Anthropomorphism in Human-Robot Interactions with Adults with Intellectual Disabilities and Robots
Saminda Sundeepa Balasuriya, Laurianne Sitbon, Alicia Mitchell · 2026 · Journal of Human-Robot Interaction · doi:10.1145/3800572
Summary
This qualitative study re-analyses video data from two prior field studies with an Australian disability service organisation, in which adults with intellectual disabilities interacted with two commercially available social robots: Pepper, a 120 cm humanoid with articulated arms, expressive LED eyes, and a chest tablet; and Cozmo, a small toy-like robot with tracked wheels, an animated-eye screen, and programmable play modes. The Cozmo study spanned five weekly game-play sessions with six participants, while the Pepper study comprised four weekly sessions with fifteen participants engaging in activities such as joke-telling, dance, healthy-eating quizzes, and a guided exercise routine. The authors, drawing on 17 Cozmo videos and 10 Pepper videos, apply Braun and Clarke's inductive reflexive thematic analysis to investigate how participants anthropomorphise the robots and co-construct their identities, an emerging concept the authors frame as robo-identity. Their analytical lens draws on Kahn et al.'s six features of perceived human-likeness in HRI (autonomy, imitation, intrinsic moral value, moral accountability, privacy, and reciprocity), the uncanny valley effect, and interpretive-research principles emphasising transferability over generalisability. Rather than tallying incidents, the authors foreground rich examples of how anthropomorphic form and behaviour shape social expectations. They organise findings into three themes: physical interactions reflecting social norms (greetings, touch, eye contact, mirroring); verbal engagement and playfulness (introductions, encouragement, humour); and anthropomorphic perceptions and identity attributions (capability attribution, companionship, expectations, and gender).
Key findings
Participants readily treated both robots as social others, but each robot's form pulled interactions in different directions. Pepper's humanoid body elicited immediate human-style behaviour: attempted handshakes, self-introductions, permission-seeking before touching, sustained eye contact, and affectionate goodbyes (one participant kissed Pepper's hand). It also raised conversational expectations that Pepper could not meet — one prospective participant dropped out after realising Pepper was limited to pre-set phrases and tablet games, and others grew frustrated when it failed to reciprocate a handshake or answer a personal question. Cozmo, by contrast, was initially treated as a toy or pet, with greetings only emerging after researchers modelled them, but its name-recognition and game-winning animations produced delight precisely because they exceeded low expectations. Reciprocity and mirroring emerged as the strongest drivers of rapport: participants bowed when Pepper bowed, cheered when Cozmo won, and interpreted routine robot movements as intentional social gestures. Gender attribution was dynamic and sometimes contested (Pepper was variously called male or female; Cozmo was consistently read as male), with disagreements between peers occasionally dampening individual engagement. Empathy flowed towards both robots: participants warned Cozmo about table edges, asked whether it was hungry, and whispered while it 'slept'. Participants assigned the robots roles including friend, companion, learning facilitator, playmate, and potential hospital visitor. Compared to neurotypical users, the adults with intellectual disabilities in the study were less focused on robot utility and more invested in relational rapport, and they tended to explain technical malfunctions through human frames (tiredness, hearing difficulty) rather than updating their mental model.
Relevance
For practitioners designing assistive or companion technologies for adults with intellectual disabilities, the paper offers four directly usable design recommendations: align appearance with capability (so conversational promise is not made by a form that cannot deliver it); embed clear reciprocal signals such as gaze, name use, and mimickable gestures to initiate rapport; support multiple interaction modalities (speech, touch, tablet, gesture) rather than forcing a single channel; and design robots that can transition between roles (peer, tutor, collaborator) because participant identities for a robot shift with activity and context. The work also surfaces ethical tensions worth raising with stakeholders, including the risk that strongly anthropomorphised robots can mislead users into believing the system understands or cares, and the consent implications for a population that the authors treat with ongoing rather than one-off consent. Limitations include small sample sizes, two specific platforms (Pepper and Cozmo), no systematic age comparison, and a re-analysis of previously collected video rather than a purpose-built study, so readers should treat the findings as transferable insights into how anthropomorphism operates in this population rather than as prescriptive rules.
Tags: intellectual disability · human-robot interaction · social robots · anthropomorphism · inclusive design · qualitative research · thematic analysis · co-design