and their preferences are related to the emotions evoked by the pet styles.

Conclusion

In conclusion, our study aimed to address the gaps in current research on virtual pets and their potential for promoting mental health and enhancing skill development in individuals who cannot keep real pets. We focused on creating virtual pets with personality traits and exploring how players perceive their personalities. Our research found that appearance variations affect users' perceptions of virtual pet personalities. Players prefer virtual pets like cats and dogs with neurotic personalities and cute appearances. We also developed a novel method for a game character design that combined traditional methods with machine learning techniques. Our study provides several design implications for virtual pet character design. It highlights the potential of using voxel pets' appearances as virtual companions to enhance the mental well-being of young individuals by reducing anxiety levels through interactive engagement with virtual pets. Overall, our study contributes to the understanding of factors contributing to the development of personalities in different species and how we can design artificial companions that mimic and respond to these traits.