Robot Characters: Co-Designing Dynamic Personalities for Cognitively Assistive Robots
Dagoberto Cruz-Sandoval, Alyssa Kubota, Connie Guan, Soyon Kim, Laurel D. Riek · 2026 · ACM Transactions on Human-Robot Interaction · doi:10.1145/3778867
Summary
The paper introduces the concept of a "robot character" — a holistic, context-sensitive framing of robot personality that extends beyond trait-based models such as Big Five by incorporating shared interests, cultural background, lived experiences, and environmental factors. To operationalise this, the authors developed R2C2 (Robot Role Character Creation), a tangible, low-cognitive-load scaffolding tool consisting of a four-quadrant design board (appearance, communication, interests, attitudes) and 43 sticker prompts. R2C2 is designed to be accessible to participants with mild cognitive impairment (MCI) and dementia — populations typically excluded from robot design because standard personality models impose cognitive and conceptual demands these users may not meet. The research is grounded in the disability-justice concept of the "whole-self," resisting narratives that reduce people with dementia (PwD) or MCI (PwMCI) to a list of deficits. A cross-cultural co-design study was conducted across senior care centres in the US and Mexico with 13 participants: five older adults (three with MCI, two with early-stage dementia) and eight geriatric care professionals (GCPs). Workshops applied scaffolding strategies — giving time to talk, think-aloud, exemplification, and tapping into prior knowledge — to design robot characters for self-chosen application contexts. Reflexive thematic analysis of transcripts and design boards examined the envisioned roles, multidimensionality, and cultural influences on robot characters.
Key findings
R2C2 effectively scaffolded ideation for PwD and PwMCI; all participants actively engaged through physical manipulation of stickers and several created their own custom attitude or interest prompts. A clear tension emerged between end-user and caregiver perspectives: PwMCI/PwD consistently designed a "proactive partner" — a friend or helper that supports behavioural and mental well-being, aids learning, and respects their autonomy — while GCPs designed a "supportive care tool," a more passive robot focused on care tasks, reminders, safety monitoring, and socialisation. Participants preferred robots with complementary traits that strengthen qualities they felt they lacked (e.g., patience), similar traits mirroring their own positive qualities, and context-dependent traits that adapt to activity type (friendly for recreation, serious for learning). Cultural background shaped character design: US participants (both primarily English-speaking) preferred motivating, supportive robots for individual activities, while Latine participants in Mexico envisioned disciplined, straightforward robots that also promote family and community interaction. All participants agreed on multimodal natural communication (speech, vision, gesture, touch). Empathetic, curious, and friendly were the most-requested attitudes across groups.
Relevance
For accessibility practitioners working on assistive technologies for cognitive impairment, R2C2 offers a concrete, open-source co-design method that treats PwD and PwMCI as primary design partners rather than proxies-for. The tangible-materials and scaffolding approach translates directly to co-design of other assistive technologies — AAC devices, reminder apps, smart-home interfaces, and caregiving software — where cognitive accessibility of the design process itself is a common barrier. The paper's contrast between end-user and caregiver visions is a cautionary finding: when proxies stand in for people with cognitive impairments, resulting designs can systematically under-serve the users themselves, reinforcing technoableist framings of disability as a deficit to be compensated for. The cross-cultural US–Mexico analysis provides rare empirical evidence that personalisation must account for culture, not just individual traits. Limitations include a small sample (n=13), heavy Latine representation, and no validation of whether co-designed characters remain acceptable once implemented on a real robot.
Tags: cognitively assistive robots · robot personality · co-design · inclusive design · dementia · mild cognitive impairment · cross-cultural research · human-robot interaction · participatory design · disability justice