Emotional distress in everyday life often takes the form of frequent, mild, yet persistent fluctuations, creating a need for low-threshold, interruptible, and repeatedly accessible forms of short-term emotional buffering. Animal-assisted intervention (AAI) and real human-pet interaction suggest that emotion regulation does not necessarily depend on language, but may instead be achieved through sustained presence, nonverbal responses, and shared interaction rhythms. Mixed reality (MR) brings virtual objects into real space, providing new media conditions for companionship mechanisms based on spatial co-presence, embodied interaction, and multimodal feedback. However, MR virtual pets designed for fragmented everyday emotion regulation still lack a systematic design framework and validation of key variables.

　　This study conceptualizes the MR virtual pet as a form of “present other” and seeks to explore design pathways through which it can foster embodied companionship and stable affective responsiveness in fragmented daily scenarios. The research combines Research through Design (RtD) with empirical evaluation. Through literature and state-of-the-art review, case studies, pet-owner interviews, and exploratory MR research, it extracts translatable nonverbal interaction elements from human-pet relationships and identifies their implementation constraints, on the basis of which an MR virtual pet design framework for emotion regulation is constructed. To examine the role of emotional cues in virtual pets, the study develops two comparative versions: a basic-drive version, which organizes presence and responses around physiological needs, and an affective-drive version, which further introduces emotional dimensions to modulate motivation and expression.

　　The evaluation consists of two stages: a pilot study and a formal experiment. The pilot study aims to examine the readability and distinguishability of the virtual pet’s different states and facial expression designs. The results show that different virtual pet states can be positioned in affective space with relatively high consistency and discriminability, and that the addition of facial expressions further enhances expressive intensity and recognition stability. The formal experiment employed a between-subjects design (N = 42; control group: n = 20; experimental group: n = 22) and used SAM, PANAS, experience questionnaires, and interviews to assess short-term emotional changes and perceived companionship. The results indicate that both versions produce significant within-group emotional improvement, while no significant difference is found in the magnitude of change between groups. Qualitative findings further suggest that the emotional support provided by MR virtual pets depends first and foremost on a clear response loop formed through spatial co-presence and embodied interaction. The incremental value of emotional modulation is reflected less in producing stronger emotional improvement than in enhancing the attribution of internal states and deepening the sense of relational experience.

　　By constructing an MR virtual pet design framework for fragmented everyday emotional support, this study clarifies the functional boundaries between basic-drive and affective-drive mechanisms, providing both design guidance and empirical reference for subsequent related design practice and research. The findings suggest that MR virtual pet design for fragmented emotion regulation should prioritize a readable action-feedback loop and reversible interaction boundaries, while supporting users’ stable attribution of the virtual pet’s behavior through the cross-validation of consistent multimodal expression. The value of emotional modulation lies not in amplifying emotion regulation effects, but in enriching relational experience and sustaining ongoing interaction through interpretable externalizations of internal state.