EmoFriends: A Modular Toolkit for Prototyping Customizable Emotion-Aware Companions

This work presents EmoFriends, a modular prototyping toolkit designed to lower the barrier for creating customizable, emotion-aware companions from familiar physical artifacts. The toolkit integrates three interchangeable modules: (1) a touch-based sensing unit for capturing affective interaction proxies, (2) an LLM-powered engine for empathetic dialogue, and (3) an embodied feedback module providing haptic and visual cues. By modularizing these components, EmoFriends supports the rapid exploration of diverse companion configurations and foregrounds technological transparency as a design resource for embodied AI. We evaluate the system through technical feasibility metrics (e.g., actuation latency and classification robustness) and an exploratory “first-contact” deployment at a public exhibition (N = 45). Our findings demonstrate the toolkit’s functional reliability in high-traffic settings and reveal a strong user valuation of customization. However, the study also highlights critical social tensions regarding sensing credibility, privacy, and the social appropriateness of toy-like form factors. We conclude by outlining how modular frameworks can advance the design and research of future embodied affective companions.