Keywords: Linguistic Communication and Dialogue, Multimodal Situation Awareness and Spatial Cognition, Multimodal Interaction and Conversational Skills
Abstract: Causal reasoning plays a key role in human cognition. As such, we hypothesize that robots provided with causal knowledge of task-relevant objects may provide human interactants with more natural and effective cognitive assistance during troubleshooting tasks. To assess this hypothesis, we present CITA: a novel causally-informed approach to robot-assisted troubleshooting. Through objective analysis and human-subjects experimentation, we demonstrate that causally-informed robot-assisted troubleshooting is not only objectively more efficient, but moreover is perceived as being of higher quality and more intelligent. Overall, our results suggest that robots designed to provide cognitive assistance to humans in task-oriented human-robot interactions should be provided with causal models of the objects involved in those interactions

Keywords: Theory of Mind in HRI, Human Factors and Ergonomics, Cooperative Intelligence in HRI
Abstract: Interactive robot learners often exhibit learning behaviors that are not intuitively interpretable by human observers, which can result in suboptimal feedback in collaborative teaching settings. Yet, how humans perceive and interpret robot learning behavior is largely unknown. In a bottom-up approach with two experiments, this work provides a data-driven overview of the factors influencing human observers' understanding of the robot learning process. A novel, observation-based paradigm to directly assess human inferences about robot learning was developed. In an exploratory interview study (N=9), four common themes in human inferences were identified: Agent Goals, Knowledge, Decision Making, and Learning Mechanisms. A second confirmatory study (N=34) applied an expanded version of the paradigm across two tasks (navigation/manipulation) and two Reinforcement Learning (RL) algorithms (tabular/function approximation). Analyses of 816 responses confirmed the reliability of the paradigm and refined the thematic framework, revealing how these themes evolve over time and interrelate. Our findings provide a human-centered understanding of how people make sense of robot learning, offering actionable insights for designing interpretable RL systems and improving transparency in Human-Robot Interaction.

Keywords: Theory of Mind in HRI, Explainable Human-Robot Interaction, Cooperation and Collaboration in Human-Robot Teams
Abstract: Successful Human-Robot Teaching assumes alignment between robot processing needs and human teaching intent. To better understand this alignment, this work seeks to uncover the underlying logic that humans intuitively apply when teaching. Through an exploratory, bottom-up study with N=34, participants observing two distinct robot Reinforcement Learning (RL) scenarios, we analyze 204 intuitive teaching responses across early, middle, and late learning phases. Results reveal that teaching decisions consist of a nuanced, interconnected network of Triggers (situational catalysts), Objectives (subjective teaching targets), Signals (communicative acts), and Strategies (high-level governance) in which teachers spontaneously adopt diverse roles, acting as coaches, engineers, or designers and prioritize different objectives. Based on these results, we introduce the TOSS Framework, which conceptualizes Human-Robot teaching as a procedural loop between robot behavior and human teaching actions, in which human teaching decisions are modeled as Trigger-Signal responses modulated by teaching Objectives and Strategies. It provides future research with an openly accessible dataset and a theoretical foundation for a) understanding teaching decisions and b) simulating realistic oracles as well as c) designing human-centered teaching settings and novel robot learning algorithms that go beyond the constraints of current robot learning settings.

Keywords: Machine Learning and Adaptation, Social Learning and Skill Acquisition Via Teaching and Imitation, Evaluation Methods
Abstract: Human-Interactive Robot Learning (HIRL) often relies on idealized oracle teachers, despite evidence that real human teaching behavior is variable and imperfect. This paper investigates whether human teaching behavior can be modeled from interaction data and reused as a realistic simulated teacher for robot learning. Using data from a prior human- robot interaction study, we train classifiers to predict when a human teacher intervenes and whether they provide evaluative feedback or demonstrations. We compare a flat multi-class classifier with a layered modeling approach that separates intervention prediction from intervention type prediction. The layered model achieves the best predictive performance under severe class imbalance and offers improved interpretability. When integrated into a Panda-gym environment, the simu- lated teacher successfully guides learning across three tasks. Although it provides fewer interventions and less variability, learning efficiency in successful sessions remains comparable to human teachers. These results show that human teaching behavior can be modeled to create realistic simulated teachers that enable scalable and repeatable evaluation of interactive learning systems.

Keywords: Social Learning and Skill Acquisition Via Teaching and Imitation, User-centered Design of Robots, Programming by Demonstration
Abstract: Human-robot teaching focuses on enabling non-technical experts to customize robots according to their needs after deployment. With recent advances in machine learning, human-robot teaching is no longer confined to offline learning where the data gathering step from a human teacher is separated from when the robot learns. Instead, more recent approaches for human-robot teaching focus on coupling human teaching with robot learning. This coupling impacts the structure, timing, and content of the teaching and learning interaction. However, it is currently unclear how such coupling dynamics affect human-robot teaching effectiveness and human perceptions towards the teaching process. Informed by human learning theories, in this paper we propose a new scale for classifying human-robot teaching interactions according to coupling dynamics present between the human teacher and robot learner. We apply this scale to a subset of the human- robot teaching literature to identify how coupling dynamics and human teacher mental model mismatches with the ground truth robot learning system affect teaching effectiveness and human perceptions towards the teaching process.

Keywords: Theory of Mind in HRI, Cooperation and Collaboration in Human-Robot Teams, Machine Learning and Adaptation
Abstract: In real-world environments where many robots and humans coexist, it is important to coordinate robots while understanding the intentions and traits of others to optimize the entire space. In this study, we propose Trait-Magnitude-Estimation Theory of Mind (TME-ToM), which estimates the sign and magnitude of another agent's trait vector as well as its next action, as a step toward whole-space optimization. Experimental results in an Overcooked environment with 76 macro-actions show that TME-ToM achieved a sign agreement rate of 87.4% between the estimated and ground-truth trait vectors, a magnitude error of 1.09, and a next-action prediction accuracy of 83.8% using trajectories from six episodes. We also confirmed that using the estimates for cooperation improves the delivery score from about 165 to 184 and the total individual score from about 155 to 226. Furthermore, we observed improvement in cooperative performance even for unseen agents and in zero-shot settings, demonstrating that the proposed method can improve cooperative performance in complex macro-action environments.

Keywords: Cooperation and Collaboration in Human-Robot Teams, Robot Companions and Social Robots
Abstract: Social continuity in heterogeneous multi-robot systems is often constrained by fragmented interaction contexts across decentralized agents. This paper proposes a role-adaptive communication framework driven by a Centralized Scenario Orchestrator. By centralizing the “Social Brain,” the system enables robots with diverse physical embodiments to adopt specialized personas while ensuring seamless contextual inheritance during task hand-offs. We introduce a multi-layered memory architecture that integrates temporal buffers and episodic Retrieval-Augmented Generation, with access filtered by Role-Based Access Control to preserve user privacy. This architecture enables reliable multi-robot workflows in settings such as reception, healthcare support, and retail assistance, where users interact with multiple platforms over time. Results indicate higher perceived social presence, clearer task delegation, and greater trust in the orchestrated scenario, highlighting the importance of centralized governance for role-based social behavior.

Keywords: Social Intelligence for Robots, Cooperative Intelligence in HRI
Abstract: Deploying large Vision-Language Models (VLMs) on human-care service robots is limited by the cost of processing video at full frame rate, even though most observation time is uneventful. We propose Dynamic Keyframe Sampling (DKS), a low-overhead, always-on front-end that gates VLM inference: it scores visual change with pixel-wise mean absolute difference and, when a person is present, refines the decision with skeleton-based relative motion, then adaptively contracts or expands the sampling interval. Using a Qwen3-VL-8B backbone, DKS cuts the average input frames by roughly 73–78% relative to dense or fixed-rate sampling while preserving accuracy, which sharply raises Intelligence Density (ID), our accuracy-per-frame efficiency measure (e.g., on MSR-VTT-QA, ID rises from 0.44 to 1.97 with under a one-point accuracy change). Beyond a single operating point, we run a controlled comparison against uniform and learned-feature samplers at matched frame budgets—an analysis rarely made explicit in prior work. It reveals that on standard trimmed benchmarks the benefit of DKS comes from adaptive budget reduction rather than from selecting better frames than uniform thinning, and that LoRA adaptation and DKS are complementary. DKS thus offers a practical, architecture-agnostic way to focus VLM computation on informative moments under tight on-robot compute budgets.

Keywords: Applications of Social Robots, User-centered Design of Robots
Abstract: Users often need to barge-in while a robot is speaking, particularly when the information being delivered is no longer relevant or desired. Existing research on barge-in interactions has primarily focused on speech-based control, while users may naturally expect to intervene through a variety of verbal and non-verbal modalities. To better understand user expectations of different modalities in barge-in interactions, we conducted a 3 (signal modality: verbal, paralinguistic, and gesture) × 2 (robot responsiveness: responsive vs. non-responsive) within-participants study (N = 87). Participants viewed video stimuli depicting a barge-in scenario in which the robot either stopped or continued its utterance. Results revealed that paralinguistic signals were less likely to be interpreted as command cues during barge-in interactions. Furthermore, when the robot ignored user signals, verbal commands led to the sharpest decline in robot evaluations, whereas gesture-based interruptions were less affected, indicating that gestures may mitigate the negative impact of robot non-responsiveness. These findings highlight the value of supporting various modalities in barge-in contexts and provide design implications for developing more natural and adaptive robot behaviors.

Keywords: LLM/Gen AI-based HRI, Social Intelligence for Robots, Detecting and Understanding Human Activity
Abstract: In exhibition halls such as science museums, a guide robot needs to decide whom to approach among multiple visitors who differ in group composition, spatial behavior, and attentiveness. This assessment requires integrating social cues such as body orientation, gaze direction, group dynamics, and ongoing activities, yet it is still commonly performed through labor-intensive manual video coding. We introduce the Visitor Selection Framework (VSF), a video-language pipeline that performs visitor-wise approach receptiveness estimation: for each visible visitor in a video clip’s decision frame, VSF predicts a 1–5 approach receptiveness score together with a cue-grounded textual rationale, yielding a ranked candidate list for robot approach decision support. The framework fuses multiple spatial attributes through structured visual overlays and a carefully designed text prompt. We validate VSF on 176 video clips collected from a real exhibition hall environment at the RoboLife Museum, containing 440 human-labeled visitors. Inter-coder agreement between Coder 1 and Coder 2 was 80.7% (QWK = 0.907), confirming a reliable annotation ceiling. We benchmark three VLMs and find that while exact 5-class label imitation remains modest (QWK ≤ 0.339), the models are far more effective as decision-support rankers: the coder-preferred target appears within each model’s top-2 shortlist in 92.0–95.4% of clips, with NDCG reaching 0.950–0.965. These results establish VSF as a practical decision-support layer for exhibition hall robots, providing reliable first-pass ranking for annotation support and approach-candidate prioritization.

Keywords: Robot Companions and Social Robots, User-centered Design of Robots, Methodologies in Social Robotics
Abstract: This paper examines how young South Koreans imagine domestic human-robot futures in the context of rapid technological development and evident demographic shifts. Nineteen participants engaged in two participatory design workshops, built around a “what-if" scenario designed to foreground social, ethical, and relational issues, including shifting notions of kinship, labor, and emotional care. Through brainstorming, design canvases, and co-creation of a Future Robot Home Catalog, participants produced speculative visions of everyday robot cohabitation. The resulting catalog pages revealed ambivalent imaginaries and highlighted tensions around productivity, intimacy, and shifting family norms, reflecting both desires for support and anxieties about cultural continuity and relational boundaries. Our work offers a case study in capturing culturally situated sociotechnical imaginaries and derives HRI design implications for more desirable human-robot futures.

Keywords: Non-verbal Cues and Expressiveness, Sound design for robots
Abstract: Robot motion sounds can shape users’ impressions of a robot, yet their influence on users’ perceived action duration remains underexplored in human-robot interaction. We conducted an exploratory within-subject study using 11 video stimuli: one silent baseline and ten sounds contrasting pitch, amplitude envelope, rhythm, distortion, and harmonic content. These dimensions were selected to represent motor-like acoustic variations in robot motion sound under controlled conditions. After each stimulus, participants reported subjective judgments of the movement duration and affective impressions of the robot. Sound significantly affected duration judgments, with overall underestimation across conditions. The results show that silence produced the greatest underestimation, whereas the rhythm-forward condition produced the smallest. Sound also influenced affective impressions, with silence rated calmer and more pleasant, and rhythm-related and rougher conditions eliciting more negative reactions and increased surprise. Overall, the results suggest that robot motion sound may influence users’ time perception and affective impressions of the robot, motivating further interactive studies of sound design for robot motion.

Keywords: Robots in Education, Therapy and Rehabilitation, Robot Companions and Social Robots
Abstract: Critical reading requires understanding textual content, organizing and articulating reasoning about arguments and evidence. While prior work has explored screen-based tools to support reading, less is known about how embodied systems may shape higher-order cognitive processes during academic reading. In this paper, we investigate how a non-humanoid desktop robot can support critical academic reading. To explore this question, we developed LUMOS, a robotic reading companion that provides ambient cues for attention regulation and dialogic prompts for reflection. Through a within-subject study comparing robot-assisted and independent reading, we found that robot-mediated scaffolding is associated with improvements in comprehension accuracy and the quality of critical writing. Notably, while participants did not generate more viewpoints, their arguments became more structured and coherent. These findings suggest that robotic companions act as interactional scaffolds that shape how readers structure and externalize their reasoning, rather than increasing the number of ideas they produce. Furthermore, qualitative findings suggest that such benefits can emerge from subtle, non-intrusive interaction, without requiring strong social presence.

Keywords: Robots in Education, Therapy and Rehabilitation, Assistive Robotics, User-centered Design of Robots
Abstract: In this work, we evaluate the feasibility of socially assistive virtual agent-based cognitive training for people with intellectual disabilities (ID) in a sheltered workshop. The Robo- Camp system, originally developed for children with Attention Deficit Hyperactivity Disorder (ADHD), is adapted based on the results of a pilot study in which we identified barriers and collected feedback from workshop staff. In a subsequent study, we investigate the aspects of usability, technical relia- bility, attention training capabilities and novelty effect in the feasibility of integrating the RoboCamp system. With this work, we contribute to the understanding of human-agent interaction within assistive contexts by providing a) assessment of the systems’ real-world applicability and b) insights into the target groups manners of interaction with a virtual agent.

Keywords: Robots in Education, Therapy and Rehabilitation, Robot Companions and Social Robots, User-centered Design of Robots
Abstract: Recent research on deploying robots in higher education has explored how social assistive robots (SARs) can support students across different learning activities and educational settings. However, many studies have not examined whether these deployments align with student preferences, leading to robotic systems being perceived as irrelevant or impractical and with limited effectiveness in supporting learning. This study aimed to investigate the preferences and perceived usefulness of SARs in higher education learning environments using a two-phase approach. In Phase 1, 30 participants were surveyed to identify learning contexts in which they would find a SAR useful (e.g., exam preparation, independent study, or time management), the types of support they would value, and their preferences regarding emotional support, adaptation and personalisation, and privacy. Insights from this informed the design of the instrument for the second phase (Phase 2), which further explored students’ learning practices, their perceptions of appropriate roles for robotic companions in different learning contexts, and the conditions under which they would accept robot-supported learning. A total of 52 participants from diverse academic backgrounds (e.g., STEM, humanities, arts, law, health, and business) were surveyed in Phase 2. Our findings indicate that SARs are preferred as learning companions rather than as emotional support companions, highlighting a potential mismatch between existing educational robotics research and students' actual support preferences. Robotic companionship was highly rated in subject-specific learning and self-regulated learning contexts, but rated low in social interactions and creativity-based learning.

Keywords: Child-Robot Interaction, Detecting and Understanding Human Activity, Social Intelligence for Robots
Abstract: Social robots deployed in children’s group activities inevitably encounter moments where interaction becomes locally problematic. These moments are often treated as system errors to be prevented, yet they are a normal feature of social interaction. Drawing on conversation-analytic video analysis of robot-mediated photo-sharing activities with children, this paper examines how misalignment unfolds in real time and how it is managed by participants. First, we identify a recurrent interactional trajectory in which a robot action produces local trouble that children detect and manage through social and embodied repair practices, while the robot does not participate in repair and may persist in the same course of action. Second, we develop an interactional model that captures this trajectory and specifies where repair-sensitive robot participation becomes relevant. The model reframes robot agency in terms of monitoring projected outcomes, recognizing participants’ orientations to trouble, and engaging in contingent repair. This perspective shifts human–robot interaction design from preventing misalignment toward enabling robots to participate as accountable agents in interactional repair.

Keywords: Applications of Social Robots, Robots in Education, Therapy and Rehabilitation, Robot Companions and Social Robots
Abstract: Effective caregiver education is essential for successful pediatric asthma management, especially during hospital discharge. To address this need, we conducted a study in an outpatient clinic involving caregivers of pediatric asthma patients. The goal was to assess caregivers’ perceived usability, engagement, and satisfaction with a human-support robot providing asthma education. The robot delivered validated educational content through a structured, interactive dialogue, and 13 caregivers and 16 healthcare professionals were subsequently asked to evaluate their experience. The majority of participants reported positive perceptions of the robot, citing high levels of usability, engagement, and satisfaction with the educational format. Caregivers also emphasized that the robot-supported interaction facilitated their understanding of asthma management strategies. Furthermore, participants provided valuable feedback regarding potential enhancements. These findings underscore the potential of socially assistive robots as a complementary tool for pediatric asthma education and highlight the importance of incorporating caregiver perspectives in the iterative design and refinement of such systems.

Keywords: Evaluation Methods, Applications of Social Robots, Degrees of Autonomy and Teleoperation
Abstract: Interactive systems are increasingly studied across research domains due to their growing importance in real-world settings, yet their capabilities are often described only at a high level, hindering systematic comparison and assessment across contexts. Existing evaluation approaches are based on subjective assessments or are too vague or broad to enable meaningful comparison.

We present a structured complexity framework that provides reproducible complexity assessments for hardware and software. The questionnaire awards up to 50 points: 25 for hardware (degrees of freedom, computing power, sensor diversity, hardware safety, and interfaces) and 25 for software (decision autonomy, adaptability to user behavior, modalities, accessibility and usability, as well as personalization capabilities). It enables consistent classification of system capabilities and transparent comparison between different platforms.

We conducted a survey of interactive systems used in the HRI community, collecting 28 system assessments across 25 distinct platforms. The results provide initial evidence for the questionnaire’s reliability, content and construct validity, as well as its usability and coherence, based on participant-reported system assessments, usability feedback, and analysis of the collected data.

Keywords: Evaluation Methods, Cooperative Intelligence in HRI
Abstract: Effective human-robot collaboration depends on seamless handover interactions, where fluency and intuitiveness are crucial for operational efficiency and user satisfaction. Despite the significance of these interactions, quantifying the quality of nuanced handovers remains a challenge. This paper introduces a novel data-driven framework designed to bridge the gap between objective robotic metrics and subjective human experiences during handovers. Our research establishes a minimal set of objective measures—contact forces, end-effector speed, and handover distance—correlated with key subjective human factors: timing, pose, speed, safety, awareness, and perceived human-likeness. We define a Handover Score (HS) to abstract human evaluation of the handover quality, and calculated from these objective measures as validated in our experiments. The Handover Score (HS) can guide handover approaches, presently often underpinned by human-centric evaluation criteria, to move towards data-driven policies that include reinforcement learning methods and low-level controller tuning. This framework offers a comprehensive approach to evaluating the fluency and intuitiveness of robot-human handovers, ultimately aiming to advance the field of collaborative robotics.

Keywords: Multimodal Interaction and Conversational Skills, Explainable Human-Robot Interaction, Cooperation and Collaboration in Human-Robot Teams
Abstract: Current HRI models provide valuable insights into multimodal communication, interaction taxonomies, and temporal changes such as trust or engagement. However, these approaches remain largely interaction-centric and do not adequately explain how repeated multimodal interactions develop into lasting human-robot relationships. This work identified three structural gaps in existing HRI theory: interaction-centric modeling, multimodality without relationship integration, and dynamic aspects in relationship development. To address these limitations, we introduce the Multidimensional Interaction Model (MIM³). The model integrates three key dimensions: multimodal behavior, meaning-sensitive adaptation, and temporal accumulation. Multimodality captures the bidirectional exchange of sensory signals between humans and robots. Meaning-sensitive adaptation describes how information is interpreted across hierarchical representational levels, and temporal accumulation explains the stabilization of expectations and roles during repeated human-robot encounters. MIM³ offers a structured perspective on how interaction evolves into relationship providing a theoretical basis for the development of long-term human-robot systems that enable stable, predictable, and socially meaningful interaction. The model is used as an analytical framework to classify current project examples involving service robots.

Keywords: Social Intelligence for Robots, Curiosity, Intentionality and Initiative in Interaction, Robotic Etiquette
Abstract: Proactive robots are increasingly deployed in public environments where people are encountered not as isolated individuals but as members of cohesive social groups. Yet whether the prevailing design paradigm in proactive human-robot interaction (HRI) accounts for the relational structure that defines a group as a social unit remains largely unexamined. Through a systematic review of 63 proactive HRI studies in group settings from 2000 to 2025, we identify a recurring tendency, the Individual-Targeting Assumption (ITA), in which robots treat co-present people as independent engagement targets. We find that ITA is present in 60.3% of the corpus, with group-aware approaches emerging almost entirely after the robot is already embedded in an ongoing interaction. Critically, how a robot should detect and negotiate entry into a pre-formed group before initiating contact remains unaddressed across the corpus. Three failure modes, engagement misdetection, social ratification blindness, and bystander neglect, emerge as recurring patterns when proactive robots interact with human groups. These findings reframe proactive HRI in group settings not as an extension of dyadic interaction but as a qualitatively distinct design problem, and they identify the robot’s approach to the group during the entry phase as a critical and understudied open challenge.

Keywords: Multimodal Interaction and Conversational Skills, LLM/Gen AI-based HRI
Abstract: As robots become increasingly integrated into human environments, their ability to perform strategic, socially grounded tasks, such as negotiation, becomes critical. Recent advancements in large language models (LLMs) have enabled robotic agents to exhibit fluent, human-like dialogue in many contexts. However, general-purpose LLMs often lack the domain-specific expertise and robust strategic reasoning required for effective human-like negotiation. Moreover, it remains unclear how the agent's expertise and physical embodiment jointly influence negotiation performance and user perception. To address these limitations, we develop an expert embodied negotiation framework that extracts linguistic and emotional features from human speech and facial cues, and uses these features together with domain knowledge retrieved via a retrieval-augmented generation (RAG) process to guide a large language model to generate synchronized verbal responses and select persuasive gestures for closed-loop human-robot negotiation. Using this framework, we instantiate three agent conditions with varying levels of embodiment and expertise: a non-expert embodied agent, an expert voice-only agent, and an expert embodied agent. We evaluate these agents in a within-subject study (n = 30) using a human-robot negotiation task to investigate how agent expertise and embodiment influence negotiation performance and user perception. Results indicate that embodiment is associated with higher perceived rapport and interaction clarity, whereas expertise tends to support more structured and strategic negotiation behavior. These findings suggest that integrating embodiment and domain expertise holds promise for designing robotic agents capable of more effective and socially grounded human-robot negotiation.

Keywords: Cooperation and Collaboration in Human-Robot Teams
Abstract: Robots are increasingly integrated into shared workspaces, making it critical that their collaboration strategies foster positive human perceptions. While fairness has been widely studied in human-human collaboration, its role in shaping perceived robot attributes in human-robot teams remains underexplored, with prior work largely focusing on equality or contribution-based allocation. In this work, we explore a need-sensitive perspective of fairness to design human-robot collaboration strategies for teams with heterogeneous capabilities. We evaluate two need-sensitive strategies, need-based and equity-based collaboration, alongside two existing strategies, equality-based and incentive-based, in a user study (n = 66). We examine how these strategies influence perceived robot attributes and willingness to work with the robot, and how perceived fairness relates to these evaluations. Our results show that collaboration strategies significantly shape perceived robot attributes and willingness to work with the robot. Strategies that balance equal treatment with adaptive support, particularly the equity-based strategy, received more positive evaluations, whereas need-based strategies were rated lower. Perceived fairness toward oneself and the partner was also positively associated with all perceived robot attributes and willingness, with differences across user strength emerging primarily for perceived intelligence, teammate evaluations, and willingness. These findings highlight the importance of need-sensitive collaboration design and perceived fairness in shaping how robots are evaluated and accepted as teammates.

Keywords: User-centered Design of Robots, Assistive Robotics, Medical and Surgical Applications
Abstract: The integration of robotics into wound care offers a promising solution to the growing patient demand amid a global nursing shortage. Formative work understanding the design requirements for wound care robotics to support clinical tasks remains an unmet need. In particular, wound care is a very hands-on and collaborative domain and therefor understanding the manipulation demands is necessary for developing robotic systems that align with wound care clinical needs. In this paper, we observe nurses performing wound care to inform the manipulation design requirements for robotics in this domain. Through a thematic analysis of observational notes, nurse manipulation approaches are grouped into six areas: grasping strategies, contact regions, bimanual manipulation, mobile manipulation, common motions, and workspace. For each area, we: 1) identify dominant techniques and the clinical context driving their use; 2) describe how techniques vary across different care tasks and subjects of interaction; and 3) demonstrate how these findings inform robotic design decisions through the design and fabrication of a novel end effector for wound dressing. Researchers can use these findings to design robotic systems that align with the manipulation needs of care tasks within wound care.

Keywords: Assistive Robotics
Abstract: As dental professionals face strenuous conditions and patients experience dissatisfaction or unsuccessful procedures, dental robots hold the promise to improve precision, speed, and satisfaction in dental care. However, there is a knowledge gap on patients’ and dentists’ expectations of robots in the realm of digital dentistry. As an exploratory study, we contribute an analysis of the implications of robotics in dental offices for patients and dentists. Given the constraints of busy and high-earning dentists, we were able to conduct an explorative study with six dentists, with a combined 86 years of practicing experience, and 12 patients in real-world dental offices. Results from dentists show that current benefits reside in treatment planning and instrument sterilization. With patients as their priority, dentists’ main concerns were a decrease in patient-doctor interactions and a lack of adaptability in unforeseen circumstances. Similarly, patients had concerns about possible malfunction and its impact on the quality of experience. Patients emphasized the benefits of improving procedures to produce the desired or lasting treatment. These findings contribute new knowledge and future directions to an interdisciplinary yet underexplored health domain in dentist-, patient-, or dentist-patient-robot interaction.

Keywords: Degrees of Autonomy and Teleoperation
Abstract: Robotic disassembly of complex mating components often renders pinch grasping infeasible, necessitating multi-fingered enveloping grasps. However, visual occlusions and geometric constraints complicate teaching appropriate grasp motions when relying solely on 2D camera feeds. To address this, we propose an affordance-guided teleoperation method that pre-generates enveloping grasp candidates via physics simulation. These Affordance Templates (ATs) are visualized with a color gradient reflecting grasp quality to augment operator perception. Simulations demonstrate the method's generality across various components. Real-robot experiments validate that AT-based visual augmentation enables an operator to effectively select and teach enveloping grasp strategies for real-world disassembly, even under severe visual and geometric constraints.

Keywords: Novel Interfaces and Interaction Modalities, Motion Planning and Navigation in Human-Centered Environments, Virtual and Augmented Tele-presence Environments
Abstract: Autonomous mobile robots operating in human-shared indoor environments often require paths that reflect human spatial intentions, such as passing through wider areas, avoiding locally undesirable regions, or following a preferred side of a corridor. However, conventional path planners primarily optimize geometric costs and provide limited support for explicit route specification by human operators. This paper presents MRReP, a Mixed Reality-based interface that enables users to draw a Hand-drawn Reference Path (HRP) directly on the physical floor using hand gestures. The drawn HRP is integrated into the robot navigation stack through a custom Hand-drawn Reference Path Planner, which converts the user-specified point sequence into a global path for autonomous navigation. We evaluated MRReP in a within-subject experiment against a conventional 2D baseline interface. The results demonstrated that MRReP improved path specification accuracy and usability while reducing perceived workload, while enabling more stable path specification in the physical environment. These findings suggest that MR-based direct path specification is an effective approach for reflecting user-intended geometric paths in mobile robot navigation. Additional material is available at https://mertcookimg.github.io/mrrep/

Keywords: Virtual and Augmented Tele-presence Environments, Explainable Human-Robot Interaction, Novel Interfaces and Interaction Modalities
Abstract: Teleoperation for contact-rich manipulation remains challenging, especially when using low-cost, motion-only interfaces that provide no haptic feedback. Virtual reality controllers enable intuitive motion control but do not allow operators to directly perceive or regulate contact forces, limiting task performance. To address this, we propose an augmented reality (AR) visualization of the impedance controller's target pose and its displacement from each robot end effector. This visualization conveys the forces generated by the controller, providing operators with intuitive, real-time feedback without expensive haptic hardware. We evaluate the design in a dual-arm manipulation study with 17 participants who repeatedly reposition a box with and without the AR visualization. Results show that AR visualization reduces completion time by 24% for force-critical lifting tasks, with no significant effect on sliding tasks where precise force control is less critical. These findings indicate that making the impedance target visible through AR is a viable approach to improve human-robot interaction for contact-rich teleoperation.

Keywords: Novel Interfaces and Interaction Modalities, Trust in HRI, Cognitive Skills and Mental Models
Abstract: Intuitive teleoperation interfaces are crucial for the safe and effective operation of robotic manipulators in challenging environments. In the nuclear industry, surface contact tasks such as swab sampling require precise path and force tracking, obstacle avoidance, and sustained operator attention, which conventional joystick interfaces struggle to support effectively. This study designs and evaluates a novel touchscreen teleoperation interface that maps continuous finger movements directly to robotic manipulator motions, provides finer velocity control, and integrates control with visualization, enabling more natural, precise, and intuitive surface interaction than conventional controllers. A comparative user study with 20 participants evaluated task performance and workload using the proposed touchscreen, a conventional joystick, and a single-click autonomous mode. Tasks simulated realistic surface manipulation using a Franka Emika Panda arm, remotely controlled from another country. Kinematic, physiological, and behavioral data were recorded to comprehensively assess task performance, cognitive load, and operator trust across each control condition. Participants completed teleoperation tasks more efficiently and accurately with the touchscreen interface, achieving a 53.5% reduction in completion time (median: 2.50 vs. 5.38 min), higher in-area coverage on the sinusoidal path (90.7% vs. 84.1%), and lower overshoot on both path geometries compared with the joystick. Cognitive load, quantified via NASA-TLX (0–100), decreased from joystick to touchscreen (mean TLX 52 to 43; -9 points, -17.3%) and was lowest under the autonomous one-click mode (31; -21 points vs. joystick, -40.4%; -12 vs. touchscreen, -27.9%). This research presents an easy-to-implement touchscree

Keywords: Multimodal Interaction and Conversational Skills, LLM/Gen AI-based HRI, Personalities for Robotic or Virtual Characters
Abstract: Multi-robot systems hold significant promise for social environments such as homes and hospitals, yet existing multi-robot systems often treat robots as functionally interchangeable, overlooking how distinct agent identities shape user perception and how such individuality changes the coordination requirements of multi-robot interaction. To address this, we introduce M2HRI, a multimodal multi-agent framework that models each robot as an identity-bearing agent through personality and long-term memory, together with a contextualized coordination mechanism that regulates agent participation. In a controlled user study (n = 105) in a multi-agent human–robot interaction (HRI) scenario, we found that most personality contrasts were distinguishable and consistently expressed. Long-term memory improved preference awareness and interaction naturalness, while contextualized coordination improved conversational flow, response appropriateness, and overlap avoidance. Together, these findings show that agent individuality and contextualized participation coordination play complementary roles in supporting coherent and socially appropriate multi-agent HRI. Project website available at https://project-m2h ri.github.io/.

Keywords: Assistive Robotics, Human–Robot Value Alignment and Safety, LLM/Gen AI-based HRI
Abstract: Safe and reliable human–robot interaction remains a challenge, especially for Large Language Model (LLM)-powered socially assistive robots in elderly care. This paper proposes a risk-aware response framework that explicitly models the safety implications of user queries. We introduce a Query Risk Assessment Module (QRAM), which computes a Query Risk Score (QRS) using structured semantic indicators to classify user inputs into different risk levels. Based on this, we develop two strategies: risk-aware response refinement and risk-aware response generation. To assess safety, we introduce the Risk-Aware Response Safety Score (RRSS), a metric that captures both the presence of risk-bearing language and the absence of necessary safety-critical guidance. Experimental results across multiple LLMs demonstrate that incorporating query-level risk awareness reduces response-level risk, with both strategies outperforming raw responses.

Keywords: Computational Architectures, LLM/Gen AI-based HRI, Linguistic Communication and Dialogue
Abstract: Enabling robots to maintain and communicate structured world knowledge through natural spoken language is essential for transparent and accessible human-robot interaction. Moreover, there are several solutions that employ knowledge graph representations as part of their behavior generation pipeline, but the knowledge creation can be a limitation for autonomous robots. This paper presents a bidirectional speech-based knowledge-grounding tool for creating and querying new knowledge for autonomous robots. The presented framework integrates speech-to-text and text-to-speech components with a locally deployed Large Language Model whose output is constrained by a context-free grammar, ensuring that every generated token sequence is a syntactically valid structured object that encodes knowledge graph operations. This grammar-constrained decoding eliminates the need for post-hoc output validation and reduces malformed responses. The system is evaluated on a purpose-built benchmark of diverse utterances spanning eight interaction categories, comparing five locally-deployed language models. Results demonstrate that the proposed solution achieves full structural validity while maintaining competitive semantic accuracy across different models.

Keywords: Cooperation and Collaboration in Human-Robot Teams
Abstract: The analysis of Operational Technology (OT) networks is increasingly important in the context of traffic consistency, cybersecurity, and real-time (RT) performance in modern industrial Ethernet systems. Existing approaches typically rely on statistical indicators, such as jitter and inter-frame intervals, or machine learning techniques applied to traffic features. While effective in many cases, these methods primarily address temporal characteristics and do not explicitly capture the structural and semantic properties of communication protocols. This paper proposes an approach for OT network diagnostics based on a large language model (LLM), specifically a GPT-based architecture, in which real-time communication is modeled as a timed formal language. Network traffic is represented as sequences of symbolic events characterized by syntactic, semantic, and temporal attributes. This representation enables the application of language modeling techniques to learn communication patterns and dependencies in OT environments. A prototype implementation is developed and evaluated on real OT traffic, focusing on RT communication in a multi-protocol environment with a PROFINET-based system and a reduced event alphabet. The GPT-based model is used to analyze and interpret communication sequences, enabling the identification of structured deviations from expected behavior. The results demonstrate the feasibility of the proposed approach and highlight its potential for semantics-aware diagnostics and anomaly detection in industrial networks.

Keywords: Applications of Social Robots, LLM/Gen AI-based HRI, Novel Interfaces and Interaction Modalities
Abstract: Narrative-Based Medicine (NBM) integrates traditional care with conversational approaches that address the psychosocial and emotional dimensions of illness. In assisted living and long-term care, social robots represent a promising medium for operationalizing the principles of NBM through continuous monitoring and interaction. However, purely generative Large Language Model (LLM)-based systems lack the structured dialogue strategies required for effective narrative elicitation. This paper presents a framework for structured narrative interaction in social robots to support older adults with chronic conditions in collecting personal narratives, monitoring emotional states, and promoting therapeutic adherence. The system adopts a hybrid architecture combining a Finite State Machine (FSM) for structured dialogue progression, LLM-driven dynamic prompting for context-aware responses, and profile-aware retrieval based on long-term user memory. A within-subject study with twenty older adults compared the proposed system with a purely generative baseline. Results show that the structured approach significantly improves perceived sociability, enjoyment, and social presence, while maintaining comparable ease of use. Additionally, the system enables the extraction of health-related and emotional parameters to support integrated monitoring. These results suggest that structured human–robot dialogue provides an effective and scalable approach for embedding NBM principles into assistive technologies.

Keywords: LLM/Gen AI-based HRI, Personalities for Robotic or Virtual Characters, Multimodal Interaction and Conversational Skills
Abstract: We introduce prompting as a design material for domestic robotics, enabling users to shape a robot's character through natural language rather than fixed, developer-defined scripts. This reframing challenges the assumption that robot character is a static design-time decision by positioning it as a user-configurable interaction layer. We demonstrate this concept through a physical prototype-a vacuum robot integrating a large language model (LLM) and a vision-language model (VLM)-that supports prompt-based character customization and context-aware interaction. In an exploratory video-based study (N=27), participants expressed diverse preferences, ranging from silent appliance-like behaviour to expressive and humorous styles, while also revealing critical risks such as socially inappropriate language and unsafe action suggestions. We discuss design implications for balancing personalization with predictability and propose a layered safety framework. Our findings highlight prompting as a powerful yet volatile design material for embodied systems, offering opportunities for personalization alongside urgent safety and ethical challenges.

Keywords: Novel Interfaces and Interaction Modalities, Anthropomorphic Robots and Virtual Humans, Non-verbal Cues and Expressiveness
Abstract: We present a hybrid interaction paradigm in which an expressive Augmented Reality (AR) avatar serves as the agent' s primary appearance while a concealed robotic arm provides physical manipulation. This design decouples social expressiveness from robot hardware and allows the virtual character to act on real-world objects. A central challenge is preserving the system' s visual coherence by ensuring the actuator remains concealed. We introduce a concealment-aware control method that optimizes actuator concealment and quantifies robot extrusion with a geometry-based metric. We evaluated the system in a 25-participant block-assembly study. Compared with an avatar only baseline, physical manipulation improved usability and engagement, while presence showed a positive but non-significant trend. In the avatar+robot condition (Trial B), Concealed and Exposed conditions produced clear extrusion differences, but whole-step gaze evidence for stronger robot-directed attention was inconclusive. Exploratory analyses suggested more robot gaze during higher-exposure moments. Overall, the results support combining software-defined expressive appearance with realworld manipulation, while the perceptual benefit of concealment requires more sensitive evaluation.

Keywords: Embodiment, Empathy and Intersubjectivity, Virtual and Augmented Tele-presence Environments, Human Factors and Ergonomics
Abstract: Body augmentation technologies aim to extend human capabilities by introducing mechanical body parts such as supernumerary limbs. Previous studies in this area have primarily focused on methods for controlling the additional limbs. In contrast, the present study focuses on the transfer of body ownership and sense of agency to an augmented limb in a virtual reality (VR) environment. Prior studies on embodiment in virtual body extension have mainly investigated the effects of multisensory congruency, such as visuomotor or visuotactile feedback. However, little attention has been paid to the role of habituation tasks that allow users to become accustomed to an augmented limb. We investigated habituation tasks designed to facilitate the embodiment of an extended second left hand. Habituation tasks refer to short manual activities performed before the main interaction to help users adapt to the augmented limb. Five different habituation tasks were designed, and their effects on body ownership and sense of agency were analyzed. A user study with 17 participants revealed that the strength of embodiment differed depending on the type of habituation task. In particular, tasks with higher difficulty tended to suppress embodiment of the augmented hand. These results highlight the importance of task design as a factor in facilitating embodiment in body augmentation systems.

Keywords: Social Touch in Human–Robot Interaction, Affective Computing, Applications of Social Robots
Abstract: This study investigates how non-human-like body designs can influence human–robot hugging interactions. While most existing robots are designed based on human-like hugging forms, i.e., two arms, we developed a robot with four arms that enables a novel hugging configuration. Specifically, we introduce a double-loop hug that simultaneously embraces the neck and waist, which is difficult to achieve with human arms. We conducted a within-participant experiment to compare three types of hugs: a neck-loop hug, a waist-loop hug, and the proposed double-loop hug. Participants experienced all conditions, evaluated their impressions via questionnaires, and indicated their preferred hug. The results showed no significant differences in questionnaire ratings across conditions. However, in terms of preference, the double-loop hug was significantly favored over the two single-loop hugs. These findings suggest that non-human-like robot designs can offer advantageous interaction experiences, particularly in the context of affective physical interactions such as hugging.

Keywords: Social Touch in Human–Robot Interaction, Robot Companions and Social Robots, Social Intelligence for Robots
Abstract: Large Language Models (LLMs) have transformed conversational AI, but an unresolved human-robot interaction question remains: if the same conversational intelligence is used in both cases, does a physical body still matter? This paper compares an embodied Pepper robot with a non-embodied laptop interface driven by the same Ollama deployment of a Llama 3.2 7B parameter model in a within-subjects, counterbalanced study (N=32). Participants engaged in open-ended dialogue with both agents and completed post-interaction questionnaires assessing interaction quality, social response, comfort, satisfaction, and overall preference. The embodied condition was rated significantly higher on naturalness, flow, relevance, understood responses, engagement, perceived social entity, connection, and comfort, with the largest paired effect on engagement (d_z = 1.12). Self-rated tech-savviness did not significantly correlate with the core outcomes reported for either condition. We argue that the contribution of the study lies in demonstrating that embodiment remains consequential even when linguistic capability is held constant at an LLM level.

Keywords: Monitoring of Behaviour and Internal States of Humans
Abstract: The "Proteus Effect" and other phenomena in which an avatar's appearance influences users' behavior and cognition have been widely reported. However, previous studies have mainly focused on temporary behavioral changes, and the underlying personality structure remains unclear. In this study, we hypothesize two personality models: the "Avatar-Self Consistency Model," which views the avatar as an extension of the self and assumes that behavioral changes caused by the avatar appear even when not using one; and the "Avatar-Self Switching Model," which views the avatar as separate from the self and assumes that behavioral changes are limited to the time of avatar use. In the experiment, we conducted an opinion-stating task on social themes using Zoom and analyzed the persistence of the effects of negative comments provided by others on speech tendencies. The purpose of this study is to examine which personality model the behavioral changes during avatar use are based on, and to clarify the characteristics of self-regulation and personality structure through avatars. The experimental results suggested a marginal trend supporting changes in speech tendencies consistent with the Proteus effect reported in previous studies. Furthermore, the changes in speech tendencies caused by avatar use were observed even after avatar use, suggesting that the effect is not limited to during use but may persist to some extent afterward. These findings are more consistent with the Consistency Model and indicate that experiences acquired through an avatar can partially carry over to the subsequent self. This study provides new insights into Proteus effect research, not only from the perspective of behavioral changes during avatar use but also from the perspective of their persistence.

Keywords: Applications of Social Robots, Machine Learning and Adaptation, Detecting and Understanding Human Activity
Abstract: Just-in-time interventions are critical in health behavior change, where messaging must occur at the moment of health-related decision making. Mobile robots can be particularly effective in supporting such interventions in multiple decision locations if they anticipate where and when users make health-related choices and can move to appropriate locations in time to intervene. We propose a Gaussian process-based trajectory prediction method that models routine user movement patterns with minimal data, sometimes from a single observation. By intersecting predicted user trajectories with spatiotemporal activity windows, a robot can anticipate target behaviors and position itself in time to deliver persuasive messages. We conducted a within-subjects study with 15 participants in a simulated office environment, where participants engaged in ten ``days'' of activities, and compared a proactive robot using our model to a robot that only reacted to current user location. The proactive robot delivered significantly more health-related interventions and accurately predicted daily routines after limited training. These findings demonstrate the feasibility of trajectory-based prediction for enabling proactive, opportunistic health interventions by mobile robots.

Keywords: Applications of Social Robots, Monitoring of Behaviour and Internal States of Humans, Detecting and Understanding Human Activity
Abstract: Prolonged sedentary behavior is a recognized occupational health risk in office environments, yet current digital interventions based on passive reminders are frequently disregarded during cognitively demanding tasks. This paper presents RHYTHM, a social robotic system for adaptive, physiologically-informed micro-break promotion in real office environments. RHYTHM integrates an autonomous mobile robot with wearable physiological monitoring and an LLM-based activity recommendation module that selects personalized interventions based on real-time heart rate deviation from each user's individual baseline and contextual availability. A pilot feasibility and acceptability study was conducted with six office workers (N = 6) across four sessions each in a real working environment, achieving an Acceptance Rate of 80.0% with a technical failure rate below 2%. Post-interaction evaluations indicated high perceived likeability and relaxation across participants. Exploratory correlation analysis suggests that acceptance appeared to be associated primarily with contextual appropriateness rather than engagement alone, and that relaxation benefits are largely independent of robot perception. These results support the feasibility of embodied, context-aware robotic interventions as a promising complement to passive digital reminders for promoting healthier workplace habits.

Keywords: Applications of Social Robots, Ethical Issues in Human-robot Interaction Research, Social Presence for Robots and Virtual Humans
Abstract: Social robots are increasingly deployed in everyday environments, raising questions about their capacity to influence human behaviour ethically and effectively. This paper presents a naturalistic field experiment in a university cafeteria, where we deployed a social humanoid robot (ARI) to encourage sustainable food choices using Nudge Theory ---an indirect persuasion framework grounded in behavioural science. Unlike previous work that relied on controlled lab settings, we explicitly apply nudging techniques---including framing, salience, and anchoring---through both static posters and dynamic robot interactions. Across four experimental conditions (baseline, posters only, robot only, combined), we analysed 4,271 purchases over 20 days, measuring carbon emissions per product, sustainability ratios, and meat-based consumption. While our overall hypotheses showed mixed support, we observed a promising downward trend in beef purchases and a statistically significant increase in vegetarian jacket potato sales when the robot was present. This suggests that robotic nudging may be most effective in contexts with low menu variability. Our findings advance social robotics by demonstrating the first real-world deployment of Nudge Theory via a robot, while also highlighting the importance of menu consistency, novelty effects, and longitudinal study design for future work.

Keywords: Non-verbal Cues and Expressiveness, User-centered Design of Robots, Multimodal Interaction and Conversational Skills
Abstract: In everyday human–robot interaction, robots may need to interrupt people while they are engaged in ongoing activities, for example, to ask questions, provide reminders, or warn about potential issues. Designing appropriate robot interruptions is challenging, as their effectiveness and social acceptability depend on a range of contextual factors. We conducted an in-person interview study in which 18 participants embodied a robot and demonstrated how it should interrupt them in six scenarios that varied in task type (communicative vs. physical) and urgency (low, medium, high). Our results revealed 15 interruption behavior clusters and 4 themes describing participants’ expectations. We found that participants' expectations differed regarding whether they wanted robots to interrupt them at all in low-urgency situations, often depending on whether the robot was perceived as a social companion or a task-oriented assistant. As urgency increased, participants generally preferred more immediate and pronounced interruptions that combined modalities such as visual cues, gestures, speech, and, occasionally, physical touch.

Keywords: Detecting and Understanding Human Activity, Monitoring of Behaviour and Internal States of Humans, Cooperative Intelligence in HRI
Abstract: Understanding human actions is critical for advancing behavior analysis in human–robot interaction. Particularly in tasks that demand quick and proactive feedback, robots must recognize human actions as early as possible from incomplete observations. Sub-actions offer the semantic and hierarchical cues needed for this, since human actions are inherently structured and can be decomposed into smaller, meaningful units. However, conventional approaches focus pri- marily on holistic actions and often overlook the rich semantic structure embedded in sub-actions, making them poorly suited for early recognition. To address this gap, we introduce SASI (Sub-Action Semantics Integrated cross-modal fusion), a novel framework that integrates existing graph convolution networks to fuse spatiotemporal features with sub-action semantics. SASI exploits a segmentation model with a traditional skeleton-based graph convolution network, capturing both fine-grained sub- action semantics and overall spatial context, while operating in real-time at 29 Hz. Experiments on BABEL, a skeleton- based dataset with frame-level annotations, demonstrate that our method improves recognition accuracy over conventional approaches, with additional gains expected as the quality of sub-action segmentation improves. Notably, SASI also achieves superior performance in understanding partial action sequences, revealing its capability for early recognition, which is essential for proactive and seamless Human-Robot Interaction (HRI). Code is available at https://github.com/SavickTso/SASI .

Keywords: LLM/Gen AI-based HRI, Cooperation and Collaboration in Human-Robot Teams
Abstract: This paper presents MEKARD, a robot system that provides context-aware, proactive assistance in environments shared with humans by querying an external database of past experiences. The database stores multimodal records, including text, images, numerical state values, and embodiment information describing the physical states and actions of both humans and robots; this embodiment information is treated as a core component. Because human states, robot states, and environmental context can combine to create diverse situations that are difficult to handle using hand-crafted rules alone, we design a retrieval-augmented generation (RAG) framework that retrieves relevant past situations to support decision making. The system integrates the database and similarity-based retrieval with a multi-stage pipeline based on vision-language models and large language models. Experimental results show that the proposed method produces the largest number of correct outputs among the evaluated methods. A user study further suggests that embodiment-aware actions are generally rated more positively than the alternatives in the tested scenarios.

Keywords: Non-verbal Cues and Expressiveness, Cooperation and Collaboration in Human-Robot Teams, Robots in art and entertainment
Abstract: Musical ensemble performance relies on implicit visual gestures, such as eye contact and head nodding, to negotiate fluid turn-taking. While the field of robotic musicianship has explored visual coordination, many existing systems treat gestures as explicit control interfaces or rely on audio-only cues that introduce significant latency. To address this, we present a multimodal turn-taking system for a robotic marimba player utilizing a markerless vision pipeline to process facial landmarks and hand keypoints in real time. We conducted a within-subjects study (N=20) comparing a traditional audio-only silence-detection interaction against three visual-based interactions: eye contact, eye contact with a head nod, and a deliberate thumbs-up gesture. Objective evaluations confirmed our first hypothesis; all visual conditions significantly reduced turn transition latency compared to the audio baseline (p < 0.001). However, only the ”natural” eye contact with a head nod condition required significantly higher mental effort (p=0.039) and resulted in the lowest overall participant satisfaction. Conversely, Eye contact only was rated as the most fluent and practical modality (p < 0.05), while the thumbs-up gesture and audio-only baseline provided the highest levels of predictability and satisfaction. These findings reveal a trade-off between social naturalness and system reliability, suggesting that for robotic musicians, simple, reliably detectable cues, whether natural eye contact or explicit gestures, are more essential for perceived fluency than the integration of complex, multi-modal social mimicry.

Keywords: Sound design for robots, Robots in art and entertainment, Cooperation and Collaboration in Human-Robot Teams
Abstract: Robotic Musicianship (RM) is the implementation of machines that listen to, interact with, and improvise music. RM often aims to design systems that listen like humans but play with the techniques and capabilities that are humanly impossible. Robots such as Shimon play like a machine by leveraging humanly impossible playing speeds and multiple arm morphologies that allow for sonorities unachievable by humans. Other robots such as the Expressive Robotic Guitarist (ERG) successfully emulated human performance but have yet to fully exploit its own mechanical embodiment. This paper introduces the latest version of the ERG, which navigates unexplored musical landscapes by leveraging techniques that are physically impossible or difficult for a human to execute. We detail the implementation of augmented guitar techniques including complex polyrhythms, variable speed tremolo, and asynchronous glissandi, which transcend human physiological constraints. To evaluate the musical and interactive potential of these techniques, we conducted online surveys with 30 guitarists and performed in-person workshop sessions with 4 professional guitarists. Participants evaluated the system based on its perceived creativity and its ability to provoke collaborative musical experiences. We contribute a practical application of the Cooperative-Creative Index by evaluating the perceived Trust, Collaboration, Creativity, and Stability of the system. Although qualitative results from the online format indicate mixed responses, quantitative responses and in-person responses indicate that our augmented techniques show potential to foster innovative human-robot collaboration. By leveraging objective technical limits as musical strengths, we demons

Keywords: Robots in art and entertainment, Sound design for robots
Abstract: We focus on how snare drummers use their fingers as compliant springs, adjusting the gripping force to control the stick's rebound intervals on the drumhead. Driven by the concept of replicating this behavior with a variable-impedance spring mechanism, this study aims to achieve dynamic alterations in rhythm synchronized with varying tempos. While existing research on drum-playing robots has successfully achieved high-speed double strokes by adjusting the rotational stiffness of the stick via pneumatic artificial muscles or tensile spring-based variable stiffness mechanisms, these systems were not designed to achieve the highly sophisticated performance of a human drummer who adaptively changes rhythms on the fly. In this study, we developed a novel variable-impedance mechanism that adjusts the spring's moment arm. Utilizing this mechanism, we developed the second-generation drum-playing robot, which successfully modulated the stick's natural frequency over a 3.4-fold range from 2.8 to 9.3 Hz, thereby achieving dynamic strike differentiation between double and triple strokes. Future work will focus on controlling the amplitude and phase at arbitrary timings to enable dynamic rhythm alterations and real-time cooperative interaction with human musicians.

Keywords: Detecting and Understanding Human Activity, Cooperation and Collaboration in Human-Robot Teams, Non-verbal Cues and Expressiveness
Abstract: To fully realize human-robot teams in unstructured environments, egocentric sensor suites should capture both verbal and non-verbal user communication, as non-verbal language provides critical contextual information. Deictic gestures provide substantial context from a single point. Detecting and grounding these cues with minimal and highly resource constrained on-body hardware is challenging, but a desire for real-world systems. This paper presents an egocentric deictic gesture grounding framework using a single monocular camera capturing the user-centric viewpoint. The system classifies deictic gestures from egocentric hand-and-arm segmentations, estimates a 3D pointing vector from 2D sensor data, and grounds the referent as a ground location. To support different compute constraints, the framework introduces two complementary pointing-vector estimation methods, 1) a deep learning point estimation (DLPE) approach that uses dense 3D reconstruction, and 2) a lightweight geometric pinhole point estimation (GPPE) approach that infers depth geometrically from segmented limb contours and calibrated camera intrinsics. Studies compare two pointing-vector estimation methods on localization accuracy and runtime performance. We validate the system end-to-end on hardware and report benchmarks demonstrating successful robot control via deictic gestures. This work provides a practical, monocular egocentric alternative to depth-sensor and third-person deictic gesture perception systems that can be deployed on resource-constrained hardware in the field.

Keywords: Virtual and Augmented Tele-presence Environments, Degrees of Autonomy and Teleoperation, Novel Interfaces and Interaction Modalities
Abstract: Deictic spatial commands—pointing, gaze direc- tion, and utterances such as “reach there”—provide a natural means of specifying manipulation targets, but they require continuous resolution from an operator’s egocentric mixed- reality (MR) frame into the robot base frame. Existing systems commonly treat this alignment as a one-time calibration or assume that target poses are already expressed in robot coor- dinates. We present a deictic shared-autonomy manipulation system built on continuous markerless egocentric frame fusion. A pose graph combines headset visual-inertial odometry and spatial depth with robot joint-state kinematics and wrist- camera observations to maintain a drift-corrected headset-to- robot transform without online fiducial tracking or external in- frastructure. Gaze, tracked-stylus, and voice inputs are resolved through this transform into end-effector targets, while planned motion and robot state are projected back as MR overlays. A within-subjects evaluation across three reaching tasks and four interface conditions measured frame alignment, target- grounding error, task performance, and workload. Continuous fusion reduced target-grounding error by 49% on motion- dependent tasks and reduced operator interventions by 52% relative to one-shot alignment. The results indicate that online frame fusion, rather than MR visualization alone, is the critical enabling layer for reliable deictic manipulation.

Keywords: Explainable Human-Robot Interaction, Multimodal Situation Awareness and Spatial Cognition, Assistive Robotics
Abstract: Many assistive robotic systems still rely on explicit user commands such as button presses or body tilts to initiate help, which can be cumbersome and unnatural. To reduce this burden, we propose an implicit multimodal intent inference framework that fuses full-body motion cues with egocentric image/depth context via an interpretable Bayesian belief model. Our system comprises a transformer-based gesture recognition module that learns pre-gesture movement patterns and an object detection-based mapping module that estimates task-relevant object proximity. These cues are probabilistically fused through a Bayesian network to produce a probability distribution over intents. For comparison, we also implement a multimodal deep model that directly maps motion and egocentric RGB-D inputs to intent labels. In a controlled living-room scenario with multiple object-related interactions, we benchmark gesture-only (GR), Bayesian fusion (BF), and end-to-end (E2E) inference. In full leave-one-subject-out (LOSO) evaluation, BF achieves the highest mean accuracy (0.906), while E2E achieves the highest macro-F1 score (0.761). BF also produces significantly earlier sustained accurate predictions than both GR and E2E, with an average prediction lead time of 0.452s before action onset. Under ambiguous gesture predictions, both multimodal methods substantially improve over GR, showing that egocentric visual context can reduce uncertainty when motion-only evidence is unreliable. The code is available at github.com/RTHMLab/context-aware-intent-inference.

Keywords: Cooperation and Collaboration in Human-Robot Teams, Machine Learning and Adaptation, Assistive Robotics
Abstract: Effective human-robot teamwork requires robots to adapt to partners, situations, and task dynamics from the start of an interaction. In the MATRX Urban Search and Rescue (USAR) environment, people can externalize collaboration patterns (CPs) they discover during teamwork through a chat and reflection interface. We study whether a robot can use such prior team experience to become a better teammate in future interactions. To this end, we represent historical CPs as knowledge-graph episodic memories and use graph representation learning with a node-classification objective to identify a representative and effective memory for reuse. We then initialize the robot with this memory before a new collaboration episode begins. Across 20 participants and 160 round-level observations, initializing the robot with a single automatically selected prior CP increases rescue success from 25.7% to 41.3% and reduces average task time by 283 seconds. The strongest gains appear at the beginning of interaction, suggesting that reusable episodic memory can help robots enter collaboration with more effective task knowledge and support smoother early teamwork.

Keywords: Cognitive Skills and Mental Models, Cooperation and Collaboration in Human-Robot Teams, Cooperative Intelligence in HRI
Abstract: Understanding the collaborative behaviour between team members and the formation of shared mental models in multi-party meetings is essential to achieving shared teaming goals and effective Human-Agent Teaming. We introduce TGN-SMM, a Shared Mental Model framework based on a Temporal Graph Neural Network, designed to model the team's cognitive states in meetings as an evolving speaker-interaction graph. Each node here represents a team member, characterised by the context of their utterances, their team role, and the intention of their speech. As the meeting unfolds, pairwise temporal events are passed to update the speakers' cognitive states, and the interpersonal alignment dynamic is directly inferred from edge representations between speaker pairs. The interactions in the meeting are treated as pairwise temporal events, with the edge representation relying on local reasoning rather than a global graph representation. This allows the framework to capture how the cognitive states and interpersonal alignment dynamic patterns emerge and evolve. Temporal adjacency pairs in meetings are represented as a list of dyadic conversations among team members in the AMI Meeting Corpus, which is used to conduct experiments in which the TGN-SMM achieves the highest macro-averaged F1 score, with particular improvement on minority alignment classes on its static and temporal SMM baselines, illustrating the benefits of incorporating temporal graph reasoning into models of team alignment.

Keywords: Linguistic Communication and Dialogue, Social Learning and Skill Acquisition Via Teaching and Imitation, User-centered Design of Robots
Abstract: Successful conversations require speakers to align on the meaning of concepts, a challenging but crucial task for human-robot interaction. Understanding the process of establishing such alignment is hindered by competing interpretations of the term and isolated, unidirectional investigations of its design space. This paper argues for a design-centric understanding of conceptual alignment as a bidirectional and co-constructive process. We introduce a taxonomy that characterizes conceptual alignment dialogues along what triggers its initiation and what level(s) of conceptual understanding it concerns. We further present a dialogue act schema as an operational tool that captures the interactional moves through which alignment is achieved. Together, these contributions provide a structured foundation for analyzing, comparing, and designing conceptual alignment in human-robot interaction.

Keywords: LLM/Gen AI-based HRI, Linguistic Communication and Dialogue, Computational Architectures
Abstract: Robots operating in shared environments must do more than execute actions. They must ground task instructions in the current world state and remain responsive to changes during execution. Large language models (LLMs) offer strong reasoning capabilities for task planning, but their reliance on static, text-only scene descriptions often leads to brittle plans that fail once execution begins. To overcome this, we detail an ontology-grounded planning framework for robot manipulation that integrates an LLM with a symbolic world model and execution-time state updates. Rather than reasoning over free text alone, the planner operates on a grounded state representation built from ontology queries and simulator observations. We evaluate the proposed approach in a simulated PyBullet table-clearing setting under nominal conditions and execution-time disturbances. Within this setting, the ontology-grounded configuration achieves higher plan feasibility, task success, and failure recovery than the implemented text-only baseline.

Keywords: Detecting and Understanding Human Activity
Abstract: We propose CLAD, a Constrained Latent Action Diffusion model for vision-language procedure planning, the challenging task of predicting a sequence of steps that lead from a start state towards an intended goal state. Procedure planning, while critical in robot skill learning and for assistive robots, has been largely neglected so far, and existing methods have not leveraged semantic information for step generation. In contrast, CLAD exploits the fact that the latent space of diffusion models trained for procedure planning contains rich semantic information. Our method uses a Variational Autoencoder (VAE) to learn the latent representation of steps and observations as constraints and integrate them into a diffusion process. As such, our method uses these latent constraints to steer the diffusion model to generate better steps in the procedural plan. We report extensive experiments on four datasets: three covering human procedure planning and one robot learning, and show that our method outperforms state-of-the-art methods by a large margin. We demonstrate that the proposed integration of the step and observation representations learnt in the VAE latent space is key to these performance improvements. Code is available at https://github.com/leishi07/clad.

Keywords: Machine Learning and Adaptation, LLM/Gen AI-based HRI, Multimodal, Multilingual and Multitask Modeling
Abstract: Unlike single-step motion planning, long-horizon embodied planning is difficult to achieve through imitation learning or vision language action models. One of the most viable methods uses foundation models to decompose natural language instructions into predefined skill sequences via in-context learning, followed by sequential execution. However, this approach is prone to logical errors and hallucinations if the in-context examples are not highly relevant to the given task, limiting its future applications. Therefore, we present ReLEP, a novel framework for Real-time Long-horizon Embodied Planning, which completes long-horizon tasks beyond table-top manipulations without in-context examples by learning implicit logical inference through supervised fine-tuning. The large vision language model formulates plans as sequences of skill functions that are selected from a carefully designed skill set. ReLEP is also equipped with a Memory module for plan and status recall, and a Robot Configuration module for versatility across robot types. In addition, we propose a data generation pipeline to tackle dataset scarcity. When constructing the dataset, we considered the implicit logical relationships, enabling the model to learn from them and dispel hallucinations. Through comprehensive evaluations across various long-horizon tasks, ReLEP demonstrates high success rates and compliance to execution even on unseen tasks and outperforms state-of-the-art baseline methods.

Keywords: Long-term Experience and Longitudinal HRI Studies, Robot Companions and Social Robots, Applications of Social Robots
Abstract: This paper explores the integration of social robots into three privately owned cafes in Japan as a case study to examine human-robot interaction in public spaces. These cafes with robots such as aibo and LOVOT provide an organic environment for real-world human-robot encounters. Through ethnographic observations and interviews with cafe owners, we analyze how robots are embedded into community gatherings, workplace practices, and service offerings. These insights contribute to a deeper understanding of how robots can coexist with humans in dynamic public settings.

Keywords: Robot Companions and Social Robots
Abstract: We present CHATBOX, a lightweight, cost efficient Socially Assistive Robot (SAR) for mental health conversations with young users. Built from 3D-printed components and low-cost servo motors, CHATBOX delivers a deployable 12-DOF expressive platform costing under USD 300. An interchangeable animal-mask system gives users agency over companion morphology and support user engagement. The system adopts a privacy-preserving edge-to-server architecture in which a Jetson Nano communicates with a local server running a customized EfficientNet V2-S Facial Emotion Recognition (FER) model (CHATBOX V1), achieving 81% validation accuracy across seven emotion classes. We integrate Retrieval-Augmented Generation (RAG) as an emotionally grounded continuity layer: session logs and emotional states are embedded after each interaction and retrieved in later sessions, enabling context-aware responses perceived as attentive rather than scripted. A pilot in-lab study with N = 10 participants (ages 10–16) reported high ratings for comfort, friendliness, and perceived continuity. We conclude with future directions, including expressive DOF expansion and culturally diverse datasets for children and adolescents

Keywords: Detecting and Understanding Human Activity
Abstract: Isolated sign language recognition remains challenging due to limited training data, large visual variation, and inconsistent label conventions across public datasets. This paper presents DeepSignV2, a ViT-based approach trained with a large-scale merged corpus covering multiple sign languages. We curate a pretraining dataset of 515,306 videos with 13,915 classes. We unify the label space of four American Sign Language datasets by merging classes, yielding 3,535 merged ASL classes, resulting in a total of 11,469 glosses. Non-ASL datasets are kept separate to preserve language specific label semantics. We apply cutout augmentation and adopt UniFormerV2 backbones initialized from InternVideo2 distilled checkpoints. We evaluate distilled Small and Base variants, and conduct ablation experiments using the small variant. On the WLASL100 benchmark, DeepSignV2 achieves 94.57% top-1 accuracy, exceeding prior published results including DSLNet at 93.70% and DeepSignV1 at 92.64%. We also observe state-of-the-art performance on additional isolated sign language benchmarks.

Keywords: Machine Learning and Adaptation
Abstract: Isolated sign language recognition is limited by the small size of available annotated datasets and the sensitivity of sign identity to fine-grained hand shape and motion. Synthetic video augmentation is an appealing remedy, but pose-guided diffusion models do not guarantee geometric faithfulness, and generated clips with incorrect finger articulation introduce label noise. We present SynthSign, a pose-faithful synthetic augmentation pipeline for isolated sign language recognition. Real RGB training clips are processed through Sapiens 1B to extract dense 133-keypoint whole-body pose sequences, which condition LTX-Video 2.3 via a ControlNet adapter. A closedloop filtering stage re-estimates pose from each generated clip and rejects samples whose keypoint trajectories deviate beyond a threshold from the source sign. On WLASL100, unfiltered synthetic data does not improve over the real-only baseline, while filtered augmentation at a 1:1 ratio improves top-1 accuracy from 93.80% to 94.57%. The method achieves state of-the-art results on popular WLASL100, WLASL300, MSASL100, and MS-ASL1000 benchmarks, without pose streams or multimodal fusion at inference time.

Keywords: Trust in HRI, Creating Human-Robot Relationships
Abstract: Large language model (LLM)-based assistants are increasingly being integrated into interactive systems and presented in different forms, such as text-based interfaces and physically embodied robots. However, when such agents produce erroneous outputs due to LLM hallucinations, it remains unclear whether different forms of embodiment and different types of hallucinations lead to different patterns of trust change. To address this issue, this study examines the effects of assistant embodiment and hallucination type on users’ trust. We conducted a mixed-design user study using a multiple-choice quiz task, in which participants interacted with either a physically embodied robot or a text-based assistant across correct trials, factuality hallucination trials, and faithfulness hallucination trials. Trust was assessed through pre- and post-task questionnaires, post-trial trust ratings, and open-ended responses. The results showed that users’ overall trust decreased after the interaction, and that trust ratings were lower in hallucination trials than in correct trials. However, the quantitative analyses did not reveal significant differences either between the embodiment conditions or between hallucination types. Nevertheless, the qualitative results suggested that embodiment may have influenced how users experienced trust loss.

Keywords: Affective Computing
Abstract: Prior evaluations of large language models (LLMs) on emotion understanding have largely focused on surface-level recognition accuracy, leaving open whether LLMs capture the appraisal-based structure underlying emotions or reproduce the distributional variability observed across human judgments. In Study 1, we compared LLM-generated appraisal ratings against human annotations across seven appraisal dimensions and seven emotion categories using a corpus of first-person emotional narratives. Models achieved high accuracy overall but consistently misclassified shame as guilt; appraisal agreement was strongest for pleasantness and responsibility but substantially lower for certainty, control, and circumstance. In Study 2, we examined whether LLMs reproduce the distributional properties of human emotion judgments using a Korean multi-label emotion corpus with 43 categories. Both models produced higher-entropy distributions over more emotion labels. Shared human--model PCA showed that human and model judgments were partly organized by common coarse emotion dimensions. However, model responses were shifted away from human annotations and more widely dispersed in this joint space. Together, these findings suggest that current LLMs exhibit partial structural alignment with human emotion understanding, highlighting the importance of evaluation beyond label-level accuracy.

Keywords: User-centered Design of Robots, Robots in Education, Therapy and Rehabilitation, Child-Robot Interaction
Abstract: Shared reading between parents and children is essential for early literacy development. Literacy programs often focus on supporting families in shared reading through coaching from trained specialists. However, such specialized and personalized coaching requires considerable resources making such coaching inaccessible to most families. Recent work in educational robotics has explored how robots can support children's reading and learning directly but do not consider how robots might support family literacy practices. This paper investigates how literacy specialists coach parents during parent-child reading interactions and how these expert practices can inform the design of robots that support parents during shared reading. Namely, semi-structured interviews and a card sorting activity was conducted with five literacy specialists. The analysis revealed four themes that characterize expert literacy specialist coaching practices: relational protection, conditional intervention, deliberate skill-building, and triadic navigation. The card sorting activity further highlighted areas of agreement and disagreement around future robot coaching behaviors. Based on these findings, we outline design implications for a robot literacy coach that supports parents during shared reading by emphasizing supportive feedback, parent control over coaching behaviors, protection of the parent’s role, and structured instructional support.

Keywords: Child-Robot Interaction, Robots in Education, Therapy and Rehabilitation, LLM/Gen AI-based HRI
Abstract: Social robots are increasingly proposed as educational companions for children, but meaningful educational dialogue depends on something many systems leave undesigned: knowing the child, and sustaining that knowledge across multiple encounters. We formalize Dialogic Educational Child–Robot Interaction (DECRI), a child-centered framework where pedagogical intent and human expertise define the interaction and large language models are used only where they add clear value. We extend DECRI into a multi-session interaction architecture: early sessions elicit a child's user model, and later sessions draw on that accumulated knowledge, justifying a personalized learning material choice, building continuity, and supporting reflective discussion, so that personalization is earned through interaction. Content stays safe and traceable through a hybrid offline authoring-and-validation pipeline: the robot delivers only pre-validated utterances within a rule-based dialogic scaffold, with no generative model accessed at runtime. We examine the framework in Robot Bookworm, a real-school deployment with 51 children (aged 8–11) discussing books with a NAO robot across four sessions, drawing on content-quality metrics and child self-reports.

Results show strong educational grounding, no unsafe content, and evidence consistent with the intended relational and cross-session experience, alongside clear limits: personalization remained bounded and affective content, especially humor, was difficult to assure. We argue that for embodied educational CRI the central design problem is not maximizing generative flexibility but architecting interaction that earns and uses what it learns about the child.

Keywords: Child-Robot Interaction, Robots in Education, Therapy and Rehabilitation, Applications of Social Robots
Abstract: Social robots are increasingly explored in education, yet their pedagogical value remains under-theorized and inconsistently realized in practice. This paper investigates how embodied social robots can support early programming education through the design and deployment of a two-week Cozmo-based curriculum co-developed with teachers and implemented in a real-world learning setting for children aged 7–10. Grounded in embodied cognition, constructionism, and social mediation, we present a mixed-methods study combining qualitative teacher interviews with quantitative engagement and affect measures. Through reflexive thematic analysis, we identify three interrelated mechanisms through which the robot supports learning: embodied feedback, affective engagement, and teacher-mediated social scaffolding. We contribute a conceptual framework of Embodied-Affective Social Mediation, positioning social robots not as standalone teaching tools but as mediating infrastructures that couple physical action, social interaction, and emotional engagement. Our findings show that while embodiment enables immediate cause–effect understanding and affect sustains motivation, the educational value of the system critically depends on teacher facilitation. We also identify key constraints related to technical reliability and scalability. This work provides design implications for integrating social robots into pedagogy and advances a conceptual understanding of how embodiment and affect shape learning in child–robot interaction.

Keywords: Child-Robot Interaction, Explainable Human-Robot Interaction, Trust in HRI
Abstract: Children often overestimate robots’ abilities and form misconceptions, highlighting the importance of explainability in child-robot interaction. Although explainable AI has been widely studied in adult-AI interaction, explainability in child-robot interaction remains under-explored, particularly before the interaction begins. To address this gap, we investigated whether short, neutral pre-interaction introductions to a robot’s capabilities and limitations influence children’s perceptions, trust, and understanding of the robot. We conducted a study with 68 children (ages 7–9) who received either a human-delivered introduction, a robot-delivered introduction, or no introduction prior to engaging in a decision-making learning activity with the QT Robot. The robot explained its decisions using two age-appropriate reasoning styles: decision trees and pattern recognition. Results show that pre-interaction introductions and their source nudged specific beliefs rather than global attitudes. These findings position pre-interaction introductions as a lightweight transparency strategy for calibrating specific misconceptions in child-robot interaction.

Keywords: Robots in Education, Therapy and Rehabilitation, Child-Robot Interaction, Narrative and Story-telling in Interaction
Abstract: Research has consistently demonstrated the effec-tiveness of social robots in supporting children’s language learning. However, there have not yet been adequate studies that systematically examine the role of social robots’ multidi-mensional human-likeness in facilitating children’s learning and perceptions. This study aims to investigate the impact of physi-cal and social human-likeness of social robots on language learning (attention, expressive vocabulary and story compre-hension) and perceptions among young children. Demonstrated through humanoid appearance, movable arms, and a robot backstory, the physical and social human-likeness of two re-searcher-developed social robots was evaluated among 76 sec-ond graders in a Hong Kong primary school, employing a 2 x 2 experimental design. ANOVA results indicated significant physical human-likeness effect on story comprehension and attention. Further analyses using Linear Mixed Models (LMM) revealed an interaction effect between physical and social hu-man-likeness where children using the both physically and so-cially more human-like robot exhibited the highest sustained attention. However, no significant effects were observed for expressive vocabulary and perceptions. The findings contribute to the understanding of the multidimensional human-likeness of social robots among young children and shed light on practical implications of the educational tool in language learning.

Keywords: Child-Robot Interaction, Robots in Education, Therapy and Rehabilitation, Applications of Social Robots
Abstract: This paper presents REMind, an innovative educational robot-mediated role-play game designed to support anti-bullying bystander intervention among children. REMind invites players to observe a bullying scenario enacted by social robots, reflect on the perspectives of the characters, and rehearse defending strategies by puppeteering a robotic avatar. We evaluated REMind through a mixed-methods play-testing study with 18 children aged 9--10. The findings suggest that the experience supported key learning goals related to self-efficacy, perspective-taking, understanding outcomes of defending, and practicing intervention strategies. These results highlight the promise of Robot-Mediated Applied Drama (RMAD) as a novel pedagogical framework to support social-emotional learning.

Keywords: Ethical Issues in Human-robot Interaction Research, Philosophical Issues in Human-Robot Coexistence, User-centered Design of Robots
Abstract: The prevalence of social robots in everyday settings makes it imperative to comprehend how liability should be assigned when they cause harm. Zoomorphic robots are a particularly challenging case due to their similar appearance and behavior to animals, but their technical nature makes them a challenging case. The aim of this study is to investigate the way people allocate responsibility when a robot dog causes harm, and whether their judgments are more closely aligned with product liability or animal liability frameworks. Through a scenario-based survey, participants analyzed five accidents that involved a robot dog and identified who should be held responsible. Our study examined the impact of previous contact with dogs, anthropomorphism tendency, and general liability attitude on a Human Liability Index, and then summarized responses across scenarios. The results indicate that the most significant factor in liability allocation was the general liability attitude. Participants' responses were different from one scenario to the next and were influenced by contextual details, particularly whether the accident appeared to be caused by owner negligence or technical malfunction. Participants often differentiate emotional attitudes toward robot dogs from legal accountability during qualitative responses, which indicate that they assign liability based on the perceived cause of the accident. Overall, people rely on a general view of liability when judging robot dog accidents, but they still adjust their judgments based on the specific situation. The empirical insight gained from this finding on public acceptance of the animal-liability analogy contributes to the ongoing discussion in robot liability research.

Keywords: Philosophical Issues in Human-Robot Coexistence, Innovative Robot Designs, Assistive Robotics
Abstract: Can a robot pray for humans? This is a question that touches the boundaries of both assistive social robotics and the realm of religion and personal spirituality. It is also the core research question of this paper, which presents a novel robot called ClémenCE, whose role is to pray for the users. This robot, the first of this kind, has been placed into a hospital chapel in France and interacted with hundreds of visitors. We present the results of this experiment and discuss the subsequent insights, including ethical aspects, at the frontier of this research field.

Keywords: Anthropomorphic Robots and Virtual Humans, Assistive Robotics, Philosophical Issues in Human-Robot Coexistence
Abstract: With the advancement of Large Language Models (LLMs) and machine perception, social robots operating within the public and private spheres are becoming increasingly lifelike, demonstrating a growing capacity to augment or replace human activity in specific contexts. These sophisticated systems, equipped with affective computing (emotion recognition) and advanced linguistic proficiencies, are already deployed across diverse scenarios, with their prevalence and functional variety expected to expand significantly over the coming decade. In some areas the objective of robotic intervention often extends beyond physical assistance, and frequently aims to substitute high-level cognitive, intellectual, or emotional human functions. This trajectory necessitates a high degree of anthropomorphism. In this study, we define this beneficial, efficiency-enhancing, yet dignity-respecting approach as "positive anthropomorphism". In contrast, current scholarly literature frequently associates anthropomorphism with negative connotations. Within the fields of Human-Robot Interaction (HRI), HRI ethics, and regulatory discourse, the prevailing consensus maintains that "deception", whereby a robot misleads a user regarding its capabilities or its ontological status as a machine, must be avoided. This stance is reflected in extant legal frameworks. Beyond transparency and anti-deception mandates, current data protection regulations also act as barriers to the implementation of positive anthropomorphism. This paper – based on the analysis of the relevant literature – argues that since the partial authorization of positive anthropomorphism is inevitable in the medium term, a more flexible and balanced approach is required in both social acceptance and legal regulation. Consequently, we propose a

Keywords: Ethical Issues in Human-robot Interaction Research, Robot Companions and Social Robots, Creating Human-Robot Relationships
Abstract: The increasing integration of social robots into daily life raises critical privacy concerns regarding the continuous collection of data by the robot. Users may have different expectations regarding their privacy, how the data is collected, used or stored by the robot. Users’ expectations are often influenced by their cultural backgrounds, but also by the regulatory landscape from which they are part of (e.g., European Union, Japan etc.). This study explores through a video-based survey how users in Norway and Japan understand perceived privacy in terms of informed consent in human-robot interaction (HRI) across: different spaces (public vs. private) and various settings (general vs. healthcare-related), and in various scenarios (e.g., guiding, gathering sensitive data such as temperature, the robot collecting data about personal items, or the robot logging data about personal routines), when using different robots (TIAGo vs. Lovot). We use a qualitative approach based on Latent Content Analysis (LCA) to investigate respondents’ expectations regarding perceived privacy, and their perception of (informed) consent. Our findings show that Norwegian respondents preferred explicit and continuous consent regardless of the space and setting, while Japanese respondents showed a higher tolerance for implicit consent and passive data collection, as long as there are transparent mechanisms and guaranteed rights to opt-out or delete their data at a later timepoint. However, in general, both groups require strict and explicit consent in private settings, whereas in public and non-healthcare settings, users accept that the robot collects data for practical benefits, but they reject unnecessary monitoring.

Keywords: HRI and Collaboration in Manufacturing Environments, User-centered Design of Robots, Assistive Robotics
Abstract: Industry 5.0 emphasizes human centric manufacturing, yet autistic individuals remain underrepresented in both the workforce and in industrial robot design processes. To address this gap by centering autistic voices in the design process, we conducted semi-structured interviews with eight autistic adults to understand their perspectives on industrial collaborative robots for assembly tasks. Grounded in Self-Determination Theory (SDT), which examines how autonomy, competence, and relatedness support motivation and well-being, participants completed surveys assessing workplace needs and robot design preferences, then engaged in interviews exploring their envisioned ideal robot collaborators. Interview data were analyzed using thematic analysis. Results revealed that participants view robots as potential enablers but emphasized critical design requirements such as user-directed operation, customizable multimodal interface, precision assistance for challenging tasks, embedded learning support, and predictable behaviors. We discuss implications for designing neuroaffirming industrial robotics that support neurodivergent workers' autonomy, competence, and relatedness.

Keywords: Ethical Issues in Human-robot Interaction Research, Anthropomorphic Robots and Virtual Humans, Trust in HRI
Abstract: As robots increasingly enter the workforce, human-robot interaction (HRI) must address how implicit social biases influence user preferences. This paper investigates how users rationalize their selections of robots varying in skin tone and anthropomorphic features across different occupations. By qualitatively analyzing 4,146 open-ended justifications from 1,038 participants, we map the reasoning frameworks driving robot color selection across four professional contexts. We developed and validated a comprehensive, multidimensional coding scheme via human--AI consensus (kappa = 0.73). Our results demonstrate that while utilitarian Functionalism is the dominant justification strategy (52%), participants systematically adapted these practical rationales that align with existing racial and occupational stereotypes. Furthermore, we reveal that bias frequently operates beneath conscious rationalization: exposure to racial stereotype primes significantly shifted participants' color choices, yet their spoken justifications remained masked by standard affective or task-related reasoning. We also found that demographic backgrounds significantly shape justification strategies, and that robot shape strongly modulates color interpretation. Specifically, as robots become highly anthropomorphic, users increasingly retreat from functional reasoning toward Machine-Centric de-racialization. Through these empirical results, we provide design implications to help reduce the perpetuation of societal biases in future workforce robots.

Keywords: Cooperation and Collaboration in Human-Robot Teams, Curiosity, Intentionality and Initiative in Interaction, Personalities for Robotic or Virtual Characters
Abstract: Robot initiative is a central challenge in multi-party human-robot collaboration. A robot that contributes without being addressed may provide timely support, but it may also disrupt coordination, divide attention, or interrupt turn-taking; a robot that waits to be addressed may preserve human control, but it may also miss opportunities to assist. We investigate this design challenge in a collaborative escape room in which pairs of participants work with a humanoid robot under either a reactive interaction model, where the robot responds only when addressed, or a proactive model, where it listens continuously, contributes autonomously, and periodically re-initiates interaction. We evaluate both models using puzzle-solving performance, interaction frequency, and participant ratings on the Godspeed and RoSAS scales. The proactive model substantially increases interaction frequency, whereas the reactive model shows a descriptively higher overall success rate (92.86% vs. 71.42%). The strongest differences emerge when prior experience and personality are taken into account: participants with LLM experience solve the early puzzles faster in the reactive condition, and participants with prior robot experience show modified evaluations of proactive and reactive interaction as do introverted participants. These findings demonstrate that the effects of robot initiative are simultaneously shaped by users’ prior experience, personality traits and more generally by the needs of the group.

Keywords: Non-verbal Cues and Expressiveness, Robot Companions and Social Robots, Creating Human-Robot Relationships
Abstract: Turn-taking in human conversation is often achieved with short gaps of approximately 200 ms, whereas human-robot dialogue frequently exhibits longer response delays due to speech-processing pipelines. This work investigates whether a simple pre-speech gesture can mitigate users’ perceived delay and improve turn-taking experience without changing physical latency, using the non-humanoid companion robot Vector. In Experiment 1, the pre-speech lift gesture shortened perceived delay and increased perceived onset predictability under identical physical latency. In Experiment 2, which employed a semi-autonomous Wizard-of-Oz (WoZ) setup in a triadic multi-party conversation (MPC), the cue was more often judged predictable and reduced perceived flow disruption. These findings concern perceived timing and conversational flow rather than objective turn-taking behavior, and suggest that simple embodied indicators can make delayed robot responses easier to anticipate.

Keywords: Non-verbal Cues and Expressiveness, LLM/Gen AI-based HRI, Embodiment, Empathy and Intersubjectivity
Abstract: Backchannels are brief listener responses that regulate turn-taking and signal engagement. In human-robot interaction, a robot that remains affectively neutral while a user is speaking may appear less attentive or empathic, even when its end-of-turn replies are appropriate. This paper investigates whether emotional feedback during a user’s speaking turn improves social and affective perception of a conversational robot. We present an emotion-aware backchanneling system with two feedback modes: Immediate Reaction, triggered by short pauses within a user turn, and Aggregate Reaction, delivered at turn end with a brief spoken reply. The system integrates voice activity detection, cloud speech-to-text, and large language model inference within a platform-agnostic architecture linking a backend processing core to a robot-side interface for facial expression rendering and speech output. We evaluated a neutral control, an immediate-plus-aggregate emotion condition, and an aggregate-only emotion condition in a balanced incomplete within-subject study with 33 participants. Mixed-effects analyses showed that the immediate-plus-aggregate condition significantly improved perceived warmth, competence, and emotional responsiveness relative to the neutral baseline, while the aggregate-only condition significantly improved competence. These findings suggest that emotion-aware backchanneling during the user’s speaking turn can strengthen key social and affective evaluations of conversational robots.

Keywords: Multimodal Interaction and Conversational Skills, LLM/Gen AI-based HRI
Abstract: Predictive turn-taking has been shown to improve conversational flow in human--robot interaction, but its application to structured scenarios such as job interview training remains underexplored. This paper presents the use of multimodal predictive turn-taking in robot-mediated job interview training using the Furhat robot. The proposed system integrates Voice Activity Projection (VAP) for acoustic prediction and TurnGPT for linguistic turn-completion estimation within a unified interaction controller, enabling earlier anticipation of turn transitions and more adaptive conversational timing. We evaluated the system in a 2X2 mixed-design user study (N = 22), comparing predictive turn-taking with a fixed silence-threshold baseline across physical and simulated robot embodiments. The results show that predictive turn-taking reduces response latency and perceived interruptions while improving conversational fluency and overall interaction experience. These findings indicate that multimodal predictive turn-taking based on acoustic and linguistic cues improves conversational timing and perceived interaction quality in robot-mediated job interview training.

Keywords: Sound design for robots, Multimodal Interaction and Conversational Skills, Personalities for Robotic or Virtual Characters
Abstract: Sound can be a very effective modality for robot to human communication. However, beyond speech generation, robot sound design remains relatively understudied — especially for zoomorphic robots that may rely more on nonverbal communication. To address this gap, we designed two different sounds — harmonious and dissonant — and ran a within-subjects study (n=31) to explore how modulating these sounds based on a participant's distance from a large quadruped robot affected their responses. Participants completed collaborative physical tasks with the robot while we measured proxemics, subjective perceptions, and heart rate. Participant distances from the robot were significantly affected by whether the sounds were designed to encourage or discourage approach and, notably, the closest approaches were when the robot was silent except for consequential sound from its fans. Proxemic results were task-dependent, appearing only for one of the three tasks. Subjective responses were consistently affected — more harmonious sounds increased perceived warmth ratings while dissonant sounds increased perceived discomfort ratings. Sound did not influence heart rate. Participants' attitudes towards robots (assessed with GAToRS) affected all response types: those with positive personal attitudes rated the robot as warmer across all conditions, while those with negative personal attitudes reported greater discomfort. Strong societal attitudes — whether positive or negative — predicted a closer approach than more moderate views. Notably, attitude effects were of comparable magnitude to sound effects and operated independently of them. Our results show the effectiveness of conscious sound design as well as the importance of considering individual human attitudes in robot--human communication.

Keywords: Cognitive Skills and Mental Models, Personalities for Robotic or Virtual Characters, Long-term Experience and Longitudinal HRI Studies
Abstract: Episodic memory (EM) may be key to intelligent systems, assisting them in daily tasks and personalizing interactions. Yet, existing EM implementations lack the flexibility of human EM, limiting their potential. Indeed, while in humans, EM storage and retrieval are shaped by personality, which influences the meaning and feelings associated with memories, the robotic literature does not provide models that explore how personality affects EM processes or its impact on agent behavior. To address this gap, we propose a flexible EM model, where episodes are recombined through a Large Language Model (LLM). We have integrated the model with a cognitive architecture for robotic personality, capable of managing Conscientiousness, Extroversion, and Agreeableness (CEA) traits in robots and digital humans. This ultimately provides a tool for examining how robotic personality influences memory encoding, storage, and retrieval. The system has been evaluated in a controlled environment, involving a digital human and ten simulated personas in multiple interaction scenarios. We compared seven conditions, including a no-personality baseline and the six extreme poles of the CEA traits, drawing valuable insights into the influence of personality on EM storage, task completion, information encoding, and EM significance.

Keywords: Applications of Social Robots, Motivations and Emotions in Robotics, Machine Learning and Adaptation
Abstract: This paper introduces a novel generative framework for synthesizing expressive robotic routines for the textbf{Haru} social robot. Our approach leverages a textbf{Flow Matching} model to orchestrate motor positions and velocities, integrated with a textbf{Feature-wise Linear Modulation (FiLM)}-based neural network to synthesize synchronized mouth LED patterns. The proposed model enables the concurrent generation of synchronized motor trajectories and mouth LED patterns, conditioned on a continuous emotion space derived from textbf{Plutchik’s wheel of emotions}. By formulating routine generation as a probability path that bridges Gaussian noise with the data distribution, the model facilitates the production of smooth, diverse, and emotionally coherent robotic behaviors, avoiding preprogrammed or hand-animated behaviors that can be repetitive and predictable.

Keywords: HRI and Collaboration in Manufacturing Environments, LLM/Gen AI-based HRI, Linguistic Communication and Dialogue
Abstract: People naturally chat while collaborating and share personal information (i.e., self-disclose) to build rapport and maintain social connections. As robots are increasingly developed to work with people, the effective use of these social behaviors to enhance engagement and support teamwork becomes ever more important. While prior work has shown that robot-initiated small talk can benefit human–robot collaboration, less is known about how best to design such small talk. In this work, we explore how self-disclosure may be designed to support small talk within a human–robot team—especially when the robot is an industrial manipulator that lacks anthropomorphic cues and performs physical work. We first developed an LLM-driven manipulator capable of partaking in small talk, adopting either a low-disclosure or high-disclosure strategy. We then conducted a user study (N = 50) to investigate how self-disclosure in small talk influences human–robot dynamics. Unexpectedly, participants disclosed more in the low-disclosure condition and reported stronger teaming and coordination than those in the high-disclosure condition. This effect was more pronounced among users with prior experience teaming with robots. These results suggest that increasing robot self-disclosure does not necessarily foster rapport, social connection, or reciprocal disclosure; other factors, such as prior HRI experience, should be considered.

Keywords: LLM/Gen AI-based HRI, Personalities for Robotic or Virtual Characters, Applications of Social Robots
Abstract: Large language model (LLM)-integrated social robots have the potential to be adopted in new application domains thanks to recent advances in conversational and behavioural capabilities. Their ability to engage in natural dialogue and adopt distinct personas makes them suitable for applications such as conversational practice, where robots can simulate human interaction scenarios, including socially complex communication contexts. However, deployment in real-world settings requires a clear understanding of user perceptions and preferences, as well as LLMs' capabilities and limitations with persona-driven prompts. In this paper, we evaluated 48 participants’ attitudes towards a social robot embodying four distinct personas (two for career advising and two for diet advising), assessed user experience, and examined whether participants' perceptions of these personas influence their experience. The study used a mixed-subject design, with one group role-playing as career advisors and the other as diet advisors. The findings provide insights into the potential of LLMs to support distinct behavioural traits through careful persona and prompt design, and how these shape user experience and perceptions in the development of student training applications. The results suggested that the users' perceptions were consistent with the intended persona design. The system's usability was rated highly, and statistical analyses indicated that participants' individual traits influenced their perceptions of the system. These results will inform our future work in the novel application area of using AI-powered social robots as communication practice tools for Optometry university students.

Keywords: Assistive Robotics, Robot Companions and Social Robots, Human Factors and Ergonomics
Abstract: Socially assistive robots hold promise for delivering health guidance, yet the impact of robot communication style on user acceptance across the lifespan remains underexplored. We examined how robotic dominance and age-related factors influence the willingness to seek advice from robots during a health-literacy task. Participants received advice from the robot, and we measured their advice-seeking and advice-following behaviors, subjective evaluations of the robot, and psychological status. Results revealed that while advice-following rates were unaffected by dominance level, low-dominance robots were perceived as significantly more competent and sociable, and increased overall advice-seeking willingness. When the high- and low-dominance robots were presented simultaneously (forced choice), younger adults showed a significant preference for the low-dominance robot. Conversely, older adults provided generally higher evaluations for both robots. The only circumstance in which the high-dominance robot was preferred was among individuals relying on disengagement coping strategies, who showed a higher likelihood of selecting it for domestic use. These findings underscore the necessity of personalizing robotic communication strategies based on user age and coping mechanisms to ensure long-term acceptance.

Keywords: Human Factors and Ergonomics, Anthropomorphic Robots and Virtual Humans, Ethical Issues in Human-robot Interaction Research
Abstract: To investigate how the general public will accept this technology in their daily life, and what factors affect such acceptance of humanoid robots, the study conducted an online survey in Japan (N = 900) to investigate the impact of dark personality traits on the social acceptance of humanoid robots by adopting dark tetrad personality traits and the Frankenstein Syndrome Questionnaire (FSQ). The result suggested that sadism had positive impacts on anxiety toward the existence and development of humanoid robots in the society and expectation of humanoid robots. Psychopathy had negative impacts on anxiety toward the usage of humanoid robots in daily life and expectation of humanoid robots. Machiavellianism had positive impacts on all the subscale scores of the FSQ. Narcissism had positive impact on anxiety toward the existence and development of humanoid robots in the society and negative impact on anxiety toward the use of humanoid robots in daily life. The result also implied that persons with high levels of these traits may create confusion regarding the introduction of humanoid robots in the organizations to which they belong.

Keywords: Human–Robot Value Alignment and Safety, Machine Learning and Adaptation, User-centered Design of Robots
Abstract: Personalizing driver-assistance systems is essential for improving usability and user acceptance in human–machine collaborative driving. However, existing methods for incorporating human preferences, such as inverse optimal control and preference-based learning, often require high-quality demonstrations or comparative feedback and frequently overlook safety, limiting their applicability in typical driving scenarios. In this work, we propose a framework for safety-aware personalization of adaptive cruise control using binary feedback that can be provided by non-expert users, making personalization practical in realistic driving settings without pairwise comparisons. To prevent unsafe behaviors during preference-based optimization, we incorporate safety constraints via a control barrier function (CBF) within a model predictive control (MPC) framework. To address the difficulty of obtaining comparative feedback in real driving, we learn a proxy model of human preference from binary feedback (“good” or “bad” labels) and use it to optimize control parameters via Covariance Matrix Adaptation Evolution Strategy (CMA-ES), which is well-suited for noisy human feedback. In simulation with aggressive and conservative user profiles across 200 unseen driving scenarios, the personalized controllers improved the good-label rate from 79.5% to 88.0% and from 75.5% to 83.5%, respectively, while maintaining safe driving behavior without collisions.

Keywords: Machine Learning and Adaptation, Human–Robot Value Alignment and Safety, Non-verbal Cues and Expressiveness
Abstract: Ensuring environmental safety during expressive humanoid motions remains challenging. Traditional obstacle avoidance methods based on reward weighting often leads to conflicts between task objectives and safety constraints and make the training process sensitive to hyperparameter tuning. This paper proposes a humanoid motion control method that combines explicit reference motion tracking with Dual-Advantage PPO (DAPPO) and models potential collision risks using a spatial risk field. A dual-critic structure is adopted to separately learn task rewards and risk costs, thereby decoupling motion imitation objectives from safety constraints. Experimental results show that the proposed method consistently outperforms a PPO baseline in both task reward and tracking reward across multiple random seeds, while enabling the robot to generate environment-adaptive and safety-aware motion behaviors.

Keywords: Multimodal Situation Awareness and Spatial Cognition, LLM/Gen AI-based HRI, Human–Robot Value Alignment and Safety
Abstract: Modern safety-critical systems increasingly rely on human–robot interaction to reduce disaster risk and support decision-making during emergencies. Vision–Language Models (VLMs) are promising for these settings because they can interpret complex scenes and communicate safety-relevant information, but they still require careful evaluation to ensure reliable safety reasoning. In particular, current evaluations often frame danger recognition as a binary decision (Safe/Unsafe), making it unclear whether a model is identifying true physical hazards or merely reacting to unusual scene elements. We address this limitation by introducing an explicit distinction between hazard and anomaly, and by separately recognizing hazardous and anomalous state. We evaluate several state-of-the-art VLMs across two datasets and multiple prompting strategies to test whether this distinction changes model behavior. Our results show that VLMs frequently misinterpret anomalousness as hazardousness, revealing an over-reliance on contextual irregularity as a proxy for danger. We further show that explicitly separating anomaly from hazard provides a more informative evaluation of VLM safety reasoning and exposes failure modes that binary safety judgments can obscure. Our public dataset is available on Roboflow.

Keywords: Sound design for robots, Non-verbal Cues and Expressiveness, Human Factors and Ergonomics
Abstract: This paper investigates how music-based sonification can improve safety in human-robot shared workspaces. We use a calm computing approach to design a music-based sound system that communicates the safety level of a robot without creating distraction or physiological stress associated with current audio notifications. By modifying higher-level music features, we manipulate the tonal tension of lo-fi music, a genre commonly used for working in environments that require focus. Through an experiment with 28 participants performing a dual-task scenario of sorting tiles while interacting with a robot, we compared two sonification approaches: music-based sonification and traditional audio alarms. Participants worked on their primary task while the safety level of a robot in a shared space changed periodically. Both conditions enabled all participants to correctly identify whether a robot was in a safe or dangerous state. Results further showed that participants using the music-based sonification system performed better on their primary task. When traditional alarms sounded, participants briefly pause and were distracted from their primary task, while transitions in the music-based sonification system had significantly less effect on primary task performance. These findings suggest that higher-level music features can communicate safety information without demanding immediate attention, supporting sustained focus over longer periods of time. These findings introduce a novel sonification technique that can serve as an effective tool where information must be conveyed without disrupting ongoing tasks.

Keywords: Motion Planning and Navigation in Human-Centered Environments, Human–Robot Value Alignment and Safety, Multimodal Situation Awareness and Spatial Cognition
Abstract: Robots navigating in public spaces should be able to account for human feedback in social environments. Recent research leverages Large Language Models (LLMs) for semantic reasoning, but their latency makes them unsuitable for the reactive, low-level control loops required to safely navigate around humans. To bridge this gap, we present LiSSEN, a lightweight framework designed to operate in the robot sensorimotor loop, grounding motion trajectories in language through the use of synthetic data. We adopt a Spatiotemporal Rapidly-exploring Random Tree (ST-RRT) method to generate collision-free trajectories in simulated dynamic scenarios. We design an autoregressive multi-modal transformer that converts natural-language instructions and semantically enriched LiDAR observations into collision-free trajectories. The augmented LiDAR provides object differentiation and velocity extraction in addition to distances. Furthermore, a short temporal window is introduced to leverage the temporal attention of the transformer model. We evaluate LiSSEN on collision avoidance and instruction following, demonstrating competitive performance against ORCA and learning-based baselines while generalizing strongly to unseen layouts and paraphrased instructions. An online user study (N = 87) demonstrates that trajectories generated by our model are legible and preferred over a reactive baseline in terms of perceived safety and social acceptability.

Keywords: HRI and Collaboration in Manufacturing Environments, Cooperative Intelligence in HRI, LLM/Gen AI-based HRI
Abstract: In high-risk environments such as disaster response, situational awareness depends not only on detecting hazards but also on communicating them clearly to human operators. Vision–Language Models (VLMs) have shown strong potential for scene understanding in safety-critical settings, yet their value as part of human-facing robotic systems remains underexplored. We present a VR-based Human–Robot Interaction framework for studying how VLM-assisted robots can support situational awareness in simulated hazardous environments. In our system, a robot explores a virtual scene and queries a VLM to identify potential hazards and annotate user-facing points of interest. These annotations are presented to a human operator through an immersive VR interface. This framework enables controlled evaluation of both robotic hazard identification and the communication of safety-critical information to users. Results from our study indicate that the annotated VR interface was preferred over the unannotated baseline and that participants reported high clarity, usefulness, and comfort when interacting with the system. These findings suggest that combining VLM-based robotic perception with immersive visualization is a promising approach for supporting situational awareness in hazardous settings.

Keywords: Explainable Human-Robot Interaction, Novel Interfaces and Interaction Modalities, Multimodal Interaction and Conversational Skills
Abstract: Explaining why a failure occurred is crucial for building user trust and making robots transparent and interpretable. Recent studies have explored multimodal explanations combining textual and visual components to communicate failure causes. However, incoherence between modalities remains a largely overlooked problem, arising from automatic explanation generation methods, such as those produced by large language models, or the independent creation of explanations for different modalities and then presented together. This work examines how two types of incoherence, contradiction and dissociation, impact the explainability of robot failures. We generate coherent and incoherent explanations using three modalities: speech, structured textual plans, and robot perception visualisations. Our findings from an online user study (N=74) indicate that while contradictions reduce perceived correctness and sufficiency of multimodal explanations, these measures for dissociation were nearly indistinguishable from those of the coherent explanations. Incoherence has also altered which modalities participants found most helpful for understanding explanations. These findings highlight the need to prioritize detecting cross-modal contradictions and motivate further study of dissociation as an alternative when contradictions are unavoidable.

Keywords: Trust in HRI, Explainable Human-Robot Interaction, Human Factors and Ergonomics
Abstract: As Maritime Autonomous Surface Ships (MASS) shift operational oversight to shore-based control centres, maintaining appropriate operator trust after system failures has become a key challenge. It remains unclear how different post-violation feedback strategies affect trust repair and cognitive cost. We conducted a between-subject experiment (N=30) in a simulated shore-based autonomous ship monitoring task to examine how apology, causal explanation, and no feedback influenced trust repair and workload after a system failure. The results showed that apology produced higher post-repair situational trust than both explanation and no feedback, whereas explanation provided no advantage over no feedback. Across all conditions, restored system performance was the primary driver of trust recovery, while dispositional trust significantly predicted post-repair trust. These findings do not imply that transparency is generally harmful. Rather, they suggest that real-time technical explanations must be carefully calibrated to operator expertise, message complexity, timing, and task workload.

Keywords: Explainable Human-Robot Interaction, Non-verbal Cues and Expressiveness, Multimodal Interaction and Conversational Skills
Abstract: Effective robot-to-human communication can increase transparency and trust, reduce uncertainty, and contribute to safer collaboration in shared workspaces. Designing and validating an effective robot communication strategy is challenging due to the varying and often limited communication modalities across robots, differences in how diverse recipients interpret messages, and the underexplored virtual-to-real gap in studies of communication legibility. We present a systematic, large-scale comparative validation of existing communication strategies for a mobile non-humanoid robot across message types and settings (online and in-person). Based on the prescribed message types in the existing standards for industrial robots, we realize and compare a low-expressive, unimodal LED-based strategy with a highly expressive, multimodal one that leverages robotic gaze, gestures, and voice. For each strategy, we analyze the communication of a turning intention, an attention request, error status, whether the robot is stuck, and whether it is functioning normally. We evaluate these strategies in replicated online and in-person experiments. We find strong evidence that highly expressive multimodal communication is perceived as more legible and intuitive than unimodal LED-based communication. Comparing the online and real-world study findings, we observe a notable decrease in overall legibility, particularly for signaling with LEDs. Similarly, confidence in message interpretation decreases during the real-world evaluation.

Keywords: Trust in HRI, Explainable Human-Robot Interaction, Cooperation and Collaboration in Human-Robot Teams
Abstract: Technology advancements are leading to a future in which service robots are fully integrated into everyday environments. As these robots transition to being powered by LLMs, their inherently non-deterministic behavior can generate responses that appear correct and well-reasoned, even when they are not, raising challenges for the development of trust and reliance. While explanations have been widely shown to improve user acceptance and appropriate interaction in human-computer interaction, their role in embodied systems, particularly under conditions of uncertainty and error, remains insufficiently explored. In this work, we investigate the effects of natural language justifications on trust, compliance, and behavior in a robot-assisted recycling task using a controlled within-subject design. We find evidence that providing explanations for decisions significantly increases user compliance and reliance, but simultaneously introduces non-trivial trade-offs. Our results indicate that explanations can impair users’ ability to critically assess and override incorrect recommendations. Moreover, their impact on trust over time is not consistent; it varies depending on users’ prior knowledge and the difficulty of the task. Finally, we observed that explanations influence the speed of decision-making, with participants acting faster when explanations were provided. These findings highlight that explanations in embodied AI systems act not only as tools for transparency but as mechanisms that introduce fundamental challenges for designing trustworthy and well-calibrated human-robot interactions.

Keywords: Applications of Social Robots, Non-verbal Cues and Expressiveness
Abstract: While the term may carry negative connotations, robot deception has been shown to elicit beneficial behavioural changes in people when trust is established. This study investigates trust and deception in human-robot interactions, focusing on how individuals recognize and respond to multimodal deceptive behaviours in a robot. In particular, we believe that a robot's confidence can positively influence people's trust and perception, even in the presence of contradictory deceptive behaviour. Eighty participants engaged in a Treasure Hunt game with a deceptive robot across four conditions. Results show that people tend to overtrust robots even when contradictions are detected, though trusting behaviours varied with confidence level. A low confidence induced participants to identify the contradiction as programming error, or a corrective action taken by the robot rather than intentional deceit. Medium confidence caused participants to doubt their own memory and interpret the robot's behaviour as human-like rather than deceptive. High confidence led participants to acknowledge contradictions yet still comply with the robot's instructions.

Keywords: LLM/Gen AI-based HRI, Multimodal Situation Awareness and Spatial Cognition
Abstract: Embodied task planning requires reasoning under uncertainty about world states that are partially observed and incrementally revealed through interaction. However, most vision–language model (VLM) planners operate in a reactive sense–plan–act loop without maintaining a persistent representation of object locations. As a result, they cannot accumulate negative evidence across time (e.g., “the object is not here”), leading to redundant exploration and repeated manipulation failures. We introduce PRISM, Planning with Belief under Incomplete World Models, a lightweight inference-time framework that augments VLM planners with an explicit probabilistic belief over candidate object locations. PRISM maintains a categorical belief over object locations and updates it online via Bayesian conditioning from visual observations and structured action feedback. To reduce initial uncertainty, the belief is initialized from instruction-aware affordance priors and integrated into the planning loop via prompt conditioning and belief-guided action filtering. We evaluate PRISM on EmbodiedBench across EB-Habitat (Habitat) and EB-ALFRED (AI2-THOR) using InternVL2.5-8B and GPT-4o-mini backbones with 50 episodes per task category. On EB-Habitat, average success improves from 21.6% to 42.8% with InternVL2.5-8B and from 30.4% to 47.2% with GPT-4o-mini. On EB-ALFRED, the belief pipeline enables InternVL2.5-8B, which achieves near-zero success under ReAct, to reach up to 30% success across multiple task categories. Beyond performance gains, the belief state provides a transparent uncertainty representation that exposes the agent’s location hypotheses at each step, supporting interpretability for human operators in shared environments.

Keywords: LLM/Gen AI-based HRI, Linguistic Communication and Dialogue, Social Intelligence for Robots
Abstract: Interpersonal relationship estimation is important for socially adaptive dialogue systems. This study investigates the estimation of dominance and formality in dyadic small-scale conversations using a framework based on large language models (LLMs). The proposed method utilizes conversational context and utterance-ending expressions to infer relational attributes under realistic conversational conditions that include recognition errors. Experimental results showed variations across conversational topics in dominance estimation, where dominant and subordinate classes were more frequently misclassified as equal in personal-experience conversations. In contrast, formality estimation consistently reflected differences between relationship types and exhibited relatively stable distributions across topics. Although speaker-mixing and automatic speech recognition (ASR) errors occurred at non-negligible rates, their impact on relationship estimation was limited and did not fully account for the observed topic-dependent misclassification. These findings provide empirical insights into language-based relational estimation in small-scale conversational settings.

Keywords: LLM/Gen AI-based HRI, Cognitive Skills and Mental Models, Multimodal, Multilingual and Multitask Modeling
Abstract: Traditional neural network models for intent inference rely heavily on observable states and struggle to generalize across diverse tasks and dynamic environments. Recent advances in Vision–Language Models (VLMs) and Vi- sion–Language–Action (VLA) models introduce common-sense reasoning through large-scale multimodal pretraining, enabling zero-shot performance across tasks. However, these models still lack explicit mechanisms to represent and update belief, limiting their ability to reason like humans or capture the evolving human intent over long-horizon. To address this, we propose a belief-aware VLM framework that integrates retrieval-based memory by leveraging prior context for improved VLM reasoning. Instead of learning an explicit belief model, we estimate prior belief from a vector-based memory database by combining relevant multimodal context. We further refine decision-making using a reinforcement learning policy over task-contextualized VLM latent space. We evaluate our approach on publicly available Visual Question Answering (VQA) datasets such as HD-EPIC and demonstrate consistent improvements over zero-shot baselines, highlighting the importance of belief-aware reasoning.

Keywords: Cooperation and Collaboration in Human-Robot Teams, Evaluation Methods, User-centered Design of Robots
Abstract: Recent advances in Artificial Intelligence are transforming robots from automated tools into collaborative partners in human–robot teams (HRTs), increasing the need for robust evaluation methods to guide safe, effective, and satisfying interaction. However, evaluating HRT performance remains challenging, as prior studies employ diverse metrics that limit comparability across collaborative settings. This paper presents a systematic review of empirical studies assessing team performance and team fluency in human-centric industrial HRT scenarios. The review distinguishes between outcome-oriented metrics describing what the team achieves and process-oriented metrics capturing how coordination unfolds during interaction. It further examines how these metrics are applied across different task contexts. By clarifying and structuring commonly used metrics and their relationship to task characteristics, this work supports more consistent and comparable evaluation practices. It also lays the groundwork for the development of structured evaluation frameworks for HRTs that explicitly account for key interactive features of collaboration.

Keywords: HRI and Collaboration in Manufacturing Environments, Detecting and Understanding Human Activity, Cooperation and Collaboration in Human-Robot Teams
Abstract: Human Action Recognition (HAR) is frequently investigated in Human-Robot Collaboration (HRC) research to understand what actions have been performed and hence the state of a collaborative task. Accurately tracking an assembly state from HAR is however not fully investigated, and in realistic scenarios is not a trivial task. This research systematically investigates and compares methods for tracking assembly state using action recognition inputs. Investigations using two diverse datasets and five state tracking approaches, including logic-based, Hidden Markov Model (HMM), and neural network (NN) methods, show that optimal approaches are not uniform across different tasks and that different methods fail under different circumstances. Testing is performed using both simulated inputs with varying noise levels and realistic inputs from a HAR model. Results show NN and HMM methods can perform well in tasks with limited variability, but for other scenarios logic-based approaches can be more robust. Methods which model expected action duration are also important for tasks with repeated actions where no additional sensing is provided.

Keywords: Affective Computing, Multimodal Interaction and Conversational Skills
Abstract: Emotion recognition is a key capability in human--robot interaction (HRI), but existing multimodal emotion recognition (MER) methods rely mainly on facial, vocal, and textual cues, which are often unreliable in practical HRI settings. In particular, occlusion and degraded or missing modalities can substantially reduce the performance of models that rely on conventional expressive modalities. This paper investigates human--object interaction (HOI), defined here as the observable interaction between a person and surrounding objects or scene elements, as a complementary modality for emotion recognition in HRI. To this end, we construct an HOI-aware emotion dataset of 464 short video clips, curated from HOIGen-1M, MER2024, and self-collected online videos, annotated with seven discrete emotion categories. We represent HOI using structured semantic descriptions of interaction attributes and integrate this modality into three representative MER baselines, namely Emotion-LLaMA, AffectGPT, and MMA-DFER. For Emotion-LLaMA and AffectGPT, HOI is introduced as additional semantic context in the textual input, whereas for MMA-DFER it is combined with the original model output through an emotion-wise gating-based fusion mechanism. Results show that HOI consistently improves all three baselines and provides particularly notable benefits under blurred-face conditions, while also yielding lower standard deviation across repeated runs in most settings. Overall, we demonstrate that HOI offers complementary cues that can enhance the robustness and stability of emotion recognition for embodied HRI.

Keywords: LLM/Gen AI-based HRI, Cognitive Skills and Mental Models, Embodiment, Empathy and Intersubjectivity
Abstract: Large language models (LLMs) exhibit a systematic empathy imbalance: they over-rely on surface-level emotional reactions while lacking the interpretive depth characteristic of human responses. We propose a preference optimization framework that leverages Theory of Mind (ToM) depth and cognitive appraisal dimensions to construct cognitively grounded preference signals for alignment. On out-of-distribution evaluation data, the aligned model reduced the empathy gap on emotional reaction by 38%, though interpretive gains were limited. Expert- annotated preference pairs produced stronger effects than larger auto-labeled sets, suggesting that annotation quality outweighs data volume. These results offer an initial step toward cognitively grounded empathic dialogue for social robots.

Keywords: Machine Learning and Adaptation, Applications of Social Robots, Detecting and Understanding Human Activity
Abstract: Home service robots must identify users quickly without requiring explicit cooperation. While face recognition is sensitive to viewpoint and distance and fingerprint identification requires physical contact, gait recognition provides a natural and contact free alternative. However, in compact home environments, close range interaction and limited camera viewpoints often prevent full body observation, making many existing gait recognition methods impractical. To address this limitation, we propose a pose landmark driven gait identification framework that focuses on lower body observations and introduce a Leg-Cycle-Aware Dynamic Parent Graph (LCA-DPG). This model enhances lower limb representation through cycle aware dynamic parent modeling, phase aware aggregation, and adaptive gating. Experiments on the CASIA-B74-50 protocol and a realistic Group5 setting show that lower body pose retains strong identity information. LCA-DPG Lower achieves the best overall performance among all compared methods, with consistent gains under appearance variation conditions (BG and CL) while remaining competitive in nominal walking (NM). These results demonstrate that lower body gait is not merely a fallback under partial visibility, but a practical and robust biometric modality for domestic robots operating at close range. Furthermore, cycle aware dynamic structural modeling provides meaningful improvements beyond strong pose driven baselines, especially in the lower body branch.

Keywords: Creating Human-Robot Relationships
Abstract: Target re-acquisition after person loss is a key challenge in indoor robotic person following, where occlusions, abrupt turns, and branching layouts often break continuous tracking. Trajectory prediction methods are limited when target motion changes rapidly, while belief based recovery methods often do not explicitly exploit environment topology. We propose FoV-Aware Belief-Guided Graph Search (FBGGS), which formulates target re-acquisition as active search over a topological graph extracted from an occupancy map, jointly considering belief and topology. FBGGS combines graph belief propagation with an FoV-aware Bayesian negative observation update, allowing the robot to iteratively refine its target belief during recovery. Candidate recovery nodes are ranked using expected detection probability and travel cost. Experiments in Gazebo simulation over four indoor scenarios show that FBGGS achieves the highest recovery success rate and the lowest recovery time among compared baselines under the evaluated conditions. The results suggest that jointly reasoning over graph topology and target belief is effective for indoor target recovery.

Keywords: LLM/Gen AI-based HRI, Linguistic Communication and Dialogue, Multimodal Interaction and Conversational Skills
Abstract: For companion and human-care service robots, a single ASR transcription error can silently redirect a spoken request to the wrong action, eroding user trust that makes social human-robot interaction meaningful. We evaluate whether combining two existing robustness mechanisms, structured N-best ASR uncertainty exposure and constrained one-turn clarification, improves final tool selection before tool execution. Using a 2x2factorial design plus an oracle upper bound, we evaluate five conditions across 22 tool categories on a 440 utterance English benchmark. Across five repeated runs, ASR degradation alone reduces tool-call accuracy by an average of 3.00 percentage points relative to oracle text. The combined N-best + clarification setting reaches a mean of 93.23% final tool success against 91.77% for the ASR 1-best baseline, a mean gain of 1.46 percentage points with positive improvement in all five repetitions. These results show that the combined N-best + clarification setting is an effective and reliable strategy for improving tool-selection reliability of speech-driven companion and home care robots.

Keywords: Human–Robot Value Alignment and Safety, Detecting and Understanding Human Activity, HRI and Collaboration in Manufacturing Environments
Abstract: Vision-Language Models (VLMs) are effective at static image reasoning, but existing VLM-based safety systems remain biased toward static posture recognition (e.g., lying or sitting) rather than dynamic event understanding. In robotics, this distinction is critical because recognizing a state alone is insufficient to determine whether an event has actually occurred. Assistive robots require continuous monitoring and prompt responses to safety-critical transitions, which demands process-level reasoning. However, single-frame inference lacks temporal context, while video-based multi-frame pipelines are often computationally heavy for low-latency, on-device deployment. To address this gap, we propose SFC-VLM (Single-Frame Compression for Vision-Language Models), a framework that transforms temporal reasoning into spatial reasoning by compressing consecutive frames into a single image for event-level understanding. SFC-VLM employs two compression methods: Grid Compression (time represented as spatial order) and Motion Trail Compression (time represented through progressive fade/blur), along with compression-aware prompting that guides the model to leverage spatial reasoning for interpreting temporal sequences. We evaluate SFC-VLM on the Le2i Fall Detection Dataset using frame-level classification at 10 FPS. Grid Compression with a 2×3 grid layout achieves 58.06 F1 score, representing a 40% improvement over the no-compression baseline (41.49), while maintaining 97.00% specificity. Latency analysis shows that Grid Compression achieves an average event response time of 0.465 s and an average detection latency of 0.158 s, suggesting its potential as a low-latency temporal compression approach for robotic perception with only one VLM inference per decision.

Keywords: Linguistic Communication and Dialogue, Machine Learning and Adaptation, Multimodal Interaction and Conversational Skills
Abstract: Low-height mobile robots are acoustically vulnerable in indoor human-robot interaction (HRI) because floor level microphones are strongly affected by reflections, clutter, occlusion, and reverberation. We formulate low-height active listening as a camera-free, map-free, speech-to-text (STT)- oriented online listening-pose search problem that uses user speech itself as a natural probe. The proposed policy-dominant controller uses an interpretable audio-only representation composed of temporal, spectral, and action-aware listening cues, while sharing the same safety-only execution guard as a deterministic heuristic baseline. In a limited-scale physical pilot study, the policy-dominant cold-start controller reached qualified STT-favorable poses in more of the reported runs than the baseline under matched runtime conditions, doing so with lower average motion cost, while maintaining comparable held-out terminal STT quality after arrival. As secondary evidence, early warm-start checkpoint reuse reduced search time and motion cost in matched pairs, though terminal quality remained scene-dependent. These results suggest that speech-derived listening cues can support practical, map-free active listening for compact service robots, while broader validation remains future work.

Keywords: Child-Robot Interaction, Robots in Education, Therapy and Rehabilitation, Non-verbal Cues and Expressiveness
Abstract: A robot's presentation influences perceptions of its traits and roles, yet little is known about how children interpret aesthetic modifications to robots. Aesthetic modification is a practical yet underexplored method for reframing robot perception, allowing a single robot platform to serve different roles and present different traits without requiring additional hardware or technical expertise from teachers and students. In this work, we investigate how aesthetic cues shape children's perceptions of robot traits, classroom roles, and their willingness to listen. Through three activities with children (aged 6-9) examining voice (N=17), decorations (N=24), and costumes (N=11), we find that aesthetic cues shaped trait impressions and classroom role assignments, but that willingness to listen was not significantly affected, instead reflecting personal associations like relationships and individual interests. We present three design guidelines to help educational robot designers use aesthetics to signal traits and roles, and invest in behavioral scaffolding to build authority where aesthetics alone are insufficient.

Keywords: Novel Interfaces and Interaction Modalities, Explainable Human-Robot Interaction
Abstract: Traditional robot programming involves distinct stages of coding, compiling, and testing, creating delays that hinder a user's ability to understand how their code affects robot behavior. We introduce a Responsive Environment for Authoring Code and Testing Robots (REACTR), a live programming environment in which users interact directly with visualizations to manipulate program behavior and receive immediate feedback. REACTR executes code continuously, automatically synchronizing code changes with visual feedback of the robot’s behavior. We perform an evaluation of REACTR (Live + Direct manipulation) and compare it to two other programming interfaces that utilize standard code editing or live feedback. In this study, participants program a robotic manipulator to complete tasks in a chemistry lab setting. The results indicate that live feedback with direct manipulation significantly improves task completion time. Participant feedback further highlights opportunities for advancing interaction through programming.

Keywords: Robots in Education, Therapy and Rehabilitation, Long-term Experience and Longitudinal HRI Studies, Assistive Robotics
Abstract: In recent years, researchers have increasingly deployed social robots designed to enhance motivation, participation, and learning in high school and university settings. Although most robotic studies focus on short-term interventions, typically within one or two sessions, there is a growing interest in developing and evaluating such robotic systems over the long term, as learning requires sustained engagement over time. To understand the educational impact of social robots in long-term settings, this review examines how these systems are implemented and used, how their impact is evaluated, and what strategies are employed to keep students engaged after the novelty effect wears off. We compile findings from eighteen empirical studies that employed multi-session or longitudinal interventions involving social robots in secondary and higher education. The results show that motivational and engagement benefits have been reported more frequently than academic gains. The review highlights several recurring limitations, such as dependence on predefined interaction scripts, restricted behavioural variation, and insufficiently robust adaptive mechanisms. Synthesising these findings, the review identifies key challenges and research opportunities to advance more adaptive and sustainable long-term human–robot interaction in higher education settings.

Keywords: User-centered Design of Robots, Personalities for Robotic or Virtual Characters, Innovative Robot Designs
Abstract: Human-Robot Interaction (HRI) research is impacted by the availability, cost and accessibility of social robots. While commercial solutions are available, these have pre-defined functionalities, behaviours and appearance. We propose a cheaper alternative, Roaming Roombas, a do-it-yourself (DIY) social robot platform which may be adapted to a variety of use cases. Drawing on principles from prior DIY platforms, our system integrates: modularity, through the use of 3D printing and cardboard prototyping techniques; real-time responses, through the coordination of facial expressions with user interactions; movements, through a robotic vacuum cleaner (Roomba); and flexibility, by defining clean software architecture linked to storyboarding and consequently robot behaviours. Rather than presenting a validated system, we position our work as a feasibility study and design exploration demonstrating the platform’s potential for enabling expressive and adaptable DIY robots. We evaluate this potential through a feasibility study in which users design new DIY robots in a series of co-design workshops. This work contributes an accessible and replicable foundation for building low-cost, expressive robots, broadening participation in robot design and supporting inclusive HRI research.

Keywords: User-centered Design of Robots, Robots in Education, Therapy and Rehabilitation, Child-Robot Interaction
Abstract: Participatory design offers a way to involve children in robot design, but many robot design concepts, such as autonomy, responsiveness, and emotional expression, remain difficult for young children to evaluate. This challenge is especially pronounced for preoperational-stage children, who rely on concrete perceptual experience rather than abstract reasoning. To address this gap, we present a framework for translating abstract robot design requirements into child-accessible design decisions. Grounded in developmental psychology, embodied cognition, and participatory design theory, the framework consists of three steps: identifying relevant design dimensions, translating them into concrete perceptual representations, and calibrating the activity structure to reduce cognitive load while preserving child agency. We demonstrate the framework through Build-a-Robot, a game-based participatory design activity for children aged 4--7 in the context of designing a reading companion robot. A case study with 14 children shows how the framework supported participation, sustained engagement, and meaningful preference expression, while also revealing practical considerations for implementation. This work contributes a structured approach for abstract-to-concrete translation in child-robot design and offers guidance for developing participatory design tools for young children.

Keywords: Robots in Education, Therapy and Rehabilitation, User-centered Design of Robots, Child-Robot Interaction
Abstract: This paper presents Parrobot, an affordable social robot designed for human–robot interaction research and educational workshops in a resource- and time-sparse environment. The system integrates ROS2 functional nodes, an HTTP communication server, and a web interface to provide a complete, yet lightweight system from user interaction to motion control.

We describe the system architecture, the design philosophy, and the deployment in workshops with children, demonstrating how such platforms can facilitate early-stage HRI experiments and collaborative robot design with children. The results highlight Parrobot as a viable low-cost tool for inclusive robotics research and education.

Keywords: Explainable Human-Robot Interaction, Trust in HRI
Abstract: When a robot is perceived as agentic, a robot's harmful actions tend to be judged more harshly. However, a robot's perceived agency can fluctuate across interactions and contexts, with important implications for fairness and trust. To better understand these constructs, we conducted a study to understand how shifts in perceived agency influence evaluations of a robot's harmful actions, with a focus on fairness judgments. Using a vignette-based experiment, we found that fairness evaluations varied depending on a robot's perceived agency at each moment. Participants judged the same action as more unfair when the robot was described as highly agentic, and adjusted their fairness judgments if momentary cues about the robot's agency were changed. Beyond fairness, we found that the shifts in perceived agency influenced blame and trust. While blame closely tracked changes in perceived agency, changes in moral and performance trust were less closely tied.

Keywords: Applications of Social Robots, Trust in HRI, User-centered Design of Robots
Abstract: Public spaces increasingly use social robots to provide guidance and customer service, raising ethical and legal concerns around transparency in Human–Robot Interaction (HRI). This paper investigates how transparency levels and users' willingness to learn about security information are associated with personal data sharing during interactions with a robot acting as a city guide in public spaces. We conducted an experiment with 61 participants who interacted with the social robot ARI under low- or high-transparency conditions, choosing between two scenarios and deciding whether to save or delete shared data. No significant differences were found between the two transparency conditions. However, participants who chose to view the optional security information showed higher rates of data saving and lower rates of data deletion than those who did not. Responses to reflection questionnaires indicated that most participants found the privacy risk labels and security information helpful, with 61.5% choosing to view the security information and approximately 70% reporting greater control over their personal data, suggesting that the transparency features supported users' perceived autonomy and trust.

Keywords: Trust in HRI, LLM/Gen AI-based HRI, Ethical Issues in Human-robot Interaction Research
Abstract: What happens when a robot can pay for its own survival? We introduce the Traveling Robot project, a multi-institutional initiative to deploy a physically embodied robot equipped with a digital wallet, blockchain-based smart contracts, and an onboard large language model (LLM) agent into uncontrolled human environments, where it must negotiate and execute face-to-face financial transactions with humans to sustain its travel. Building on our prior work establishing Robonomics, the study of robot-human peer-to-peer financial transactions, and enabled by the recent emergence of production agentic payment protocols, we present: (1) the system architecture integrating an LLM agent, smart contracts, and four wallet governance models with a four-state operational lifecycle; (2) a three-axis autonomy model decomposing robot capability into physical, cognitive, and economic dimensions; (3) an eight-phase agreement lifecycle and transaction taxonomy; (4) a phased experimental protocol with adversarial scenarios grounded in behavioral game theory; and (5) a systematic articulation of the societal questions this experiment probes, from trust formation and moral hazard to the legal personhood of transacting machines. The Traveling Robot tests a fundamental hypothesis: that economic capability changes how humans interact with, protect, exploit, or destroy autonomous robots.

Keywords: Creating Human-Robot Relationships, Affective Computing, Philosophical Issues in Human-Robot Coexistence
Abstract: With the increasing deployment of robots in public spaces, encounters between robots and incidentally copresent persons (InCoPs) are becoming more frequent. However, InCoPs remain largely underexplored in the literature, particularly from a cross-cultural perspective. Therefore, the present study investigates differences in InCoPs' existence acceptance (EA) of autonomous cleaning robots in public spaces among Japanese and German participants. Online survey results revealed that Germans showed significantly higher EA. Social Norms and Trust were the strongest positive EA predictors across cultures. More specifically, for Germans, EA was directly influenced by Usefulness, Interest and Anger, showing a functional-affective pattern where functional perceptions boost EA and anger suppresses it. For Japanese participants, Trust, Surprise and Fear were the direct associational factors, forming a trust-emotion pattern. These findings suggest that the cognitive and emotional drivers of public robot acceptance may vary across cultural contexts, emphasizing the need for adaptive robot design.

Keywords: Robot Companions and Social Robots
Abstract: This anthropological study examines two different hospitality settings in Japan: a café and an airport that have implemented robots in their daily work. Using concepts of staged and unstaged settings, we observed robots and interviewed staff on how workplaces with a staged setting, people going to a robot café, differed from an unstaged setting, an airport, where experiencing robots were not expected by visitors. We draw on social scientific theories in our analysis, namely imaginaries and impostering, to examine how expectations shape the integration of robots into the workplace, both in staged and unstaged environments. Through a thematic analysis, the findings show considerable differences between staged and unstaged settings, including the dismissal of robots by staff in unstaged settings and the more apparent positive perceptions and integration of robots in the workflow in staged settings. By analysing how robots potentially are integrated into existing workflows, this paper highlights the importance of anthropological methods within HRI, particularly for understanding the gap between imagined functions and everyday practices.

Keywords: Methodologies in Social Robotics, Applications of Social Robots
Abstract: Mobile robots are increasingly deployed in commercial, institutional, and public settings. In these environments, some of the challenges that accompany real-world use stem less from what a robot can do than from ongoing deployment demands like monitoring, updating, and governance. These operational demands are shaped by early design and commercialization choices, though such choices may not be discernible in a robot's final form. This paper draws on semi-structured interviews with sixteen senior engineers at established robotics companies to examine how industry professionals reason about these design choices. Triangulating these interviews with a qualitative content analysis of public-facing materials from twenty mobile-robot manufacturers, we trace how industry design reasoning prefigures the kinds of oversight that later become feasible or difficult. From these data, we identify recurring reasoning patterns organized along two dimensions: (1) flexibility versus specialization, contrasting robots optimized for narrow, repeatable functions with platforms designed to be repurposed across settings; and (2) product versus system orientation, distinguishing stand-alone deployments designed for minimal vendor dependence from integration-heavy deployments that embed robots in broader software, logistics, and operational infrastructures. We operationalize these dimensions as an analytic framework linking design reasoning to downstream oversight needs such as monitoring capacity, update governance, responsibility allocation, and the production of deployment-relevant evidence. Making these patterns explicit can help HRI researchers and host institutions anticipate where friction, opacity, or accountability gaps are likely to arise in shared human–robot environments.

Keywords: Embodiment, Empathy and Intersubjectivity, Robot Companions and Social Robots, Robotic Etiquette
Abstract: Everyday voice-based agents are often feminized and are increasingly exposed to inappropriate or aggressive user behavior. While most agents respond with polite and accommodating strategies during conflict, it remains unclear how alternative response styles are perceived, particularly by observers. In a within-subjects video study, participants observed a female-voiced humanoid robot, Ameca, being verbally mistreated after a misunderstanding, with the robot responding either apologetically or assertively. Results show that the apologetic response was perceived as warmer and more effective in resolving the conflict, whereas the assertive response elicited higher negative affect and perceived mistreatment. Individual differences showed only limited effects, with a compromising conflict style associated with more favorable evaluations of the apologetic response in terms of communication and social skills. These findings suggest that polite and accommodating responses may remain socially effective in situations of mistreatment, but also raise important design questions for feminized robots. Consistent accommodating robot responses may reinforce submissive norms, whereas assertive responses may challenge these expectations but risk being perceived as more conflictual. This work highlights the need for robot response strategies that balance de-escalation, boundary-setting, and socially responsible interaction design.

Keywords: Evaluation Methods, Methodologies in Social Robotics, Social Intelligence for Robots
Abstract: Humanoid robots evoke ambivalent attitudes, combining perceived benefits with concerns such as lack of control and human replacement. Prior research on attitudes towards robots has largely focused on present-day robot scenarios. Less is known about how people evaluate robots across different temporal expectations. To close this research gap, we conducted an online survey to investigate ambivalent attitudes towards current and prospective scenarios of humanoid robots. Fifty participants evaluated six robot scenarios across different use cases. Using linear mixed-effects models, we explored how media exposure, robot-related knowledge, general attitudes towards robots, and the expected time of occurrence of the scenarios relate to positive and negative robot evaluations. Results revealed that positive attitudes were associated with more favorable general attitudes towards robots, lower exposure to robot-related news, higher prior knowledge about robots, and a more proximal expected time of occurrence. Negative attitudes were associated with less favorable general attitudes towards robots, lower robot-related knowledge, and a more distant expected time of occurrence. Taken together, the results highlight the importance of considering temporal distance and ambivalence when studying perceptions of humanoid robots.

Keywords: Anthropomorphic Robots and Virtual Humans, Social Presence for Robots and Virtual Humans
Abstract: As household robots such as ‘roombas’ become more commonplace, there is a risk that their users will anthropomorphise them in ways that perpetuate domestic gender stereotypical roles, despite their non-humanoid design. Further, social media aesthetic trends may cause gender perceptions of robots to vary based on accessorising. This study investigates whether the function of a household robot and the presence and placement of bows on the robot impact how gender identities are ascribed and impact the naming and associations made with the robot. We find that the presence of a bow significantly increases the ascription of feminine traits to the robot, regardless of placement. We also find that domestic robots are perceived as masculine by default, but accessorising can change this perception for some robots. Our work indicates that domestic gender stereotypes are not perpetuated on the basis of domestic robot function, but any bow can increase the perception of feminine identity traits.

Keywords: Explainable Human-Robot Interaction, Anthropomorphic Robots and Virtual Humans
Abstract: A robot's appearance is a known factor influencing user's mental model and human-robot interaction, that has not been studied in the context of its influence in expected robot explanations. In this study, we investigate whether and to what extent the human-like appearance of robots elicits anthropomorphism, which is conceptualised as an attribution of mental capacities, and how the level of anthropomorphism is revealed in explanations that people expect to receive. We designed a between-subject study comprising conditions with visual stimuli of three domestic service robots with varying human-like appearance, and we prompted respondents to provide explanations they would expect to receive from the robot for the same robot actions. We found that most explanations were anthropomorphic across all conditions. However, there is a positive correlation between the anthropomorphic explanations and human-like appearance. We also report on more nuanced trends observed in non-anthropomorphic explanations and trends in robot descriptions.

Keywords: Non-verbal Cues and Expressiveness, Anthropomorphic Robots and Virtual Humans, Embodiment, Empathy and Intersubjectivity
Abstract: Humanoid robots are increasingly endowed with expressive capabilities; however, most existing implementations rely on articulated motions or audiovisual channels. By contrast, fluid-based physiological expressions in humanoid embodiments remain largely unexplored. This study investigates tear expression as a fluid-based expressive modality and presents Tearoid, a tear-expressing humanoid head prototype developed for controlled tear generation. Inspired by the human lacrimal system, Tearoid implements a simplified fluidic mechanism consisting of upper inflow, lower drainage, and a tear accumulator, enabling visible tear expression through regulated water flow. To support visually recognizable droplet generation, tear flow is controlled using servo-driven pinch valves, and the control parameters are adjusted based on the relationship between the valve opening duration and water output. The system successfully generated visible tear droplets in the eye region, demonstrating the feasibility of integrating tear expression into a humanoid platform as a fluid-based expressive modality. Through a user study, we examined how the presence and form of tear expression affect perceived emotionality and human-likeness. Our findings suggest that tear expression functions not only as a visual feature but as a socially meaningful cue that reshapes the interpretation of the robot’s internal state.

Keywords: Applications of Social Robots, Social Presence for Robots and Virtual Humans, Creating Human-Robot Relationships
Abstract: As robots become increasingly integrated into everyday environments, people are likely to experience more than a single encounter with one robot. In such contexts, responses to a robot may depend not only on its absolute attributes but also on how it is perceived relative to previously encountered robots. Drawing on anchoring theory from psychology, we examined whether the impact of a robot’s height is relational and shaped by prior interactions. We focused on the robot’s height, as this attribute has been shown to influence humans’ responses to robots, particularly their compliance with a robot’s request. We used a modular service-table robot that could be configured as either Short or Tall. Participants experienced two consecutive interactions. The first Anchor robot delivered materials required for the experiment. After participants completed a computerized task without a robot present, another Target robot approached them and made a voluntary request to complete a 300-item questionnaire. We measured compliance with the Target robot’s request as a function of its height (Short/Tall) and its relation to the Anchor robot (Same/Different). Results revealed that compliance was shaped by the relative context rather than by absolute height alone. Participants complied more with a Short Target than with a Tall Target. Critically, this effect was significantly amplified when the Target robot followed a robot of a different height. These findings suggest that height effects in HRI are relational and subject to anchoring effects across sequential encounters, highlighting the need to move beyond single-robot paradigms and consider how the relative perception of design features shapes robots’ social influence.

Keywords: Degrees of Autonomy and Teleoperation, Cooperation and Collaboration in Human-Robot Teams, Detecting and Understanding Human Activity
Abstract: Teleoperated fragile object manipulation is a critical capability in high-stakes domains such as remote hazardous-environment operations and surgical robotics, where perceptual uncertainty can lead to irreversible damage. To understand the perception-control correlation, we present a dataset for analyzing human behavior in this setting, comprising 221 trials from 11 participants manipulating objects with 20 visually encoded fragility levels (50--1000~gf), indicating the maximum force can be exerted onto the object. The dataset includes synchronized multimodal observations: robot joint trajectories, gripper force signals, multi-view RGB video, users' perceived object fragility, confidence ratings, and trial outcomes. Our analysis reveals key behavior patterns: a 48% overall success rate under challenging fragile object constraints, generally under-estimating bias when perceiving object fragility, weak correlation between perception confidence and accuracy, substantial early-contact failures, and failure due to suboptimal control. We further show that pre-contact kinematics encode users' perception of object fragility while exhibiting strong user-specific baselines. We release this dataset as an open platform to support the development of perception-aware shared autonomy, where systems explicitly model operators' internal estimates of object properties rather than treating observable actions alone as ground truth. This enables research in user behavioral profiling, personalized user modeling, proactive intervention design, and adaptive robot control for damage-sensitive manipulation.

Keywords: Anthropomorphic Robots and Virtual Humans, Explainable Human-Robot Interaction, Embodiment, Empathy and Intersubjectivity
Abstract: Teleoperation is a promising technology in applications such as telenursing and search & rescue missions. Nevertheless, there are significant challenges when human-robot control is performed on remote sites with variable network latency. In recent years, network development has improved reliable low-latency interactions. However, since end-to-end system latency is affected by various factors, including sensing and motor actuation, in addition to network latency, we need to understand how latency affects task performance and user experience. In this paper, we study how users manipulate objects using a robotic system under various latency conditions and analysed both objective performance and subjective experiences regarding perceived delay, difficulty, agency, and embodiment. We present a teleoperation set-up consisting of a robotic arm and an anthropomorphic robotic hand that follows the operator’s hand movement using real-time motion capture. 45 subjects completed two standardised manipulation tasks while receiving direct natural visual feedback. We isolated the effect of delay in the command-execution pipeline on task performance and user experience by injecting controlled latencies from 0 ms up to 200 ms to the base system latency of 220 ms. Higher latencies significantly degraded task performance and increased perceived difficulty, particularly in the fine dexterity task. We also observed that participants performed better over time, showing a clear learning effect. These results highlight the importance of minimising delays in teleoperated interactions.

Keywords: Multimodal Situation Awareness and Spatial Cognition, Virtual and Augmented Tele-presence Environments, Cooperation and Collaboration in Human-Robot Teams
Abstract: This paper presents a bandwidth-efficient teleoperation system that enables effective environmental perception and object manipulation in CBRNe (Chemical, Biological, Radiological, Nuclear, and Explosive) environments under constrained communication conditions. The proposed framework combines a mobile robotic platform equipped with a manipulator and perception sensors with a virtual reality operator interface developed in Unity. Instead of streaming raw video, the system transmits compact spatial and semantic information to reconstruct a simplified yet informative scene representation. Object size and pose are estimated using 6D pose estimation to support reliable manipulation of potentially contaminated objects. Experimental results demonstrate that the proposed approach maintains operational awareness while reducing bandwidth requirements by up to 15 times compared to standard RGB video streaming.

Keywords: Degrees of Autonomy and Teleoperation, Programming by Demonstration, Machine Learning and Adaptation
Abstract: Dynamic camera control is essential for effective teleoperation, yet it often competes with manipulation for operator attention, increasing cognitive load and disrupting task flow. We present POVFormer, a transformer-based framework that learns autonomous point-of-view control from demonstrations of coordinated manipulator and camera motion. The model is trained on data collected in a simulated dual-device Jenga task and evaluated under manual, autonomous, and fixed camera control conditions. Results from a simulation study involving 7 participants show no significant difference in task completion time, while the autonomous system improves manipulator visibility and is rated as comparable or superior to manual control in orientation accuracy and ease of use.

We further validate the approach on an embodied teleoperation system using a low-cost robotic platform equipped with a novel vortex gripper. A physical user study with 5 participants produced similar findings, showing consistent performance and reduced cognitive load as measured by NASA-TLX scores. These results support transformer-based shared autonomy for task-synchronous hand-eye coordination in teleoperation.

Keywords: Explainable Human-Robot Interaction, Degrees of Autonomy and Teleoperation, Novel Interfaces and Interaction Modalities
Abstract: Robot failures are inevitable and often require human review so that underlying issues can be resolved. This is a challenging task for human supervisors, who must interpret data split across multiple modalities, with limited context for the situation at hand. Performing accurate and efficient failure attribution typically requires purpose-designed interfaces and procedures. We propose an approach that integrates models learned from limited data with feature-level interpretability tools to accelerate the review of multimodal data, highlighting relevant times and modalities in a review interface. We implement this method in a warehouse shelf-picking workcell setting using community-standard visualization tools. In user study with twelve participants, we observe that the assistance improves overall labeling recall and shows a trend toward reducing cognitive workload. We discuss participants' strategies for using the assistance, and suggest future directions for developing context-aware robotics visualization tools.

Keywords: Novel Interfaces and Interaction Modalities, Degrees of Autonomy and Teleoperation, Evaluation Methods
Abstract: The increase of drones in industry and domestic use has made human-drone interaction a significant focus in user interaction research. However, drone interaction design still relies heavily on traditional controllers like joysticks and mobile apps, with limited exploration of natural modalities such as voice. We conducted a within-subject study (N=25) comparing voice commands with a traditional mobile app touch interface, in both collocated and remote settings. Our findings provide insights into user experience and interaction design beyond touch. Participants found the mobile app more responsive, but favoured the minimal device interaction and continuous focus on the drone offered by voice. We highlight the limitation of mobile apps requiring frequent attention switching between touchscreen and drone. We also present experimental design guidelines to aid researchers and designers in exploring different interaction modalities. Our findings and design considerations aim to enhance human-drone collaboration and contribute to the growing field of drone interaction.

Keywords: LLM/Gen AI-based HRI, Assistive Robotics, Robots in Education, Therapy and Rehabilitation
Abstract: Physically Assistive Robots require personalized behaviors to ensure user safety and comfort. However, traditional preference learning methods, like exhaustive pairwise comparisons, cause substantial physical and cognitive fatigue for users with severe motor impairments. To solve this, we propose a low-burden, offline framework that translates unstructured natural language feedback directly into deterministic robotic control policies. To safely bridge the gap between ambiguous human speech and robotic code, our pipeline uses Large Language Models (LLMs) grounded in the Occupational Therapy Practice Framework. This clinical reasoning decodes subjective user reactions into explicit physical and psychological needs, which are then mapped into transparent decision trees. Before deployment, an automated "LLM-as-a-Judge" verifies the code's structural safety. We validated this system in a simulated meal preparation study with 10 adults with paralysis. Results show our natural language approach significantly reduces user workload compared to traditional baselines. Additionally, occupational therapists confirmed the generated policies are safe and accurately reflect user preferences.

Keywords: HRI and Collaboration in Manufacturing Environments, Human–Robot Value Alignment and Safety, Motion Planning and Navigation in Human-Centered Environments
Abstract: Robots operating in human environments must not only ensure physical safety but also exhibit behaviors that are understandable, fluent, and acceptable to human partners. This paper investigates motion generation strategies that combine safety guarantees with interaction quality considerations, such as motion smoothness and human comfort. While the design of robots capable of ensuring safety in shared human-robot environments has enabled closer and more advanced forms of interaction, these new proximity-based tasks require moving beyond purely technical considerations. In particular, robot behavior must also be addressed from psycho-cognitive and social perspectives. In this context, we argue for the relevance of integrating social-aware motion control into robotic systems. First, we identify the motion parameters that influence human perception and operator experience. Then, we implement a Model Predictive Control (MPC) framework that generates four distinct socially-informed robot behaviors. Finally, we conduct a user study to evaluate and validate these behaviors and assess their social impact on non-expert participants. The results demonstrate that variations in robot behavior significantly affect the perceived social acceptability of the system. These findings highlight the importance of incorporating human-centered considerations into motion generation strategies for robots operating in shared environments.

Keywords: Medical and Surgical Applications, Cooperation and Collaboration in Human-Robot Teams, Robots in Education, Therapy and Rehabilitation
Abstract: Trocar insertion is a critical yet high-risk step in pars plana vitrectomy (PPV), where unstable tool motion and post-puncture overshoot can compromise surgical safety. This paper presents a comparative study of bimanual ophthalmic trocar insertion, evaluating manual insertion by a surgeon versus cooperative robot-assisted insertion using the well-established steady-hand eye robot (SHER) platform. To ensure a fair comparison, we first standardize the trocar insertion procedure and develop a dedicated data collection pipeline for both insertion modalities. Task-specific metrics are further introduced to quantify surgical safety, tool motion quality, and bimanual asymmetry. In the implemented series of ex vivo porcine eye experiments, robotic assistance improved overall insertion safety and tool motion stability compared with manual insertion, particularly by mitigating post-puncture overshoot. Robot-assisted operation also altered the interaction dynamics during scleral puncture by reducing tool orientational drift and revealing hand-dependent differences in cooperative manipulation. These findings demonstrate the potential of cooperative robotic assistance to improve the precision, stability, and safety of ophthalmic trocar insertion, while providing new insight into human-robot interaction in bimanual microsurgical tasks.

Keywords: Degrees of Autonomy and Teleoperation, Medical and Surgical Applications, Explainable Human-Robot Interaction
Abstract: Tele-ultrasound systems based on bilateral architectures can provide the physician with both visual and haptic information, but their stability is compromised by communication delays. To address this issue, passivity-based control strategies such as the Time Domain Passivity Approach are commonly adopted at the cost of system transparency. In this context, the present study investigated whether the introduction of TDPA in a bilateral tele-ultrasound architecture affects ultrasound (US) image quality during renal acquisitions. Experiments involved one physician and 14 healthy volunteers. US image quality was evaluated through three objective image-based metrics, namely Tenengrad, entropy, and frame-to-frame SSIM, and through subjective questionnaires addressing perceived image quality, workload, and usability. Results showed that the passivated architecture satisfied the non-inferiority criterion relative to a non-delayed architecture for all three image-based metrics, indicating that TDPA did not degrade US image quality beyond the empirically defined acceptable margins. However, subjective evaluation showed higher workload and lower usability, but also a more favorable perceived evaluation of the acquired US images. Overall, these findings suggest that TDPA can preserve image quality while improving perceived image stability, despite a less favorable user experience.

Keywords: Medical and Surgical Applications, Assistive Robotics, Innovative Robot Designs
Abstract: In robotic surgery, assistant support is limited to point-contact holding by forceps, hindering stable organ retraction and complex field exposure by the assistant alone. Therefore, by analyzing HALS hand movements, we developed a 15 mm port-compatible, 3-fingered, 7-DOF robotic forceps. This device consolidates human hand functions to replicate five core motions: power grasp, precision grasping, hook grasping, spreading, and retraction. Basic motion force measurements and simulated partial nephrectomy tasks demonstrated that the proposed hand provides sufficient output for surgical support. Furthermore, it can independently assist the surgeon by coordinating with the da Vinci Surgical System. This system has the potential to reduce surgeon workload and shorten operative times.

Keywords: Cooperation and Collaboration in Human-Robot Teams, HRI and Collaboration in Manufacturing Environments, Human Factors and Ergonomics
Abstract: This paper presents the design and evaluation of a hybrid human--robot collaborative workcell for automotive assembly. The proposed system integrates a hyper-redundant robotic platform, consisting of a UR10e collaborative manipulator mounted on a KUKA KR150, a sensing architecture based on proximity and tactile sensors (ProxySKIN) integrated into the cobot to ensure safe human--robot interaction (HRI), and a passive upper-limb exoskeleton. Operator experience is assessed through a multidimensional human-centered approach encompassing cognitive workload measured using the NASA Task Load Index (NASA-TLX), technology acceptance, usability measured through the System Usability Scale (SUS), and qualitative feedback. The proposed assembly scenario was evaluated in a user study with ten industrial workers across six operational conditions. The results suggested that the exoskeleton may redistribute rather than reduce workload, with higher Effort and Physical Demand scores observed in exoskeleton-related configurations while perceived performance remained stable. The findings also pointed to a gap between functional and affective acceptance, as the exoskeleton was perceived as useful but less satisfying and less usable. In contrast, the cobot was consistently perceived as intuitive, safe, and non-intrusive, and its combination with the exoskeleton improved perceived usability despite increasing coordination demands. Overall, these findings highlight the importance of evaluating assistive technologies at the level of the entire workcell through a multidimensional human-centered approach.

Keywords: Trust in HRI, Non-verbal Cues and Expressiveness, Anthropomorphic Robots and Virtual Humans
Abstract: Anthropomorphism is a key factor for trust in Human-Robot Interaction (HRI). Yet little is known about how anthropomorphic cues interact with behavioural experience of a robot's actual trustworthiness. In this study, we examined the effect of two anthropomorphic features (voice and joint attention) on implicit and explicit trust attributions to a NAO robot, which was programmed to behave either trustworthily or untrustworthily in an iterated investment game. Participants' investments in the game served as a behavioural measure of trust, while follow-up questionnaires assessed perceived robot trustworthiness, likeability, and anthropomorphism. Results showed that participants overall preferred a synthetic, congruent voice for the robot, but invested more money with the natural-voiced robot when the robot behaved untrustworthily. This suggests that voice sets expectations against which subsequent behaviour is evaluated, with less trust being placed in the robot when those expectations are violated. Joint attention had no effect on trusting behaviour, but influenced implicit questionnaire ratings after the game. Implicit and explicit measures of trust diverged, suggesting that they capture different aspects of trust. These findings highlight the importance of congruent, context-sensitive voice design in HRI, taking into account that initial expectations of a robot's behaviour based on voice influence people's trusting behaviours over time.

Keywords: Robots in Education, Therapy and Rehabilitation, Trust in HRI, Anthropomorphic Robots and Virtual Humans
Abstract: Socially assistive robots (SARs) are increasingly deployed in educational and information-sharing contexts, supported by advances in large language models that enable fluent real-time interaction. Despite the growing diversity of robot embodiments, it remains unclear whether a single robot appearance is appropriate across different interaction tasks or whether trust depends primarily on contextual factors. In this study, we examine how robot appearance and task type jointly influence trust in robots. Using a within-subjects videobased experiment (N = 81), participants evaluated three robots with distinct appearances while performing three educationally relevant tasks: teaching, procedural instruction, and personalinformation discussion. Results from repeated-measures analyses show a strong main effect of task on trust, with participants reporting the highest trust during instructional guidance, moderate trust during teaching activities, and significantly lower trust when robots requested personal information. In contrast, robot appearance showed no significant main effect, and the interaction between appearance and task was marginal. These findings suggest that trust in human–robot interaction is shaped more strongly by task context than by physical embodiment alone. By focusing on future educators as end users, this work contributes empirical evidence toward task-aware robot deployment in educational environments and highlights the importance of aligning robot roles and behaviors with interaction goals rather than relying solely on anthropomorphic design.

Keywords: Applications of Social Robots, Motivations and Emotions in Robotics
Abstract: How people interpret robot errors may depend not only on the observed behavior, but also on the robot’s morphology and on whether the error is framed as unintentional or intentional. Prior works showed that human-like robots can elicit stronger mind attribution and deceptive robot behavior can encourage intentional interpretations, but their joint effect on the same erroneous action remains unclear. We addressed this question in an online video-based study in which participants (N = 71) watched four robots differing in human-likeness (within-subject factor: Temi, Pepper, Miroka, and Ameca) produce the same incorrect response in an escapegame scenario. The error was framed either as a failure or as an intentional error, namely deception sustained by simulated social motives. Quantitative results showed that Ameca received higher mind attribution ratings than the other robots, while deceptive framing increased mind attribution across all robot morphologies. No interaction emerged between morphology and framing. Thematic analysis showed that unintentional errors were mainly interpreted in terms of compliance, programming, and system limitations, whereas deceptively framed errors elicited more social and relational explanations.

Keywords: Non-verbal Cues and Expressiveness, Affective Computing, Linguistic Communication and Dialogue
Abstract: Robots operating in human-centered environments frequently encounter interaction errors due to limitations of underlying machine learning models, potentially negatively affecting user trust and adoption. Therefore, detecting when the models fail is critical for robust HRI systems. Existing approaches mainly rely on model uncertainty (confidence) or supervised classifiers requiring extensive labeled data. In this work, we explore humans' affective reactions as implicit feedback signals for error detection by a robot. We develop a system to extract facial action units during situated verbal interaction with a robot and analyze data collected from 30 participants. Despite some individual variability in the reactions, a consistent pattern emerged across participants: lid tightening and left outer brow raising were more frequent during error situations. Although automatic speech recognition confidence scores provide a strong baseline for word-level error detection, combining them with facial cues significantly improves performance. Visual feedback of transcribed speech by the robot elicits more informative facial responses than spoken feedback. These findings highlight the benefit of combining implicit human feedback with model-based uncertainty for more accurate error detection, supporting the development of adaptive recovery mechanisms.

Keywords: Explainable Human-Robot Interaction, Trust in HRI, Human–Robot Value Alignment and Safety
Abstract: Autonomous vehicles increasingly require effective communication with passengers, particularly in critical situations where unexpected decisions occur. In such contexts, providing explanations can help users understand system behaviour and calibrate their trust. In this paper, we investigate how passengers' satisfaction with explanations varies across different levels of situational risk, ranging from routine driving to near-accident and high-risk scenarios. We further examine the role of textit{visual information availability}, comparing conditions in which passengers either have access to the same visual input as the vehicle or are deprived of it. We conducted a mixed-design study with 420 online participants, in which situational risk (low, medium, high) was manipulated within subjects, and explanation type (causal vs. intentional) and visual information availability (available vs. unavailable) were between subjects. Our findings show that when passengers have access to visual information similar to the vehicle, they tend to prefer intentional explanations. In contrast, when such information is unavailable—particularly in high-risk scenarios—they significantly favour causal explanations. These results highlight that explanation effectiveness is context-dependent, shaped by both situational risk and information asymmetry, and provide design implications for explainable autonomous systems.

Keywords: Assistive Robotics, Multimodal Situation Awareness and Spatial Cognition, Machine Learning and Adaptation
Abstract: Hand gestures are a primary output of the human motor system, yet the decoding of their neuromuscular signatures remains a bottleneck for basic neuroscience and assistive technologies such as prosthetics. Traditional human-machine interface pipelines rely on a single biosignal modality, but multimodal fusion can exploit complementary information from sensors. We systematically compare linear and attention-based fusion strategies across three architectures: a Multimodal MLP, a Multimodal Transformer, and a Hierarchical Transformer, evaluating performance with unimodal and multimodal inputs on two publicly available datasets: NinaPro DB2 (sEMG and accelerometer) and HD-sEMG 65-Gesture (high-density sEMG and force). Across both datasets, the Hierarchical Transformer with attention-based fusion consistently achieved the highest validation accuracy of 97%, surpassing the multimodal and best single-modality linear-fusion MLP baseline by over 10% on NinaPro DB2 and 3.7% on HD-sEMG. To investigate how modalities interact, we introduce an Isolation Network that selectively silences unimodal, cross-modal, or individual token attention pathways, quantifying each group's contribution to downstream decisions; on NinaPro Exercise B this enables removal of 62.5% of sensors with only a ~2% reduction in validation accuracy, permitting a smaller model size, faster inference times, and reduced hardware complexity. Further ablations reveal that cross-modal interactions contribute approximately 30% of the decision signal across transformer layers, highlighting the importance of attention-driven fusion and providing a causal, layer-resolved guide for sensor selection and practical model design for neuromuscular medical applications.

Keywords: Machine Learning and Adaptation, Assistive Robotics, Medical and Surgical Applications
Abstract: This work addresses key limitations in state-of-the-art hand-landmark detection to enable reliable, gesture-based human-robot interaction in healthcare settings. Current gesture recognition applications fail to account for the unique characteristics of clinical scenarios, such as the requirement for healthcare professionals to wear medical gloves throughout critical stages of patient care. To mitigate this domain gap, we introduce the novel medical-glove dataset MedGlove1k and demonstrate its effectiveness in adapting hand-landmark detection models to the medical domain, while meeting crucial requirements for low latency and edge-computing deployment. After fine-tuning on MedGlove1k, our YOLO-based detection model achieves a pose mAP₅₀ of 97.4% while maintaining an inference speed of ≈ 200 frames per second, indicating an improvement of 42.4% in pose estimation performance over non-domain-adapted baselines. We further validate our approach under edge-deployment conditions, observing full-architecture latencies of 330.02 ms on an NVIDIA Jetson Nano. We demonstrate the system’s practical viability by successfully integrating it into a robotic platform for real-time human-robot interaction, achieving a 96.8% success rate for end-to-end gesture-based robot control. As a foundation for future applications in healthcare settings, we make our MedGlove1k dataset publicly available and provide our detection architecture as a ROS2-based landmark and gesture recognition framework.

Keywords: HRI and Collaboration in Manufacturing Environments, Machine Learning and Adaptation, Multimodal, Multilingual and Multitask Modeling
Abstract: This paper investigates wearable-based recognition of human activities and gestures to support Human–Robot Interaction (HRI) in object-handover and assembly-like scenarios. Electromyography (EMG) and Inertial Measurement Unit (IMU) signals were collected using a Myo armband, culminating in a novel dataset introduced as MAGIC-HRI (Multimodal Activity, Gesture and Intention Collection) with a large taxonomy of 53 movement classes, including Brazilian Sign Language (LIBRAS) numbers (0–9), hand gestures, object/tool handover actions (pick up/give/hold), tool-manipulation tasks, and generic assembly/idle motions, collected from 11 participants with 10 samples per class (530 samples per participant). Signals are segmented by detecting muscle activation via an EMG energy envelope, then processed using sliding windows; time- and frequency-domain features are extracted. Multiple classical classifiers are tuned via cross-validated grid search, with Random Forest as the strongest baseline. A Leave-One-Subject-Out (LOSO) protocol reveals a large generalization gap, indicating substantial subject dependence. A personalized adaptation experiment suggests that injecting a small number of samples from a new user can markedly improve recognition. Overall, the study contributes a broad, HRI-driven multimodal dataset, a rigorous evaluation emphasizing generalization, and practical evidence that personalization is likely required for robust deployment in practical HRI.

Keywords: Machine Learning and Adaptation, Detecting and Understanding Human Activity, Non-verbal Cues and Expressiveness
Abstract: Deploying hand gesture recognition in human-robot collaboration requires building systems for application-specific vocabularies while operating in resource-constrained settings. We propose a general pipeline that facilitates data collection and enables model improvement through a guided approach based on active learning. The framework combines deterministic heuristics for validation and ambiguity filtering with a lightweight model that can be continuously improved through targeted enhancement. The acquisition process uses diversity metrics to prevent redundant samples and interactive augmentation to reduce annotation effort. We identify weaknesses during inference and guide the capture of ambiguous cases. The system is deployed on an NXP i.MX 8M Plus evaluation kit, processing continuous gesture streams using sliding window inference with majority voting to handle variable user pace. Experimental evaluation demonstrates that targeted dataset refinement improves accuracy by +34.25% and unknown gesture recall from 20% to 95%. These findings position the pipeline as a practical foundation for non-verbal human-robot interaction without requiring specialized hardware or large-scale annotation. The pipeline code and the datasets we collected are available at https://github.com/v33to/DHG.git.

Keywords: Virtual and Augmented Tele-presence Environments, Multimodal Situation Awareness and Spatial Cognition, HRI and Collaboration in Manufacturing Environments
Abstract: Industrial environments increasingly deploy robots, creating a need for intuitive spatial interaction. Augmented Reality (AR) supports this, but can become difficult to use in large or occluded spaces where users lack direct views and fixed cameras are impractical. We address this by using a mobile robot as a live, repositionable viewpoint for AR workspace editing. We present a multi-perspective AR system and compare joystick and touch-based gesture control of a mobile robot viewpoint during AR object manipulation. In two user studies, we isolated task-dependent effects (navigation-only vs. manipulation-supported) and evaluated performance in realistic editing tasks. Results indicate that mobile robot viewpoints can support AR interaction in inaccessible areas without fixed camera infrastructure. Joystick control showed advantages for navigation-only viewpoint repositioning, whereas no clear performance difference emerged when viewpoint control supported a touch-based manipulation task. These findings indicate a task-dependent trade-off between control efficiency and interaction integration. We discuss design implications for task-dependent viewpoint control for AR interaction without fixed infrastructure.

Keywords: Gesture, Posture and Facial Expressions, Intention Recognition, Human-Robot Collaboration