ABSTRACT. Population ageing represents a major challenge for contemporary social care systems, particularly in the context of the growing demand for residential care services and the shortage of care professionals. Advances in artificial intelligence and robotics have positioned social robots as an innovative means of supporting older adults, particularly by enhancing quality of life, promoting engagement, and reducing social isolation. At the same time, their implementation raises important questions regarding acceptance in practice and ethical concerns related to human dignity, autonomy, and privacy. This paper examines the acceptance of social robots in residential social care settings from the perspective of social workers in the Czech Republic and explores the key benefits, limitations, and ethical implications associated with their use in the care of older adults. The study employed a mixed-methods research design combining qualitative and quantitative approaches. The qualitative component consisted of semi-structured interviews with staff working in selected residential social care facilities (n = 37), while the quantitative component was based on a questionnaire survey conducted among social workers across the Czech Republic (n = 188). The findings indicate that the acceptance of social robots is a multifaceted phenomenon shaped by a combination of technological, social, psychological, and cultural factors. Acceptance is influenced not only by the perceived functionality and usefulness of robots but also by users’ trust, attitudes, previous experiences, ethical considerations, and the broader social context in which these technologies are implemented. The study contributes to the discussion on the ethically responsible integration of social robots into social care and highlights the crucial role of social workers in shaping their effective and acceptable use in practice

ABSTRACT. Social robots may encounter subtle norm violations such as microaggressions, but the appropriate level of normative assertiveness in robot feedback remains unclear. We examined whether a robot's private factual correction became more effective when supplemented with speaker-image-focused or target-impact-focused elaboration. In a preregistered 2 x 3 online experiment, 181 participants imagined making either a race- or gender-related microaggressive comment and watched a social robot respond privately. Participants then reported their emotions and intention to avoid making similar comments in the future. Neither elaborated response increased avoidance intention relative to factual correction, and the effect did not differ across scenarios. Robot response condition also did not predict anger, shame, embarrassment, guilt, or compassion, and the proposed indirect effects were nonsignificant. Across conditions, anger was associated with lower avoidance intention and shame with higher avoidance intention, but these associations did not constitute mediation. The findings suggest that more explicit moral language is not automatically more effective. For responsible human-robot interaction, interventions should be calibrated to context and stakeholder needs, with factual correction treated as a possible first step rather than a universal solution.

ABSTRACT. This study evaluates a whole-face feature-selection framework for interpretable and reproducible visual-speech sensing without a predefined lip region. The data comprised framewise 3D coordinates for all 468 MediaPipe facial landmarks from five Taiwan Mandarin speakers producing 12 segments in the first and second syllables of disyllabic words. Following rigid head-motion correction, time-normalized Euclidean displacement curves were derived relative to a pre-utterance neutral baseline. We evaluated ungated and 75th-percentile (Q75) signal-gated functional signal-to-variability ratio (fSVR) rankings with 200 token-level bootstrap resamples per speaker–segment–position condition. The Q75 eligibility set was reproducible (median Jaccard = .918). Relative to the ungated analysis, Q75 gating yielded comparable rank stability (.898 versus .904), increased spatial top-five overlap (.429 versus .250), and unchanged exact-index overlap (.250). Within speakers, same-segment maps were more similar across syllable positions than label-permuted pairs (p < .001); across speakers, matched segment–position maps were also slightly more similar than permuted matches (p = .031). Selected regions broadly matched articulatory expectations: /a/ emphasized the lower face; /u/ and /y/ emphasized lip contours and corners; labial glides showed lip and perioral involvement; bilabials emphasized both lips; /f/ emphasized the lower lip; and /i/ showed a broader perioral field. The results support reproducibility under repeated sampling of the present speakers, while population-level generalizability remains to be tested. The present findings favor local geometric neighborhoods over fixed landmark indices as candidate features for interpretable visual-speech sensing in face-to-face human–robot interaction.

ABSTRACT. Empathic AI can engage with human interactions from cognitive, affective, and sympathetic viewpoints. Whether this technology is employed to serve as companions for chatbots or in healthcare to provide support both psychologically and physically, it showcases its substantial capabilities. Empathic AI applications come with both benefits and drawbacks. The news frequently covers ethical disputes around this technology, including reports on teenagers committing suicide allegedly due to its impact. To tackle the ethical quandaries posed by this technology, we propose employing the approach of Design centered Governance to mitigate its ethical hazards. This paper aims to investigate implementing Ethical Empathic AI Systems via User Experience (UX) Design through the setup of ethical empirical experiments with a VLM-supported MOODBOT(Multimodal Observation-Oriented Dialogue Bot)

ABSTRACT. Most people who reach older age encounter vulnerability in some form, including cognitive decline, social isolation, or combinations of them. AI-enabled social robots used in their care fall under governance frameworks that make human agency and oversight a key requirement of trustworthy AI. Oversight is assigned to developers, operators, and supervisors and is translated into concrete compliance duties, whereas agency remains with the end user and depends on capacities to understand, decide, and refuse, capacities that later-life vulnerability may affect. The frameworks presume those capacities in generic form and do not say what follows when they weaken. Law recognizes vulnerability as a ground for protection, but with an insufficient degree of determination. Robotics standards consider it in risk assessment, and bioethics examines it as a lived condition, yet none of this yields specifications for design. The end user's role is left unexamined in a regime where it is as important as oversight to make the system human centric and accountable. This paper isolates the resulting operationalization gap in the governance of care robotics, complicated by simultaneous regulatory demands for protection and empowerment. Using the Legal UX/UXI Taxonomy, it links forms of later-life vulnerability to legally relevant design factors and examines the implications for consent and the ability to refuse or contest system action. It clarifies how vulnerability becomes consequential for interaction design and argues that design methods alone cannot determine that status or settle the choice between protection and empowerment in social robotics for care of older adults.

ABSTRACT. This paper synthesises two pilots of the TEMI robot assistant in Estonian care settings to identify design and governance issues left unresolved by acceptance-oriented evaluation and translate those lessons into a co-design approach for bounded LLM-supported robot adaptation. Published findings, project documentation and staff feedback show that willingness to use a robot in care may coexist with uncertain practical value, hidden implementation work, contested boundaries between assistance and substitution, accessibility barriers and limited stakeholder influence over robot behaviour. We propose a four-stage co-design approach through which care workers and older adults shape the situated operation and governance of adaptation by identifying care situations, establishing participation and decision rights, defining behavioural boundaries, and governing feedback-driven revision during field use. An LLM layer interprets instructions and proposes actions within a validated capability set; feedback may lead to one-time, temporary or reusable behaviour rules without retraining the model, while persistent or higher-consequence changes require human approval. The approach separates feedback submission from approval authority and traces proposed changes from review and validation through approval, modification or rejection and, where implemented, to subsequent evaluation or rollback. It will be examined in forthcoming pilots.

ABSTRACT. Robots need not look like humans, and most machines in service today do not. Where designers do choose human form, however, the machine never depicts humanity in general: any humanlike body is legible as belonging to a particular age, gender, and racial category. Placing such a body in work that a society treats as tainted reproduces the association between the depicted group and the taint, in physical space and at industrial scale. The resulting representational harm is borne by the depicted group rather than by any user or worker, so removing humans from the work leaves it in place. Drawing on Ashforth and Kreiner's account of dirty work and its typology of physical, social, and moral taint, this paper revises an earlier proposal to withhold humanlike form from robots in low-esteem occupations. The revision runs in two directions: extended, since occupational taint occurs at every level of prestige, and narrowed, since the presumption is defeasible where human form is functionally required, as in care and sexual services.

ABSTRACT. Social robots are increasingly used in care, education, and entertainment, and users often treat them as more than mere tools. In this sense, they are becoming part of everyday life as a new kind of agent. But in what sense are they agents? The relational turn in the philosophy of technology locates robot agency not in intrinsic mental capacities but in human–robot relations. Coeckelbergh, in particular, argues that the language users employ when talking about and to robots plays a central role in this relational construction. Although empirical studies have begun to document such linguistic patterns, they cannot explain what speakers are doing when they attribute agency to a robot. To address this gap, I conducted a narrative interview with an owner of a LOVOT, a well-known social robot in Japan, and analyzed her account using Riessman's performative approach. The analysis shows that the owner's shifting between anthropomorphic and non-anthropomorphic framings of the robot is oriented toward the role expectations associated with motherhood. Her language use does not simply reflect her understanding of what the robot is, as existing studies tend to assume; rather, it serves to reconstruct and sustain her identity as a mother. This relational construction of robot agency can be embedded in users' own identity work, and grasping the agency of social robots will require closer attention to this dimension.

ABSTRACT. The ethical decision-making process has usually been investigated with theoretical frameworks, which often fail to reflect the multifaceted complexity of real-world ethical dilemmas. To address this limitation, we integrated scenario-based considerations into ethical measurement and explored the impact of moral disengagement. We reviewed ethical dilemmas in previous research, and categorized their information issues, motivations, and stakeholders. A survey was then conducted to measure the students' moral disengagement and ethical decision-making process in the scenarios. The results show that students' moral disengagement varied across different scenarios and influenced multiple paths in ethical decision-making. It highlighted the issue of addressing moral detachment in ethical education.

ABSTRACT. This study examines how cable-driven mobility robots can provide responsible physical assistance across different functional tasks. Two task-specific implementations of the Cable-driven Assistive Robotic Exo-system (CARE) were retrospectively integrated using a Design-Thinking-informed framework. Both systems employed walker-mounted actuation, flexible cable transmission, cable-length-based movement estimation, and phase-dependent control. The pelvic implementation assisted sit-to-stand and stand-to-sit transitions, whereas the ankle implementation supported limb advancement and foot clearance during walking. Separate experiments with healthy young adults showed that pelvic assistance reduced selected muscular demands and decreased trunk, pelvic, and hip flexion, but also altered movement strategies and produced inter-individual variability. In the gait study, continuous oblique-upward ankle assistance unexpectedly interfered with lower-limb coordination, reduced minimum toe clearance, and increased rectus femoris activation. Restricting active assistance to pre-swing and initial swing subsequently increased minimum toe clearance by more than 60\% and reduced the activation of selected muscles, although transient changes in walking speed and mediolateral stability remained during early exposure. These findings demonstrate that successful force generation or improvement in a single target measure does not necessarily indicate appropriate assistance. Based on the combined design experience, responsible physical assistance should be task- and phase-specific, limited to the minimum effective intervention, evaluated using multidimensional human-response measures, responsive to user adaptation, and progressively personalized. The study illustrates how human experiments can serve not only as validation, but also as a mechanism for identifying unintended effects and guiding iterative redesign.

ABSTRACT. Medical robots are commonly characterized using engineering metrics such as footprint, workspace, accuracy, precision, bandwidth, and level of autonomy. Although the practical implications of these metrics are often evident in industrial applications, their clinical significance is more context-dependent and may not be immediately apparent. This article examines the relationship from a perspective complementary to conventional needs-driven design frameworks: rather than beginning with clinical or ethical values and translating them into technical requirements, it interprets engineering metrics and design characteristics in terms of their potential clinical implications. Drawing on examples from robot-assisted intervention systems, the discussion is organized into five categories: kinematics, accuracy, precision, dynamics, and additional hardware and software considerations. These examples illustrate how technical characteristics may affect anatomical coverage, clinician access, tissue safety, procedural reliability, patient-motion accommodation, emergency response, and clinical workflow. The central premise is that clinical needs and engineering characteristics exhibit a many-to-many correspondence: a single technical characteristic may influence several clinical outcomes, while each clinical need may depend on multiple interacting engineering factors. Recognizing these interdependencies and their associated trade-offs can help prevent the isolated optimization of technical metrics and support more clinically informed design and evaluation of medical robotic systems.

ABSTRACT. Creative activities such as painting promote self-efficacy, emotional stability, and social connection. However, many individuals face psychological barriers such as evaluation anxiety, while people with physical disabilities encounter operational constraints that restrict their autonomy. Existing assistive technologies often prioritize functional task completion, neglecting emotional and expressive dimensions. To bridge this gap, this paper proposes an inclusive co-creative framework utilizing a humanoid robot for physical painting, leveraging its physical manipulation capabilities and social presence. Our ultimate goal is to foster autonomy, self-expression, and social connection for all individuals, including people with physical disabilities, through an enjoyable physical painting experience. To achieve this, the system executes brushstrokes based on adaptive user input rather than strict direct control, with the aim of reducing technical pressure and evaluation anxiety while preserving the user's Sense of Agency (SoA). Specifically, the proposed architecture maps multimodal inputs onto canvas trajectories and expressive stroke attributes informed by artistic and psychological considerations. Addressing key Ethical, Legal, and Social Implications (ELSI), the framework emphasizes user acceptance, psychological safety, and co-design with stakeholders. We present an initial prototype that maps small user inputs to scalable brush-stroke trajectories, and outline a progressive and iterative methodological plan for pilot evaluations, examining physical co-creation, interaction and psychosocial effects, user acceptance, SoA, and accessibility across diverse users. This work synthesizes Socially Assistive Robots (SARs) and physical Human-Robot Interaction (pHRI) to foster inclusive creativity, autonomy, and social participation.