Robot-Assisted Social Dining as a White Glove Service
Atharva S Kashyap, Ugne Aleksandra Morkute, Patricia Alves-Oliveira · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3790481
Summary
This CHI 2026 paper addresses a gap in the robot-assisted feeding literature: existing systems are almost exclusively evaluated in the lab or the home for solitary dining, leaving public, social contexts such as restaurants largely unstudied. The authors argue that social dining is central to how people with disabilities (PwD) maintain relationships, and that robots intended for those contexts must handle dynamic, unsupervised environments with additional participants (dining companions, waitstaff, caregivers). The study uses speculative participatory design with six PwD who require upper-limb assistance to eat (self-described disabilities include quadriplegia, osteoarthritis and shoulder injury, spinal muscular atrophy, Parkinson's with partial paralysis, intention tremors, and multiple sclerosis). Each participant attended a 75-90 minute session combining avatar creation, six-panel storyboarding of an ideal dining scenario, drilling into nine key interaction moments, imagined robot character design, and a semi-structured exit interview. A custom tool, Speak2Scene, let participants generate speculative images via voice-driven prompts to OpenAI gpt-image-1 through a Vision Language Model pipeline. Transcripts were analyzed using reflexive thematic analysis (Braun and Clarke) with recruitment guided by the information power principle rather than saturation. Four themes structure the results: interaction ecology (user-to-robot, robot-to-user, robot-to-others, others-to-robot channels), context-sensitive robot behavior, robot role during and outside of mealtime, and perceived user relationship with the robot. The authors frame their synthesis using the hospitality metaphor of "white glove service."
Key findings
Participants envisioned robots that support multimodal input (haptics, buttons, finger pads, eye gaze, head movement, voice, tablets, wands) chosen dynamically based on fatigue and body state, paired with minimal, unobtrusive output (lights, beeps, haptics, brief screen cues) rather than verbose speech. A strong finding across participants was the need for pre-action signaling — a light or sound before the robot moves — as a trust-building mechanism that prevents surprise. For social behavior, participants wanted robots to "fade into the background," support but not initiate shared table practices (passing, toasting), adapt personality to the user, maintain awareness of surroundings, and periodically assess the user without interrupting conversation. They explicitly did not want robots to infer or expose disability information. Robot roles extended well beyond feeding: self-serving from shared dishes, menu reading and ordering, navigation (pulling out chairs, path-clearing to restroom/exit), and payment handling (retrieving cards, opening folders, pulling receipts). Participants consistently framed caregiver-robot teaming — caregivers correcting or redirecting robot errors — rather than robots replacing caregivers. Design preferences around embodiment (color, size, personality) varied widely, underscoring that one-size-fits-all is inadequate. The authors condense findings into five white-glove design implications: anticipatory assistance, discretion and privacy, attention to detail, personalization, and seamless problem resolution.
Relevance
For practitioners building assistive robots, AAC systems, or any public-facing AT, this paper offers a sharp reminder that social dignity — not feeding mechanics — drives adoption. The white-glove framing gives designers a concrete vocabulary for trade-offs between responsiveness and discretion, and the emphasis on pre-action signaling and symptom-level (not diagnostic) sensing has direct implications for privacy-preserving sensor design. The multimodal, switchable-input finding aligns with broader ability-based design principles and applies beyond robots to any interactive product used in mixed social settings. Limitations worth flagging: six participants, all from the US, mostly white and female, and only PwD were consulted — caregivers, waitstaff, and dining companions were not interviewed, which matters because the white-glove metaphor depends on orchestrating all of them. Findings are speculative: no working robot was deployed. GenAI storyboarding occasionally misrepresented participants' disabilities, and one participant with visual impairment could not fully engage with image outputs, a useful cautionary note for teams using GenAI in accessibility research.
Tags: assistive robotics · robot-assisted feeding · speculative design · participatory design · generative AI · motor disability · social dining · qualitative research