Program

A detailed workshop program will be published soon.

Below, we list all accepted contributions. Each accepted contribution went through a peer-review process that included two human reviews, supported by AI-generated reviewer-assistance material, as well as two additional AI-generated reviews. All listed papers will be published in the ISMAR 2026 Adjunct Proceedings.

Poster Session: Rethinking XR Evaluation and Practice 

Beyond Task Performance: Lessons Learned from Evaluating an Exploratory VR Interaction Technique

Nevzat Demirseren, Corey Pittman, Isayas Adhanom, Karthikeyan Umapathy, and Kevin Pfeil

Dense virtual environments present significant challenges for object selection and manipulation due to visual occlusion and limited target accessibility. Novel interaction techniques are commonly evaluated using objective performance measures such as task completion time and accuracy. Although these measures are essential, they may not fully capture the potential of unfamiliar interaction concepts during initial use. This paper presents RodCast as an exploratory case study investigating explicit trajectory visualization for interaction in dense virtual environments. We conducted a within-subjects user study comparing RodCast with Go-Go Hand and FlowerCone across three representative interaction tasks using both objective and subjective evaluations. The results revealed that the proposed implementation incurred performance costs on more demanding manipulation tasks, while subjective evaluations and qualitative feedback highlighted benefits in spatial awareness and target accessibility that were not fully reflected by conventional performance measures. These findings suggest that early objective performance should be interpreted alongside subjective evaluations and qualitative feedback when assessing exploratory XR interaction techniques, allowing the potential of an interaction concept to be distinguished from the limitations of its implementation.

InspectAR: User-Centered Design and Evaluation of Outdoor Augmented Reality and Computer-Vision to Support Bridge Inspection

Hamid Tarashiyoun, Ishan Pradhan, Alan Smith, John Luksas, Rodrigo Sarlo, and Joseph Gabbard

Bridge inspections are critical for maintaining infrastructure safety and reliability, requiring inspectors to conduct thorough evaluations of structural components and document their conditions. However, traditional inspection workflows rely heavily on manual measurement and data entry, which introduces inefficiencies, inconsistencies, and transcription errors. The adoption of the Specification for the National Bridge Inventory (SNBI) has increased the complexity of these assessments, further highlighting the need for modernized inspection methods.

This research explores the application of Augmented Reality (AR) and Computer Vision (CV) in bridge inspections through the development of an AR-based head-mounted display system. By overlaying digital annotations, enabling hands-free interaction, and integrating automated measurement tools, the proposed system supports enhanced data collection, visualization, and more efficient data conversion in the office. Through iterative design, development and field testing with professional bridge inspectors, we evaluated the usability and practicality of various AR-based inspection techniques.

Key findings from real-world testing indicate that AR improves spatial awareness by anchoring annotations and data directly to defects of interest on the bridge structure. The integration of CV for crack quantification has value for facilitating measurements at a distance, but its usability as a fully automated tool is low. The team determined that the best way to ensure reliability in assessments was a hybrid approach where CV is guided by the inspector, with flexibility to bypass it altogether if preferred.

Beyond the Beep: Comparing Visual-Only and Multimodal Notification Design for Improved Noticeability in AR

Aditya Raikwar, Zahra Borhani, Lucas Plabst, Anil Ufuk Batmaz, Mayra Donaji Barrera Machuca, Florian Niebling, and Francisco Ortega

Augmented Reality (AR) notifications must balance attracting user attention with the need to avoid distraction or cognitive overload, particularly in performance-critical tasks. While prior work suggests that multimodal notifications combining visual and audio cues can improve user attention, the effectiveness of visual-only notification designs remains less understood. This study investigates whether visual-only notifications can support user performance and attention as effectively as multimodal notifications in AR environments. We conducted a within-subject user study to compare four notification conditions: Notifications on Object With Sound (NoO WS), Notification on Viewport (NoV), Notification on Object With Arrow (NoO WA), and a Control condition (No Notifications). The study was implemented in ARtisan Bistro, an open-source AR sandbox environment simulating restaurant tasks. Results indicate that user performance did not differ significantly across notification conditions, suggesting that multimodal designs do not necessarily outperform visual-only notifications in all scenarios. In addition, combined visual-audio notifications did not significantly outperform NoV in terms of attention capture or user preference. According to the findings, despite their practicality, not all visual-only notifications can function as well as multimodal ones. The study emphasizes the need for more research into increasingly difficult AR activities, improvements in visual notification design, and the effects of background noise on notification effectiveness.

The Elephant in the Ear: Auditing Audio-Chain Reporting in XR Presence Research

Magdalena Igras-Cybulska and Artur Cybulski

Audio can shape presence, immersion, plausibility, and affect in extended reality (XR), yet auditory conditions are often reported as background implementation details instead of as methodological variables. This paper argues that the problem is not necessarily poor audio but unknown audio: readers often cannot determine how sound was rendered, delivered, calibrated, synchronized, or constrained and therefore cannot reconstruct the auditory exposure or assess its influence on the findings. Thus, the “elephant in the ear” is not that all XR studies use inadequate sound, but that many provide insufficient audio-chain information for interpretation and replication. We introduce the Audio Reporting Completeness Index (ARCI), a lightweight coding scheme for assessing how completely XR studies report auditory conditions. We conducted LLM-assisted full-text ARCI coding of a balanced, non-probability sequential quota sample of 30 studies. A domain expert, blinded to the LLM ratings, independently recoded a balanced six-paper subset comprising three studies from each analytical group. Exact human LLM agreement was 90.3%, and linearly weighted Cohen’s κ was .887. Coding followed a conservative explicit-evidence rule: audio chain information received credit only when it was explicitly stated in the inspected full-text materials, not inferred from hardware names or generic stimulus descriptions. ARCI scores varied widely (M =45.4, Md =41.7, SD=27.5, range =0.0–95.8). Audio/multisensory studies reported the audio chain more completely than general presence/UX studies (M = 63.6 vs. M = 27.2). We therefore propose MARC-XR, a minimum audio reporting checklist for XR presence studies.

Oral Talks: Rethinking Peer Review in the Age of AI 

A token for your thoughts: AI, Agency, and the Future of Peer Review.

Adalberto Simeone

As scholarly publication volumes grow, the peer-review process faces an unprecedented crisis of scale. The integration of Large Language Models (LLMs) is often presented as a logistical solution to this deluge. However, this position paper argues that such a transition represents a way to prioritise efficiency over rigour. We contend that the uncritical adoption of AI in peer review poses significant threats to intellectual agency, enabling the outsourcing of critical judgment to algorithms. By treating a systemic crisis of quality as a mere problem of logistics, AI integration risks subsidising the production of poor-quality academic work and eroding the human expertise central to scientific validation. To counter this trajectory, we propose a paradigm shift away from attempting to optimise the efficiency of the peer-review process towards rethinking peer review itself and making it more attractive to both authors and reviewers.

ReVoicer: Conversational Voice Annotation for Human-Centered, LLM-Assisted Peer Review

Matt Gottsacker, Ahinya Alwin, Hiroshi Furuya, Robert W. Lindeman, Gerd Bruder, and Gregory Welch

We present ReVoicer, a prototype system that supports peer reviewers by letting them converse with a paper as they read it. The reviewer highlights a passage and speaks (or types) a train-of-thought comment. A large language model then cleans the comment using the surrounding prose as context, tags it by comment type, and anchors it to the passage. After the reviewer finishes reading, ReVoicer checks the accumulated notes against a venue-specific rubric, reports coverage gaps, and drafts a review composed only from the reviewer's own comments, written to a style guide distilled from the reviewer's past reviews. The system introduces no critiques of its own. We describe the system's design rationale and implementation, and we outline plans for future evaluations. With the ISMAR community, we will gather feedback and discuss the system design and ideas for additional features and evaluations.

Two Last Human Sections? A Case Study of Largely AI-Generated Paper Writing in Empirical XR Research

Anusha Devanga, Gerd Bruder, Daniel Zielasko, Alexander Giovannelli, Zubin Choudhary, Hiroshi Furuya, Matt Gottsacker, Rob Lindeman, and Gregory Welch

Generative AI is rapidly transforming scientific research, particularly the analytical activities surrounding empirical studies. While experiment design, data collection, and statistical analysis remain largely human-driven in extended reality (XR) research, the introduction, related work, discussion, conclusion, abstract, and title of a paper are increasingly susceptible to AI generation. In this paper, we investigate through a practical example how an AI can generate all analytical components of a scientific paper when provided only with human-authored empirical sections. Starting from an accepted but yet-to-be-published ISMAR 2026 paper, we retained only the experiment and results sections and iteratively used a state-of-the-art AI system to generate the remainder of the manuscript. Comparing the resulting AI-generated paper with the original human-authored version, we find that the AI produces a coherent, persuasive, and publication-ready narrative, but one that systematically reframes the work around the most prominent findings rather than the researchers' original motivations. Our results show that fully AI-generated analytical framing of empirical research is already feasible, raising questions about scientific authorship, researcher intent, publication incentives, and the future role of human judgment in scientific communication.

When Papers Read Papers: Scientific Communication in the Age of AI-Generated Writing and Reviewing

Daniel Zielasko and Gerd Bruder 

Generative artificial intelligence is moving through scientific publishing in stages. It began as an aid for spelling, grammar, translation, and stylistic revision. It is now routinely used to transform notes and argument structures into academic prose, retrieve and synthesize literature, draft methods and introductions, analyse data, formulate discussions, and generate reviewer-style feedback. End-to-end systems already demonstrate that ideation, experimentation, manuscript production, and simulated peer review can be integrated into one automated pipeline. We argue that this progression creates a problem more fundamental than whether individual sentences were written by a human: the scientific paper may cease to be the natural unit of scholarly communication. If papers are generated by AI, reviewed by AI, summarized by AI, and consumed mainly as input to subsequent AI-generated work, maintaining the paper as the central interface of science becomes increasingly difficult to justify. This position paper develops the trajectory from language assistance to a recursive AI-mediated publication loop and asks what forms of human epistemic agency, accountability, and empirical grounding must remain. We do not argue for prohibiting generative AI. Instead, we argue that scientific communities should proactively redesign publication and review around inspectable research objects, explicit provenance, and accountable decisions before an informal and largely invisible automation of scholarship becomes the default.