Designing Socially Assistive Robots for Perinatal Depression Screening: Insights and Ethical Considerations from Two Exploratory Studies
Mengyu Zhong, Lux Miranda, Fotios C. Papadopoulos, Katie Winkle, Alkistis Skalkidou, Ginevra Castellano · 2026 · ACM Transactions on Human-Robot Interaction · doi:10.1145/3797257
Summary
This paper reports two exploratory studies investigating the use of Socially Assistive Robots (SARs) for Perinatal Depression (PND) screening. PND affects up to 10% of individuals during pregnancy or postpartum, yet 50% of antenatal depression cases and 69% of PND cases go undiagnosed. The authors' prior work examined stakeholders' perspectives through interviews with psychiatrists, PND experts, and gender scholars; this article extends that by directly involving primary users—women with lived experience of PND. The research comprises two studies. First, a participatory design study with seven Swedish women with previous PND experience, using semi-structured interviews and storyboard-based design exercises with a Furhat robot to explore use cases, design factors, and acceptance. Second, a user study with five women who had previously screened positive on the Edinburgh Postnatal Depression Scale (EPDS), evaluating two interaction contexts: (1) an autonomous SAR administering the EPDS questionnaire, and (2) a SAR remotely operated by a psychiatrist administering the MINI International Neuropsychiatric Interview. Both studies used hybrid thematic analysis in NVivo 14, combining inductive and deductive coding. The robot prototype, built on the Furhat platform, featured a gender-ambiguous default appearance with personalisation options across eight face presets varying in ethnicity, gender, and level of anthropomorphism. This hybrid methodological framework—combining participatory design with prototype-based in-person evaluation—is particularly suited to investigating sensitive topics with vulnerable populations, providing both conceptual depth and ecological validity.
Key findings
Most participants in both studies were receptive to robot-administered PND screening. Four of five user study participants were open to future robot-delivered EPDS, valuing the robot's neutrality, non-judgement, and the reduced sense of human evaluation compared to clinician-administered screening. The robot-administered EPDS was generally rated superior to digital self-administration for verbal engagement and perceived information completeness. Questionnaire results showed strong willingness to follow the robot's suggestions (M = 4/4) but only moderate willingness to recommend the robot to others (M = 2.4/4) and moderate comfort sharing personal data (M = 3/4). Acceptance was not universal: some participants strongly preferred human interaction, citing emotional needs and data security concerns. Transparency and human oversight emerged as critical requirements. The tele-operated MINI condition created confusion about robot agency, with participants uncertain whether they were engaging with the robot or the psychiatrist. This highlights the need for explicit role transparency mechanisms in human-robot teaming. Moderate anthropomorphism was preferred; excessive human-likeness risked setting unmet emotional expectations and eroding trust. Privacy, informed consent, data minimisation, and immediate escalation pathways to human clinicians were identified as essential ethical safeguards, particularly given the vulnerability of the target population.
Relevance
This paper is directly relevant to the intersection of assistive technology, mental health, and vulnerable populations. PND intersects with accessibility in that affected individuals face barriers to help-seeking—stigma, emotional burden, systemic gaps—that technology can either reduce or inadvertently exacerbate. The paper's ethical framework is particularly instructive: the authors move beyond checklist-style AI ethics to argue for relational, sociotechnical accountability that considers distributed responsibility across users, clinicians, and robot systems. The participatory design methodology demonstrates how to ethically involve vulnerable users in technology co-design—transferable to any assistive technology work with people with mental health conditions, cognitive impairments, or other sensitive characteristics. Four design implications have broad applicability to any conversational AI or SAR in healthcare: calibrated anthropomorphism, explicit role and agency clarity, human complementarity (not replacement), and transparent data handling aligned with GDPR and trustworthy AI principles.
Tags: socially assistive robots · mental health · perinatal depression · participatory design · human-robot interaction · ethics · healthcare technology · vulnerable populations · trustworthy AI
Standards referenced: GDPR · EU Ethics Guidelines for Trustworthy AI · BS 8611