Posters

High consequence procedural training in Virtual Reality currently relies on reactive Intelligent Tutoring Systems that correct novices only after a physical error is committed. Allowing erroneous physical execution fosters negative training transfer and incorrect motor schema consolidation. We introduce Semantic Dissonance Detection, a cross-modal machine learning framework designed to predict and intercept cognitive action slips before physical contact occurs. By mapping a user's verbalized intent into a semantic latent space using a frozen Vision-Language Model, and simultaneously projecting their real-time 13-element OpenXR kinematic state vector through a custom Contrastive Transformer, the system continuously estimates intent-action divergence. To solve spatial overfitting, we apply Relative Position Normalization across 45-frame rolling temporal windows. The kinematic encoder is trained using Supervised Contrastive Learning to mitigate false negative penalties among kinematically similar linear motions. In a controlled study (N=50, 1500 trajectories) evaluating a four-action spatial protocol, our framework achieved a 93.7% Leave-One-Subject-Out (LOSO) cross-validation accuracy. During applied pedagogical testing, the framework isolated a stable congruence baseline that successfully identified action slips in our experimental setting within the first 0.25 seconds of movement. This 750ms predictive safety buffer allowed users to correct their trajectories mid-reach, associated with a 95.3% relative reduction in completed physical errors compared to reactive baselines. This work offers a focused yet rigorous demonstration of supervised cross-modal kinematic-semantic alignment within a constrained spatial training environment for early predictive modeling in immersive environments.

TurniBlocks is a head-mounted mixed reality system designed to support turn-taking training for children with autism spectrum disorder (ASD). Inspired by the role collaboration and turn-taking mechanisms of LEGO®-Based Therapy (LBT), it transforms block-based cooperative roles into virtual building tasks embedded in a familiar real therapeutic space, forming a multi-party collaboration model. To evaluate its usability, we conducted a user study. Results suggest that TurniBlocks can partially enhance children’s engagement, collaborative turn-taking behaviors, and social participation, highlighting the potential of head-mounted MR for structured autism social training.

Traditional sports training often struggles to deliver real-time, personalized, and biomechanically accurate feedback during rapid physical execution. This paper presents XR Badminton, a system architecture engineered for adaptive coaching with high-fidelity physics for virtual reality badminton training running on standalone head-mounted displays. The framework introduces three core contributions: (1) a real-time tracking pipeline that utilizes a Holt-Linear filtered state machine to segment complex overhead strokes into a seven-phase kinematic sequence, scoring performance across eleven expert-derived biomechanical rules via distinct mathematical mapping profiles; (2) a high-fidelity training environment featuring a procedural, aerodynamic shuttlecock simulation with nose-heavy physical balance and quadratic drag trajectory prediction; and (3) an adaptive coaching layer that isolates a player's most critical mechanical flaw via priority-weighted deficit selection and dynamically routes tailored corrective feedback through seamless modality substitution across visual HUDs, spatial audio, and haptic cues. We outline the core system components, demonstrate a non-blocking telemetry logging infrastructure used for high-frequency player tracking, and present a verification protocol validating the framework's tracking calculations and trajectory physics

We present a 4D human-scene reconstruction method for precise human-scene alignment in bouldering scenarios. 4D bouldering replay can convey complex climbing movements from different perspectives and support asynchronous collaborative training. However, it is nontrivial to accurately reconstruct and align human motion with climbing-wall geometry from monocular video. To address this problem, we introduce a contact-aware fine-tuning method that builds upon state-of-the-art human-scene reconstruction methods. By incorporating human-scene contact and penetration constraints, our method improves alignment between reconstructed climbers and climbing walls. We further develop an AR bouldering system for optimal climbing-wall anchoring and motion replay.

Augmented Reality has been used to support users in a range of scenarios, often showing advantages over traditional approaches such as paper-based instructions. However, existing systems rarely consider the user's expertise when presenting instructional content. We investigate how expertise can be predicted from data available in a consumer-grade head-mounted display and use these predictions to implement an adaptive augmented reality assembly tutorial that tailors instructional support to the user. Expertise classification is performed on a public dataset using different machine learning algorithms. Results show motion-based expertise classification is feasible, demonstrating the potential for adaptive instructional systems.

Communication with emotionally expressive parents may induce negative emotions in early childhood teachers, challenging emotion-management training. Although VR can simulate such interactions, prior work relies on passive observation or scripted dialogues. We evaluated an LLM-driven virtual parent dialogue approach in VR for eliciting negative emotions in pre-service early childhood teachers. Twenty-eight participants completed dialogues with anxious and angry virtual parents while subjective and physiological data were collected. Both dialogues elicited negative subjective changes and physiological responses; participants recognized affective profiles, and angry parents produced stronger negative responses. Findings provide an experimental basis for teacher training and VR emotion-elicitation research.

As social media has become a central part of daily life, algorithmically driven recommender systems can now shape many human decisions, including what they view, what they focus on, and what they do. These recommendations introduce algorithmic bias based on the assumptions and choices made during the system's development and data use. The benefits of using recommender systems to enhance the social media experience are substantial, but so too is the potential to affect users' cognitive processes and decision-making through embedded algorithmic biases. Most current studies addressing algorithmic bias focus on model design, auditing, and regulation, with little to no research on how users interact with and dispute these systems. To address this gap, this study proposes to develop an interdisciplinary Human-Computer Interaction (HCI) method using Extended Reality (XR) to create a fully immersive virtual environment that renders abstract algorithmic biases as tangible, manipulable spatial elements. Drawing on self-organization in ant colonies as my model, I am developing the A-N-T-S (Apprehension–Negation–Transition–specialization) model of user behavior to understand what causes users to detect and oppose algorithmic bias. The ANTS model is guiding the development of an XR prototype that brings together geographic information systems and multimodal interaction to represent weighted algorithmic decisions as dynamic "crack zones" and bias maps. In the XR environment, users can navigate a virtually distorted city, experience the spatial manifestation of recommender logic, and test different paths to escape or intervene in the recommendation process (see Fig. 1). My planned user study will investigate whether using this XR environment based on the ANTS model can improve users' ability to perceive, resist, and interpret algorithmic bias, providing a design framework for more transparent, participatory, and user-focused algorithms.

Brain cancer affects 76,000 Americans annually. Approximately 80% of brain cancer survivors (BCSs) experience mild cognitive impairment (MCI), affecting memory and executive function (MEF). Although studies show that paper/computerized cognitive training mitigates MCI, meta-analyses confirm that virtual reality can improve MEF in BCSs. We developed a Virtual Reality Cognitive Rehabilitation Training (VR-CRT) platform with immersive cognitive exercises. As a follow-up to our preliminary study, our current four-week intervention/control clinical study (N = 36) will evaluate the feasibility and effect of VR-CRT. Cognitive performance is measured at baseline/pre-post using HVLT/COWA/TMT tests. Anticipated outcomes are modest-to-significant MEF improvements in BCSs with MCI.

As neurotechnology is converging with Extended Reality (XR), advances in neurophysiological monitoring are expanding the ability of XR systems to gain insight into users' attentional, cognitive, and affective states. Existing Responsible XR frameworks address risks associated with behavioural and physiological data, yet Neuro-XR (i.e., XR systems augmented with neural sensing) introduces a different challenge; that is, access to information about users' internal mental processes, raising concerns around mental privacy, cognitive liberty, neuro-surveillance, and algorithmic influence. We argue that this convergence necessitates a new research agenda centred on Responsible Neuro-XR. Drawing on literature in XR ethics, neuroethics, and neurotechnology governance, we identified four core ethical challenges and propose four design considerations for the ISMAR community. We also highlighted research opportunities in neuroadaptive learning, collaboration, wellbeing, and inclusive design where the community can (re)shape responsible development of future brain-aware immersive systems.

Industrial assembly and maintenance tasks require access to complex technical knowledge, yet consulting manuals or experts is often impractical during hands-busy and safety-critical work. These challenges particularly affect non-knowledge workers (e.g., technicians and operators) who must resolve issues in situ under time pressure. While Augmented Reality (AR) has been used to support industrial training and guidance, many existing systems rely on predefined instructions that cannot adapt to workers’ ad hoc questions. We present a prototype AR system that integrates an LLM-powered conversational assistant to provide hands-free assistance, and illustrate it through a pipe assembly and maintenance use case.

Dynamic balance assessment is important in sports and rehabilitation. Mixed Reality (MR) offers new opportunities for portable diagnostics. This study investigated whether the standardized Y-Balance Test (YBT) can be reliably and validly administered in MR environments. Twenty-four participants performed the YBT in a real environment (RE), Augmented Reality (AR), Augmented Virtuality (AV), and Virtual Environment (VE) on two test days. Results confirmed validity across all MR conditions (p > 0.05), with good test-retest reliability in AR and VE and moderate-to-good reliability in AV. Despite higher perceived demands, MR-based YBT assessment showed outcomes comparable to RE, supporting its feasibility for diagnostics.

Conversational and augmented reality technologies offer new opportunities for classroom learning support, yet their design often overlooks the social and emotional conditions that shape students’ willingness to ask for help. This paper presents a co-design study of a conversational AR avatar intended to support low-pressure help-seeking and emotional regulation in classrooms. Through workshops and surveys with 47 students and 13 teachers, we found that the main barriers to help-seeking were social rather than purely cognitive, especially fear of judgement, embarrassment, and unwanted visibility. Participants preferred the avatar to act as a low-authority peer or guide, provide scaffolded support rather than direct answers, and offer multimodal interaction options that preserve privacy. Based on these findings, we derive design requirements for conversational AR avatars as socially aware, classroom-integrated support tools.

eXtended Reality (XR) is a key technology for industrial digital twins, yet adoption is constrained by inefficient content authoring and management. We propose an XR content authoring and management strategy based on an XR Contents Fabric architecture, which extends the data fabric concept to XR by separating functional and semantic content elements to enable reusable, maintainable pipelines. We demonstrate it through an XR-based airport vehicle checkpoint security screening training service. A pilot with airport security instructors confirms that the framework supports instructor-centered authoring, continuous content operation, and service sustainability, providing a foundation for future AI-driven intelligent XR content platforms.

Endovascular procedures such as thrombectomy require hands-on training constrained by limited access to expensive simulation hardware. We present a mixed reality training simulation for endovascular thrombectomy on Meta Quest 3. Real patient CT-A scan data is rendered holographically using a custom HLSL raymarching shader, enabling trainees to visualize patient-specific vascular anatomy in their physical environment. The system supports interactive guidewire and catheter navigation through reconstructed vascular geometry, with haptic feedback during procedural steps. A fluoroscopy simulation mode and radiation exposure monitoring replicate clinical workflow conditions, demonstrating feasibility of patient-specific holographic anatomy visualization on consumer MR hardware for accessible endovascular training.

Claws of Change is an immersive virtual reality educational experience that places learners in the body of an American lobster exploring the Atlantic Ocean. Participants travel through time to witness the effects of climate change using a novel claw-based swimming locomotion system that strengthens embodiment. The experience aims to foster empathy for marine life through non-human perspective-taking and environmental storytelling. A pilot study in a university sustainability classroom found the experience engaging, easy to learn, moderately immersive, and effective for climate education. Future work will improve accessibility, locomotion, environmental interactions, and educational content.

Extended Reality (XR) is an umbrella term that includes various immersive technologies, including Virtual Reality (VR), Augmented Reality (AR), and Mixed Reality (MR) \cite{whatisxr,mrdef}. Recently, XR has gained tremendous traction and application in the educational sector, particularly for training professionals aiming to refine their skills and deepen their expertise in specific disciplines. This transformative tool has proven to be exceptionally beneficial across numerous industries, with agriculture standing out as a crucial area of application \cite{agrivr}. By offering immersive and interactive simulations that replicate diverse farming environments and the operational workflows inherent to them, XR significantly enhances the educational experience.

We present an approach for extracting 3D object representations and their spatial positions from egocentrically captured panoramic video recordings. The system processes panoramic videos captured by a head-worn camera, with a focus on instructional scenarios in which object configurations and interactions are central to learning. The extracted 3D objects are tracked over time and replayed in augmented reality (AR). By decoupling instructional information from the original camera viewpoint, learners can observe and follow procedures from their own perspectives. The primary motivation is to reduce constraints during capture, permitting natural head movement while maintaining spatial fidelity required for AR-based instructions.

We present a VR rowing exergame that uses static cooperative feedforward avatars to support pacing, motivation, and performance. After data cleaning, 26 participants were included in a repeated-measures study across baseline, +3\%, and +5\% conditions. Both avatar conditions significantly increased stroke rate and power output compared with baseline. Interest/Enjoyment and Effort/Importance differed significantly across conditions, while Pressure/Tension did not. Qualitative feedback suggested that +3\% felt smoother and more motivating, whereas +5\% was sometimes perceived as more effortful or tiring. These findings suggest that cooperative avatar pacing can support exercise without relying on competitive pressure.

This paper investigates the effectiveness of a real-time parametric virtual reality (VR) environment for teaching the buckling behaviour of concrete columns in civil engineering education. The system enables users to interactively modify structural parameters, including column height, cross-section, applied load, and boundary conditions, while observing real-time deformation and failure behaviour. A between-subject experiment compared VR-based learning with traditional video instruction, evaluating conceptual understanding, memory retention, cognitive load, and learner confidence. Results showed that the VR group experienced lower perceived cognitive load and greater increases in confidence and engagement. Although no statistically significant differences were found in learning outcomes between the two methods, the confidence measures and qualitative findings indicated that the interactive features of the VR environment provided meaningful pedagogical benefits. The work contributes a Predict–Observe–Explain-based VR approach that integrates symbolic and visual representations to support the teaching of structural behaviour that is difficult to demonstrate through physical experimentation.

Stress inoculation training (SIT) builds psychological resilience by progressively exposing individuals to controlled stressors. We present Resilience Protocol, a first-person virtual reality application designed as an immersive SIT experience built in Unreal Engine 5.6. A MetaHuman-driven AI antagonist pursues the participant through three escalating stages of increasing challenge, incorporating physics-based interactive puzzles and an adaptive audio system that responds dynamically to in-game events. The system translates established cognitive-behavioral SIT methodology into an interactive VR format, offering an engaging and scalable alternative to traditional stress inoculation approaches. We describe the system architecture, design rationale, and potential applications in resilience training contexts.

Professional study often requires learners to connect abstract concepts with concrete experience, but conventional review rarely preserves the everyday cues that could support those connections. We present spatially anchored knowledge graphs, a mixed-reality authoring workflow that turns captured objects into source-grounded memory anchors. The prototype captures object snapshots in passthrough MR, generates role-aware association paths through exploratory AI or textbook-grounded retrieval, and transfers selected paths into a VR knowledge graph for spatial organization and source inspection. It positions MR as situated capture and VR as embodied graph authoring, providing a concrete basis for studying associative recall and review workload.

Augmented reality (AR) can support vocabulary learning by placing mnemonic cues in the physical contexts where new words are encountered. Prior work has augmented real objects with bilingual labels, audio, keywords, and 3D visualisations of those keywords interacting with the objects, but these cues are typically predefined by system designers. This limits scalability and reduces learners' ability to construct personal mnemonic links. By grounding mnemonics in objects that learners already recognise, use, and associate with prior experiences, the system can connect new vocabulary to existing mental models rather than requiring learners to adopt designer-defined associations. We propose a generative AR approach in which learners create personalised 3D mnemonics by transforming familiar objects in their surroundings into situated mnemonic anchors. We position this concept within a timeline of related work and argue that recent advances in generative 3D asset creation enable a new design point: personalised 3D mnemonics that do not merely place content in the physical learning environment, but modify familiar real-world objects so that they become part of the mnemonic itself.

Subsea infrastructure faces an increased risk of damage as geopolitical tensions rise in the Baltic Sea region, necessitating cooperation between public and private actors. Current monitoring systems used by national coast guards and cable solutions providers are based on 2D sea chart data visualization, with limited capability for displaying contextual information, and do not effectively support collaboration and knowledge sharing. Based on an iterative user-centered design process, we introduce a concept design for sea-scale situation awareness, integrating 3D bathymetry, satellite, ship transponder, geodata, and cable system sensor data into an integrated view, evaluating immersive vs non-immersive XR modalities.

Wayfinding support in mixed reality (MR) is usually delivered as fixed, turn-by-turn overlays that substitute a visitor’s spatial reasoning. This is efficient for routing but poorly suited to cultural environments, where orientation and exploration matter alongside reaching a target. We present an early-stage framework that treats embodied navigation behaviour (dwell, hesitation, backtracking, heading change from head-mounted display tracking) as signals for lightweight inference of a visitor’s navigation state, then adapts MR spatial prompts accordingly, from in-situ cues to withheld guidance. We describe the pipeline, a museum walkthrough, and a planned within-subjects evaluation. No user study has been conducted.

Outdoor heritage and pilgrimage sites present navigation challenges that existing AR systems have not addressed: GPS-grade localization without structured geometric models, culturally specific destinations expressed in informal or vernacular language, and landmark-rich but semantically unannotated spaces. We present HARI-AR, an AR navigation system deployed at the Tirumala Venkateshwara temple complex, integrating a five-agent LLM orchestration pipeline over a custom 3,650-node OSM navigation graph enriched with 155 hand-curated semantic POIs to deliver landmark-anchored AR guidance via natural language voice queries. Rather than proposing a new AR rendering paradigm or novel agent architecture, the primary contribution is the deployment, empirical evaluation, and validation of semantically grounded landmark-anchored guidance in a real outdoor heritage environment at scale. A within-subjects study (N=24) showed landmark-referenced instructions significantly reduced directional errors compared to GPS-arrow-only guidance (21/24 vs. 14/24 correct turns; Fisher's exact p<0.05), with a SUS score of 74.2, establishing that natural language landmark navigation is both feasible and measurably effective for first-time users in complex cultural environments.

Collaboration is a core component of augmented reality (AR) usage, especially in co-located settings where shared visualizations can improve performance and communication. While symmetric AR collaboration has been widely studied, asymmetric settings where only one or a subset of users have access to AR content remain underexplored. To investigate this phenomenon, we conducted a user study with 24 participants working in pairs on a tangram puzzle task. Participants either used an AR headset (Hololens 2) with AR guidance in the form of puzzle outlines in the workspace or, when having no access to the AR instructions, utilized Aria glasses for non-intrusive eye tracking. Results demonstrate that AR improves objective performance; however, the asymmetric constellations significantly change team communication, gaze behavior, and subjective feelings about the collaborative process. This situation reveals challenges for supporting effective collaboration when AR access is unequal.

Most eXtended Reality (XR) applications for industrial scenarios are developed and validated under controlled laboratory conditions, where environmental variability, operational constraints, and real- life workflows are partially represented. Although such evaluations provide important evidence of feasibility and usability, they often limit understanding of how XR performs when integrated into industrial routines, where safety requirements, task interruptions, and contextual factors can influence adoption. This work reports an XR framework designed for industrial application domains, exploring individual and collaborative contexts (co-located/remote), through virtual and augmented features interconnected with a factory digital twin to support human technicians. The framework was developed through a human-centered process conducted in cooperation with industry partners across successive evaluation stages, from laboratory studies to shop floor deployment. Early findings indicate promising potential for XR adoption while revealing key design considerations for transferring XR solutions beyond controlled settings.

Social VR creators film inside VRChat while recording controls remain on the desktop. VRSlate brings recording control, tally feedback, scene/shot/take labels, and post-take handoff into the camera operator's in-headset avatar-menu workflow. It bridges VRChat OSC parameters with OBS WebSocket, records crash-resilient MKV takes, remuxes H.264/AAC media to MP4, and writes a CSV shot-list index. In one author-led project timing, an off-HMD manual control step took 20--34 s, versus about 3 s with VRSlate.

This experimental study investigates how characteristics of virtual avatars build trust and encourage user (excessive) self-disclosure within a VR-based digital interview context. Thus, we manipulate agents’ social cues and conversational reciprocity in an adult lifespan sample. We specifically focus on older adults, whose age-related changes in motivational priorities and cognitive control may increase their susceptibility to oversharing. Self-disclosure is operationalized via response rates and elaboration during conversations about personal information. By evaluating these dynamics, this research clarifies privacy boundaries and rapport-building in human-agent interactions across different age groups. Data collection is ongoing, with final results expected in August 2026.

We present three locomotion adaptation approaches: Motion Acceleration, Turn Acceleration, and Motion Deceleration to improve postural stability during body-state transitions in virtual reality (VR). The system detects standing-to-walking, turning, and walking-to-stopping transitions and applies adaptative locomotion smoothing. Motion Acceleration gradually increases locomotion speed when users begin walking, Turn Acceleration smooths rotation while turning, and Motion Deceleration gradually reduces movement speed before stopping. We evaluated these techniques in a virtual navigation task using objective balance measures. Preliminary results show reduced center of pressure (COP) velocity and improved balance confidence. These findings suggest that locomotion adaptations can improve balance and navigation experience.

Journaling has been widely used in mental well-being contexts to support self-awareness, emotional reflection, and emotion regulation. Digital journaling extends this practice by reducing writing effort and enabling interactive, guided, and more engaging reflection. In this paper, we present DayScape, a digital journaling system that combines an LLM-driven virtual agent with generative VR. Instead of relying only on text-based entries, DayScape transforms users’ daily reflections into immersive, supportive virtual spaces with generated environments, background soundscapes, and music. We conducted a pilot study with eight participants to evaluate system usability, the perceived quality of the generated VR experiences, and users’ views of VR-based journaling. Our findings provide early insights into how generative VR could support daily reflection and inform the design of a future full-scale study and the development of generative VR systems for mental well-being contexts.

Inspecting large or hazardous facilities requires scarce on-site experts. Video inspection locks the expert's viewpoint; point clouds lack the fidelity to judge fine surfaces. We present a shared MR–VR teleinspection system where an on-site MR user and a remote VR expert co-inhabit one photorealistic 3D Gaussian Splatting replica—shared, metric, and touchable at once, unlike single-user or unregistered 3DGS. A lightweight offline pipeline (single-marker scaling, 2DGS collider extraction, image-based localization) registers users and assets into one common metric frame, enabling on-surface pointing, metric measurement, and cross-reality annotation. We validate each component in a proof-of-concept room.

In MR environments, virtual avatars play an important role in supporting users’ understanding of smart home device states, communicating behavioral intentions, and enhancing emotional experience. However, existing research has offered relatively limited discussion on the visual expression of virtual avatars for smart home products.This study first conducted semi-structured interviews to identify the suitable application scope of virtual avatars for smart home devices. It then employed participatory design to explore a design framework for anthropomorphic avatars in MR environments from the perspectives of product attributes and user emotions. Based on these findings, an initial MLLM-based prototype was developed to explore the workflow from physical device recognition to virtual avatar generation. This study aims to move beyond the design of individual visual forms in smart home anthropomorphism and develop a strategy-oriented framework grounded in design metaphor. By providing structured rule constraints for MLLMs, the proposed approach supports more systematic and context-aware avatar generation, offering a new perspective for the design and application of smart home avatars in MR environments.

This study investigates the effect of avatar design on users’ perception in social virtual reality (VR). While it is a common understanding that avatar usage enhances user perception in VR applications, little is known about the effect of different avatar designs on the user perception within social scenarios with seated positions. We conducted a between-subjects user study with 70 participants (35 acquainted dyads) attending a virtual opera event. Bayesian statistical analysis revealed equivalence for hand-and-hands vs. full-body avatars for all measures of user perception applied. Systematic content analysis of post-experiment interviews show 90% of participants preferred full-body representations, nonetheless.

As history shapes current perspectives, the way we present past events should be carefully curated, enabling responsible societies. In this work, together with the relevant communities (ex-inmates, contemporaries, experts), we co-design and evaluate a social VR experience showcasing the history of the Franco regime’s political prisoners. The results, analysed alongside current literature, show that depending on the type of memory the groups hold, they focus on different aspects of the experience during the co-design and evaluation. This finding suggests that to create meaningful memory-based experiences, all groups shaped by that memory should be engaged in the design process and decisions.

Impossible Spaces in virtual reality enable navigation of large virtual environments within limited physical areas by connecting overlapping spaces through portals. We investigated whether spatial audio propagating through such portals helps users to navigate efficiently. We conducted a user study with 31 participants comparing navigation in impossible spaces with and without propagated audio. No significant differences were observed, however, qualitative feedback varied: some participants found audio helpful, and others perceived it as overwhelming. This suggests that spatial audio through portals may not consistently improve navigation in impossible spaces, and benefits may depend on individuals' cognitive and auditory processing skills.

Existing clinical communication training is limited by cost, scalability, and static interactions. We present CrisisX, an AI-driven VR system featuring an emotionally adaptive virtual patient. Driven by a novel Emotional State Engine and large language models, the agent dynamically escalates or de-escalates in response to the trainee’s voice interactions and provides explainable, rubric-based feedback. A two-phase evaluation, culminating in a clinical pilot with 14 health-profession students, demonstrated high immersion, perceived usefulness, and good usability (UMUX-LITE 76.8/100). Crucially, workload analysis revealed a high-effort, low-frustration cognitive "sweet spot," proving particularly effective at bridging the practical experience gap for novices. These results suggest CrisisX offers a scalable, emotionally realistic approach to healthcare communication training.

The increasing demand for mental healthcare, and recent advances in large language models (LLMs), have led to the widespread adoption of intelligent virtual agents (IVAs). IVAs can be designed with a diverse range of visual features, and LLMs offer customization options for their communication. In our 2x2 within-subject study, we examined how visual human-likeness and conversational style of IVAs in virtual reality (VR) influence perceived trust and rapport. 35 participants engaged in psychotherapeutical self-help exercises with four different IVAs that varied in embodiment (human-like vs. abstract) and conversational style (human-like vs. mechanical). Human-like conversational style had the strongest positive effect.

Even with the same VR collaboration system and task, collaboration outcomes can vary by participant characteristics. This study examined how empathic concern relates to performance and dialogue structure in a VR director–matcher Tangram task. Thirty-two dyads completed the task, and completion time and matcher turn count were measured per round. Matcher's empathic concern was found to be positively associated with longer completion time, and this relationship was statistically explained by increased matcher turn count. These findings suggest that empathic concern may be associated with greater conversational involvement, which can appear as a communication cost in time-sensitive VR collaboration.

Communication barriers for Deaf individuals persist in everyday scenarios where interaction relies primarily on speech or text, limiting accessibility and increasing the risk of social exclusion. While translation frameworks exist for widely documented sign languages, real-time pipelines for Italian Sign Language (LIS) remain scarce. We present a real-time text-to-LIS framework combining a Large Language Model-driven multi-agent architecture with a motion-capture dataset for 3D avatar animation. Our approach handles out-of-vocabulary signs, supports multiple avatars, and is highly extensible to other sign languages and vocabularies. A user evaluation with Deaf users provided overall promising feedback, suggesting potential extension to Extended Reality applications.

Older adults are a growing demographic group who could bene fit from communication technologies such as telepresence robots and augmented reality avatars. However, they constitute a distinct user group due to age-related changes and life-course experiences. We present a pilot study exploring how older adults perceive (self-reported measures) and behave (gaze and speech behavior) in avatar- and robot-mediated communication, and contrast these results with face-to-face (F2F) interaction. Our initial findings suggest that representation-specific characteristics shape gaze and speech behavior: in robot-mediated communication, multiple robot components may compete for gaze, whereas in avatar-mediated communication, behavioral realism and novelty may influence gaze. We further observe a divergence between self-reported and observed gaze behavior, alongside subtle intra- and inter-individual differences. Together, these initial insights point to representation- and user-dependent effects relevant to the design of future mediated communication systems for older adults and that possibly extend to other demographics.

We present a preliminary study on how the visual presentation of a virtual environment influences performance on a Fitts’ law selection task, which is known to be repetitive and tedious. We compare a "neutral" or sterile environment with a detailed arcade environment suggestive of a competitive atmosphere. Results suggest that the environment had a limited impact on performance, but participants had a more positive impression of the more detailed environment.

This work-in-progress presents an eXtended Reality (XR) interface for Unmanned Traffic Management (UTM). As new airspace regulations emerge to accommodate automated drone flights, representing dense traffic on 2D displays limits spatial understanding of complex urban environments. Designed based on stakeholder input, we developed a cross-reality XR interface that combines Mixed Reality (MR) tabletop overviews with immersive Virtual Reality (VR) to support visualization, monitoring, and planning of drone operations. We integrate our prototype with a real UTM system to show technical feasibility. Preliminary expert feedback indicates promising potential for XR in unmanned airspace management and reveals important considerations for future work.

We present Couch-AR, a spatial augmented reality system that transforms soft furniture into interactive surfaces for lean-back seated interaction. While prior spatial AR systems have focused on rigid surfaces such as tables, floors, and walls, soft furniture remains underexplored. In this work, we showcase seated interaction via touch, swipe, and mid-air gestures for couch-based content navigation and manipulation. A preliminary evaluation revealed a positive user experience across UMUX, UEQ+, and NASA-TLX, which creates the premises for future investigations on seated AR experiences.

VR collaboration allows users to share non-verbal expressions through visual cues. Existing techniques often project these expressions to various locations within the VR space, which can cause confusion as this is not common in real-world interactions. To address this, we introduce EmpathicMirror, a mirror that orients itself near the user's perspective to enable sharing expressions. It uses reflections to reproject VR users’ gestures and emotions to support a VR collaboration. It operates in three modes: Player Dominant, Fixed, and Partner Dominant. These modes manipulate the mirror's position based on the users' view directions or a fixed location in the VR space. Additionally, it rotates itself to allow users to see each other through reflections. Finally, it can facilitate sharing emotions through facial expressions by rendering visual cues and effects on or around the mirror.

VR game development remains largely desktop-centric, requiring developers to repeatedly switch between conventional authoring tools and head-mounted displays for testing. This paper proposes the VR-Native Game Development Framework, comprising four design principles: In-situ Spatial Editing, World-Embedded Visual Logic, Continuous In-Headset Iteration, and Scaffolded Complexity. We instantiated the framework through a functional Unreal Engine 5 environment supporting immersive level editing, visual scripting, integrated Build–Link–Play workflows, and persistent world management, and evaluated the resulting environment through an exploratory user study with 20 participants. Results showed promising usability (SUS = 73.88), low workload (NASA-TLX = 5.03/20), and high ratings for immersion, responsiveness, and satisfaction. The findings suggest that the proposed framework provides a complementary approach to immersive authoring and early-stage gameplay prototyping.

This work explores the early design of Augmented Reality Head-Up Displays (HUDs) to support astronauts during lunar surface missions, using a simulated Extravehicular Activity task in Virtual Reality for user-centered development. Based on preliminary interviews, two HUD concepts were designed, implemented, and tested in a simulated deployment task on the lunar surface. Both HUDs were well received and considered propitious for providing assistance, particularly through a task procedure step list. Early stage feedback provides a foundation for future system iterations and indicates the target group's desire for user customization.

ORCESTRA is a mixed-reality system for programming robot digital twins through no-code waypoint teaching and language-guided control. In a passthrough mixed-reality workspace, users place robot twins on real surfaces, teach trajectories, save robot-relative episodes, or issue spoken/typed commands that a vision-language model converts into structured digital-twin plans. Both interaction modes share a backend for metric grounding, embodiment-aware validation, preview, confirmation, and digital-twin execution. The system supports heterogeneous robot embodiments, including fixed-base manipulators, a mobile base, and a humanoid robot, demonstrating MR validation as a safety layer for language-guided robot programming before physical deployment.

This paper reports a controlled laboratory user study of input modalities for a functional LLM-based AI agent integrated into an Augmented Reality (AR) quality-inspection prototype. The system combines spatial AR inspection guidance with a conversational assistant grounded through Retrieval-Augmented Generation. A within-subject study with 32 participants compared virtual keyboard and speech-to-text input across three inspection-related query tasks. Voice input was significantly faster, achieved higher usability scores, and was preferred for hands-busy scenarios. Results suggest that speech-to-text should serve as the primary modality for rapid operator queries, while virtual keyboard input remains useful as a precision fallback.

Embodied XR assistants need room knowledge that can be reused during dialogue and connected to spatial behavior. A room image can describe visible objects, but the assistant also needs a representation that links object descriptions to approximate locations so it can retrieve, answer, and point or move toward relevant objects later. We propose RoomAnchor, an object-level room memory pipeline for embodied XR assistants. RoomAnchor takes a single egocentric RGB room capture, produces object-level 3D hypotheses, labels detected crops, stores the resulting object records in a retrieval index, and uses an agent to answer room-related questions and select reference anchors for embodied behavior. We evaluated RoomAnchor on 24 room scenes and 1440 manually checked room-related question-answer trials. Across this evaluation, RoomAnchor reliably answered questions about mapped room content while usually retrieving the supporting mapped objects and selecting appropriate embodied reference anchors. Memory construction took 2.381 s on average. The results indicate that compact object-level room memory can support reliable room-question answering without continuous reconstruction or a device-specific depth pipeline.

We present an autonomous robotic portrait-generation system combining real-time face detection, AI-based sketch generation, and robotic drawing. The system captures video frames, extracts facial regions, converts them into minimalist single-line sketches using the Gemini Vision API, optimizes stroke order through graph-based path planning, and executes smooth trajectories on a 6-DoF collaborative manipulator. This perception-cognition-action pipeline integrates computer vision, neural artistic abstraction, motion optimization, and robot control. User ratings on a 5-point scale were high for sketch quality (4.33), perceived execution (4.53), and user experience (4.65), indicating recognizable, appealing, and engaging robotic portraits.

Augmented reality (AR) instructions help people understand how to operate or assemble objects by overlaying visual guidance directly onto the physical environment. However, repeatedly presenting detailed instructions can negatively impact performance. We introduce an interactive AR instruction system that groups instructions based on recurring action patterns, providing detailed guidance on the first occurrence of actions and abstracting repeated sequences. We introduce a novel workflow that automatically organizes narrated expert demonstrations into instruction structures using a multimodal large language model, improving the scalability of AR guidance systems. Our user study shows that grouped instruction visualization is clearly preferred.

Low-cost mobile virtual reality (MVR) devices, such as Google Cardboard, support accessible virtual museum tours, but their limited input channels make interaction challenging. In such systems, head movement serves as both natural viewing and control input, which may cause intention ambiguity and false triggers. This paper proposes a Layered Head-Intention Interaction method for low-cost MVR museum tours. The method identifies five recurring interaction types and translates them into five head-intention layers: natural viewing, spatial movement, viewpoint adjustment, system operation, and exhibit information access. These layers are separated through spatial, angular, temporal, and object cues as design rules. We implemented the method in a Cardboard-based virtual museum prototype and conducted a preliminary user study, showing acceptable usability and feasibility.

Early CMF exploration in product design often begins with informal design intent but becomes fragmented across generated images, comparison boards, and downstream 3D previews. This fragmentation makes it difficult to trace how color, material, finish, and perceived product qualities are interpreted, compared, selected, and transformed across media. This work presents a speech-driven AR concept-board workflow that structures speech-like CMF intent into a design brief, supports LLM-assisted shortlisting and designer-led board comparison, and links a selected candidate to 3D preview inspection. A portable-speaker walkthrough and preliminary designer feedback illustrate its potential for traceable CMF reasoning in early design.

We present an early-stage VR-based system for embodied crochet-style designs through freeform line sketching in 3D space. Unlike existing crochet modeling approaches with surface tiling or pattern-based workflows, our method focuses on sparse, lace-like openwork structures characterized by branching lines and spatial voids. The system translates user-drawn strokes into procedural crochet arrangements and generates models suitable for FDM and SLA 3D printing. Our initial exploration suggests that VR enables intuitive spatial creation of decorative crochet forms, supporting rapid prototyping of open structures. We discuss fabrication limitations and highlight opportunities for embodied digital craft design in VR environments.

Current LLM-powered augmented reality systems rely on dedicated perception pipelines to bridge semantic understanding and real-world coordinates. However, recent advances in AI increasingly render this division obsolete, enabling scene analysis to be reasoned about entirely within the model itself --- given the right tools. We present WorldMCP, an approach that lets VLMs view a raw scene, spatially ground references within it, and then register and anchor virtual objects directly in an AR environment. A preliminary evaluation using frontier models confirms the feasibility of our approach, paving the way for general-purpose assistants that leverage unrestricted scene understanding through the use of tools.

We present a lightweight geometry-based pipeline for segmentation of depth-sensor meshes on standalone mixed reality headsets. The method combines mesh simplification, normal-based plane extraction, concavity decomposition, and adaptive bounding-box refinement to obtain a compact representation of indoor scenes directly on resource-constrained mobile hardware. The approach was evaluated on indoor data acquired with the HTC Vive XR Elite. Runtime measurements show a total processing time of $4009 \pm 239$ ms for the evaluated mesh, while 3D Intersection over Union (IoU) is used to evaluate the influence of the segmentation threshold on bounding-box quality. The results show that a normal-similarity threshold of $0.99$ provides the best trade-off between spatial coverage and over-segmentation in the evaluated scene. The proposed pipeline provides a practical basis for asynchronous scene preprocessing, object replacement, collision handling, and spatial interaction in MR applications.

We introduce the concept of Voluntary Muscle Effort (VME) and further proposes the design perspective of “VME-as-Interaction”, exploring whether users can construct an experience of effort solely through the active engagement of bodily resources in the absence of real physical resistance. We used surface electromyography (sEMG) as a technical medium for detecting active muscle activation, designed a VME-based VR interaction system for a construction safety training scenario, and evaluated it through a within-group experiment (N = 15). The results indicate that VME significantly enhanced immersion whilst increasing self-reported workload; however, participants generally perceived this additional effort as a genuine, natural and valuable experience. This study demonstrates that the experience of effort does not necessarily require actual physical force or extensive bodily exertion that reproduces real-world fatigue, but may emerge through users' active bodily engagement, bodily performance, and bodily imagination.

Virtual reality enables avatars that depart from the human body, but it remains unclear how extended non-human limbs trade off reach, control, embodiment, workload, and manipulation behavior. We explore this design space through three deliberately contrasting avatar-control probes: a naturalistic hand, telescopic robotic arm, and octopus tentacle. In a within-subjects study (N=21), participants completed an obstacle-based sorting task while we logged trajectories, performance, VEQ, and NASA-TLX. Results show that reach extension reduces controller movement while expanding tip coverage, workload follows mapping complexity, and subjective embodiment remains comparable despite divergent behavior.

The adoption of mixed extended technologies as authoring tools is increasing, as XR and AI-assisted systems simplify the creation of immersive content and narratives and make the process accessible to users without advanced technical expertise or the need for a traditional desktop interface. Existing tools have the ability to capture user intent and rapidly translate it into immersive creative outputs. However, their editing capabilities remain limited, as they often treat objects as isolated entities, failing to preserve semantic and contextual coherence across the generated scene. To preserve scene coherence during actions, we designed a novel Spatial Dependency Graph (SDG) structure for managing inter-object relationships (support and dependency) inferred from bounding-volumes and object-type semantics. SDG is traversed during action to preserve scene coherence through novel cascaded operations on dependent objects: (i) co-motion in case of movements, (ii) re-support find a compatible support in scene, and (iii) remove dependents when no coherent support exist and (iv) reverse which rollback to any prior scene states. We implemented SDG in a MR demonstrator integrating Unity3D and Meta Quest 3 with a text LLM-based command parser. A preliminary study with two participants achieved an adapted SUS score of 82.5/100, indicating promising initial usability and supporting broader dependency-aware XR authoring.

UltraArUco - a lightweight multilingual library and framework for low-latency, real-time marker-based tracking in mobile augmented reality. Unlike standard OpenCV-based implementations, UltraArUco introduces an optimized multilingual wrapper that reduces per-frame latency in four times, while maintaining high accuracy. Distributed Wi-Fi architecture provides portability, connects mobile device (camera input and detection) with a PC-based visual application, enabling responsive interactions. The framework is validated through an interactive piano simulation, where static ArUco markers on keys enable occlusion-based note triggering, and hand-mounted markers provide spatial gesture recognition. UltraArUco’s system requirements makes it perfect for resource-constrained mobile AR applications, demonstrating viable AR music application without specialized equipment.

Virtual reality (VR) has become a powerful medium for cultural heritage, yet many systems emphasize spatial reconstruction while offering limited support for interpreting the diverse evidence on which heritage narratives depend. We propose a mapping method for designing embodied data storytelling in VR, transforming heterogeneous heritage evidence according to its narrative role rather than its media type: temporal data become embodied narrative checkpoints, event data become interpretive data islands, and geographic data become situated scaled environments. We instantiate the method in a prototype based on the Silk Road and conduct a preliminary study with six participants. Results indicate that the approach connects interaction, spatial experience, and data interpretation, yet embodiment alone is insufficient, users still rely on explanatory anchors such as narration, captions, and mini-maps to understand what each action or spatial representation means.

Assembling physical robots from two-dimensional printed instructions is a persistent challenge in STEM education: novice learners must mentally translate flat, static diagrams into three-dimensional spatial actions, a process that increases cognitive load, generates assembly errors, and creates high dependence on instructor support. We report preliminary field evidence from a classroom deployment of AR-semble, a mobile Augmented Reality application supporting robot assembly in middle school STEM education. AR-semble overlays step-by-step 3D animated guidance onto printed assembly manuals using marker-based tracking (Unity/Vuforia), enabling learner-controlled progression. Sixteen students completed a full six-step robot assembly session (M = 42:28 min), generating detailed navigation logs. A post-session survey (n = 8) assessed perceived usability, learning support, and engagement. Results indicate strong usability (easy to follow: M = 4.12/5; easy to learn: M = 4.00/5) and positive perceived usability and learning support (improved understanding: M = 3.75/5; learning efficiency: M = 3.75/5). Log analysis identifies pages 3 and 4 as navigationally challenging, informing targeted content redesign. These findings bridge the gap between design validation and a planned controlled comparative experiment.

As XR integrates into everyday workflows, users face interruptions causing spatial cognitive disorientation. Mitigating this requires a structured understanding of interruption management. We introduce a framework to map this design space, categorizing interruptions by their source location (collocated vs. remote) and delivery timing (immediate vs. deferred). We illustrate this matrix using select examples to demonstrate its structure, discuss initial design strategies for mediating these disruptions, and provide a starting point for future exploration.

Augmented Reality (AR) Head-Up Displays (HUDs) advance driver assistance by directly displaying critical information into the dynamic road environment. Current simulator study (N=56) evaluated driver performance by comparing a static HUD and an eye-tracked AR HUD along a route with sudden hazardous events, analyzing safety, navigation, cognitive workload and user experience. Although high data variance limited statistical significance, distinct descriptive trends emerged. The AR HUD improved navigation efficiency and hedonic acceptance without altering reaction times or cognitive load, though it reduced pragmatic clarity. Consequently, while safety remains equivalent, AR enhances navigation but demands careful interface design to preserve simplicity.