Keywords: Cooperation and Collaboration in Human-Robot Teams, LLM/Gen AI-based HRI, Explainable Human-Robot Interaction
Abstract: In Human--Robot Collaboration, failures that are not explained or handled with alternative strategies can undermine user trust and interaction quality. Despite recent advances in LLM-based task planning, failures are often treated as terminal events, leading to non-transparent robot behaviors. This work investigates how explanatory and cooperative responses to planning failure impact human-robot collaboration. We propose a modular architecture that integrates symbolic planning in PDDL with LLM-based reasoning to diagnose planning failures, generate recovery strategies, and negotiate solutions with the human partner instead of halting the execution of the task and terminating the cooperation. We conducted a Human-Robot Interaction video-based study with 33 participants to investigate how different robot behaviors across controlled failure scenarios affect the human's perception of the robot. Results show that transparent explanations and cooperative negotiation significantly improve perceived intelligence and likability, while unexplained errors are rated most negatively. Our findings highlight that, in collaborative settings, managing interaction during failure is a key factor for effective human–robot collaboration.

Keywords: Assistive Robotics, Degrees of Autonomy and Teleoperation
Abstract: Assistive robotic systems have shown growing potential to improve the quality of life of those with disabilities. As researchers explore the automation of various caregiving tasks, considerations for how the technology can still preserve the user's sense of control become paramount to ensuring that robotic systems are aligned with fundamental user needs and motivations. In this work, we present two previously developed systems as design cases through which to explore an interaction paradigm that we call implicit control, where the behavior of an autonomous robot is modified based on users' natural behavioral cues, instead of some direct input. Our selected design cases, unlike systems in past work, specifically probe users' perception of the interaction. We find, from a new thematic analysis of qualitative feedback on both cases, that designing for effective implicit control enables both a reduction in perceived workload and the preservation of the users' sense of control through the system's intuitiveness and responsiveness, contextual awareness, and ability to adapt to preferences. We further derive a set of core guidelines for designers in deciding when and how to apply implicit interaction paradigms for their assistive applications.

Keywords: Multimodal Interaction and Conversational Skills, LLM/Gen AI-based HRI, Multimodal Situation Awareness and Spatial Cognition
Abstract: Embodied AI aims to develop agents capable of perceiving, acting, and collaborating with humans in dynamic environments. However, existing Vision-Language Model (VLM)-based approaches to Human–Robot Interaction (HRI) often assume accurate shared understanding and depend on task-specific finetuning, limiting robustness and generalization. In practice, both human instructions and robot perceptions are inherently error-prone, leading to misalignment that can significantly degrade task performance. To address this challenge, we propose INPERAL, an interactive framework that detects and resolves perception mismatches between humans and embodied agents using pretrained VLMs. INPERAL incorporates a Perception Error Detector to identify inconsistencies between human intent and agent perception, and leverages interactive conversation alongside environment exploration to iteratively refine understanding and resolve these errors. Once alignment is achieved, INPERAL generates an executable action plan to complete the task. Evaluations in two real-world HRI scenarios with diverse perception misalignment settings demonstrate that INPERAL effectively mitigates misalignment, improves task success rates, and enhances robustness without requiring task-specific finetuning, underscoring the importance of interactive error resolution for reliable HRI.

Keywords: Explainable Human-Robot Interaction, Degrees of Autonomy and Teleoperation, Cooperation and Collaboration in Human-Robot Teams
Abstract: Effective human-robot interaction requires robots to not only identify objects referred to in natural language, but to recognise when they cannot do so reliably and request clarification before acting. We present an introspective grounding framework that combines neural object detection with procedural membership functions over colour, shape, and spatial relations. A conflict score χ, computed as the normalised variance of evidence across neural and symbolic sources, serves as a principled measure of cross-source disagreement. When χ exceeds an Adaptive Self-Confidence threshold αt, the Introspective Decision Gate initiates a clarification request rather than executing an uncertain action. The threshold αt adapts over repeated interactions from binary human feedback, calibrating the robot’s clarification frequency to its actual task performance. A spatial relation module resolves references such as “the cup near me” or “the mug away from me” using person-anchored 3D distance, enabling reliable disambiguation of spatially distributed objects. Experiments on a real tabletop setup across 212 trials demonstrates that χ reliably quantifies cross-source disagreement at decision time, that αt remains responsive to feedback while operating well within its safety bounds, and that Adaptive Self-Confidence outperforms static threshold baselines on query efficiency.

Keywords: Multimodal Situation Awareness and Spatial Cognition, LLM/Gen AI-based HRI, Assistive Robotics
Abstract: When humans issue high-level natural language requests to robots in assistive environments, they implicitly assume a shared situational awareness. However, a robot's internal world model may not align with the user's expectations nor with the current physical reality, which is subject to changes and often lacks the granularity required to assess goal feasibility. Grounding human intention depends on nuanced object states and environmental configurations that cannot be assumed a priori or captured by abstract representations alone. We present a visual-semantic approach for situated intent evaluation and alignment, emphasizing the need for an active perceptual check before committing to a goal. Our approach integrates a dual-stream process: a bottom-up scene analysis that extracts grounded properties and constraints from the environment, and a top-down goal validation that aligns intentionality with this perceptual evidence. By identifying discrepancies, the robot ensures a coherent alignment with human intention while identifying the physical and semantic limitations that govern its capacity to assist. Results show that this grounding process improves transparency, enabling the robot to justify its commitment through human-aligned, causal explanations.

Keywords: Motion Planning and Navigation in Human-Centered Environments, LLM/Gen AI-based HRI, Cognitive Skills and Mental Models
Abstract: Robots are finding wider adoption in human environments, increasing the need for natural human-robot interaction. However, understanding a natural language command requires the robot to infer the intended task and how to decompose it into executable actions, and to ground those actions in the robot's knowledge of the environment, including relevant objects, agents, and locations. This challenge can be addressed by combining the capabilities of large language models (LLMs) to understand natural language with 3D scene graphs (3DSGs) for grounding inferred actions in a semantic representation of the environment. However, many 3DSGs lack explicit spatial relations between objects, even though humans often rely on these relations to describe an environment. This paper investigates whether incorporating open- or closed-vocabulary spatial relations into 3DSGs can improve the ability of LLMs to interpret natural language commands. To address this, we implement two pipelines using off-the-shelf models: an LLM-based pipeline for target object grounding from open-vocabulary language commands and a vision language model-based pipeline to add open-vocabulary spatial edges to 3DSGs from images captured while mapping. Finally, we evaluate two LLMs across 14 scenes using 905 natural language statements (786 procedurally-generated, 119 human-authored) to assess performance on the downstream task of target object grounding. Our study demonstrates that explicit spatial relations improve the ability of LLMs to ground objects, and while open-vocabulary relation generation with vision language models proves feasible from robot-captured images, our analysis did not yield evidence favoring either open-or closed-vocabulary relations.

Keywords: Human Factors and Ergonomics, Evaluation Methods, Detecting and Understanding Human Activity
Abstract: Improving walking ability of the elderly is an important factor in extending healthy life expectancy. In this paper, we use inverse dynamics analysis of a musculoskeletal model to estimate the support force required for ankle foot orthoses (AFOs) with elastic frames. A hypothetical frail elderly person with muscle weakness of lower leg was modeled in simulation and attached with a linear elastic model of AFO. The energy consumption of the orthosis was compared for each condition by changing the spring constant and initial angle of the orthosis in steps. The conditions with the lowest energy consumption were identified, and the quasi-optimal torque characteristics for AFO were investigated. The design method proposed in this study can be applied not only to the design of AFOs but also passive wearable devices that utilize elastic forces.

Keywords: Assistive Robotics, Machine Learning and Adaptation, Robots in Education, Therapy and Rehabilitation
Abstract: Frailty and mobility decline reduce tolerance for sustained walking, limiting opportunities for daily exercise and rehabilitation. Wearable ankle-assist devices can improve walking economy and support continued activity, but their effectiveness depends on reliable real-time gait phase recognition and stable assistive timing during prolonged walking. In our previous approach, gait phases were estimated by PSO-optimized linear fusion of IMU and plantar-pressure signals for phase-dependent assistance. However, this method could produce physiologically implausible phase transitions, such as discontinuous phase jumps and boundary chattering, and its limited robustness to walking-condition variation could reduce timing reliability. This paper presents an ankle-assist framework combining neural-network (NN)-based gait phase recognition with a walking-cycle update strategy incorporating Hopf-oscillator-based central pattern generator (CPG) updating during stable walking. The NN exploits nonlinear multimodal feature extraction to improve temporal consistency in five-phase gait decoding, thereby reducing abnormal phase transitions and boundary chattering for more reliable control scheduling. Building on gait-event timing, the proposed update strategy retains gait-event-based synchronization while using CPG-based updating to moderate target walking-cycle increase during stable walking under fatigue. Offline evaluation showed that the proposed NN-based recognizer improved phase recognition accuracy from 70.70% to 87.30% and reduced abnormal phase transitions. In treadmill experiments, the proposed condition increased total walking distance from 641 m to 656 m, reduced heart rate change (HRC) from 4.93% to 2.76%, and decreased estimated active energy consumption from 110.0 kcal to 1

Keywords: Social Learning and Skill Acquisition Via Teaching and Imitation, Robots in Education, Therapy and Rehabilitation, Human Factors and Ergonomics
Abstract: This study proposes utilizing Series Elastic Actuators (SEA) in exoskeletons for high-acceleration tasks, such as tennis swings. This approach minimizes the influence of the device's inertia and viscous forces. Consequently, these design choices enhance transparency, preserving the user's natural motion while enabling accurate transmission of interaction forces. Building on these benefits, we connect the instructor and student through bilateral control. This configuration allows them to physically sense movement deviations, thereby facilitating effective sports teaching. We developed both 1-DOF and 3-DOF wrist exoskeletons and conducted two experiments: bilateral teaching for the tennis forehand stroke, and motion teaching using an ideal trajectory. The results confirm that our 3-DOF exoskeleton and teaching strategy are effective for motion guidance and teaching.

Keywords: Assistive Robotics, Human Factors and Ergonomics, Machine Learning and Adaptation
Abstract: This research demonstrates the development of a virtual human simulation to evaluate exoskeleton influence using deep reinforcement learning. Exoskeletons can enhance elderly individuals' daily living activities. Our objective is to train an AI system capable of preventing user falls through reinforcement learning. However, simulating user reactions when losing balance is challenging. To address this, we propose a virtual human integrated exoskeleton evaluation framework for testing controllers within a simulator. A musculoskeletal model from MyoSuite is trained via a neural network to walk like a human using Soft Actor-Critic combined with SEED-RL for parallel computing. To accelerate training, a natural human gait model from a database is introduced. After training, the musculoskeletal controller imitates human reactions during locomotion when training the exoskeleton controller. The controller is compared with real human %MVC signals, which represent muscle activation, showing high similarity. We demonstrate how the framework trains the exoskeleton controller to consider user reactions. Results show the exoskeleton can support humans even when muscle ability is insufficient for locomotion. The relationship between required exoskeleton torque and user muscle ability is reciprocal. However, the maximum required torque remains similar when the exoskeleton assists the user's balance. Our framework allows developers to rapidly evaluate assistive device effects when attached to users, enabling broader usage by training different controllers to simulate various patient types.

Keywords: Human Factors and Ergonomics, Innovative Robot Designs, Novel Interfaces and Interaction Modalities
Abstract: This study investigates a single-actuator approach for rendering force and vibration associated with dynamic friction by pulse-width modulating a magnetorheological fluid brake (MRB). Conventional methods often rely on separate actuators for force and vibration, which increases both mechanical and control complexity. To address this issue, we superimpose vibration on a base torque by applying a rectangular-wave current to the MRB. First, the MRB output characteristics under PWM control are evaluated, confirming that the average output torque varies with duty ratio and that vibration can be presented at 25--100 Hz. Next, a vibration perception experiment shows that participants can discriminate the presented frequencies with an average accuracy of 87%. Finally, the proposed method is implemented in a lower-limb exoskeleton to present frictional haptic cues while users walk and drag an object in virtual reality. Significant differences are observed among conditions in subjective ratings of dragging and weight sensations, and conditions combining load torque and vibration improve both sensations. When the base torque was held constant, the additional effect of vibration was not statistically significant.

Keywords: Human Factors and Ergonomics, Interaction Kinesics, Assistive Robotics
Abstract: Falls among older adults should be understood not only as a medical issue but also as a problem of everyday environmental design. Rather than relying on conspicuous assistive devices, an alternative approach is to embed subtle supportive features directly into the living environment. In this study, we investigate the potential of small environmental protrusions that can be distributed throughout everyday spaces to enhance balance-support behavior in an unobtrusive manner. We conducted balance tasks with contact to environmental surfaces in 62 older adults, using small protrusions (3 mm and 5 mm) attached to horizontal (table) and vertical (wall) surfaces. The results demonstrate that millimeter-scale protrusions can influence balance-support behavior. However, their effects were not uniform and depended on the contact surface orientation (horizontal table vs. vertical wall) and varied across participants. In the table condition, protrusions—particularly at 5 mm—were associated with reduced postural sway, suggesting improved stability without increasing the magnitude of applied force. In contrast, in the wall condition, the effects were less consistent, and the 3 mm protrusion was associated with decreased usable force and increased postural sway. These findings suggest that the effectiveness of small environmental features depends on their role as tactile cues and on whether they enable stable and controllable interaction with adequate supportive force. This study provides insights into the design of subtle environmental modifications that support balance in everyday environments.

Keywords: Multimodal Interaction and Conversational Skills, Machine Learning and Adaptation, Non-verbal Cues and Expressiveness
Abstract: Turn-taking prediction is a key requirement for social robots involved in human-human interaction, particularly in mediator settings, where the robot must anticipate conversational dynamics rather than merely react to pauses. This work presents a Multimodal Voice Activity Projection (MM-VAP) framework that extends the original audio-only VAP formulation to synchronized audio-visual inputs while preserving its self-supervised future-projection objective. The proposed approach builds on pretrained audio-visual backbones originally optimized for speech-related tasks and adapts them through Low-Rank Adaptation to the multimodal turn-taking problem. After independent speaker encoding, an inter-speaker attention stage models the relational dynamics required to project future voice activity. In addition, a semantic consistency loss is introduced to regularize the 256-state output space according to higher-level dialogue activity patterns. Experiments on NoXi and NoXi+J showed improvements over the current baselines, particularly for some turn-taking events. Additional evaluation on the Haru EDR corpus further supported the suitability of this direction for mediation-oriented human-robot interaction.

Keywords: Social Intelligence for Robots, Monitoring of Behaviour and Internal States of Humans, User-centered Design of Robots
Abstract: This study explores the requirements for socially acceptable robot behaviors based on a user's state, aiming to develop robots that can serve as good partners to humans. As the first trial in an iterative research process, we developed a prototype robot after simultaneously acquiring multimodal data, including EEG, blood pressure, facial expressions and voices. To examine the appropriateness of state-aware behaviors in a ``welcome home'' scenario, we implemented two rule-based behavior patterns—proactive support and minimal assistance—triggered by the user's smile level. Then, we used the recorded interactions as stimuli for a third-party questionnaire. The results showed that respondents' evaluations of the robot's behavior were influenced by the respondents' estimations of the user's level of tiredness. These findings suggest the technical feasibility of our multimodal sensing platform and highlight the importance of state-aware behavior design for socially acceptable in-home robots.

Keywords: Assistive Robotics, Multimodal Interaction and Conversational Skills, LLM/Gen AI-based HRI
Abstract: Designing effective Human-Robot Interaction in task-oriented settings requires carefully balancing user engagement with socially acceptable levels of robot intrusiveness. In this paper, we examine how different robot intervention strategies shape user experience, interaction dynamics, perceived intrusiveness, and sense of support. We compare two approaches: a continuous engagement-seeking robot strategy, and a context-aware strategy that selectively intervenes based on the user’s state and task context. Both approaches rely on multimodal behavioral cues, including body orientation and attentional cues, to guide robot actions. We evaluate these strategies in a user study with 32 participants performing a task in a simulated hospital environment. Our findings show that higher interaction frequency does not necessarily lead to better engagement. Instead, we observe a systematic trade-off between perceived support and intrusiveness, influenced by factors such as physical proximity and user effort. These results provide empirical evidence that effective engagement in HRI depends on adaptive, context-sensitive intervention policies.

Keywords: Robot Companions and Social Robots, Long-term Experience and Longitudinal HRI Studies, Evaluation Methods
Abstract: Companion robots (CRs) have been proposed as alternatives to companion animals such as dogs and cats. Continuous assessment of owner attachment is important for developing and improving CRs that are loved over the long term. However, previous methods based on questionnaires are costly because they need to be conducted periodically to monitor the owners’ subjective attachment levels over time. Therefore, we investigated alternative behavioral indicators of subjective attachment levels using low-cost and continuously accessible CR log data. In our previous studies, log-based analyses found only weak associations between behavioral features and subjective attachment, with correlation coefficients below 0.3. In this study, we analyzed behavioral log data from 224 owners by grouping them into tertiles based on total hugging duration. Moderate correlations (r ≥ 0.4) between subjective attachment and several behavioral features were observed in the low- and medium-duration groups. In the low-duration group, the numbers of days with hugging behavior and cuddling-to-sleep behavior before the CR’s bedtime were correlated with subjective attachment, whereas in the medium-duration group, the frequency of nose-touching behavior exhibited a moderate correlation. These results suggest that grouping based on total hugging duration may help identify behavioral features associated with the subjective attachment level from behavioral log data.

Keywords: LLM/Gen AI-based HRI, Robot Companions and Social Robots, Social Intelligence for Robots
Abstract: Human-robot interaction (HRI) evaluation relies almost exclusively on human-completed questionnaires, leaving the robot's perspective unexamined. We propose an textit{inverted evaluation}, in which LLM-powered robots complete the same standardized instruments from their own perspective, and test whether these ratings agree with human ground truth. In Study~1, five LLMs completed HRI-CUES, Godspeed, and RoSAS questionnaires for 25~interactions (N = 1{,}522 evaluations) from the HRI-CUES dataset. LLMs achieved moderate-to-strong agreement on engagement dimensions (satisfaction r up to .65 and enjoyment r up to .72) with excellent test-retest reliability (ICC geq .82), but textit{systematically inverted} the comfort/strangeness dimension (r = -.44 to -.67, all p < .05), conflating engagement with comfort. In Study~2, a Nao robot running Claude~Sonnet~4.5 replicated these patterns in live interactions (N = 4), including real-time turn-by-turn assessment. The strangeness failure persisted across five models, synthetic controls, and embodied deployment. We argue that current LLM-based robots lack access to the internal affective states needed to assess constructs like strangeness, and that inverted evaluation requires supplementary modalities (e.g., physiological signals, gaze, proxemics) to move beyond behavioral proxies. These findings establish boundary conditions for using LLMs as interaction evaluators in HRI.

Keywords: Robot Companions and Social Robots, Applications of Social Robots
Abstract: The use of robotic companions in everyday environments is increasing as these systems are developed to support social interaction and assist with routine activities. However, there is still a limited understanding of how the presence of robotic companions influences humans during everyday activities. This study investigated how different forms of companionship—none, human, and robot—affect emotional responses, physiological measures, and gait during a shared walking activity and routine task. Participants completed three trials consisting of five minutes of walking followed by folding a basket of clothes under each companionship condition. Emotional responses were measured using a questionnaire, while heart rate and spatiotemporal gait parameters were recorded using a smartwatch and motion capture suit, respectively. Results showed significant differences in emotional responses between human and robot companionship, with the robot condition consistently eliciting less desirable responses. No significant change in heart rate was found during both the walking and task completion portions of the experiment. Participants also altered their gait when walking with the robot compared with walking with a human, including a decrease in walking speed and walking speed variability, along with increases in stride time, stride time variability and stride width variability. These findings suggest that robotic companionship can influence both emotional experience and physiological behaviours during everyday activities.

Keywords: User-centered Design of Robots, Applications of Social Robots, Linguistic Communication and Dialogue
Abstract: Modern, neural-based ASR and TTS systems for speech communication are now very sophisticated, but they are frequently designed for average users in more idealized scenarios, which can underperform when interacting with users having specialized listening needs, or who speak in non-standard ways. Here we review recent TTS research, and note that industrial, production-level implementations of adaptive TTS systems are scarce. We describe how we implemented changes to TTS systems based on prior research, and then report qualitative results from two case studies, where adapted TTS techniques are applied to real-world HRI settings: First, in a older-adult care home, and second, in an airport, where noise levels create comprehension problems. Results suggest promising directions using relatively simple modifications to pitch, duration, and intensity of the robot's speech, and offer a framework for larger, more formal studies of how TTS adaptations can improve HRI.

Keywords: Assistive Robotics, Multimodal, Multilingual and Multitask Modeling, Machine Learning and Adaptation
Abstract: This paper proposes a sign language translation system that converts spoken language into sign language avatar animations to improve information accessibility for Deaf and Hard of Hearing (DHH) individuals. The proposed system consists of two components: (i) a spoken language to sign language translation module that converts spoken language input into a sequence of sign glosses accompanied by auxiliary labels representing spatial placement and non-manual signals (mouthing and facial expressions), and (ii) a motion blending and rendering module that synthesizes temporally coherent hand, body, and facial movements. In this study, we implement a prototype system targeting Japanese breaking news and conduct technical evaluations using actual breaking news text as input. We report the resulting translation outputs and processing latency, and discuss design decisions, system constraints, and future directions. The results demonstrate that the proposed architecture provides a practical foundation for delivering sign language accessibility in time-critical public information scenarios such as emergency broadcasts.

Keywords: Multimodal Situation Awareness and Spatial Cognition, Machine Learning and Adaptation
Abstract: Sound source distance estimation (SDE) is a critical capability in human–robot interaction. An inappropriate interaction distance not only reduces the reliability of speech acquisition and understanding, but also compromises the naturalness and comfort of the interaction. Most existing SDE methods rely on microphone arrays, however, multi-microphone systems typically require careful hardware synchronization, geometric calibration, and additional space and computational resources, which limits applicability to size-constrained and computability-limited embodied platforms. To alleviate these issues, we propose Fast-SDE, a lightweight single-microphone SDE framework that is suited for deployment on robot platforms with limited computational resources and strict size constraints. Specifically, Fast-SDE employs a subband-based backbone that decomposes the frequency axis into multiple subbands, rather than processing the entire spectrum with a wide full-band backbone. A shared subband encoder then maps each subband to a compact latent representation and learns the relationship between acoustic structure and time–frequency patterns. Finally, a lightweight regression head converts the fused subband representations into the estimated distance. Extensive simulation and real-world experiments demonstrate the merits of the proposed method. To benefit the broader research community, we have open-sourced our code at https://github.com/JiangWAV/FAST-SDE.

Keywords: Child-Robot Interaction, Robots in Education, Therapy and Rehabilitation, Robot Companions and Social Robots
Abstract: The introduction of social robots in early education contexts faces a range of complex challenges, as robots must adapt to each setting’s pedagogical principles, the diverse needs of children, and teacher involvement must be considered. We address these complexities by presenting a long-term, in-the-wild study aimed at introducing a social robot in early childhood settings to foster intercultural education, while actively involving teachers in the design and research process. The project involved 165 children aged 3 to 5 who interacted with the robot once a week for six weeks in their classrooms. The robot acted as a peer, designed to stimulate children’s self-expression and curiosity toward different cultural perspectives. We used behavioral grids to measure relational, emotional, and intercultural dimensions of children’s observed behaviors during interactions with the robot. Results show that behaviors related to openness toward peers and the sharing of personal identity emerged over time and were correlated with participation in the activities proposed by the robot. Thanks to its strong methodological framework, this study’s main contribution is its adaptability and reproducibility, even across different educational contexts.

Keywords: Social Touch in Human–Robot Interaction, Innovative Robot Designs, Novel Interfaces and Interaction Modalities
Abstract: In human-robot interaction (HRI), active touch initiated by a robot has the potential to influence human psychological states and perceived intimacy. However, existing approaches are constrained by the robot being stationary, limiting expressive behavior and physical proximity. Inspired by the attachment behavior of pets jumping onto their owners, this study proposes a novel form of active touch in which a robot moves, jumps onto a human forearm, and grasps it. We present: (1) a dual-robot system comprising a launcher robot and a lightweight 52 g grasper robot with soft grippers based on the Fin Ray effect, (2) a MuJoCo physics simulation environment that reproduces the jumping and grasping behavior, and (3) a Soft Actor-Critic (SAC)-based deep reinforcement learning framework that stabilizes the jumping posture and generates grasping motions. The learned policy maintained pitch within ±8° during flight, achieving superior posture stability compared to a fixed-arm baseline (-9.85°). Reproducibility was confirmed across five different seeds with a 100% target contact success rate, with a mean pitch at contact of -5.56 ± 8.00°. Furthermore, a preliminary real robot experiment using the learned parameters confirmed successful grasping of a cylindrical target for over 60 s, validating the sim-to-real applicability of the proposed approach.

Keywords: Social Touch in Human–Robot Interaction, User-centered Design of Robots, Robot Companions and Social Robots
Abstract: Zoomorphic Socially Assistive Robots (SARs) offer an alternative source of social touch for individuals who cannot access animal companionship. However, current SARs provide only limited, passive touch-based interactions and lack the rich haptic cues, such as warmth, heartbeat or purring, that are characteristic of human-animal touch. This limits their ability to evoke emotionally engaging, life-like physical interactions.

We present a multimodal tactile prototype, which was used to augment the established PARO robot, integrating thermal and vibrotactile feedback to simulate feeling biophysiological signals. A flexible heating interface delivers body-like warmth, while embedded actuators generate heartbeat-like rhythms and continuous purring sensations. These cues were iteratively designed and calibrated with input from users and haptics experts in two pilot studies. We outline the design process and offer reproducible guidelines to support the development of emotionally resonant and biologically plausible touch interactions with SARs.

Keywords: Social Touch in Human–Robot Interaction, User-centered Design of Robots, Robot Companions and Social Robots
Abstract: While quadruped robots are increasingly deployed as social companions, their tactile design remains underdeveloped, typically featuring uniform shells that prioritize durability over affective engagement. To address this gap, this paper introduces "tactile zoning," defined as the differentiation of surface materials across a robot’s body regions, as an exploratory approach to understanding users’ tactile expectations for quadruped companion robots. A formative questionnaire (N = 120) suggested moderate interest in differentiated rather than uniform robot surfaces. Subsequently, a physical mapping experiment (N = 21) asked participants to evaluate seven materials and assign them to body regions of a quadruped robot under two ownership contexts (a personal home companion and a shared therapy assistant). Across conditions, soft and biomimetic textiles (Fur, PET Fibre) were more frequently assigned to primary interaction regions including the top of head and back, whereas more industrial or durable materials (Leather, Silicone Rubber) were more often placed on peripheral regions (legs). Generalized Linear Mixed Models showed that higher companion-role congruence ratings were associated with selection for the Back (β = 0.621, p < .001) and Top of Head (β = 0.473, p < .001), but not for the Tail (β = 0.046, p = .663). Ownership context produced modest shifts in material assignment patterns, with a significant Preference × Context interaction for the Abdomen (β = 0.395, p = .011). These findings provide exploratory evidence that users hold differentiated tactile expectations across robot body regions, offering a basis for future work on multi-material design in quadruped companion robots.

Keywords: Novel Interfaces and Interaction Modalities, Creating Human-Robot Relationships, Multimodal Interaction and Conversational Skills
Abstract: Thermal sensing and control play an important role in enhancing the quality of contact and enable richer physical communications between humans and robots. Thermal measurement, as a modality of robotic tactile sensing, can obtain information such as human body temperature and the material properties of contacted objects, and thus serves as a key element for realizing delicate contact. In this study, to further improve the accuracy of thermal sensing and make it more applicable to human–robot touch interactions, the influence of thermal flow resistance at the contact interface (thermal contact resistance) is evaluated. This factor has not been sufficiently considered in most existing studies on robots' thermal sensing. Thermal contact resistance depends not only on the thermal properties of the contacted object but also on contact conditions such as variations in contact surface roughness. As a result, thermal measurement becomes sensitive to environmental changes and more difficult to utilize compared with other sensing modalities, which is one of the reasons why its practical use has not progressed significantly. This paper presents case-study experiments demonstrating both the usefulness and challenges of thermal sensing. Furthermore, based on a physical model of contact thermal resistance and fundamental experimental results on material recognition, insights are discussed for applying thermal sensing to human–robot interaction.

Keywords: Virtual and Augmented Tele-presence Environments, Degrees of Autonomy and Teleoperation
Abstract: Teleoperated robots are widely used in environments that are difficult or hazardous for humans to access. To improve operability and safety, teleoperation systems incorporating haptic feedback have been extensively studied. However, existing systems typically provide only limited haptic feedback and are often implemented as stationary setups, resulting in poor portability. In this study, we have developed handheld haptic interfaces using a miniature magnetorheological fluid device (MR-HHI). The proposed device weighs 122 g, which is approximately 50 g lighter than the previous one. To evaluate haptic feedback performance, experiments were conducted using the developed MR-HHI in combination with a virtual reality system. The experimental results demonstrated the effectiveness of the haptic feedback provided by the proposed device. However, improvements in operability through contact-based haptic feedback were not clearly observed. Future work will investigate the effects of different grasping conditions on system performance.

Keywords: LLM/Gen AI-based HRI, Personalities for Robotic or Virtual Characters, Motivations and Emotions in Robotics
Abstract: We present a framework for emotionally agile robots that support adaptive, non-deterministic interaction with humans. Unlike existing approaches based on fixed sentiment-to-action mappings and limited modes of expression, our method enables robots to adjust both their emotional responses and proactive behaviors over time. The framework integrates the Big Five personality traits, Appraisal Theory, and Dynamic memory through Large Language Models (LLM). The LLM generates a parameterized personality, interprets human language and sentiment, evaluates behavior, and selects actions informed by past interactions. Validation with three robots sharing the same context but different personalities showed that personality, appraisal, and memory substantially affect interaction quality and adaptability. Ablation studies further confirmed the contribution of each component. This approach can enable more meaningful and personalized interactions for assistive, educational, pet, and collaborative robots.

Keywords: LLM/Gen AI-based HRI, Creating Human-Robot Relationships, Linguistic Communication and Dialogue
Abstract: Humor plays a central role in human social relationships, and recent advances in computational humor create new opportunities for integrating humor into human–robot interaction (HRI). While large language models (LLMs) can generate diverse forms of humor, it remains unclear how humor style, joke content, and language preference shape perceptions of robot-delivered humor in group settings. In this exploratory study, we employed a mixed factorial design in which participants evaluated AI-generated jokes delivered by a robot in a university classroom. We examined the effects of humor type (Affiliative, Self-Enhancing, Aggressive, Self- Defeating) and joke content (person-related vs. political) on perceived funniness and appropriateness, as well as preferred language. Results show that humor type significantly influences funniness, with Aggressive and Affiliative humor rated higher, while joke content primarily affects appropriateness, with person-related jokes preferred over political ones. Language preference was shaped by both joke content and participants’ self-reported fluency and humor practices.

Keywords: LLM/Gen AI-based HRI, Narrative and Story-telling in Interaction, Robots in Education, Therapy and Rehabilitation
Abstract: Disclosure is a key step in helping individuals facing psychosocial difficulties, yet many refrain from speaking about such difficulties due to fear, shame, or related concerns. We present “Draw My Life”, a robot-mediated visual storytelling activity designed to support self-disclosure and detect intimate conversations. The activity relies on two modalities, speech and drawings, with the aim of providing the user an unconstrained and relaxing way to express themselves. The robot uses a Vision Language Model to autonomously process both modalities and interact with the user following an active listening approach. We conducted a two-conditions within-subject study, with 25 participants, manipulating the level of intimacy of the conversation topics to disclose about with the robot, and exploring whether features extracted from the participants' multimodal data could be used to predict back the level of intimacy of the conversation topics (RQ1). A logistic regression model achieved the best performance in this task (accuracy = 0.74, ROC-AUC = 0.765, permutation test p-value = 0.002). A feature importance analysis (RQ2) revealed that the most predictive feature lied within the participants' speech content. Features extracted from the drawing content and drawing behavior also showed good predictive power, demonstrating the importance of multimodal analysis when trying to detect intimate conversations. Features extracted from the speech behavior did not showcase important predictive power. Lastly, a thematic analysis of the participants' open feedback indicated an overall positive perception of the developed system, but also identified areas for technical improvement (RQ3).

Keywords: Creating Human-Robot Relationships, Robot Companions and Social Robots, Social Intelligence for Robots
Abstract: As social robots become increasingly present in everyday environments, establishing rapport with users is critical for enabling sustained and meaningful human–robot interaction. In human–human communication, self-disclosure is a well-established mechanism for building rapport; however, its role and impact in human–robot interaction (HRI) are not yet well understood. Although recent advances in large language models (LLMs) have improved the linguistic competence of conversational robots, such systems are typically deployed as reactive information providers, offering limited social reciprocity and relational depth. In this article, we investigate the role of robot self-disclosure in fostering rapport during face-to-face human–robot interaction. We introduce a dialogue system that leverages retrieval-augmented generation (RAG) to enable a robot to retrieve and reference simulated autobiographical memories, allowing first-person self-disclosure to be incorporated naturally into ongoing conversations. We evaluate the proposed approach through an in-person user study, comparing interactions with a self-disclosing robot against a baseline non-self-disclosing robot. Our results show that robot self-disclosure significantly influences conversational dynamics and enhances users’ perceived rapport with the robot, suggesting that self-disclosure is a valuable social behavior for long-term and socially engaging human–robot interaction.

Keywords: Applications of Social Robots, Androids, LLM/Gen AI-based HRI
Abstract: Previous research has shown that a human-like robot’s acceptance heavily depends on the setting in which it operates and its ability to perform relevant tasks. This paper, first, reports on how our robot processes natural language to generate a multimodal, verbal response integrating emotional expressions based on an emotion simulation backend. Then, it describes how visitors were invited to speak with our robot in their own language at three different, public locations, where the robot was running continuously for several days. The Technology Acceptance Model Version 2 (TAM2) questionnaire results reveal that on average users were motivated to use the robot and found it rather useful and easy to use regardless of the specific location. However, public spaces like the tourist information and the city library seem to be a better fit for our interactive, robotic head than an office environment such as the building authority, where the willingness to interact was lower. Overall, the robot’s multi-lingual responses were very much appreciated, but every fifth user found the response time too slow impeding the dialog flow, which remains to be improved in future work.

Keywords: LLM/Gen AI-based HRI, Human–Robot Value Alignment and Safety, Methodologies in Social Robotics
Abstract: Large language models (LLMs) are increasingly used for verbal interaction in social robots, yet prompt design in human-robot interaction (HRI) remains underspecified. As a result, robots may present hallucinated capabilities, unclear behavioural boundaries, and misleading personas. This paper develops a framework for prompt design in LLM-based robots and introduces a structured prompt template comprising eight functional components through which robot behaviour can be specified, bounded, and adapted. The framework is grounded in a review of prior LLM-based HRI work and complemented by survey and discussion data from HRI experts gathered at the Robo-Identity workshop at IEEE RO-MAN 2025 (N=27). The qualitative findings highlight limited legibility of robot personality, the need for user adaptation, and strong ethical concerns about safety, deception, and governance. Based on these findings, we present prompting guidelines accompanied by proof-of-concept template as a structured design and reporting aid for HRI research. We argue that prompt design should be treated as a socio-technical problem rather than a minor implementation detail, requiring explicit capability boundaries, transparent behavioural assumptions, and context-sensitive safeguards to support reliable and interpretable HRI.

Keywords: Motion Planning and Navigation in Human-Centered Environments, Applications of Social Robots
Abstract: Maritime transportation accounts for the most of global trade volume, and improving fuel efficiency while reducing carbon dioxide emissions has become a critical research topic. One of the key technical challenges in enhancing fuel efficiency is the prevention and removal of marine biofouling on ship hulls. Therefore, effective and environmentally responsible hull cleaning technologies are increasingly required not only for improving fuel efficiency but also for preventing ecological impacts. Conventionally, hull cleaning is performed either during periodic inspections in dry dock or by divers in harbor environments. While frequent cleaning is desirable to maintain optimal fuel efficiency, dry-dock inspections are typically conducted only once per year, and diver-based cleaning is both costly and associated with significant safety risks. To address these challenges, underwater robotic systems have been proposed as a promising solution for automated hull cleaning. Our research group has been actively developing ship hull cleaning robots and has demonstrated the feasibility and potential of robotic cleaning through a series of studies and experimental validations. However, a major challenge in achieving fully autonomous operation is accurate localization in underwater environments, where GPS signals are unavailable and sensing conditions are severely degraded. In particular, reliable position estimation near large ship hulls is difficult due to multipath effects, limited visibility, and the lack of distinctive features. In this paper, we focus on this challenge and present a localization method for a ship-hull cleaning robot using a one-way Long Baseline (LBL) positioning system. The effectiveness of the proposed approach is demonstrated through experimental validation.

Keywords: Evaluation Methods
Abstract: 。無人水中ビークルのような水中ロボット (UUV)は安定した無線通信が不可欠です 操作、制御、センシング、データを含む 感染。しかし、コミュニケーションパフォーマンスの評価 水中環境では通常、高価なフィールドが必要です 実験や大規模施設、反復的化 設計や検証が難しい。 本論文では、回路ベースのエミュレータを提案します。 水中無線通信システムの事前評価 ロボットプラットフォームで使用されます。提案されたエミュレーターは再現されています 観測される全体の伝導特性 水中環境、両方

Keywords: Degrees of Autonomy and Teleoperation
Abstract: To improve the efficiency of underwater structure inspections, the deployment of Autonomous Underwater Vehicles (AUVs) is highly anticipated. However, accurate self-localization relative to feature-poor structures remains challenging due to the cumulative errors of dead reckoning (DR). Furthermore, acoustic data from mechanical scanning sonars contain severe noise caused by multipath reflections and suspended particles, hindering reliable feature extraction. This paper proposes a novel framework integrating robust signal processing of acoustic sonar data and hierarchical self-localization. First, multi-stage filtering combining soft gain adjustment, a Laplacian filter, and dynamic thresholding using CA-CFAR successfully extracted only the surface edges of the structure as a clear point cloud, even under severe environmental noise. Next, by introducing a computationally efficient analytical rectangular model, we integrated a "high-frequency update" via a likelihood field-based MCL per beam and a "low-frequency update" via Likelihood Field Optimization using KD-Tree and L-BFGS-B algorithms per scan. In a concrete block inspection experiment using the AUV Tuna-Sand2 in an actual sea environment, the proposed method successfully reduced the localization error (RMSE) by approximately 75% compared to the uncorrected DR trajectory, demonstrating its effectiveness for stable long-term autonomous navigation.

Keywords: Innovative Robot Designs, Evaluation Methods
Abstract: Underwater radio communication has the potential to complement conventional underwater communication methods, such as acoustic and optical systems, which suffer from limited bandwidth and strict alignment requirements, particularly for short-range and real-time robotic applications. However, its practical deployment in underwater robotic platforms is hindered by electromagnetic noise generated by onboard components, which significantly degrades communication performance and has not been systematically evaluated. In this study, we experimentally investigate electromagnetic noise generated by a thruster system in an underwater robot from the perspectives of conducted and radiated emissions. Conducted noise is measured using a line impedance stabilization network (LISN) with common-mode (CM) and differential-mode (DM) separation, while radiated noise is evaluated using a loop antenna in an anechoic chamber. The results reveal that the thruster system generates strong comb-shaped noise spectra with a spacing of 160~kHz, corresponding to the switching frequency of the motor controller. These findings demonstrate that the thruster system is a dominant source of electromagnetic interference (EMI) that directly degrades underwater radio communication performance. To mitigate these effects, line filters and a shielding structure are introduced. The line filters substantially reduce conducted noise over a wide frequency range, with reductions exceeding 30~dB at many frequency components, while the proposed shielding achieves up to 20~dB reduction in radiated noise. These results provide a systematic understanding of thruster-induced electromagnetic noise and offer practical design guidelines for achieving reliable underwater radio communication in underwater robotic systems.

Keywords: Interaction with Believable Characters, Detecting and Understanding Human Activity, Creating Human-Robot Relationships
Abstract: We present a real-time upper-body human-to-humanoid motion imitation framework driven by neuromorphic event-based vision. This work addresses practical perceptual bottlenecks of conventional frame-based RGB sensors—specifically their difficulty in high dynamic range (HDR) scenes and rapid motions due to fixed integration times. By leveraging the Prophesee EVK4 event camera, which operates asynchronously with high temporal resolution and a dynamic range exceeding 120~dB, our system supports stable tracking in conditions where standard vision pipelines degrade, such as severe backlighting and very low light (<5 lux) environments. The architecture integrates a low-latency Perception Module, utilizing optimized event accumulation and gravity-aligned inertial fusion, with a causal Motion Module (TWIST) that performs online kinematic retargeting. We validate the system on an embedded NVIDIA Booster T1 platform and an 18-DoF humanoid upper-body setup, demonstrating an end-to-end photon-to-action latency of 23--34~ms and advantages over RGB baselines under our experimental setup. The results indicate a practical trade-off: events can be preferable for fast or poorly lit upper-body teleoperation, whereas well-lit static scenes may favor RGB or hybrid sensing.

Keywords: Detecting and Understanding Human Activity, HRI and Collaboration in Manufacturing Environments, LLM/Gen AI-based HRI
Abstract: Manual tasks are a fundamental form of human interaction, and recording and analyzing such motion is essential to understand daily activities, measure skilled performance, and build demonstration data for robot control. However, sensing interactions between both hands, objects, and the surrounding environment requires careful consideration of multiple factors, including occlusions, sufficient proximity to handled objects, and the visibility of each hand’s state. This paper investigates camera configurations for an upper-limb-mounted sensing system designed to capture manual tasks. We focus on how camera placement influences the availability of hand and environmental information, with particular attention to both the Hand-Occlusion Rate (HOR), which directly affects the reliability of visual data acquisition, and MLLM-based object recognizability. Based on these considerations, we provide a recommended configuration of upper-limb-mounted cameras to support manual task analysis.

Keywords: Programming by Demonstration, User-centered Design of Robots, Motion Planning and Navigation in Human-Centered Environments
Abstract: We present a human-intuitive Learning from Demonstration (LfD) framework that enables non-expert users to teach a Delta robot complex manipulation trajectories through hand-guided motion, requiring no physical robot contact and no specialised hardware. A monocular vision pipeline using ArUco fiducial marker tracking acquires 3D demonstration data; Gaussian Mixture Model and Gaussian Mixture Regression (GMM/GMR) generalise nine demonstrations into a smooth, reproducible trajectory; and an analytical inverse kinematics (IK) solver converts the predicted path into servo commands. The GMR prediction achieves a trajectory RMSE of 1.74 pm 0.12~mm, outperforming a direct GMM mean baseline (3.21 pm 0.34~mm), and end-effector reproduction error remains below 1.5~mm across all axes with no cumulative drift. Validated on a PCB path-tracing task representative of industrial solder inspection workflows, this work directly addresses the human-robot interaction challenge of lowering the barrier to robot skill transfer in unstructured manufacturing environments.

Keywords: Embodiment, Empathy and Intersubjectivity, Cooperation and Collaboration in Human-Robot Teams, Innovative Robot Designs
Abstract: Robots are entering everyday life in increasingly diverse forms, from mobile manipulators and humanoids to robotic furniture, objects, and spaces. We study how such embodiments shape user perception during physical collaboration through an exploratory Wizard-of-Oz study (N = 16), comparing object-embodied manipulation (a toolbox with arms, a kitchen bookstand with arms) with generic standalone arms, across mixed-initiative and human-initiative-only styles. Users characterized object embodiments as unified entities with contextual expertise (10 of 16 participants), whereas generic arms were perceived as separate tools. Object embodiments were rated more likable (p = 0.049), and perceived intelligence increased after collaboration only for object embodiments (p < 0.01). Initiative style did not significantly affect fluency, but participants valued assistance that fell into two functional categories cutting across styles, supporting user decisions through guidance and confirmation, and reducing user workload through anticipatory action, secondary assistance, and independent task contributions. These findings suggest that embedding robotic capabilities into task contexts can change how users conceptualize robot collaborators.

Keywords: Monitoring of Behaviour and Internal States of Humans, LLM/Gen AI-based HRI, Assistive Robotics
Abstract: Proactive robot assistance in household environments requires accurate prediction of human activities and object usage under dynamic and noisy conditions. Existing approaches often rely on complex spatio-temporal models, which can be computationally expensive and sensitive to environmental variability. In this paper, we propose GLOBE, a lightweight framework that combines n-gram Markov models for capturing temporal behavioral patterns with uncertainty-guided large language model (LLM) reasoning. The framework performs sequential prediction efficiently while selectively invoking LLM reasoning only when the model confidence is low. To evaluate performance under realistic conditions, we introduce HOMER-Noise, a noisy extension of the HOMER+ dataset that simulates structured disturbances such as object movements caused by humans, pets, and toddlers. Experimental results show that GLOBE achieves competitive performance with state-of-the-art methods while improving robustness and computational efficiency across both clean and noisy settings. The framework is further validated through a proof-of-concept integration with a Stretch 3 mobile manipulator, demonstrating its potential application in real-world human–robot interaction scenarios.

Keywords: LLM/Gen AI-based HRI, Assistive Robotics, Machine Learning and Adaptation
Abstract: Robotic assistance in household environments requires not only predicting where objects should be placed, but also reasoning about when objects should not be placed at all. Existing approaches to personalized object rearrangement primarily focus on placement decisions under the assumption of clean observations and complete actionability, limiting their applicability in realistic, cluttered, and partially erroneous settings. In this paper, we introduce APOLLO, a hybrid framework for abstention-aware personalized object rearrangement that combines a lightweight, personalized embedding model (PEM) with selective large language model (LLM) assistance. PEM is trained for each user–environment pair using a small number of demonstrations, operates entirely on CPU, and produces uncertainty estimates, which are used to selectively invoke LLM-based reasoning only for ambiguous decisions, balancing efficiency, privacy, and reasoning capability. To evaluate this formulation beyond existing benchmarks, we introduce APOR, a synthetic, LLM-generated dataset that captures room-level, multi-furniture environments, diverse organizational profiles, explicit abstention behavior, and noisy partial scene context. Extensive experiments on both PARSEC and APOR provide initial evidence that APOLLO improves over prior LLM-based baselines in controlled benchmark settings while substantially reducing LLM usage. Code is available at https://github.com/PaInt-Lab/APOLLO.

Keywords: LLM/Gen AI-based HRI, Multimodal Situation Awareness and Spatial Cognition, Explainable Human-Robot Interaction
Abstract: Robots operating in human environments must often make decisions based on perceptual information that is uncertain, incomplete, or ambiguous. This paper proposes a reliability-aware reasoning framework that enables large language models (LLMs) to account for perceptual uncertainty when selecting actions in human-robot interaction scenarios. The environment is represented as a structured scene composed of detected objects enriched with confidence estimates, attribute reliability, and spatial uncertainty information. Using this representation, the LLM evaluates candidate objects through a reliability scoring mechanism that integrates multiple sources of perceptual evidence and supports uncertainty-aware decision-making. The proposed approach is evaluated using perception episodes with controlled levels of uncertainty and compared with a baseline LLM-based matching strategy that ignores perceptual reliability. Experimental results show that incorporating uncertainty-aware reasoning substantially improves decision robustness under medium and high uncertainty conditions while reducing safety-critical behaviors caused by overconfident autonomous decisions.

Keywords: Explainable Human-Robot Interaction, Non-verbal Cues and Expressiveness, LLM/Gen AI-based HRI
Abstract: Mobile manipulators are increasingly deployed in human-centered environments to perform tasks. While completing such tasks, they should also be able to communicate their intent to the people using expressive robot behaviors. Prior work on expressive robot behaviors has used preprogrammed or learning from demonstration based expressive motions, and/or large language model-generated high-level interactions. These approaches have not considered human-robot interactions (HRI) where users may interrupt, modify, or redirect a robot’s actions during task execution. In this paper, we develop the novel ExpressMM framework that integrates a high-level language-guided planner based on a vision-language model for perception and conversational reasoning, with a low-level vision-language-action policy to generate expressive robot behaviors during collaborative HRI tasks. Furthermore, ExpressMM supports interruptible interactions to accommodate updated instructions by users. We demonstrate ExpressMM on a mobile manipulator assisting a human in a collaborative assembly scenario and conduct an HRI study. Results show that ExpressMM helped observers interpret the robot’s actions and intentions while supporting socially appropriate and understandable interactions. Participants also reported that the robot was useful for collaborative tasks and behaved in a predictable and safe manner during the demonstrations, fostering positive perceptions of the robot’s usefulness, safety, and predictability.

Keywords: Linguistic Communication and Dialogue, Degrees of Autonomy and Teleoperation, Multimodal Interaction and Conversational Skills
Abstract: Voice control offers an intuitive alternative to manual drone piloting, yet most existing systems rely on rigid command vocabularies that fail to handle the spontaneous, disfluent speech of naive users. This paper addresses this gap by proposing an End-to-End Spoken Language Understanding architecture for real-time human-drone interaction in French. Our model combines a frozen Self-Supervised Learning acoustic encoder with a lightweight LSTM-based classification head, augmented by a cross-modal knowledge distillation objective that aligns acoustic representations with semantic embeddings from a text teacher, without requiring transcription at inference time. We evaluate our approach on VoiceStick, a novel French corpus of spontaneous speech collected during real teleoperation sessions with 29 nonexpert dyads. On simple voice commands, our best configuration achieves 93% accuracy at 7 ms inference latency, outperforming cascade baselines (79%, 202 ms) with a 29× speedup. On the full spontaneous speech test set, our architecture reaches 82% accuracy, with crossmodal distillation consistently improving robustness across all configurations. These results demonstrate that End-to-End architectures are not only feasible but preferable for spontaneous voice-guided UAV teleoperation, combining semantic robustness, low latency, and calibrated confidence.

Keywords: Detecting and Understanding Human Activity, Cooperation and Collaboration in Human-Robot Teams, Machine Learning and Adaptation
Abstract: Human-robot interaction (HRI) requires robots to interpret human actions early in their execution in order to respond safely, efficiently, and naturally. However, many existing approaches to human action recognition rely either on sparse skeletal representations, which lack fine-grained motion cues, or dense optical flow, which can be computationally expensive for low-latency perception pipelines. In this paper, we propose PoseOFF, a pose-anchored optical flow representation that captures local motion information around human joints to support earlier human intent understanding. By conditioning motion feature extraction on human pose, PoseOFF encodes localised motion dynamics at semantically meaningful body locations, forming a structured motion representation that is explicitly aligned with human kinematics. We evaluate PoseOFF across multiple benchmark datasets and backbone architectures for action anticipation, demonstrating consistent improvements in recognition accuracy, particularly at early observation ratios. Our results show that PoseOFF enables models to achieve comparable or improved performance while observing less of the action sequence, highlighting its effectiveness for early prediction. Importantly, these gains are achieved without requiring full-frame motion processing, making the approach practical for real-time and resource-constrained settings. These findings suggest that pose-centred motion representations such as PoseOFF can enhance the ability of interactive robot systems to infer human actions earlier, supporting more responsive and anticipatory behaviour in human-robot interaction scenarios.

Keywords: Cooperation and Collaboration in Human-Robot Teams, Cooperative Intelligence in HRI, Explainable Human-Robot Interaction
Abstract: Shared autonomy requires principled mechanisms for allocating and transferring control between a human and an autonomous agent. Existing approaches often rely on blending control inputs or heuristic switching rules, which lack theoretical guarantees and fail to account for the human's likelihood of intervention. This paper presents Flip-Team, a cooperative framework for authority switching in shared autonomy. We formulate the control switching problem as a team-optimal decision problem in which authority transitions are embedded into the system dynamics, yielding optimal switching policies rather than ad hoc rules. We model the human's override propensity, the likelihood of exercising override authority, and derive a critical threshold that determines when human intervention is cost-effective. For linear-quadratic systems, we derive closed-form switching conditions and value function recursions, enabling efficient computation independent of the continuous state dimension. We further propose an online method to estimate override propensity from observed interventions, enabling adaptive switching without prior calibration. Evaluation on a quadrotor altitude regulation task demonstrates that Flip-Team achieves lower cost than always-human and always-autonomous baselines, with selective authority switching that balances human adaptability and autonomous efficiency.

Keywords: Linguistic Communication and Dialogue, LLM/Gen AI-based HRI, Trust in HRI
Abstract: Robots that follow natural-language instructions must decide not only what action to execute, but also whether the current evidence is sufficient to act at all. We present Ask-to-Act, a framework for grounded robot instruction following that treats clarification as a cost-sensitive decision under uncertainty. The system either executes immediately or requests one minimal grounding cue before acting. We evaluate Ask-to-Act on SCOUT++, a grounded HRI benchmark with command-label-image examples from situated human-robot interaction. In a text-only setting, Ask-to-Act improves grounded action prediction from 0.8225 to 0.8308 accuracy while asking in 0.2350 of episodes. In a multimodal setting, command-time visual grounding improves the always-act baseline to 0.8503 accuracy, and Ask-to-Act further improves accuracy to 0.8545 while asking in 0.1277 of episodes. These results show that Ask-to-Act can adapt its interaction rate to the available evidence. It asks more often when language alone leaves action choices underdetermined and asks less often when visual grounding strengthens the base action predictor. To analyze this behavior, we define ask-beneficial ambiguity as cases where the gold action is in the model's top-k candidates but not ranked first, and evaluate uncertainty-based ambiguity detectors with AUROC and AUPRC. Overall, Ask-to-Act positions clarification as an adaptive component of grounded robot learning rather than a fixed requirement of language interaction.

Keywords: Motion Planning and Navigation in Human-Centered Environments, Evaluation Methods, Applications of Social Robots
Abstract: Social, also called human-aware, navigation is a key challenge for integrating mobile robots into human environments. The evaluation of such systems is complex, as factors such as comfort, safety, and legibility must be considered. Human-centered assessments, typically conducted through surveys, provide reliable insights but are costly, resource-intensive, and difficult to reproduce or compare across systems. Alternatively, numerical social navigation metrics are easy to compute and facilitate comparisons, yet the community lacks consensus on a standard set of metrics.

This work explores the relationship between numerical metrics and human-centered evaluations to identify potential correlations. If specific quantitative measures align with human perceptions, they could serve as preliminary benchmarking tools, providing a human-aligned assessment when large-scale surveys are not feasible. Our results indicate that while current metrics capture some aspects of robot navigation behavior, important subjective factors remain insufficiently represented, necessitating new metrics.

Keywords: Interaction Kinesics, Evaluation Methods, Curiosity, Intentionality and Initiative in Interaction
Abstract: Research on interpersonal distances, known as proxemics, suggests that the nature of our relationships is often reflected in the distance at which interaction partners feel comfortable and safe. Interpersonal distance is also known to be influenced by demographic and cultural factors. However, these factors have not been well studied in the context of human-robot interaction (HRI). We describe an online task which we developed to study proxemic preferences during HRI among Japanese and UK participants. A group of Japanese (N=48) and UK (N=44) adults watched 36 video clips of NAO robot walking towards the camera, which varied in: speed, angle of approach, camera height/eye level, or indoor vs outdoor setting. Participants stopped each video clip when NAO was at an appropriate distance to begin a conversation, deploying a “Say-when” paradigm. While Japanese participants were slower to stop the video clip, thus allowing the robot to advance further towards them, both cultural groups were quicker to stop the video for trials featuring head-on approaches. Further, head-on approaches were rated as more polite and interpersonal by both groups. Although limited within the online setup and in need of further stimuli refinement and increased statistical power, the findings may serve as a precursor to develop a robust method in future for capturing systematic variations in visually mediated robot proxemics of HRI.

Keywords: Motion Planning and Navigation in Human-Centered Environments, Evaluation Methods, Non-verbal Cues and Expressiveness
Abstract: Ensuring that robot navigation is safe and socially acceptable is crucial for comfortable human-robot interaction in shared environments. However, existing validation methods often rely on a bird's-eye (allocentric) perspective, which fails to capture the subjective first-person experience of pedestrians encountering robots in the real world. In this paper, we address the perceptual gap between allocentric validation and egocentric experience by investigating how different perspectives affect the perceived sociability and disturbance of robot trajectories. Our approach uses an immersive VR environment to evaluate identical robot trajectories across allocentric, egocentric-proximal, and egocentric-distal viewpoints in a user study. We perform this analysis for trajectories generated from two different navigation policies to understand if the observed differences are unique to a single type of trajectory or more generalizable. We further examine whether augmenting a trajectory with a head-nod gesture can bridge the perceptual gap and improve human comfort. Our experiments suggest that trajectories rated as sociable from an allocentric view may be perceived as significantly more disturbing when experienced from a first-person perspective in close proximity. Our results also demonstrate that while passing distance affects perceived disturbance, communicative social signaling, such as a head-nod, can effectively enhance the perceived sociability of the robot's behavior.

Keywords: Assistive Robotics, Cooperative Intelligence in HRI, Creating Human-Robot Relationships
Abstract: Person following in crowded indoor environments remains challenging owing to occlusion-induced tracking loss and identity confusion among bystanders. We propose a crossmodal person-following system that couples three components: First, a video object segmentation tracker, initialized by a single click on the target person, propagates identity-stable masks without requiring repeated per-frame detection. Second, a parallel 2D LiDAR leg tracker projects its position estimate onto the image plane to support recovery when visual tracking becomes unreliable. Third, a dual-model Kalman filter with depth-scaled measurement noise maintains an accurate threedimensional state estimate, adapting between steady walking and sudden maneuvers. In experiments across various occlusion conditions, the proposed system achieved up to a 31-point improvement in tracking availability and up to a 72-point reduction in non-tracking time ratio compared to conventional camera-based tracking. The proposed system yields a highly robust tracking pipeline that effectively reduces the chance of permanent target loss, even during prolonged visual interruptions.

Keywords: Detecting and Understanding Human Activity
Abstract: Autonomous mobile service robots, such as lawnmowers or cleaning robots, operating in human-populated environments need to reason about local human-human interactions to support safe and socially aware navigation. For such systems, interaction understanding is not primarily a fine-grained recognition problem, but a perception problem under limited sensing quality and computational resources. Many existing approaches focus on holistic group activity recognition, often relying on complex and computationally expensive models that are not well suited for mobile robotic platforms. In this work, we argue that pairwise human interactions constitute a minimal yet sufficient perceptual unit for robot-centric social understanding. We study the problem of identifying interacting person pairs and classifying coarse-grained interaction behaviors sufficient for downstream group-level reasoning and robot decision-making. To this end, we adopt a two-stage framework in which candidate interacting pairs are first identified using lightweight geometric and motion cues, and interaction types are subsequently classified using a relation network. We evaluate the proposed approach on the JRDB dataset, where it achieves competitive performance with significantly reduced computational cost and model size compared to appearance-based methods. Additional experiments on the Collective Activity Dataset (CAD) and zero-shot evaluation on a lawnmower-collected dataset further demonstrate the generality of the proposed framework. These results suggest that simple geometric and motion cues provide a practical and efficient basis for interaction-aware perception in mobile service robots.

Keywords: Motion Planning and Navigation in Human-Centered Environments, Social Intelligence for Robots, Multimodal Situation Awareness and Spatial Cognition
Abstract: Autonomous navigation in unknown and dynamic environments remains a major challenge in robotics, particularly in spaces shared with humans. Traditional metric mapping approaches suffer from high computational costs and rapid map obsolescence. This paper proposes a human-flow-driven skeleton framework for incrementally constructing a topological representation of the environment. The method uses pedestrian motion to extract directional information, segmenting corridors into circulation lanes and building the representation asymmetrically during exploration. Experiments in simulated bidirectional corridors show that the proposed method reduces lane invasions while maintaining stable navigation aligned with human flow. A hybrid variant allowing controlled crossings near the goal further improves path length, lane compliance, and trajectory smoothness. These results suggest that incorporating directional information from human flow enables trajectories better aligned with movement organization in shared environments, promoting safer and more socially compliant navigation.

Keywords: LLM/Gen AI-based HRI, Social Intelligence for Robots, Robot Companions and Social Robots
Abstract: This paper presents a multimodal dialogue framework that enables gaze-triggered, object-centered conversation on the social robot Haru. Unlike conventional dialogue systems that rely on explicit user verbal queries, the proposed approach allows the robot to initiate interaction based on the user’s sustained visual attention, supporting non-verbal turn-taking and shared attention. This framework combines gaze with object detection models based on deep learning, which integrates 3D spatial constraints and 2D gaze features to reliably identify the user’s object of attention. Once the robot detects sustained user’s eye gaze on a specific object, it autonomously initiates an object-grounded conversation with the human user. In addition, the proposed approach allows the robot to generate contextually relevant responses via vision- language models (VLMs), while aligning the robot’s head and gaze toward the referenced object, reinforcing joint attention through embodied feedback. A User study comparing robots in gaze-aligned and non-aligned conditions demonstrates that gaze-triggered dialogue significantly increases perceived social awareness, affection, and willingness to interact

Keywords: HRI and Collaboration in Manufacturing Environments, Trust in HRI, Monitoring of Behaviour and Internal States of Humans

Keywords: Detecting and Understanding Human Activity, Multimodal Interaction and Conversational Skills, Social Intelligence for Robots
Abstract: Physiological signals are increasingly relevant to estimate the mental states of users in human-robot interaction (HRI), yet ROS 2-based HRI frameworks still lack reusable support to integrate such data streams in a standardized way. Therefore, we propose Sense4HRI, an adapted framework for human–robot interaction in ROS 2 that integrates physiological measurements and derived user-state indicators. The framework is designed to be extensible, allowing the integration of additional physiological sensors, their interpretation, and multimodal fusion to provide a robust assessment of the mental states of users. In addition, it introduces reusable interfaces for timestamped physiological time-series data and supports synchronized logging of physiological signals together with experiment context, enabling interoperable and traceable multimodal analysis within ROS 2-based HRI systems.

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: Virtual and Augmented Tele-presence Environments, Machine Learning and Adaptation, Monitoring of Behaviour and Internal States of Humans
Abstract: In challenging environments, AI-powered robots have potential to be helpful partners when humans become incapacitated. Wayfinding is a key task in which human guides often rely on nonverbal cues to decide to offer their partners help, yet current robot assistants lack this ability. As a first step to address this gap, we created an immersive virtual environment (IVE) designed to elicit help-seeking behaviors in a diving navigation task under environmental stressors. Seventy-two participants completed goal-directed wayfinding tasks while we recorded multimodal behavioral signals, including locomotion speed, gaze, head and body movement, and electrodermal activity (EDA). We then trained predictive models to infer when participants would seek help from a robot partner. We found that interpretable features such as increased gaze scanning and locomotion deceleration emerging as robust predictors of help-seeking in the virtual environment. This work offers preliminary design insights for proactive agent assistance in high-stakes scenarios.

Keywords: Human Factors and Ergonomics, HRI and Collaboration in Manufacturing Environments, Monitoring of Behaviour and Internal States of Humans
Abstract: Collaborative robots (cobots) are increasingly deployed in human-centered environments, where robot motion parameters such as speed influence not only productivity but also the cognitive experience of human partners. While velocity limits for safety are well established, less is known about how robot speed interacts with task difficulty to shape human mental workload. To investigate this relationship, we developed a controlled human–robot collaboration paradigm in which a Sawyer cobot manipulates dice on a 3×3 grid while participants observe and recall movement sequences. Task difficulty was manipulated through sequence length, and robot speed was varied across three levels. Mental workload was assessed using both subjective ratings and ECG-based physiological measures. A user study with 29 participants revealed a difficulty-dependent pattern. At medium task difficulty, subjective workload showed a non-linear dip at medium robot speed, while physiological workload increased steadily with speed. Easy tasks were largely unaffected by speed, whereas hard tasks produced the highest workload overall, with subjective ratings increasing but physiological indices remaining nearly flat, suggesting possible ceiling effects or disengagement. These findings indicate that the effect of robot speed on workload depends strongly on task complexity. The results highlight the importance of designing workload-aware cobot behaviors in which robot speed is adapted not only for safety and efficiency but also for cognitive ergonomics in human–robot collaboration.

Keywords: Motion Planning and Navigation in Human-Centered Environments, Multimodal Situation Awareness and Spatial Cognition, Sound design for robots
Abstract: The ability to navigate toward sound sources extends a robot's reach beyond its visual field, enabling response to auditory events in unknown environments. To equip robots with this capability, existing methods couple acoustic and visual information through joint audio-visual learning in acoustic simulators. However, acoustic simulation is both low-fidelity and expensive, producing a domain gap that prevents reliable real-world deployment, while the discrete action spaces inherited from grid-based simulators introduce an additional kinematic gap on physical robots. To alleviate these issues, we propose BCNav, a decoupled framework that separates the acoustic module from the learned navigation policy using direction-of-arrival (DOA) estimation: an estimator provides a scalar bearing to the sound source, so the navigation policy only processes depth images and a bearing angle, two inputs whose domain gaps are well characterized. We collect shortest-path demonstrations with calibrated bearing noise injection and train the policy via imitation learning to output continuous velocity commands directly executable on ground robots. The effectiveness and advantages of our method have been demonstrated in simulation and on a physical robot, allowing robots to navigate in unknown environments without any acoustic fine-tuning, prior mapping, or real-world audio data collection. Code is available at https://github.com/york1to/bcnav.

Keywords: Detecting and Understanding Human Activity, Machine Learning and Adaptation, Multimodal Interaction and Conversational Skills
Abstract: Sign Language Recognition (SLR) is a crucial technology for enabling social humanoid robots to interact with Deaf and Hard-of-Hearing (DHH) individuals. However, a critical capability gap exists in current SLR-equipped systems: models cannot recognize Out-of-Vocabulary (OOV) glosses -- signs not seen during training --causing steep performance drops when deployed with real-world users whose vocabulary extends beyond the training set. This gap is particularly severe in human-robot interaction (HRI) contexts, where domain-specific terminology, proper names, or newly coined signs must be handled on-the-fly without costly system retraining. In this paper, we develop a customizable SLR method that enables robots to flexibly extend their sign vocabulary at inference time, without a significant training cost. The incapability of recognizing untrained signs involves two main factors: (1) the data scarcity of cheremes, the minimum units that compose sign language, leads to difficulties in defining new glosses, and (2) adding new glosses requires retraining the model with a significant cost. To address these issues, we propose automatic extraction of chereme-inspired sub-units via clustering for issue (1), and vocabulary extension based on Deep Biasing for issue (2). Experiments show that the proposed method improved Word Correct (WC) for OOV glosses by 39.1 points while minimizing degradation on trained glosses to 3.2 points, demonstrating its effectiveness for robust and customizable sign language interaction in real-world deployment.

Keywords: Non-verbal Cues and Expressiveness, Multimodal Interaction and Conversational Skills, Detecting and Understanding Human Activity
Abstract: Social robots must interact robustly not only with users assumed by speech-centered systems but also with diverse users whose communication relies on different modalities, e.g., sign language. One important capability gap is predictive turn-taking with sign language users. In sign language interaction, a robot must estimate whether a signer is likely to continue or yield the floor, while respecting the visual and embodied organization of sign language interaction. However, computational targets for continuous future activity prediction in sign language interaction are not yet well established. This paper presents an initial study toward Signing Activity Projection, a framework for modeling future signing activity in dyadic sign language interaction. We draw on the activity projection formulation of Voice Activity Projection as a computational reference. Using interaction recordings from the Public DGS Corpus, we derive binary lexical signing activity streams from lexical sign annotations and formulate proxy tasks for turn-taking prediction. The model uses pose-derived hand, eye-region, and mouth-region features extracted for each signer. The results show that SHIFT/HOLD prediction is promising, especially with hand cues, while SHIFT-prediction remains difficult. These findings provide initial evidence for both the promise and the current limitations of modeling future signing activity from visual interaction data. Predictive modeling of sign language interaction still requires sign-language-specific event definitions and richer multimodal representations.

Keywords: Affective Computing, Linguistic Communication and Dialogue, Evaluation Methods
Abstract: Emotion Recognition in Conversation is a core component of affective computing, while current sign language emotion datasets primarily focus on isolated sentences and lack conversational context. Models trained exclusively on these isolated utterances demonstrate degraded performance in real world scenarios because they cannot utilize historical dialogue flow. To address this structural limitation, we introduce the ERC task to sign language video analysis and propose the eJSL Dialog dataset. Constructed using the scripts from the STUDIES corpus, the dataset contains 1,920 video samples organized into 480 unique dialogues. We conduct systematic benchmarking on this dataset using models ranging from isolated visual networks to multimodal conversational architectures. The results reveal a domain gap when applying generic multimodal conversational emotion recognition models to sign language. These findings demonstrate the explicit need for context-aware visual extractors specific to sign language and indicate that constructing larger conversational datasets to support large-scale pre-training is a necessary next step for future research.

Keywords: Social Touch in Human–Robot Interaction, Human Factors and Ergonomics, Novel Interfaces and Interaction Modalities
Abstract: Haptic devices based on exoskeleton mechanisms, which are capable of providing proprioceptive feedback, have been actively studied; however, their inherent mechanical resistance often constrains natural finger motion and limits usability in interactive tasks. To overcome this issue, we propose a wire-driven haptic device that provides proprioceptive and pressure feedback to multiple finger segments while achieving high mechanical transparency. The device was designed to minimize motion interference and enable localized sensory presentation. We first evaluated the kinematic influence of the device by comparing finger range of motion and motion time with and without the device. A study with eight participants showed that when wearing the device, the range of motion remained at 95 ± 3.65% of the bare-hand condition, and the time required for rapid reciprocal finger motion was 99.1 ± 11.0% of the bare-hand condition. We then evaluated the perceptual and functional effects of the presented sensations with 10 and 7 participants, respectively. In a force-location discrimination task, participants correctly identified the stimulated finger segment with an accuracy of 85.6 ± 12.9 [%], demonstrating reliable spatial perception. Furthermore, task performance significantly improved when proprioceptive and pressure feedback were provided compared to the no-feedback condition (Wilcoxon signed-rank test, W = 1.0, p = 0.046). These results show that the proposed device simultaneously achieves low mechanical resistance and effective multi-segment sensory feedback, highlighting its potential for dexterous manipulation and immersive human–robot interaction in VR and teleoperation scenarios.

Keywords: Assistive Robotics, Robots in Education, Therapy and Rehabilitation, Virtual and Augmented Tele-presence Environments
Abstract: Existing gait rehabilitation systems are frequently constrained to indoor environments and lack multimodal feedback for complex outdoor trajectories. This paper presents an architecture and implementation of an outdoor Mixed Reality Visual-Haptic Gait Training System. By integrating an MR headset, wearable motion trackers, and vibrohaptic shoes, the system establishes a real-time "perception-guidance-action" closed loop. The architecture introduces a dynamic curved path guidance mechanism driven by a joystick-controlled arc-generation algorithm for outdoor turning maneuvers. This mechanism effectively overcomes the rigid straight-line constraints inherent in conventional rehabilitation systems. The system visually projects adaptive footprints based on a walking ratio model and maps real-time gait deviations to directional vibrohaptic cues. Experimental evaluations demonstrate that this integrated framework successfully supports complex spatial trajectories, improves gait symmetry, and enhances user engagement, validating its technical viability for real-world rehabilitation.

Keywords: Degrees of Autonomy and Teleoperation, Explainable Human-Robot Interaction, Virtual and Augmented Tele-presence Environments
Abstract: Bilateral teleoperation stands as a fundamental technology for extending human capabilities into remote and hazardous environments. In such systems, achieving high haptic transparency while guaranteeing stability under communication delays is a critical challenge for ensuring an immersive human-robot interaction. The Time Domain Passivity Approach (TDPA) has emerged among stabilizing methodologies as a robust benchmark. Recent advancements have targeted its intrinsic conservatism through two distinct paths: the Energy Reflection-based TDPA (TDPA-ER) optimizes the observation phase whereas the redundant TDPA (rTDPA) handles the dissipation phase. Despite these individual breakthroughs, the stateof- the-art lacks an integrated approach that leverages both. Current systems still suffer from excessive force jittering and conservatism, as the synergy between top-tier observation and dissipation algorithms remains unexplored. This paper presents the Redundant Energy Reflection-based Time Domain Passivity Approach framework, designed to enhance the perceptual fidelity of the operator. By combining the TDPA-ER observation method with the rTDPA dissipation scheme, the objective is to push the boundaries of haptic rendering by synergizing a low conservatism observer with a null-space-prioritized controller, thus providing an unprecedented level of haptic fidelity. The framework was validated on a dual 7-DoFs Franka Emika Panda setup under high-latency (up to 600 ms round-trip) and, crucially, unpredictable time-varying delays.

Keywords: Multimodal Interaction and Conversational Skills, Assistive Robotics, Novel Interfaces and Interaction Modalities
Abstract: Despite the recent proliferation of eldercare robots, it is challenging to know where exactly a user would prefer physical support for a given task. We thus introduce a multimodal framework for human--robot handlebar positioning, to allow elderly users to adjust handlebar locations suggested by a robot. The proposed framework represents touch interaction, natural-language commands, and visual feedback within a single time-varying artificial potential field (APF), allowing multiple communication modalities to shape the same control landscape. In our experimental setup, a handlebar mounted to a UR5e robot arm is controlled through APF-guided impedance control. We introduce a tunneling mechanism that interprets sustained uphill user force as implicit intent and locally reshapes the field, allowing users to progressively establish a new preferred resting position. We also develop a novel language-to-APF mapping pipeline in which synthetic sentence embeddings are regressed to parameters of localized functions that augment the APF, enabling spoken commands to affect handlebar motion. A ring of embedded LEDs provides real-time visualization of the local field gradient to improve transparency of robot intent. Human subject trials with ten young adults comparing five modality combinations indicated that the full multimodal system improved handlebar placement accuracy, reduced sit-to-stand duration, and received the highest overall user ratings. These results support multimodal APF-based interaction as a promising paradigm for assistive robotic support.

Keywords: Cooperation and Collaboration in Human-Robot Teams, Applications of Social Robots, Evaluation Methods
Abstract: Human-robot object handovers are especially challenging without vision, yet many assistive settings demand exactly that (e.g., blind or visually impaired users, surgeons or mechanics working who can visually focus on object handovers). We present a two-cycle, user-centered study of nonvisual handovers. Cycle~1 (Wizard-of-Oz) compared six concepts (user-passive, voice, guiding arm, beeping, haptic, wind): participants preferred autonomy-preserving designs with clear, multisensory spatial cues. Results show that wind and concise voice ranked highest, while beeping and haptics underperformed as standalone channels. We found that physical guidance felt efficient yet autonomy-limiting, while concise voice improved anticipation but added duration. In Cycle~2, we implemented two finalists on a Pepper robot: (i) robot-passive and user-active with airflow (``wind'') and optional speech, and (ii) robot-active and user-passive with autonomous reaching and speech. With 15 participants (within-subject design), wind was faster and more reliable. However, trust, satisfaction, competence, and comfort did not differ.

Keywords: Applications of Social Robots, Non-verbal Cues and Expressiveness, Personalities for Robotic or Virtual Characters
Abstract: In human-robot interaction, achieving a goal may require a robot to engage in competitive behaviors toward its human partner. In some cases, it is important for the robot to balance firmness with an amicable attitude to avoid causing discomfort to the user. In others, it might be necessary to push the user to act more assertively or competitively. In this work, we investigate whether a robot's competitive social behavior, operationalized through verbal and nonverbal cues, can influence humans' decision-making in a strategic game such as an auction. Participants competed in an English auction game against a humanoid robot, which displayed either an amicable or a confrontational behavioral profile, through verbal language and nonverbal gestures. Competitiveness was evaluated through behavioral outcomes and perceived robot attributes, with biometric measures derived from posture and facial expression analysis included as exploratory indicators. Results indicate that robot attitude significantly alters perceived warmth, competence, and discomfort, while confrontational behavior selectively enhanced game performance in individuals with higher baseline competitiveness, as revealed by a significant interaction between robot condition and competitive trait. No significant effects emerged in biometric indicators, suggesting that social framing primarily affects strategic and perceptual dimensions of competition.

Keywords: Affective Computing, Applications of Social Robots, Robot Companions and Social Robots
Abstract: An individual's handedness has an influence on preferences for objects’ direction. In the context of human-robot interaction, the head-tilting direction of a robot plays a vital role in enhancing the feeling of kawaii, the Japanese concept of cuteness, which contributes to the social acceptance of robots. Although various past studies investigated the effective behavior designs for such social robots, these studies focused less on the relationship between the user’s handedness and the perceived feeling of kawaii in observing robots’ behaviors. Therefore, in this study, we conducted a web-based survey experiment to clarify this relationship. Moreover, we also investigated the effects of the cultural differences on the perceived feeling of kawaii on head-tilting behaviors by comparing participants from Japan and the U.S. The experimental results showed that participants judged the dominant-hand side head tilts (i.e., right-handed: rightward, non-right-handed: leftward) as more kawaii/cute than the non-dominant-hand side head tilt (i.e., right-handed: leftward, non-right-handed: rightward), regardless of cultural differences.

Keywords: User-centered Design of Robots, Human Factors and Ergonomics
Abstract: A robot's appearance informs human expectations about its capability, but this relationship is not straightforward and may need separate consideration of motoric, sensing, and social capabilities to be properly understood. In this paper, we report on our online study (N=40) to elicit human expectations of robot sensing capabilities of 98 different robot models. We found that robots were not only thought incapable of possibly more human-associated senses such as taste or smell, but also that participants doubted robots' ability to sense humidity, temperature, and light, even with technical sensors for these factors being widely commercially available. Even for senses such as vision that can easily be realized using a camera, participants only believed 42 out of 98 robots to be capable. We also conducted exploratory data analysis to examine if visual appearance features could serve as predictors for these senses, however it was inconclusive. Our results show that users are mostly skeptical about robot sensing capabilities, which are an important factor when determining a robot's ability to complete a delegated task or collaborate on a task with a human.

Keywords: Explainable Human-Robot Interaction, Motivations and Emotions in Robotics, LLM/Gen AI-based HRI
Abstract: Attribution styles shape how people interpret success and failure, yet their counterparts in Human-Robot Interaction (HRI) remain underexplored. This study investigates a 2×2 attribution style framework, categorizing robot feedback into internal, external, benevolent, and narcissistic styles, to examine how causal framing interacts with positive and negative feedback during a robot-guided dance learning task. 64 participants engaged with a humanoid tutor that delivered controlled, pre-scripted feedback. Results indicated that feedback valence dominated learner responses: positive feedback consistently outperformed negative feedback in acceptance and influence across all attribution conditions. Furthermore, exploratory analyses of psychometric reliabilities revealed a critical phenomenon: while affective measures remained highly robust, self-evaluative cognitive scales fragmented under complex causal attributions. We introduce the Elaboration Likelihood Model (ELM) to propose that this evaluative difficulty represents a fundamental methodological boundary for static attribution framing. Our findings suggest that to bridge this cognitive divide, next-generation adaptive agents must leverage generative models to dynamically align robotic reasoning with learners' internal reality.

Keywords: Motivations and Emotions in Robotics, Non-verbal Cues and Expressiveness, Personalities for Robotic or Virtual Characters
Abstract: We explore how an anxious robot can foster prosocial responses in humans. We developed a proof-of-concept multimodal anxiety expression on a rover robot to show that the perception of robot anxiety could induce key motivators of prosocial behavior such as empathy and compassion towards the robot. We found that our anxious expression elicited empathy and compassion towards the robot. Interestingly, we did not find a significant difference in actual helping behavior. Our qualitative results reveal that while the expression of the robot might lead to engagement, the appropriateness of them with respect to the context of interaction should also be considered. This demonstrates that negative emotional expressions, or at least robot-expressed anxiety can be leveraged to elicit empathy while underscoring the need for future work on the effects and design of negative emotions in HRI.

Keywords: Non-verbal Cues and Expressiveness, Motivations and Emotions in Robotics, Interaction Kinesics
Abstract: This paper examines whether abstract robotic motion can convey emotional information, and whether factors such as object shape and movement source influence how these emotions are perceived. We report a controlled study in which 98 participants evaluated short video clips of two non-humanoid robotic forms (a sphere and a cube) performing motion sequences designed to express six basic emotions. Participants were asked to identify the emotion conveyed by each movement, allowing us to assess recognition accuracy across different conditions.

The results show that participants were able to recognize emotions at levels significantly above chance, although overall accuracy remained modest. Recognition performance varied substantially across emotions, with sadness and surprise identified more reliably than anger, fear, disgust, and happiness. Object shape did not produce a significant effect on recognition accuracy, and no interaction between shape and emotion was observed. In contrast, movement source did not affect overall performance but interacted significantly with emotion, indicating that certain movement styles were more effective for specific emotions.

The study highlights that emotion recognition in abstract robotic systems is primarily driven by the type of emotion and the characteristics of the motion, rather than the physical form of the robot. These findings suggest that, in abstract robotic systems, emotional legibility depends more on the design of motion dynamics than on geometric form, offering practical guidance for affective expression in non-humanoid robots. At the same time, the relatively low accuracy underscores the limitations of motion as a standalone communication channel.

Keywords: Robots in Education, Therapy and Rehabilitation
Abstract: 。水中無線通信はしばしば問題を抱えています。 伝送遅延の大きな変動と高いパケット 損失率。そのような環境下で、遅延トレラントの使用 ネットワーク(DTN)は情報として有望と考えられています 配達方法。DTNはデータ配信を メッセージをノードとして機会的に転送する 動け。しかし、マルチホップ通信を実現するには 水中環境における複数の移動ノード、 エンドツーエンドのデータ配信性能の向上は以下の通りです。 必須。本研究では、この課題に次のように対処します。 冗長パケット割り当て方法ӛ

Keywords: Cooperative Intelligence in HRI, Explainable Human-Robot Interaction, Creating Human-Robot Relationships
Abstract: Carrying out effective underwater missions using Marine robots require close and consistent coordination between human operators and the robotic in challenging environments. Real-time interaction between divers, surface operators, and the robotic platform is however difficult to achieve due to underwater communication constraints. This paper presents a multimodal human-robot interaction framework designed to support diver-in-the-loop marine robotic operation that is based on commercially available smartphone devices and accessories. By wirelessly enabling real-time communication between two divers operating underwater and a surface operator, the proposed system proffers a multimodal interface that integrates bidirectional video streaming, unidirectional voice communication, and real-time telemetry feedback. The mode-based interaction architecture allows the operator to coordinate divers and robotic platforms through a unified control interface while maintaining their situational awareness in the mission. The systems implementation has been deployed to commodity mobile devices and validated through both indoor and real-world experiments. Experimental results demonstrate the practicality of the proposed framework for supporting interactive underwater missions under constrained wireless conditions and in real operational scenarios. The work highlights the potential of multimodal interaction technologies for enabling effective human–robot collaboration in marine robotics.

Keywords: Degrees of Autonomy and Teleoperation
Abstract: Diver safety during underwater technical work remains challenging because emergencies can occur while a diver is submerged, weakly visible from the surface, and temporarily separated from nearby support. This paper presents a standby acoustic emergency-response concept using a BlueBoat unmanned surface vehicle (USV) that remains near the work area and begins active motion only after an SOS event is detected. In the proposed architecture, each diver carries an acoustic beacon with a distinct identifier, while the USV uses a Water Linked short-baseline localization system to estimate the relative target position and move toward the diver's x-y location. Unlike continuous diver-following systems, the proposed vehicle is not required to always remain above a single diver, which is intended to reduce energy consumption and allows a single platform to support multiple divers in the same operational area. The concept was evaluated in controlled pool experiments using an acoustic locator as the target beacon: it was fixed during the stationary localization trials and carried by a diver during the final approach demonstration. Localization accuracy was examined at near and far target positions under both stable and manually disturbed surface conditions. The three-dimensional RMSE ranged from 0.77 m to 1.59 m, while the horizontal RMSE ranged from 0.52 m to 1.16 m. A tracking experiment further showed that the USV could reduce the estimated horizontal range from about 37.1 m to 1.31 m despite the acoustic source’s limited 2 Hz update rate. These results support the feasibility of a standby USV as a first-response safety layer for diver support, while open-water validation remains future work.

Keywords: Innovative Robot Designs, Cooperation and Collaboration in Human-Robot Teams
Abstract: In this study, an Unmanned Underwater Vehicle (UUV) equipped with underwater radio communication is developed, with a focus on noise-aware system design for stable communication. While underwater radio communication enables low-latency transmission and reduced sensitivity to device orientation, it is highly susceptible to electromagnetic noise generated by the vehicle itself and its onboard electrical/electronic components. To address this issue, an integrated noise mitigation approach is proposed, including a separated power architecture, EMC filters, optimized wiring design, structural grounding, and electrical isolation of the communication system. The proposed system is expected to reduce noise influence and contribute to stable underwater radio communication, providing a basis for future experimental validation.

Keywords: Assistive Robotics, Robots in Education, Therapy and Rehabilitation, User-centered Design of Robots
Abstract: Social robots have been widely explored as tools for autism intervention, yet this literature has focused predominantly on children and has rarely involved autistic adults as active contributors to design. This creates a mismatch between existing systems and the social-cognitive challenges autistic adults actually face in everyday life, including navigating ambiguous interpersonal contexts, managing conversational timing, and interpreting implied emotional meaning. To address this gap, we conducted an online focus group and co-design session with five autistic adults to explore what a social robot for social-cognition training should do, how it should interact, and under what conditions it would be genuinely useful. The 90-minute session combined open discussion with structured co-design activities on a shared digital whiteboard, and the resulting verbal and visual data were analysed using reflexive thematic analysis. The analysis yielded seven themes that define core design requirements: the robot should function as a scaffold rather than a substitute, prioritise authenticity over comfort, provide personalised and user-controlled feedback, accommodate emotional self-awareness gaps, respect privacy and contextual boundaries, support rehearsal for real-world social situations, and remain configurable in identity, form, and expression. Together, these findings suggest that autistic adults envision a robot not as a persistent companion, but as a private and adaptive rehearsal partner designed to promote independence over time. The study offers a co-designed, first-person grounded foundation for adult-focused social robot training systems.

Keywords: Personalities for Robotic or Virtual Characters, User-centered Design of Robots, Medical and Surgical Applications
Abstract: Eye imaging in children, especially high-resolution imaging like Optical Coherence Tomography, is notoriously difficult. Limited cooperation in pediatric patients (often driven by fear, lack of engagement, or developmental constraints) prevents ophthalmologists from obtaining the high-quality imaging necessary for accurate diagnosis. While ophthalmologists use child-friendly tools to encourage cooperation, failure of these methods often requires examinations under anesthesia, exposing young patients to the risks of general sedation. Our key insight is that transforming the Optical Coherence Tomography examination equipment into an expressive robotic character may enhance the pediatric patient experience and improve diagnostic accuracy for eye diseases. In this work, we developed a panel of character prototypes for a robotic eye examination system and presented video demonstrations of these prototypes to caregivers in clinic. Building on caregivers’ insights on desired features, we establish design directions for robot characters in pediatric eye exams, using a Research through Design methodology that integrates animation theory, domain expertise, and guardian feedback.

Keywords: Medical and Surgical Applications, Child-Robot Interaction, Applications of Social Robots
Abstract: Anxiety screening is essential in pediatric care, yet standardized instruments like the STAIC rely on trained human administrators, creating a bottleneck in resource-constrained settings. However, validated clinical anxiety assessment by robots remains an unexplored area. In this paper, we present a system in which a social robot autonomously administers the STAIC state anxiety scale to children in pediatric oncology settings. Our system design mirrors clinical assessment practice: rapport building, visual answer cues, answer confirmation, and graduated error recovery derived from psychologist techniques.

We deployed across three sites with 81 children, comparing robot to human clinician administration, and investigate two research questions: whether a robot can reliably administer the STAIC (RQ1), and how robot and human scores compare (RQ2). All children completed the assessment, with item response classification accuracy of 94.2% on the first attempt and 100% after retry, with a median session duration of 8 minutes. Field observations indicate that children expressed enthusiasm and requested future sessions. Robot scores were significantly lower than human scores overall (p = 0.032), with the effect strongest in healthy children. We discuss potential explanations and implications for interaction design and clinical application.

Keywords: Child-Robot Interaction, Robots in Education, Therapy and Rehabilitation, Social Learning and Skill Acquisition Via Teaching and Imitation
Abstract: Emotional understanding is a key component of Theory of Mind (ToM), yet many autistic children struggle with recognizing and interpreting emotions. At the same time, they tend to enjoy interacting with robots, which are perceived as predictable and less socially complex. This pilot study investigates the use of age-tailored socially assistive robots in a rehabilitation setting to support emotion comprehension and enhance social engagement in autistic children. A novel approach was employed, combining social stories with multimodal expressive cues to train multiple components of emotional understanding within a single session. Six verbally able children without higher support needs participated in 4–8 individual sessions of 30 minutes over two months. Activities were differentiated by developmental stage: younger children worked with explicit, context-based emotional scenarios, while older children engaged with implicit emotional cues embedded in social narratives. Preliminary results suggest improved recognition of emotions, increased conversations with adults, and empathetic responses, suggesting the therapeutic potential of age-sensitive robot interaction design.

Keywords: Robots in Education, Therapy and Rehabilitation, User-centered Design of Robots, Child-Robot Interaction
Abstract: Children with Autism Spectrum Disorder (ASD) face significant social challenges. While interactive robots have emerged as a promising tool for ASD intervention, most existing platforms struggle to adapt to the specific preferences of children with ASD and the requirements of clinical intervention methodologies. Adopting a Participatory Design (PD) approach, this study collaboratively formulates robot design strategies tailored to the unique preferences and therapeutic needs of young children with ASD. Through semi-structured interviews with six clinicians and iterative evaluations involving 19 experts and 35 children, a design framework was established, identifying two core dimensions: Variation (emphasizing progressive and individualized content) and Generalization (advocating for anthropomorphic and natural designs to facilitate skill transfer). Building on these co-designed principles, a robot prototype integrated with the Picture Exchange Communication System (PECS) was developed. A case-based usability evaluation using a one-group pretest-posttest design was then conducted to validate the functional robustness and clinical efficacy of the system. Findings from a preliminary evaluation with three children indicated that the robot-assisted intervention facilitated communicative intent across all participants. This research innovatively extends PECS theory into digital contexts and provides a professional roadmap for future robotic development, offering a promising avenue to alleviate the shortage of clinical resources.

Keywords: Robot Companions and Social Robots, Creating Human-Robot Relationships, Robots in Education, Therapy and Rehabilitation
Abstract: This interdisciplinary research paper analyses the encounters of high school students with a Catholic robot SanTO in the school chapel. Based on conversational data – logs of verbal interactions between students and the robot, observational data – behavior demonstrated by the students, and round discussions data – impressions students had about the robot, the paper indicates potential and limitations for using a robotic device in a religious space (chapel) and for a religious context. The study demonstrates the complexity of such a setting; behavioral ambivalence among participants with a lack of explicit negative behavior; and a tendency to prefer the robot interaction when seeking factual information, and human interaction when seeking religious support and guidance.

Keywords: Sound design for robots, Non-verbal Cues and Expressiveness, Robot Companions and Social Robots
Abstract: Vocal expressiveness is a critical determinant of social presence in Human-Robot Interaction, particularly in assistive domains. While Large Language Models have significantly advanced conversational capabilities, standard Text-to-Speech engines often lack the emotional depth required for empathic interaction, leading to a perceptual mismatch between the robot's role and its voice. This study investigates the impact of integrating an emotionally adaptive neural Text-To-Speech into the UBTECH Alpha Mini humanoid robot. A two-stage methodology is proposed, comprising a preliminary survey to ensure visual-auditory congruency based on the robot's appearance, followed by a user study comparing a standard baseline against neural vocal variants across different emotional scenarios. Results indicate that neural voices significantly outperform the baseline in terms of perceived warmth, anthropomorphism, and likability. Furthermore, the findings highlight that adapting vocal timbre to the specific emotional context is essential for enhancing the communicative effectiveness and social acceptance of assistive robots.

Keywords: Machine Learning and Adaptation, Non-verbal Cues and Expressiveness
Abstract: Accurate and low-latency sound source direction estimation is essential for natural human--robot interaction. However, existing deep learning based direction-of-arrival (DoA) estimation studies have primarily focused on improving estimation accuracy, while validation of real-time operation and spatial robustness in practical interactive environments remains limited.

In this paper, we propose a lightweight real-time 3D DoA estimation framework, termed Direct-Vector Sound Source Localization (DV-SSL). The proposed model directly regresses a Cartesian direction vector, thereby removing grid-based post-processing and reducing computational cost. In addition, the architecture incorporates a stateful GRU-based streaming structure, frequency pooling, and frequency attention pooling to improve efficiency and robustness in continuous inference scenarios.

The proposed system was trained on simulation data and evaluated under various azimuth and elevation conditions in both simulation and a humanoid upper-body torso robot testbed. Experimental results show that the proposed method achieves stable real-time operation and robust 3D sound source direction estimation performance, demonstrating its applicability to humanoid robot based human--robot interaction systems.

Keywords: Linguistic Communication and Dialogue, Multimodal Interaction and Conversational Skills, Non-verbal Cues and Expressiveness
Abstract: Humans heavily rely on subtle pragmatic cues, such as sarcasm, in their social interactions, including with robots. When robots fail to recognize nonliteral intent, their responses can appear rigid or socially inappropriate.

This work proposes a real-time sarcasm detection system that integrates textual and audio cues to enable appropriate responses by the robot. Multiple sarcasm detection approaches were selected and comparatively evaluated with respect to their latency under real-time conditions. Results from an online study show that robots employing text and tone-based sarcasm detection and reacting according to social rules were perceived as significantly more anthropomorphic and intelligent than those not understanding sarcasm. The findings suggest that sarcasm-awareness through text and tone input enhances the perceived social intelligence and human-likeness of robots. Furthermore, the style of sarcasm used by the human interaction partner influences how authentic the robot’s responses are perceived to be, but does not significantly affect likeability.

Keywords: Linguistic Communication and Dialogue, Applications of Social Robots, Robot Companions and Social Robots
Abstract: Speech is one of the most important communication modalities due to its ability to convey complex information efficiently. In Human-Robot Interaction, Text-To-Speech (TTS) systems play a crucial role, as they allow the robot to emit verbal messages similar to human speech. However, these systems can suffer when faced with complex elements like named entities, foreign words, or the presence of acronyms. This work presents the MultiLanguage Speech Enhancement Module (ML-SEM), a system for enhancing the pronunciation of complex elements in Spanish. This module analyses the transcription of the robot's speech, flagging foreign words, named entities and acronyms. Foreign words are phonetically transcribed so they can be properly processed by a Spanish TTS module. Pronunciation of named entities and/or acronyms is decided through local searches in a database or queries to LLM models if it is an item never encountered before. To evaluate the performance of the proposed solution, we have conducted quantitative and qualitative analyses of the proposed system. Quantitative results show that the proposed solution is able to perform at a satisfactory level while abiding by the constraints present in real-world interactions. In addition, we conducted a user study to analyse the impact that the ML-SEM module has on how users perceive a social robot. The results showed a general improvement in this perception, particularly in the sociability dimension.

Keywords: Machine Learning and Adaptation, Linguistic Communication and Dialogue, Assistive Robotics
Abstract: We introduce an end-to-end pipeline for customizable wake word detection which does not rely on preexisting dataset and designed for robotics and smart interactive devices. Leveraging the pre-trained OpenAI Whisper encoder and a LoRA adapter, our method trains with only a few dozen user-provided audio samples, removing the need for large datasets typically required by state-of-the-art systems. The automated pipeline handles sample collection, synthetic audio generation, training, and real-time evaluation, achieving low false acceptance and rejection rates while maintaining efficient GPU inference. We validate the framework on three different wake words and test its competence with real-time evaluation runs. Our results highlight the system’s potential for deployment of personalized, high-accuracy voice activation in robotics and other branded or individual-based agent solutions, while using a fraction of the data needed by other frameworks.