Deco: Extending Personal Physical Objects into Pervasive AI Companion through a Dual-Embodiment Framework

Problem
This preprint addresses the gap in the literature regarding the emotional connection individuals have with physical objects, such as plush toys, which typically lack the capability to sense or respond to emotions. While AI companions can provide responsiveness and personalization, they do not maintain a connection to these physical objects. The authors aim to explore how digital agents can inherit and extend the emotional bonds associated with physical companions, thereby enhancing user experience and emotional engagement.

Method
The authors propose the Dual-Embodiment Companion Framework, instantiated as Deco, which integrates multimodal Large Language Models (LLMs) and Augmented Reality (AR) to create synchronized digital embodiments of users’ physical companions. The framework is built on four design principles derived from a formative study with nine participants: Faithful Identity, Calibrated Agency, Ambient Presence, and Reciprocal Memory. Deco’s architecture leverages LLMs for natural language processing and AR for visual interaction, allowing users to engage with their physical objects in a digitally augmented manner. The study involved a within-subjects design with 25 participants comparing Deco against a personalized LLM-empowered digital companion baseline, focusing on perceived companionship and emotional bond metrics.

Results
Deco significantly outperformed the baseline across multiple metrics, including perceived companionship and emotional bond, with all results yielding p-values less than 0.01. In a subsequent seven-day field deployment involving 17 participants, Deco demonstrated sustained user engagement and a statistically significant improvement in subjective well-being (p=0.040). The study identified three key relational patterns: (1) digital activities revitalized the emotional significance of physical objects, (2) the depth of emotional engagement was a stronger driver of bond deepening than interaction frequency, and (3) users maintained their emotional connections while navigating the AI nature of their digital companions.

Limitations
The authors acknowledge several limitations, including the small sample sizes in both the formative study and the field deployment, which may affect the generalizability of the findings. Additionally, the reliance on self-reported measures for companionship and emotional bond may introduce bias. The study does not explore long-term effects beyond the seven-day deployment, nor does it address potential privacy concerns related to the use of LLMs and AR technologies.

Why it matters
This work has significant implications for the design of digital companions, suggesting a shift from creating entirely new relationships to a dual embodiment approach that enhances the emotional history of physical objects. By integrating AI with physical companions, this framework could lead to more meaningful interactions and improved emotional well-being for users. The findings may inform future research on human-AI interaction, emotional design, and the development of pervasive AI systems that resonate with users’ personal histories.

Authors: Zhihan Jiang, Mengyuan Millie Wu, Ruishi Zou, Shiyu Xu, Xun Qian, Emma Macmanus, Steven Liao, Ping Zhang et al.
Source: arXiv:2605.03882
URL: https://arxiv.org/abs/2605.03882v1