"Let's Face It": Investigating User Preferences for Virtual Humanoid Home Assistants

While a growing number of households contain home assistants, they mainly remain voice-only devices where the virtual agent is not represented visually. The visual representation of the agent is limited to the device's housing and abstract light animations that signify the assistant's state to its users. However, the audio channel is limited in conveying information beyond semantic content. Embodied virtual assistants can enhance interaction with conversational interfaces by adding a visual layer to further convey personality and human characteristics. In this work, we conducted an online survey (N= 78) to explore people's preferences for visualizing humanoid assistants. Our findings suggest that participants prefer an agent who appears mature, healthy, competent, and attractive. Furthermore, demographic similarities between the users and agents are wished for the agent to look more relatable. We discuss these findings and their implications for the design of virtual humanoid home assistants.