A primary limitation of early empirical research on the perception of architecture lies in its treatment of static decontextualized images as adequate proxies for architectural space. This methodological reduction flattens architectural experience both representationally and conceptually, overlooking spatial volume, material presence, multidirectional enclosure, and embodied affordances-qualities fundamental to how architectural environments are perceived and inhabited. When isolated visual features are treated as sufficient stimuli, architectural form is effectively abstracted from its spatial, sensory, and functional contexts. Consequently, the reported preference for curvilinear forms reflects judgments about pictorial representations rather than inhabitable environments, resulting in an unsubstantiated shift in the object of judgment.

Temporal flattening further compounds this limitation. Practical constraints common in fMRI protocols, such as brief stimulus presentations (e.g., approximately 3 s), prioritize momentary neural responses over the extended temporal unfolding characteristic of real-world spatial encounters. Such compression constraints exclude dwelling, memory accumulation, attentional shifts, and emotional resonance, which are central to architectural experience, long emphasized in architectural theory (Bachelard, 1994; Heidegger, 1996; Pallasmaa, 2012). These considerations highlight the need for experimental approaches that better accommodate the durational and reflective dimensions of inhabiting space.

The second critique centers on the semantic inconsistency of the stimuli. For example, Vartanian et al. (2013) study used (a) found images from architectural databases, (b) two architects as the raters, and (c) classified these images into “curvilinear” and “rectilinear” groups. However, the selected interiors vary substantially in design intent, spatial program, material articulation, and construction resolution. Many curvilinear examples depict expressive, high-design interiors, whereas rectilinear examples depict more utilitarian or conventional settings (Vartanian et al., 2013). This typological imbalance risks conflating contour with broader cues of design quality and aesthetic capital (Vitruvius, 1999; Rowe and Slutzky, 1963).

The instability of this curvilinear–rectilinear classification further compounds this issue. The two categories are derived from expert judgments of perceived contours in photographs, rather than from measurable properties of the underlying spatial geometry (Chuquichambi et al., 2022). Participants, by contrast, evaluate the same images by imaginatively reconstructing space through everyday experience, memory, and cultural association. Prior work also shows that experts and non-experts can diverge in aesthetic judgments (Cotter et al., 2017; Vartanian et al., 2019; Corradi et al., 2020; Palumbo et al., 2022). Consequently, the relationship among designed geometry, photographed representation, and lay reconstruction remains indeterminate, limiting the interpretability and reproducibility of curvature effects.

This semantic ambiguity also reflects a broader methodological lineage. Bar and Neta (2006) contrasted curvilinear and angular objects in 2D images and reported a general preference for curvature. Vartanian et al. (2013) extended this image-based approach to photographs of architectural interiors, substituting rectilinearity for angularity—a shift that changes the comparison from perceptual extremes to a contrast between conventional and non-conventional design typologies, thereby embedding semantic and aesthetic biases into the experimental logic itself.

A follow-up study by Vartanian et al. (2024) offers a more cautious framing, emphasizing that curvature perception depends on higher-order, context-sensitive mechanisms and underscoring that conceptual and operational concerns remain about the clarity and stability of the “curvilinear” versus “rectilinear” distinction within this experimental context.

In many neuroaesthetics studies, including Vartanian et al. (2013), behavioral responses, such as approach/avoid judgments, are collected through discrete, survey-based, forced-choice tasks rather than through action-oriented engagement. When static, decontextualized images are paired with abstract response formats, perceptual evaluation is effectively decoupled from embodied spatial experience. Participants' “enter” or “exit” decisions are therefore likely speculative and symbolic, detached from genuine spatial inclination.

In real-world architecture, perception and action are tightly intertwined; spatial preference emerges through embodied engagement, affective orientation, and perceived affordances rather than isolated visual appraisal (Spence, 2020; Djebbara and Kalantari, 2023). By constraining response to disembodied judgments, such experimental designs fragment cognition, separating what is seen from how space invites action. This limitation is underscored by a recent follow-up study, which reports a neural dissociation between predicted beauty ratings and approach–avoidance behavior, indicating that visual preference derived from static images may not reliably predict spatial desirability (Vartanian et al., 2024).

Beyond the structure of experimental inputs, the translation of empirical findings into architectural design presents a further methodological challenge. Scientific results are often interpreted as objective and generalizable truths; within neuroaesthetics, reported preferences—such as those favoring curvilinear forms—can therefore be overextended and implicitly elevated to design prescriptions. This raises a critical concern regarding the constraint inherent to experimental reach.

Empirical rigor denotes a specific methodological orientation characterized by variable control, reproducibility, conditional isolation, and reduction as method. It does not claim to provide a fully representative account of situated experience, which is inherently complex, contextual, and irreducible (Deleuze and Guattari, 1987). Empirical rigor and situated experience, therefore, are non-isomorphic and operate at different methodological resolutions that are not yet systematically connected.

Consequently, the pursuit of a direct equation between a single architectural attribute and observers' perceptual responses may be less insightful than the identification of recurring perceptual-experiential patterns that emerge across varied situations. The notion of pattern, as opposed to isolated stimulus–response regularities, refers to relational configurations between spatial form, perception, and use, most prominently articulated in Alexander et al. (1977) A Pattern Language. Generating such patterns requires the accumulation of situated evidence over time, as well as methodological pathways capable of scaling inquiry beyond isolated laboratory conditions. Without mechanisms that reconnect empirical findings to architectural design processes, experimental approaches risk remaining analytically precise yet practically remote. Addressing ecological validity requires not stronger causal claims, but differently structured methodological frameworks—ones that remain open to complexity, context, and iterative interpretation.

To generate controlled visual stimuli while limiting stable and often implicit variations introduced by architectural design style, the study constructs a prototype corpus from approximately 150 architectural projects designed by the first author.

The selection proceeded through a two-stage filtering process. An initial screening identified projects that shared three characteristics: (a) emotional-intent-guided design approach, as defined by the emotive design methodology, which emphasizes intuition-driven formal exploration to elicit affective and experiential qualities of space (Wei, 2024; see also Wei, 2019) to acknowledge the earlier foundational work. (b) dynamic composition characterized by geometric articulation; and (c) a fully digital modeling workflow in Rhinoceros 3D, enabling systematic decomposition and recomposition.

From this subset, four projects were selected based on comparability and formal clarity, including: (1) a compact spatial scale (under 100 m²) to minimize variation due to size; (2) a shared cultural and programmatic context; (3) spatial function including both transitional and resting areas; and (4) a clear emphasis on either curvilinear or angular spatial form. Completed in 2009, 2013, 2020, and 2024, respectively, the selected projects exemplify four distinct geometric typologies: planar angularity (2009), volumetric angularity (2013), single-surface curvature (2020), and compound curvature (2024). Each project displays a clear figure–ground relationship, further accentuated by a high-contrast color scheme.

From these built projects, architectural elements were systematically extracted in Rhinoceros 3D through a process of geometric decomposition, focused on isolating spatial components central to architectural experience. Elements were then categorized along multiple dimensions beyond geometry: (1) figure–ground relationships, distinguishing foreground elements—referring to spatial components that directly engage bodily proximity and action (e.g., seating, thresholds, frames)—from background elements that establish spatial context and orientation; and (2) functional–spatial configuration, classifying elements as point, line, plane, or volume, understood not merely as geometric primitives but as perceptual agents that guide navigation and establish visual hierarchy, consistent with foundational architectural design methodologies (Sherwood, 1981; Ching, 2014; Hedges, 2017). For example, elongated linear volumes (lines) define spatial axes and thresholds, whereas surface-bound enclosures (planes) demarcate zones of orientation. Through this decomposition, a modular library of spatial components was established.

While component extraction followed the logic of formal clarity and spatial hierarchy, recomposition was guided by behavioral affordances and spatial use patterns through an assemblage-based strategy: parts are irreducibly merged into a whole (Deleuze and Guattari, 1987; Elderfield, 1992; DeLanda, 2016). These assemblies were structured around functional cues, including thresholds, bounding walls, seating surfaces, and niches.

That is, component identification was form-driven, whereas spatial arrangement was use-driven, allowing behavioral affordances to emerge from relational composition rather than isolated formal features.

As a result, seven spatial scenarios were created as 3D models, each combining compositional clarity with behavioral flow and representing spatial modes including passage, pause, or rest. Final models were hybridized, trimmed, and Boolean-simplified to ensure coherence and perceptual legibility.

The seven spatial scenarios were then rendered as a series of controlled images that served as the Phase 1 stimuli. Each scenario was presented in eight variations −2 (Background: curved vs. angular) × 2 (Foreground: curved vs. angular) × 2 (Color: white vs. orange)—yielding 56 images in a within-participants factorial design. All renderings were produced using V-Ray for Rhino (Chaos Group, version 6) under standardized lighting, material, and camera parameters (Figure 2).

Rather than relying on photographic interiors or schematic geometric figures, this designed rendering strategy deliberately brackets extraneous semantic associations while preserving compositional hierarchy, spatial depth, maintaining cues for bodily engagement, and enabling subsequent cross-phase calibration.

Phase 1 employed an online survey administered via Qualtrics (Qualtrics, Provo, UT). The survey instrument was adapted and expanded from established neuroaesthetics protocols. In particular, Chatterjee et al. (2021), building on the interior photograph set introduced by Vartanian et al. (2013) identified four psychologically meaningful affective dimensions of the architectural experience: pleasure, coherence, fascination, and hominess. Participants rated each image on a 1–5 scale across core perceptual and evaluative measures, including Liking, Beauty, Curvature, Angularity, Fascination, Coherence, and Hominess (Coburn et al., 2020; Weinberger et al., 2021).

In the present study, additional judgments were collected to contextualize image-based appraisal and to establish observation-linked comparators for subsequent phases. For emotional aspects, participants selected experienced emotions from a predefined list adapted from the Positive and Negative Affect Schedule (PANAS) (Watson et al., 1988). For behavioral aspects, participants estimated the imagined duration of comfortable occupancy (0–200 minutes), judged whether each space appeared more likely to function as public or private, and indicated regions of visual engagement using a uniform 3 × 3 grid task. Together, these questions constituted a shared survey protocol across all phases and provided reference points within the framework for aligning reported appraisal with observed physiological and behavioral responses.

Stimuli were presented in randomized order to mitigate sequence effects. Data analysis, conducted in the R environment for statistical computing (R Core Team, 2024), indicated that the intended geometric distinctions were perceptually salient, with substantial inter-participant agreement in ratings of curvature and angularity. Foreground geometry exerted stronger influences on several evaluative dimensions, consistent with the intended figure–ground compositional logic.

A full statistical report of Phase 1 is presented in a separate publication (Curved background and foreground elements enhance the appeal of indoor spaces; manuscript in revision). All datasets and materials are publicly available on the Open Science Framework at doi: 10.17605/OSF.IO/B42DC.

Phase 1 findings serve to assess the stimulus design strategy and to inform the translation of stimuli from 2D image representations to 3D spatial form, guided by two principles: (1) foreground elements are prioritized as the primary carriers of geometric manipulation; and (2) geometric variation is translated from surface appearance to spatial affordance, shifting the focus from how curvature is visually recognized to how it structures movement, pause, and occupation. This translation yields a passage–pause–rest schema as a spatial structuring logic. These modes are employed here as architectural operations rather than semantic labels, describing distinct patterns of bodily engagement: passage as directed movement, pause as transitional slowing or reorientation, and rest as sustained occupation.

Foreground elements from the seven spatial scenarios were re-extracted, decomposed into modular kits, and recomposed into two spatial zones: curved and angular. Each zone measured 3 × 3 × 3 meters and was assembled following the same kit-creation principles described previously, with spatial organization guided by anticipated behavioral flow and the affordances associated with the three engagement modes. The two zones were positioned diagonally within a 3 m (width) × 6 m (length) × 3 m (height) bounding volume to ensure spatial symmetry, formal alignment, and structural balance.

The resulting linear composition was further refined to support build ability and functional clarity. Standardized spatial parameters, such as entry width, ceiling height, and seating elevation, were applied to both zones to preserve analytical comparability while supporting perceptual consistency across conditions.

The outcome, a simplified curvature–angularity continuum named Suffused Blossom #8, serves as the basis for both the 1:10 scale prototype in Phase 2 and the full-scale immersive installation introduced in Phase 3.

In Phase 2, image-based evaluation extends to object-based spatial interaction through the development of a 1:10 scale physical prototype of Suffused Blossom #8, thereby enabling geometric properties such as curvature and angularity to be experienced volumetrically rather than pictorially.

The prototype was deployed across multiple institutional settings: an art department, an architecture department, and a public library, thereby extending the diversity of participation. These deployments were conceived as methodological extensions for comparison, allowing the prototype to function as a shared perceptual anchor across disciplinary and public contexts.

Participants were not subject to time constraints and were encouraged to freely explore the object from multiple viewpoints. Preference assessment was conducted through two complementary procedures. First, participants indicated their immediate spatial preference in situ by placing a marker on boards corresponding to either the curved or angular zone of the prototype, capturing intuitive, embodied judgments elicited through direct physical encounter. Second, following engagement with the prototype, participants completed an online survey using the same two-dimensional image stimuli employed in Phase 1 (Figure 3).

By reintroducing the original image set after object-based encounter, Phase 2 functions as a methodological hinge within the framework, testing whether perceptual patterns identified in image-based evaluation remain consistent after the introduction of contextual information. This structure maintains a common evaluative baseline while enabling reflective comparison across representational modes.

In Phase 3, Suffused Blossom #8 was further scaled into a full-scale installation, enabling preference, engagement, and use to be examined under immersive conditions (Figure 4). A detailed account of the design and fabrication process is documented in a separate publication [Suffused Blossom #8: A Full-Scale Installation Bridging Design and Neuroscience, accepted conference proceeding forthcoming (2026)].

Phase 3 extends this exploratory framework into a fully immersive experimental infrastructure designed to support multimodal investigation of architectural experience across physical and virtual conditions.

Two immersive laboratory settings are employed in parallel. In one setting, participants encounter the full-scale physical installation co-located with a precisely aligned digital twin, enabling experience through a combined physical–immersive condition. In the second setting, participants engage with the same architectural configuration exclusively within a virtual environment. This dual setup enables direct comparison between physical and virtual embodiments of an identical architectural configuration.

Within each setting, participants will be randomly assigned to one of two task conditions: (1) unguided exploration, allowing free movement, path selection, and self-directed dwelling; or (2) directed engagement, prompting navigation toward predefined spatial features. These complementary tasks are designed to elicit patterns of movement, attention, and occupation under contrasting modes of engagement.

Across both settings, the immersive environment functions as an adjustable experimental layer through which contextual parameters—including surrounding environment, lighting, and material appearance—can be systematically varied while preserving geometric consistency. This structure supports comparative analysis of how spatial form interacts with environmental modulation to shape experiential and behavioral responses.

During all task phases, multimodal data will be collected concurrently, including physiological signals via wearable biosensors and behavioral measures such as movement trajectories, dwelling time, and directional choice. Following completion of all immersive tasks, participants will complete an image-based survey using the Phase 1 protocol, with two-dimensional stimuli revised and calibrated to reflect the spatial conditions encountered during laboratory testing. This alignment enables cross-referencing across representational, behavioral, and physiological measures.

Details of the Phase 3 research design and task rationales will be reported in a separate publication. Curvature is treated here as a test case rather than an endpoint, and the Phase 3 platform is structured to support future extensions to additional spatial typologies, functional programs, participant populations, and cultural contexts.