Source-linked AI summary
Generative AI and Extended Reality in Collaborative Architectural Design Education: An Exploratory Studio Study
Yao Xiao, Max Chen, Yichen Li, Nathaniel Powers, Maxwell Wiesenfeld, Gillian Smith, Soroush Farzin, Shichao Liu
TL;DR
Architectural design education relies on shared visual representations, but combined GenAI–XR use in authentic studio workflows remains underexplored. This study deployed GenARch, a GenAI-assisted XR pipeline, in an undergraduate design studio and evaluated it with mixed methods. Outcomes were mixed, with reported implementation challenges as well as complementary roles for visual ideation and spatial evaluation.
Problem
Combined GenAI–XR use in ongoing architectural design courses alongside conventional tools remains relatively underexamined, despite the importance of authentic workflow integration.
Source: §1 · p. 2, §2.3 · p. 4Method
The study deployed GenARch in an undergraduate architectural studio and combined surveys, panel evaluation of team presentations, and group interviews.
Source: §1 · pp. 2–3Results
Findings were mixed, with no consistent improvements across self-efficacy, collaborative learning, teamwork, or panel-rated design outcomes despite reported benefits for ideation, spatial evaluation, and communication.
Source: §6 · p. 15Takeaways & Limitations
Different technologies supported different activities, with GenAI contributing rapid visual exploration and XR supporting spatial evaluation and shared review within this study’s workflow.
Source: §7 · p. 18Generated models did not reliably encode scale, and missing or outdated site information constrained spatial reasoning about adjacency, lighting, and context.
Source: §5.1.3 · p. 13
Abstract
from arXiv · showhide
Architectural design education relies heavily on visual ideation and representation to support collaborative learning in studio environments. Recent advances in generative artificial intelligence (GenAI) and extended reality (XR) offer new opportunities for rapid idea exploration and immersive spatial visualization. This exploratory mixed-methods classroom study investigated how GenAI-assisted multi-user XR influenced collaborative architectural conceptual design. We developed GenARch, a pipeline that integrates GenAI-based visual generation with collaborative XR environments, and deployed it in an undergraduate architectural design studio. Twenty-seven students formed seven self-selected design teams; four teams incorporated GenARch into their usual course workflow to support collaborative ideation and visualization, while three teams continued the same course workflow without GenARch. Pre- and post-intervention surveys assessed design self-efficacy, attitudes toward collaborative learning, and teamwork; a seven-member panel evaluated team design presentations; and GenARch teams participated in group interviews. The quantitative results showed larger relative declines in confidence and outcome expectancy for the GenARch condition and a positive difference-in-differences estimate for perceived conflict management, while panel-rated presentation outcomes were not significantly different between conditions. Interviews indicated complementary roles for the technologies: GenAI supported idea externalization and visual reference generation, whereas XR supported spatial, contextual, and scale-based evaluation. Students also reported challenges related to control, dimensional fidelity, shared attention, and motion comfort. These findings highlight both opportunities and limitations when GenAI and XR are incorporated into collaborative design education.
1. Introduction
Architectural design education depends on visual representations for shared critique and decision-making, while GenAI and XR offer new ways to explore ideas and spatial relationships. This study examines their integration into an authentic studio workflow through GenARch and mixed-methods evidence.
1. Introduction
- GenAI can support iterative visual ideation, while XR can support collaborative spatial exploration and shared spatial reasoning.These capabilities align with architectural design’s reliance on two- and three-dimensional representations.Source: §1 · p. 2
- Combined GenAI–XR use in architectural design education remains underexplored, particularly in authentic studio courses where tools are integrated with existing workflows.The study therefore examines GenAI-assisted XR within an ongoing collaborative design course rather than as a stand-alone tool.Source: §1 · pp. 2–3
- GenARch was developed with studio instructors and embedded in the usual course workflow to support collaborative ideation and visualization.Quantitative and qualitative methods examined students’ learning-related experiences, collaboration, and design processes.Source: §1 · pp. 2–3
- The study asks how GenAI-assisted XR relates to self-efficacy, engagement, collaborative learning, teamwork, communication, and shared decision-making.It also examines how the roles and perceived value of GenAI and XR vary across conceptual-design stages and relate to processes and outcomes.Source: §1 · p. 3
2. Related Work
The related work frames architectural collaboration as a process of negotiating ideas through shared representations, while noting limits in current generative and immersive tools. It identifies authentic course integration of GenAI and XR as a comparatively understudied setting.
2.1. Collaborative learning in architectural design
- Architectural studio teams negotiate alternatives and integrate individual ideas; shared sketches, drawings, diagrams, and models enable comparison, feedback, and collective decisions.Representations also externalize tentative concepts for inspection and discussion.Source: §2.1 · p. 3
2.2. 3D Generative AI for Architectural Conceptual Design
- Architectural applications need fidelity, consistent scale, and controllability, making current 3D generation more suited to early concept exploration than development-ready models.A studio study found design development, particularly 3D volumetric detailing, least supported by GenAI.Source: §2.2 · p. 4
2.3. Generative XR in Design and Collaboration
- Generative XR research explores immersive design and multi-user collaboration, but controllability, spatial realism, and creative agency remain challenges.Combined GenAI–XR use in ongoing architectural courses alongside conventional tools remains less understood.Source: §2.3 · p. 4
3. Method
The study deployed GenARch, a GenAI-assisted XR workflow, in an undergraduate architectural studio and used surveys, presentation ratings, and interviews to examine its role in collaborative conceptual design.
3. Method
- The study combined questionnaire-based self-reports, panel evaluation of team presentations, and semi-structured group interviews to examine GenARch’s role in collaborative conceptual design.The deployment took place in an undergraduate architectural design studio course.Source: §3 · p. 5
3.1. Course Context
- Students designed a community gathering space emphasizing daylight, artificial lighting, and inspiration from nature and everyday experience.The course used lectures, field observation, drawings, physical models, and simulation tools; GenARch addressed idea generation, visualization, and shared evaluation needs.Source: §3.1 · p. 5
3.2. GenARch Intervention Design
- GenARch generated 2D images from prompts, converted selected images to 3D, and supported shared review and XR inspection in Arkio.Students used it mainly for conceptual ideation and collaborative visualization, then continued with their usual tools for design development.Source: §3.2 · p. 5
3.3. Technical Implementation
- The in-house web and Quest 3 clients used DALL·E 3 for text-to-image generation and Shap-E, TripoSR, and Tripo3D APIs for 2D-to-3D reconstruction.Photon networking and Meta spatial anchors supported multi-user spatial alignment.Source: §3.3 · p. 6
3.4. Participants
- Seven self-selected teams participated: four GenARch teams (N = 16) and three Control teams (N = 11); assessed pre-study variables showed no significant condition differences.Two teams declined participation after a member reported severe motion sickness; the remaining teams were randomly assigned by team.Source: §3.4 · pp. 6–7
3.5. Procedure and Measures
- The staged intervention preserved the course syllabus; surveys measured self-efficacy, collaborative-learning attitudes, and teamwork before and after, while GenARch teams also completed a post-use tool evaluation.GenARch teams completed ideation and asset generation, then collaborative XR visualization; all teams presented under the same format, and a seven-rater panel assessed recorded presentations.Source: §3.5 · pp. 7–8
3.6. Statistical Analysis
- Pre–post change was compared between conditions using difference-in-differences, with participant-clustered standard errors and cautious interpretation given the small number of teams.The analysis also assessed presentation-rater concordance and used team-level Spearman correlations, treated as exploratory because they involved four GenARch teams.Source: §3.6 · pp. 8–9
3.7. Semi-Structured Interviews
- Four GenARch team interviews were analyzed using six-phase reflexive thematic analysis and inductive coding to identify patterned meanings, not prevalence estimates.The qualitative analysis complemented quantitative patterns by examining experiences, plausible mechanisms, and boundary conditions.Source: §3.7 · p. 9
4. Quantitative Results
Self-reported changes were mixed: the GenARch condition had larger relative declines in confidence and outcome expectancy, alongside a positive conflict-management estimate. Panel-rated presentations did not differ significantly between conditions.
4.1. Self-Perceived Outcomes Across Groups and Time
- The only baseline difference was lower self-efficacy anxiety in the GenARch group than Control (p = 0.005).Pre-intervention outcomes were otherwise largely comparable across the assessed instruments.Source: §4.1 · p. 10
- β = −1.675 for self-efficacy confidence and β = −2.088 for outcome expectancy indicated larger relative declines for GenARch; TQ_D had a positive estimate (β = 0.479).No significant DiD estimates appeared for collaborative-learning attitudes, overall teamwork quality, or the remaining teamwork subscales.Source: §4.1 · pp. 10–11, Table 2 · p. 11
4.2. Panel Evaluation of Student Design Presentations
- GenARch teams had lower median scores than Control on all five presentation criteria, but differences were not statistically significant; overall scores were 3.45 [2.89, 3.92] versus 3.87 [3.63, 3.88].The comparison was descriptive given seven team products and low-to-moderate rater agreement.Source: §4.2 · p. 11, Figure 4 · p. 12
- For four GenARch teams, perceived tool quality and presentation score were associated at r = 1.00 (p < 0.001), an exploratory, hypothesis-generating result.This association was not treated as evidence that perceived tool quality improved design performance.Source: §4.2 · p. 11, Figure 4 · p. 12
5. Qualitative Results
Students described GenAI and XR as serving complementary roles in collaborative conceptual design, while reporting challenges with control, fidelity, and coordination. Their preferences varied by design stage and reflected concerns about authorship and transparency.
5.1.1. GenARch as visualization tools
- GenAI supported ideation and visual references, while XR supported contextual and spatial understanding, including realistic-scale perception of architectural elements.Students wanted generated visuals to work with existing site models, sketches, photographs, precedents, and work-in-progress geometry.Source: §5.1.1 · pp. 12–13
5.1.2. GenARch as a mediator of collaboration and communication
- GenAI images gave teams shared visual anchors they could compare, criticize, and revise, making the discussion around an image part of its value.Students also envisioned XR for demonstrating work to teammates, reviewers, and stakeholders, but simultaneous headset use did not ensure shared attention.Source: §5.1.2 · p. 13
5.1.3. Challenges, limitations, and implications for future tool design
- Students reported motion sickness, unfamiliar interactions, prompt-learning effort, and early-use mistakes such as moving objects unintentionally or triggering generation repeatedly.These challenges added effort while students were also working on the design task.Source: §5.1.3 · p. 13
- Generated models did not reliably encode scale, while missing or outdated site information constrained reasoning about adjacency, lighting, and building context.Students therefore could not depend on these models alone for spatial reasoning.Source: §5.1.3 · p. 13
- Students wanted reproducible, controllable generation that establishes context and supports iterative editing of smaller components rather than one-step output.Prompt history offered some traceability but did not make outputs predictably reproducible or adjustable.Source: §5.1.3 · p. 13
5.2. Needs for GenAI and XR tools during the various design stages
- GenAI was viewed as more suitable for early exploratory prototyping, whereas XR became more valuable once a 3D artifact could be inspected spatially.Generation could take time, so students often prepared assets before or alongside immersive inspection.Source: §5.2 · p. 14
- GenAI posed a stage-dependent tension: early teams lacked ideas to prompt, while later teams found it difficult to honor constraints and often preferred direct modeling when precision mattered.XR's value increased when teams had a meaningful 3D representation to inspect for scale, site context, lighting, and spatial relationships.Source: §5.2 · p. 14
5.3. Students’ perspectives on the impact of GenAI and XR tools on architectural design
- Students worried that GenAI could displace human creative direction and described it as less transparent and reproducible than XR.They favored editable, reproducible outputs connected to their own artifacts so they could understand, modify, and take responsibility for resulting representations.Source: §5.3 · p. 14
6. Discussion and limitations
The discussion reports mixed outcomes: students valued GenAI for visual ideation and XR for spatial evaluation, but the intervention did not yield consistent improvements. It emphasizes aligning tools with design phases while accounting for technical, collaborative, and study-scope constraints.
6. Discussion and limitations
- Students identified benefits for visual ideation, spatial evaluation, and communication, but quantitative results showed no consistent improvements across learning perceptions, teamwork, or panel-rated design outcomes.The effects were not uniform across the intervention.Source: §6 · p. 15
6.1. Integrating GenAI and XR in design education and team collaboration
- GenAI images gave teams a concrete object for comparing and negotiating ideas, but multi-user XR did not guarantee shared attention across viewed or manipulated areas.Teams could struggle to focus together even though each participant had an individual headset.Source: §6.1 · p. 15
- The GenARch condition showed larger relative declines in confidence and outcome expectancy, and the study could not distinguish whether unfamiliar tools, implementation challenges, or greater awareness of limitations contributed.Reported frictions included motion discomfort, inconsistent scale, latency, and limited control.Source: §6.1 · p. 15
- GenARch did not lead to higher panel-rated presentation scores; ratings reflected the broader course process, since teams also used conventional tools and some generated artifacts remained exploratory.Both conditions used the broader course workflow to develop final designs.Source: §6.1 · p. 15
6.2. Users’ needs in different phases of creative projects
- As designs developed, prompts became easier but expectations grew; GenAI fit directed exploration, XR fit review of existing 3D artifacts, and abandoned outputs could still support learning.XR helped teams evaluate scale, site context, adjacency, lighting, and spatial presence.Source: §6.2 · p. 16
6.3. Technical considerations for developing GenAI-in-the-loop XR design tools
- Generation delays and headset discomfort shaped workflow, while students wanted realistic scale, accurate site context, and the ability to build from existing geometry.Using web and XR interfaces let students initiate generation without remaining in headsets while waiting.Source: §6.3 · p. 16
6.4. Educational Innovation
- GenARch brought rapid GenAI visual ideation and multi-user XR evaluation into an ongoing studio project, and the findings point to aligning tools with design needs and representation shifts.The study examined learning perceptions, teamwork, design outcomes, and student experiences in an established classroom workflow.Source: §6.4 · pp. 16–17
6.5. Limitations
- Evidence is bounded by a small single studio, short deployment, self-selected intact teams, and retrospective interviews; the authors call for longitudinal multi-studio studies with richer process measures.The study also could not isolate GenAI or XR effects from the other tools available in the course.Source: §6.5 · p. 17
7. Conclusion
The conclusion reports mixed outcomes from initial classroom use of GenARch, alongside complementary roles for GenAI and XR. It emphasizes timing, scaffolding, and coordination across representations in collaborative learning workflows.
7. Conclusion
- Initial classroom use produced mixed outcomes: some learning-related perceptions declined, conflict management improved significantly, and overall teamwork or design outcomes showed no clear advantage.The technologies also introduced practical frictions, including prompting, control, dimensional fidelity, shared attention, and motion comfort.Source: §7 · pp. 17–18
- GenAI supported rapid visual exploration, while XR supported spatial evaluation and shared review after representations had been created.The perceived value of the technologies changed across design stages.Source: §7 · p. 18
- The educational implications emphasize timing, scaffolding, representational continuity, and collaborative coordination when integrating emerging technologies into authentic learning workflows.Students’ ability to formulate intentions, control generated outputs, and maintain shared attention also shaped their experience.Source: §7 · p. 18
CRediT authorship contribution statement
The contribution statement assigns software, investigation, methodology, analysis, writing, supervision, and funding roles across the authors.
CRediT authorship contribution statement
- Xiao and Chen contributed to conceptualization, software, methodology, analysis, and writing; Li contributed software and drafting, while Powers and Wiesenfeld contributed conceptualization, software, investigation, and drafting.Smith and Farzin contributed review and supervision; Liu contributed review, supervision, and funding acquisition.Source: CRediT authorship contribution statement · p. 18
Data availability
Supporting materials are available through the OSF repository, while participant-level data sharing is restricted to protect confidentiality.
Data availability
- Anonymized supporting materials are available through OSF; qualified researchers may contact the authors about data access where the ethics protocol permits.Additional participant-level data are not publicly shared to protect confidentiality.Source: Data availability · p. 19