ARAM: a threshold model for understanding occupant acceptance of architectural robot adaptations

1 Introduction

Recent technological advancements are enabling everyday architectural elements to become increasingly roboticised in support of occupant needs (Ju and Takayama, 2009; Lello and Saffiotti, 2011; Sirkin et al., 2015; Fink et al., 2014). Previous studies have shown that elements such as chairs (Agnihotri and Knight, 2019), desks (Takashima et al., 2015), sofas (Spadafora et al., 2016), beds (Kapusta et al., 2019), ceilings (Denis et al., 2026), and partitions (Nguyen and Vande Moere, 2024) can now physically move or change shape to “adapt” indoor spaces. Beyond adapting according to predefined schedules, environmental conditions, or occupant commands, these elements can now autonomously initiate adaptations at appropriate moments to spark joyful engagement among occupants (Agnihotri and Knight, 2019), nudge them toward more ergonomic behavior (Takashima et al., 2016), encourage social interaction (Sadka et al., 2022), or maintain social harmony among multiple occupants (Grønbæk et al., 2017). We thus refer to these roboticised building elements as architectural robots: interior components that can physically move or change shape to adapt indoor space in an architectural way.

Despite this potential, building occupants do not always accept adaptations performed by architectural robots. This is partly due to a misalignment between the traditional perception of architectural elements as functional, static components of the spatial backdrop, and their newly unfamiliar capacity to actively perform adaptations (Sirkin et al., 2015; Grönvall et al., 2014). Such misalignment is amplified when occupants perceive adaptations as reducing their sense of control (Schweiker and Wagner, 2016; Pimminger et al., 2024), which can occur when these adaptations create discomfort (O'Brien and Gunay, 2014) or disrupt ongoing activities (Yoon et al., 2024) in ways that are unpredictable (Sirkin et al., 2015) or difficult to understand (Takashima et al., 2016). Consequently, occupants may refuse to engage with the adaptation (Sirkin et al., 2015), manually override it (O'Brien and Gunay, 2014), or even attempt to disable the robotic system altogether. Considering that a single unacceptable action that a robot performs can already damage human trust in it (Malle et al., 2020), it is crucial to determine when and how an architectural robot should initiate an adaptation in a way that supports occupant needs while also ensuring their acceptance.

Various theories and models from related disciplines already provide insights into how people accept technologies and adaptations. Established acceptance models, including the Technology Acceptance Model (TAM) (Davis, 1989) and the Unified Theory of Acceptance and Use of Technology (UTAUT) (Venkatesh et al., 2003), explain technological adoption and use through constructs related to perceived usefulness, effort, social influence, and enabling conditions. These models have subsequently been extended to consumer technologies and different robotic contexts (Venkatesh et al., 2012; Heerink et al., 2010; Bröhl et al., 2019). However, they primarily explain behavioral intention and use at the level of the overall system, rather than how occupants respond to individual, situated adaptations once an architectural robot is already present in an everyday space.

The field of human-robot interaction, on the other hand, offers substantial knowledge on how people accept autonomous social robots (Heerink et al., 2009; de Graaf et al., 2019) as companions or even social partners (Henschel et al., 2021), particularly when their behaviors adhere to social norms (Niemelä et al., 2019), provide enjoyment (de Graaf et al., 2019), or offer support in daily life (Heerink et al., 2009). However, it remains unclear how this mechanism transfers to architectural robots. Unlike social robots, architectural robots cannot be understood solely as human-like partners. They must also function as spatial elements that occupants use, inhabit, and perceive as part of the built environment. By contrast, the fields of environmental and social psychology offer insights into how occupants in shared spaces accept environmental adaptations performed by other occupants, particularly when such adaptations involve perceived responsibility (Xu et al., 2020) or social pressure (Ajzen, 1991). Yet it also remains unclear how these perceptions transfer to architectural robots. Unlike human occupants, architectural robots lack an established social role through which their adaptations can be readily interpreted, negotiated, or justified.

Lastly, the field of anthropology offers a complementary perspective by conceptualizing human-technology interactions as situated and relational, in which agency and control are continuously negotiated through practice (Suchman, 2006; Pink et al., 2022). From this perspective, acceptance is not necessarily a static attitude toward a technology, but can develop and change through continued encounters with automation in everyday settings (Lindgren et al., 2021). However, it remains unclear how this situated and evolving understanding of acceptance can explain occupants' responses to specific adaptations performed by architectural robots. Taken together, these perspectives suggest that architectural robot adaptations are distinctive because they may simultaneously be experienced as robotic behaviors, building automation, and socially situated environmental adjustments.

Our research aims to develop a framework for explaining occupant acceptance of adaptations performed by architectural robots. To do so, we qualitatively re-analyzed interview and observational data from three of our previous studies on the same typology of architectural robot: robotic partitions (Nguyen et al., 2021, 2022; Nguyen and Vande Moere, 2024). Across these studies, occupants experienced a wide range of adaptations performed by three similar robotic partitions, each deployed in a different architectural context and social situation, resulting in diverse acceptance responses among occupants. While the three previous studies were originally conducted and analyzed by the first and last authors in relation to their respective research questions, the present re-analysis specifically focuses on identifying the reasoning underlying occupants' different acceptance responses to robotic partition adaptations. By involving an interdisciplinary co-authoring team with diverse expertise in human-building interaction, human-robot interaction, environmental and social psychology, anthropology, and human environmental comfort, this re-analysis offers multiple perspectives for unpacking the factors that govern occupant acceptance in a comprehensive way.

We argue that these three previous studies on robotic partitions are particularly well-suited to our research question for four reasons. First, the adaptations performed by the robotic partitions were highly purposeful: they supported occupants' visual, acoustic, and privacy needs in different situations by reconfiguring the spatial layout. This purposefulness enables us to examine how individual needs shape acceptance, a factor already recognized as critical in human-building interaction (O'Brien and Gunay, 2014). Second, the materiality of the robotic partitions as heavy, seemingly static, and functional spatial division elements contrasted strongly with their robotic autonomy, as they could move freely within the spatial layouts while remaining detached from building structures. This contrast foregrounds an acceptance challenge previously identified in human-robot interaction studies, where everyday functional elements become roboticised (Sirkin et al., 2015; Grönvall et al., 2014). Third, the architectural scale of the partitions meant that a single adaptation could affect multiple co-located occupants simultaneously, potentially revealing tensions between personal needs and social responsibility, as discussed in environmental psychology (Ajzen, 1991; Xu et al., 2020). Finally, all three studies consistently captured fine-grained details of occupant experience, enabling a unified qualitative analysis from an interpretivist perspective (Ponterotto, 2006).

Through an interpretivist reflexive thematic analysis (Braun and Clarke, 2006, 2021), we developed three themes describing how individual occupants evaluated robotic partition adaptations in our studies, which we subsequently abstracted into three acceptance criteria. In particular, occupants were more likely to endorse an adaptation when they (1) perceived an unmet personal need that motivated them to adapt their surrounding space, (2) expected the architectural robot to be capable of addressing that need through its architectural and robotic affordances, and (3) considered the adaptation socially justifiable to co-located others. Depending on how these criteria were met, occupants adopted one of four acceptance responses to an adaptation. Specifically, occupants endorsed an adaptation when all three criteria were met, meaning that the adaptation was capable of addressing their personal need in a socially justifiable way. They refrained from engaging with an adaptation when it was capable of addressing their need but was not socially justifiable. They repurposed an adaptation when it was socially justifiable but not directly capable of addressing their need, for example by using it to signal that need to others. Finally, they rejected an adaptation when they did not perceive a personal need for it, or when they perceived it as both incapable of addressing their need and socially unjustifiable.

Because individual occupants weighed these criteria differently across adaptations and situations, we theorize that they operate as three distinct acceptance thresholds along a single continuum of perceived stress severity. We call them thresholds because each criterion changes from one state to another as perceived stress increases; for example, from an unrecognized to a recognized need, or from an unjustifiable to a justifiable adaptation. As perceived stress increases, an unmet need may shift from unrecognized to recognized, marking the personal need threshold, and an adaptation may shift from socially unjustifiable to justifiable, marking the social justification threshold. By contrast, adaptation capability acts as an upper threshold: beyond a certain level of perceived stress, an adaptation that was previously considered capable may become insufficient for addressing the occupant's need.

Following these findings, we introduce the Architectural Robot Acceptance Model (ARAM), which explains how the relative positions of the three acceptance thresholds along a continuum of perceived stress severity shape the likelihood that an occupant will endorse, reject, refrain from, or repurpose a given adaptation performed by an architectural robot. Because our results show that these thresholds vary between occupants, adaptations, and personal needs, ARAM should be interpreted situationally: it describes how an individual occupant evaluates a specific adaptation in relation to their needs at a particular moment. We then relate ARAM to interdisciplinary knowledge to discuss how it can explain and anticipate occupant acceptance and inform the design of acceptable architectural robot adaptations, before outlining how it could be operationalised and applied in future work.

Our contribution is to enhance the understanding of occupant acceptance of robots that aim to behave in an architectural way, with the aim of informing technological development and ensuring that acceptance-related factors are adequately considered, particularly whether adaptations are perceived as purposeful in relation to a recognized need, and capable of addressing that need in a socially justifiable way. By acting in ways that are acceptable to occupants, architectural robots can integrate more seamlessly into the built environment to support occupant needs. Lastly, although ARAM was developed from robotic partitions, we hypothesize that its appraisal mechanism may also transfer to other robotic, automated, intelligent, and smart technologies embedded in the built environment that are perceptible to occupants. The extent of this transferability, however, remains to be evaluated across technologies that differ in materiality or autonomy, or are embedded in different social contexts.

2 Related work

Our study is grounded in prior theories and knowledge from technology acceptance, human-robot interaction, human comfort research, and environmental and social psychology. Together, these bodies of work help explain how occupants evaluate adaptations to their surroundings, whether such adaptations are initiated by a robot, a building control system, or another co-located occupant.

2.1 Technology acceptance models

Technology acceptance has been studied extensively through models that explain why individuals adopt and use new technologies. The Technology Acceptance Model (TAM) identifies perceived usefulness and perceived ease of use as fundamental determinants of user acceptance (Davis, 1989). Perceived usefulness refers to the extent to which a person expects a technology to improve their performance, whereas perceived ease of use concerns the effort expected when using it. Building on TAM and other behavioral theories, the Unified Theory of Acceptance and Use of Technology (UTAUT) integrated eight established acceptance models into a unified framework (Venkatesh et al., 2003). UTAUT proposes four core determinants of behavioral intention and use: performance expectancy, effort expectancy, social influence, and facilitating conditions, whose effects may further vary with individual and contextual moderators like gender, age, experience, and voluntariness of use. UTAUT2 subsequently extended this framework to consumer technologies by incorporating hedonic motivation, price value, and habit (Venkatesh et al., 2012).

Technology acceptance models have also been adapted to robotic contexts. The Almere model extends UTAUT to explain the acceptance of assistive social agents by older adults, supplementing functional evaluations with factors related to social interaction and examining both the intention to use and the actual use (Heerink et al., 2010). In industrial human-robot collaboration, the Human-Robot Collaboration Acceptance Model (HRCAM) extends TAM-family models by retaining perceived usefulness and perceived ease of use as central acceptance variables, while incorporating person-specific anchor variables such as robot anxiety and technology affinity, and context-specific adjustment variables such as perceived safety, enjoyment, and ethical, legal, and social implications (Bröhl et al., 2019).

Together, these models suggest that technology acceptance is shaped by functional, social, and contextual considerations. However, established models generally examine acceptance at the level of the overall technology through behavioral intention and use, and therefore provide a more aggregated account across interactions and situations. Yet it remains unclear how these considerations combine to shape an occupant's response to a specific adaptation at a particular moment, especially as their needs, expectations, and social context may change between adaptations. This gap is particularly relevant for architectural robots, whose adaptations can physically reconfigure the space occupants inhabit and affect multiple co-located occupants at once.

2.2 Human-robot interaction

Previous human-robot interaction studies on architectural robots have shown that occupants adopt various acceptance responses to an architectural robot depending on the characteristics of its adaptations. Architectural robots that performed subtle, gradual adaptations—such as a shifting computer monitor (Kronenberg et al., 2022; Shin et al., 2019), a tilting chair (Fujita et al., 2021), or a transforming workstation (Wu et al., 2018) designed to promote ergonomic posture during prolonged sitting—were accepted when occupants perceived them as minimally disruptive (Yoon et al., 2024) and trusted their ergonomic benefits (Lee et al., 2019). Conversely, architectural robots that performed more overt adaptations to “nudge” occupant behavior—such as a trolley robot encouraging office workers to exercise (Ren et al., 2022), a robotic chair inviting passers-by to sit (Agnihotri and Knight, 2019), a robotic door prompting individuals to approach it (Ju and Takayama, 2009), or a toy-box robot urging children to tidy up (Fink et al., 2014)—were more likely to be accepted when occupants felt that their sense of control was maintained (Bailly et al., 2016). Previous research therefore suggests that architectural robots should support occupant autonomy by allowing voluntary engagement (Ren et al., 2022), making adaptations visible and predictable (Sirkin et al., 2015), permitting occupants to override inappropriate adaptations (Takashima et al., 2016), clearly acknowledging occupant requests (Sirkin et al., 2015), and progressively adapting to individual preferences (Tsai et al., 2022).

In shared environments, occupants were more likely to accept architectural robots when their adaptations aligned not only with personal needs, but also with the perceived needs of co-located others and the social norms of the situation (Krogh et al., 2017). For instance, occupants endorsed the adaptations of a shape-shifting display that optimized viewing ergonomics according to social arrangement (Takashima et al., 2016), a shape-changing table that adjusted between vertical and horizontal configurations to support different group sizes and collaborative tasks (Grønbæk et al., 2017), a pair of mobile tables that facilitated collaboration by guiding occupants to approach each other (Takashima et al., 2015), and a set of robotic partitions that separated occupants in a shared space for individual work (Onishi et al., 2022). Occupants also expected architectural robots to manage adaptations in ways that maintained social harmony, such as by synchronizing movement speed with task duration while anticipating spontaneous task changes (Grønbæk et al., 2017), distributing control among co-located occupants (Grønbæk et al., 2020), or preserving openness to the surrounding environment when needed (Onishi et al., 2022).

Together, these studies suggest that occupant acceptance of architectural robots depends not only on whether an adaptation is useful, but also on whether it is minimally disruptive, controllable, predictable, and socially appropriate. Yet it remains unclear how such considerations combine at the level of a specific adaptation to produce qualitatively different occupant responses.

2.3 Human environmental comfort

Previous research on human environmental comfort indicates that occupants evaluate environmental adaptations according to their personal level of comfort, including when such adaptations are performed by autonomous building control systems. While shared environments have traditionally been optimized across separate domains—such as thermal, visual, acoustic, or air quality—for an average occupant profile, individual occupants often experience these supposedly “optimal” conditions differently (Altomonte et al., 2020). As a result, they may adjust thermostats, windows, blinds, lighting, or other environmental controls to restore their comfort (Hong and Lin, 2013; O'Brien and Gunay, 2014).

Autonomous building control systems have been developed to adapt environmental conditions on behalf of occupants, aiming to maintain their level of comfort while improving energy efficiency. Yet studies suggest that occupants do not always accept such autonomous adaptations. For instance, automated shading or blind systems were more likely to be accepted when occupants perceived their adaptations as improving comfort, being understandable, and preserving a sense of control through feedback or override possibilities (Meerbeek et al., 2014, 2016). Conversely, when these systems misinterpreted occupant needs, created discomfort, or reduced perceived control, occupants sometimes overrode or rejected their adaptations (O'Brien and Gunay, 2014; Tamas et al., 2020; Park et al., 2019).

Recognizing these challenges, the multi-domain comfort theory emphasizes the need to consider personal, contextual, and environmental factors, including the combined and cross-modal effects of multiple physical stimuli on individual comfort (Schweiker et al., 2020). Studies informed by this theory suggest that comfort may be inferred through physiological responses and wearable sensing technologies, enabling near real-time assessment without relying solely on explicit occupant feedback (Mansi et al., 2021). Such approaches are particularly relevant to occupant-centric control, where Personal Comfort Models (PCM) (Kim et al., 2018; Bogatu et al., 2023) and Personalized Environmental Control Systems (PECS) (Shinoda et al., 2022) aim to tailor environmental adaptations to individual occupants, thus improving both comfort and energy performance.

Together, these studies suggest that occupant acceptance of autonomous building control systems depends not only on whether a system improves measurable environmental conditions, but also on whether occupants perceive a personal need to restore or maintain comfort, trust the system's capability to address that need, and retain sufficient understanding or control over the adaptation. Yet it remains unclear how these factors can explain occupant acceptance of adaptations performed by architectural robots, particularly when such adaptations typically reconfigure shared space physically rather than only adjusting environmental settings.

2.4 Environmental and social psychology

Previous research in environmental and social psychology suggests that adaptations performed by occupants in shared spaces are evaluated not only as individual actions, but also as social interactions.

At the individual level, occupants' personalities may influence how they interact with and adapt their indoor environments. For instance, Schweiker et al. (2016) found that neuroticism, extraversion, and openness to experience were associated with differences in behavioral patterns such as window opening, blind adjustment, ceiling-fan use, and clothing adjustment. Personality may also shape how occupants perceive environmental conditions themselves. For example, (Chiucchiù et al. 2026) found that personality contributed to differences in thermal and acoustic perception, including how occupants evaluated different soundscapes under changing thermal conditions. These findings further suggest that occupants may experience the same environmental conditions differently according to their individual sensitivities and preferences. In shared spaces, however, such perceptual and behavioral differences may also affect co-located occupants, introducing an additional social dimension in which occupants evaluate whether an adaptation is appropriate in relation to others.

This social evaluation is reflected in the theory of planned behavior (Ajzen, 1991), which explains planned action through three main factors: an individual's attitude toward the action, perceived social pressure to perform or avoid it, and perceived behavioral control over whether the action can be performed. Applied to environmental adaptations, this means that occupants may not only consider whether an adaptation benefits them personally, but also whether others expect, tolerate, or disapprove of it. A recent extension of this theory further showed that occupants' energy-saving actions in offices are shaped by their ascription of responsibility, referring to the perceived obligation to act, and by ease of access to controls, referring to the physical and social accessibility of shared environmental controls such as windows, shades, lighting, or thermostats (Xu et al., 2020).

These considerations become especially important when an adaptation is performed by another occupant. In such cases, affected occupants may interpret the adaptation as an expression of someone else's need or preference, or as a territorial claim over the shared environment (Altman, 1975; Brown, 2009). Norm activation theory similarly proposes that prosocial or environmentally responsible actions depend on awareness of consequences, ascription of responsibility, and personal norms (Schwartz, 1977; Steg and de Groot, 2010).

Together, these studies suggest that occupants may accept an adaptation performed by someone else when they understand the need behind it, perceive it as legitimate, and consider its impact on others to be socially justified. Such judgements may also differ between occupants because the same environmental condition or adaptation can be perceived differently according to individual sensitivities and preferences. Yet it remains unclear how these social-psychological mechanisms transfer to adaptations initiated by architectural robots rather than humans, especially when those adaptations still need to be perceived as socially justifiable by co-located occupants.

3 Methodology

We qualitatively re-analyzed data from three of our previous studies (S1 Nguyen et al., 2021, S2 Nguyen et al., 2022, and S3 Nguyen and Vande Moere, 2024) to develop a framework for explaining how occupants accepted adaptations performed by robotic partitions as one type of architectural robot. From these studies, we extracted: (1) photographic or textual descriptions of the adaptations; (2) questionnaire results in which occupants evaluated these adaptations; (3) raw interview transcripts from all occupants; and (4) observation notes from the studies.

To enable cross-study re-analysis, we treated each robotic partition adaptation as a unit of analysis when it was sufficiently documented. Each adaptation was therefore analyzed through the perspective of the specific occupant who experienced and evaluated it. Specifically, an adaptation was considered sufficiently documented when it could be linked to a specific spatial configuration in the observation data, an occupant evaluation, and an explanation from interview or questionnaire data. Questionnaire responses and observation notes were also used to contextualize the interview data, rather than as stand-alone evidence for interpreting the reasoning behind occupants' acceptance responses to adaptations.

3.1 Study overview

As shown in Table 1, the three previous studies differed in their architectural-social contexts, ranging from private homes to a semi-public breakout space and a shared office. This contextual variation likely shaped how occupants perceived the robotic partition adaptations, helping us capture a broader range of reasoning behind their acceptance responses.

Table 1. Comparison of the three previous studies (S1–S3).

Although the studies differed slightly in the embodiment of the robotic partitions, including size, materials, and colors, as well as in fidelity, ranging from semi-immersive simulation to physical reality, these differences still allowed a comparative qualitative re-analysis because all three studies examined occupants' reasoning about adaptations performed by the same architectural robot typology with the same core spatial functionality, rather than directly comparing the technical performance or perceptual realism of the prototypes. Across all three studies, occupants consistently encountered the partitions as mobile robots with some degree of autonomous agency that shared the same core functionality: dividing spatial layouts to support occupants' visual, acoustic, and privacy needs. ¹

3.1.1 S1. Private homes

S1 (Nguyen et al., 2021) involved eight occupants (S1.O01–S1.O08), who remotely experienced semi-immersive cross-reality simulations of their own homes augmented with a robotic partition that autonomously adapted the layout, as illustrated in Figure 1. Each occupant experienced adaptations that they had previously co-designed with the researchers through a mobile interface, such as a laptop or tablet, while freely moving around their actual physical home. After the simulations, occupants completed a questionnaire and took part in a semi-structured interview to reflect on: (1) how they experienced the robotic partition adaptations; (2) whether, how, and when they would accept these adaptations in their daily routines; and (3) how they perceived the robotic partition moving autonomously in their homes.

3.1.2 S2. Semi-public breakout space

S2 (Nguyen et al., 2022) involved 26 occupants (S2.O01–S2.O26), who experienced adaptations performed by a physical robotic partition in a semi-public breakout space of an office building, as illustrated in Figure 2. The partition appeared to act autonomously while dividing the spatial layout to support occupants' work tasks, although its movements were actually Wizard-of-Oz controlled by a remote researcher observing camera feeds. The sequence of each adaptation was randomized between two conditions. In the first condition, the partition performed an adaptation before the occupant freely chose a location for their work task. In the second condition, the occupant first chose a location, after which the partition performed an adaptation nearby. Occupant acceptance was initially captured through their immediate behavioral response, such as whether they remained near the partition or moved away after an adaptation occurred. Occupants were then invited to reflect on these decisions through a questionnaire and a semi-structured interview, which focused on: (1) whether and how the partition adaptations supported their work tasks; and (2) how they perceived the robotic partition moving autonomously in the breakout space with different levels of proactiveness.

3.1.3 S3. Private shared office

S3 (Nguyen and Vande Moere, 2024) involved 13 occupants (S3.O01–S3.O13), who shared control of a semi-autonomous robotic partition in their office over 5 weeks, as illustrated in Figure 3. The partition moved within the shared office to support occupants' needs by adapting the spatial layout. Through a web-based interface, occupants could initiate or override an adaptation at any time by choosing from a set of adaptations they had previously co-designed with the researchers. When initiating an adaptation, occupants also provided a reason that was visible to their colleagues. Throughout the study, occupants used the interface to give feedback on whether each adaptation met their expectations, regardless of whether it was initiated by themselves, by their colleagues, or seemingly autonomously by the Wizard-of-Oz researcher. At the end of the deployment, occupants completed questionnaires and took part in semi-structured interviews to reflect on: (1) whether and how the partition adaptations supported their needs; (2) how they accepted adaptations initiated by themselves vs. by others; and (3) how they perceived the robotic partition moving autonomously in their shared office.

3.2 Data analysis

Ten authors conducted a qualitative analysis of the interview transcripts and observational data using an interpretivist reflexive thematic analysis approach (Braun and Clarke, 2006, 2021). The analysis focused on understanding the factors that influenced occupant acceptance of robotic partition adaptations and identifying interdependencies between these factors. Descriptions of the adaptations and questionnaire results were used alongside the interview and observational data to contextualize the reasoning that occupants highlighted in their self-reported experiences.

We structured the analysis through an iterative and recursive process of initial coding, theme development, and theoretical integration. The first author, who specializes in human-robot interaction and architectural design and personally conducted the three previous studies, re-coded the data with a focus on how robotic and architectural dimensions influenced occupant acceptance, producing 42 codes. The second author, with expertise in environmental and social psychology, coded the data with a focus on the psychological processes through which occupants experienced the adaptations, generating 36 codes. The two coding passes were treated as complementary analytical perspectives rather than as an inter-coder reliability assessment. The resulting codes were brought together for comparison, with overlapping codes consolidated and complementary interpretations retained, resulting in a provisional working set of 56 codes.

Theme development then involved seven authors, who participated in workshops where the 56 codes were discussed, refined, and organized into candidate thematic clusters from interdisciplinary perspectives. To support reflexivity and transparency, analytical memo writing and group discussions were used throughout the process. Through theme development, the 56 codes were organized into 13 candidate thematic clusters representing broader patterns of meaning across the data.

Lastly, theoretical integration involved all 10 authors in examining how these candidate thematic clusters related to one another across different adaptations, occupants, and contexts, with particular attention to how combinations of factors corresponded to different acceptance responses. This process brought together perspectives from human-robot interaction, human-building interaction, environmental and social psychology, anthropology, and human environmental comfort through group discussions and iterative comparison across adaptations, occupants, and contexts. The 13 candidate thematic clusters were then further developed and refined into three themes, each comprising two contrasting appraisal states, as reported in the Results section.

3.3 Reflexive positionality

Because the first and last authors had designed and conducted the original studies, their familiarity with the data provided contextual knowledge but also shaped the perspectives they brought to the re-analysis. We therefore treated researcher positionality explicitly during the analysis. Initial coding by the second author provided a differently situated disciplinary perspective, while theme development and theoretical integration workshops were used to question prior interpretations, compare competing explanations, and examine cases that did not readily fit the developing analysis. Analytical memos documented changes to codes, candidate thematic clusters, themes, and interpretations throughout this process.

Consistent with reflexive thematic analysis, we did not use inter-coder reliability statistics such as Cohen's κ. Such measures assume that coding can be evaluated according to agreement between researchers, whereas reflexive thematic analysis treats coding as an interpretive process shaped by researchers' theoretical positions, disciplinary perspectives, and engagement with the data (Braun and Clarke, 2019, 2021). Accordingly, the involvement of multiple researchers in our analysis was not intended to establish a single “correct” coding through consensus. Instead, differences between interpretations were used reflexively to question assumptions, consider alternative readings of the data, and develop a richer interpretation of the data.

4 Results

Out of the 247 robotic partition adaptations collected across the three studies, we selected 125 adaptations (50.6%) for thematic analysis. We excluded 42 adaptations (17.0%) for which acceptance responses from the occupants were not documented, and a further 80 adaptations (32.4%) because they lacked sufficient detail about occupants' reasoning relevant to the research question.

We refer to acceptance response as a description of how an occupant observably or reportedly responds to a specific adaptation, independently of the reasoning behind that response. Across the adaptations, we distinguish four responses. Endorsing refers to engaging with an adaptation as intended. Rejecting refers to actively avoiding or disengaging from it. Refraining refers to choosing not to engage with an available adaptation without actively opposing it. Repurposing refers to engaging with the adaptation in an alternative or unintended way, giving it a different functional or communicative role.

Through the thematic analysis, we developed three themes that captured recurring patterns in how occupants evaluated robotic partition adaptations. Each theme comprised two contrasting appraisal states: (1) recognized vs. unrecognized personal need, (2) capable vs. incapable adaptation, and (3) socially justifiable vs. unjustifiable adaptation. The first theme thus concerned whether occupants perceived a personal need that motivated the adaptation; the second concerned whether they expected the adaptation to be capable of addressing that need; and the third concerned whether the adaptation could be socially justified to co-located others. We subsequently abstracted these three thematic patterns into three acceptance criteria: (1) personal need, (2) adaptation capability, and (3) social justification.

Our thematic analysis indicated that the acceptance responses were systematically associated with how occupants appraised the three acceptance criteria, as shown in Table 2. Occupants endorsed 56 adaptations (44.8%) when all three criteria were fulfilled: they recognized a personal need, perceived the adaptation as capable of addressing that need, and considered it socially justifiable. They rejected 22 adaptations (17.6%) when they either did not recognize a personal need for the adaptation, or perceived it as both incapable of addressing their need and socially unjustifiable. Occupants refrained from engaging with 38 adaptations (30.4%) when they expected the adaptation to be capable of addressing their need but considered it socially unjustifiable. Finally, occupants repurposed nine adaptations (7.2%) when they perceived the adaptation as socially justifiable but not directly capable of addressing their need, using it instead to indirectly communicate that need to co-located others.

Table 2. The association between the acceptance responses that individual occupants adopted to robotic partition adaptations and the three acceptance criteria resulting from our thematic analysis.

The reported frequencies are included to transparently contextualize the distribution of these cases within the dataset, rather than to indicate statistical confidence in the findings.

Cell colours indicate the acceptance criterion, consistently with Figures 4, 5, and 6: pink = personal need, blue = adaptation capability, yellow = social justification, and gray = appraisal not considered. Darker and lighter shades distinguish the contrasting appraisal states as labelled in the cells.

Although repurposing was the least frequent response, frequency alone was not used as a criterion for retaining response categories. This approach is consistent with reflexive thematic analysis, in which analytic significance is not determined only by prevalence, but by whether a pattern captures something important in relation to the research question (Braun and Clarke, 2006). Here, repurposing represented a qualitatively different configuration of occupant appraisal. Eight occupants across two studies explicitly perceived the intended adaptation as incapable of directly addressing their need, distinguishing repurposing from endorsement; however, they nevertheless engaged with the adaptation purposefully to address that need indirectly, distinguishing repurposing from rejection. Collapsing these cases into either of the other two responses would therefore obscure the combination of perceived incapability and continued purposeful engagement that characterizes repurposing.

The following subsections detail how occupants explained their acceptance responses to robotic partition adaptations through the three acceptance criteria. ²

4.1 Personal need

When occupants recognized a personal need for an adaptation, they could endorse, refrain from, or repurpose it, depending on how they perceived the other acceptance criteria. Conversely, when they did not recognize a personal need, they were more likely to reject the adaptation.

4.1.1 Recognized

Occupants recognized two main types of personal need that motivated robotic partition adaptations. The first need involved reducing disturbances that originated from either environmental factors: “The [partition] helped covering the glare from the window for me“ (S3.O04, endorsing), “It covered the distracting view” (S2.O17, refraining); or activities of co-located occupants: “It really helped reduce the noise from behind me” (S3.O11, endorsing), “I could read at the desk undisturbed when another person was sitting at the sofas” (S2.O09, endorsing); “When my mother comes, she can go to sleep early, and I can still work here without having to turn off the light” (S1.O08, endorsing).

The second need involved improving the quality of the spatial contexts: “[The partition] guarded me from my most vulnerable direction when resting” (S2.O22, endorsing), “When it reflects the ceiling light, it is […] much more suitable to focus and work” (S1.O06, refraining), “The sofa, the painting, and the plant made it feel like a cozy, relaxing place” (S2.O13, endorsing); to the extent that occupants would accept these adaptations in the longer term: “I think a real wall should be installed [to cover the cafeteria] here because [the partition] helped me a lot to focus” (S3.O12, endorsing).

4.1.2 Unrecognized

When occupants were satisfied with the existing spatial context and thus did not recognize a personal need for an adaptation, they perceived any such adaptation as an unnecessary disruption: “[The partition] covered the warm sunlight that I was enjoying” (S2.O20, rejecting), “I could not move freely when it was too close, I had to be cautious of my arm movement” (S2.O23, rejecting), “It did not allow me to look anywhere else but the [architectural] model” (S2.O01, rejecting). This disruption led them to move away from the location of the adaptation: “The desk became so claustrophobic, so I moved to the sofa” (S2.O05, rejecting), “I sat at that corner where it was too small for the [partition] to come in” (S2.O11, rejecting), or reject engaging with the partition altogether: “I felt threatened when it was approaching me” (S2.O21, rejecting).

4.2 Adaptation capability

When occupants expected an adaptation to be capable of addressing their personal need, they endorsed it only if it was also socially justifiable; otherwise, they refrained from engaging with it. When they expected an adaptation to be incapable, they rejected it if it was also socially unjustifiable, but sometimes repurposed it when it could still be justified to others.

4.2.1 Capable

Occupants evaluated the capability of an adaptation based on their past experience: “It felt very cozy sitting next to the [partition] […], so I told it to move there again” (S3.O11, covering distraction, endorsing); recommendations from colleagues: “[O08] told me that the [partition] helped her with the [glare], so I tried it out and it worked!” (S3.O05, endorsing); or their own perception of the robotic partitions: “I believe it tries to make good spaces” (S2.O11, improving privacy, refraining); “It moves slowly so you feel really safe around it” (S3.O12, reducing noise, endorsing).

When occupants had experienced capable adaptations, they became more willing to accept subsequent adaptations, even by following the partition's suggestions: “[The partition] was telling me not to go to the kitchen but to the living space, so I followed it” (S1.O02, improving privacy, endorsing); “The diagonal position shows you which way to go to. It is pretty useful.” (S1.O05, covering distraction, endorsing).

4.2.2 Incapable

Occupants perceived an adaptation as incapable for two reasons. First, they believed that the architectural or robotic affordances of the partition limited its ability to address their needs, including small dimensions: “It couldn't reduce noise because it's too small” (S3.O08, repurposing); insufficient height: “It's not high enough to cover the glare from up there” (S3.O01, rejecting); inability to create full enclosure: “I can't make an enclosed space that is fully private” (S3.O04, repurposing); or slow response speed: “The [partition] moved too slow to help me on time” (S2.O06, distracting sound, repurposing).

Second, occupants perceived some adaptations as introducing new distractions that outweighed their intended benefits. These distractions stemmed either from the partition itself: “I was concerned that it might suddenly move” (S2.O23, covering distraction, rejecting), “The white color of the [partition] made the space look cold” (S2.O19, improving privacy, rejecting); or from the spatial outcomes of the adaptations: “It made me feel claustrophobic because it was very narrow” (S2.O10, covering distraction, rejecting); “It feels stressful when [the partition] blocks the views of the windows” (S1.O01, improving privacy, rejecting); “It interrupts the flow between cooking and dining” (S1.O04, reducing cooking smell, rejecting); “They didn't see me behind the [partition], so they kept talking loudly about private matters” (S2.O11, improving privacy, rejecting).

When occupants found an adaptation less capable of addressing their needs than expected, they often rejected engagement with it: “I used it once, but it couldn't help with the sound of people talking” (S3.O10, rejecting); “[The partition] did not cover the view of the desk, which was stressful because it reminded me of office work” (S2.O18, rejecting); “The confinement reminds me of the intense practice room in music school” (S1.O05, reducing noise, rejecting).

4.3 Social justification

When an adaptation was perceived as socially justifiable, occupants endorsed it if it was also capable of addressing their personal need, or repurposed it if it was not directly capable but could still be used to communicate that need. In contrast, when an adaptation was perceived as socially unjustifiable, occupants either refrained from it despite its capability, or rejected it when it was also incapable of addressing their need.

4.3.1 Justifiable

Occupants primarily viewed an adaptation as socially justifiable when it did not disturb co-located others: “It was better to make my own corner beforehand so that I did not interrupt anybody's work” (S3.O11, covering distraction, endorsing); “I made it protect myself […] because I cannot expect everyone […] to be quiet for me” (S3.O12, reducing noise, endorsing). Conversely, when an adaptation did affect others, occupants considered it socially justifiable only when it addressed a crucial need: “I need a very quiet space to rest, so the [partition] was helpful for me […] although people had to go around it” (S2.O21, endorsing), “It was a very important meeting with a person from the government, so I did not want anyone to interrupt” (S3.O08, endorsing); or when it was collectively agreed to benefit multiple people: “We all benefited from covering the noise” (S3.O12, endorsing); “Of course to stop the entrance of reception, everybody has to first agree” (S3.O08, improving privacy, endorsing); “Their movements in the meeting room affected all three of us, so it felt logical to use the [partition] at that moment” (S3.O05, covering distraction, repurposing).

When occupants perceived this social justification, some felt empowered to repurpose an otherwise incapable adaptation in unintended ways to address their needs, such as by redirecting the adaptation toward co-located occupants: “There was only one [partition] and it could not protect all three of us, so it was logical to move it to the [other co-located occupants] who were creating the distraction” (S3.O12, covering distraction, repurposing) to subtly encourage them to modify their behaviors: “[The partition] could not make a sound barrier for me, so I used it to send a message to colleagues to please [keep quiet]” (S3.O13, repurposing). These occupants reflected that they had considered other alternative options before finally deciding to repurpose, such as by suppressing their personal needs: “I tried to ignore the distraction because I did not want to annoy the colleagues […] but then they got too loud” (S3.O04, reducing noise, repurposing); or attempting other adaptations: “I chose to cover the meeting room instead of covering our [sub-space] because it was more effective” (S3.O09, covering distraction, repurposing).

4.3.2 Unjustifiable

Occupants deemed an adaptation socially unjustifiable for three main reasons. First, they prioritized the needs of other co-located occupants over their own, considering adaptations that affected others as unjustifiable even when these adaptations were capable of addressing their personal need: “Putting the wall there would make the room dark for them” (S3.O03, blocking glare, refraining), “Covering the meeting room would disturb the meeting of my colleagues” (S3.O05, covering distraction, refraining). As a result, they refrained from engaging with these adaptations: “I did not move [the partition] there any longer, because it would be like I put my needs above others” (S3.O06, reducing noise).

Second, occupants avoided adaptations that would make them socially responsible for disturbing the shared space: “I did not follow [the partition] because I didn't want people to think I was responsible for disrupting the room” (S2.O26, improving privacy, refraining). In other cases, the adaptation drew unwanted attention from others, undermining the very need it was meant to address: “Colleagues who passed by got curious and started asking about the [partition], and I could not focus” (S2.O03, covering distraction, rejecting), “I couldn't avoid acknowledging all the passers-by, so I couldn't really work” (S2.O10, blocking glare, rejecting).

Third, occupants perceived some adaptations as violating social norms. They described these adaptations as antisocial: “It feels antisocial when dividing the space, like it's signaling that I want to be left alone” (S1.O01, covering distraction, refraining); unwelcoming: “Covering the door made the office feel more private, but might give the wrong impression to visitors” (S3.O13, improving privacy, refraining); or controlling: “The [partition] created a corridor that steered people away” (S2.O23, covering distraction, refraining).

5 Architectural Robot Acceptance Model (ARAM)

Based on these findings, we developed the Architectural Robot Acceptance Model (ARAM), shown in Figure 4, as a situated representation of how the three acceptance criteria combine to shape an occupant's response to a specific adaptation. In ARAM, transitions between the appraisal states of personal need, adaptation capability, and social justification are represented as thresholds along a common continuum of perceived stress severity. Here, perceived stress severity refers to how consequential an unmet or anticipated need becomes for an occupant in a particular situation, ranging from a mild concern to an urgent need for adaptation. It therefore encompasses both disturbances that occupants seek to mitigate and desired states that they seek to achieve. Although the three criteria remain conceptually distinct, their threshold positions can be compared because each refers to the severity of the same focal need at which an appraisal changes state.

The relative positions of these thresholds, together with the occupant's current perceived stress severity, correspond to four possible acceptance responses: endorsing, rejecting, refraining from, and repurposing. Because the threshold positions are situated rather than fixed, each configuration represents a specific occupant-adaptation-need-context relation and may shift as that relation changes.

The thresholds operate in different directions. Personal need and social justification act as lower thresholds. The personal need threshold marks the level of perceived stress at which an unmet or anticipated need becomes consequential enough for the occupant to recognize a need for adaptation (see Section Personal need). The social justification threshold marks the level at which that need becomes consequential enough for the occupant to consider the adaptation justifiable despite its potential impact on co-located others (see Section Social justification). By contrast, adaptation capability acts as an upper threshold: beyond a certain level of perceived stress severity, the adaptation may no longer be perceived as sufficient to address the need through the architectural and robotic affordances of the partition (see Section Adaptation capability). The relative positions of these lower and upper thresholds determine which combinations of appraisal states, and therefore which acceptance responses, become possible as perceived stress severity changes.

5.1 Theoretical grounding

We draw on previous psychological and social behavioral theories to provide conceptual grounding for this threshold-based interpretation. First, the transactional model of stress and coping proposes that stress depends on how an individual appraises their relationship with the environment (Lazarus and Folkman, 1984). Through primary appraisal, individuals evaluate whether a situation has implications for their wellbeing, for example as a harm, threat, or challenge; through secondary appraisal, they evaluate what can be done to manage that situation. This emphasis on the subjective appraisal of whether an environmental condition is sufficiently consequential parallels our personal need threshold: an occupant may experience a condition without recognizing a need to adapt until that condition becomes sufficiently salient to them. Second, protection motivation theory proposes that responses to a perceived threat depend not only on its severity and likelihood, but also on the perceived efficacy of the available protective response (Rogers, 1975). This notion of response efficacy parallels our adaptation capability threshold, as occupants evaluate whether an architectural robot can sufficiently address the need they perceive; beyond this perceived capability, the adaptation may no longer be considered an adequate response. Finally, Granovetter's threshold model of collective behavior defines an individual's threshold as the point at which the perceived benefits of taking an action outweigh its costs under particular social conditions (Granovetter, 1978). Although developed to explain collective behavior, this notion provides a useful analogy for our social justification threshold: an occupant may consider an adaptation unjustifiable when its social costs to co-located others outweigh the personal need it addresses, but may consider the same adaptation justifiable when that need becomes sufficiently important relative to those social costs. These theories thus provide theoretical precedents for interpreting acceptance as a situated appraisal process, in which changes in perceived need, response capability, and social costs can produce qualitative transitions.

5.2 Positioning against established acceptance models

ARAM shares several considerations with established technology acceptance models. The adaptation capability threshold relates most closely to perceived usefulness in TAM, which concerns whether using a technology is expected to improve performance (Davis, 1989). Perceived ease of use is less directly related, as it concerns the effort involved in using a technology (Davis, 1989). For architectural robots that can initiate adaptations autonomously, this effort may also involve how occupants understand, accommodate, or override adaptations performed without direct input. Within ARAM, control or interaction effort may nevertheless affect whether occupants can make effective use of an adaptation's capabilities. Similar considerations are represented in UTAUT through performance expectancy, with effort expectancy and facilitating conditions further accounting for the effort and resources involved in technology use (Venkatesh et al., 2003). UTAUT2 retains these core considerations while extending UTAUT to consumer technology contexts through hedonic motivation, price value, and habit (Venkatesh et al., 2012). These additional constructs concern broader motivations and patterns of technology use and therefore have no direct equivalent among the three ARAM thresholds. The Almere model extends these considerations through perceived usefulness, perceived ease of use, and perceived adaptability (Heerink et al., 2010), while HRCAM incorporates perceived usefulness and perceived ease of use alongside context-specific factors such as safety and enjoyment (Bröhl et al., 2019). The social justification threshold also relates to the social dimensions of established acceptance models, particularly social influence in UTAUT and UTAUT2 (Venkatesh et al., 2003, 2012) and social considerations incorporated in HRCAM (Bröhl et al., 2019). Finally, personal need has no direct equivalent in these models, although it relates more broadly to whether engaging with a technology is perceived as relevant or beneficial, as reflected in perceived usefulness (Davis, 1989) and performance expectancy (Venkatesh et al., 2003).

ARAM, however, differs primarily in how these considerations are defined in relation to a specific adaptation and how they interact within the model. The personal need threshold represents the point at which an occupant recognizes a situational need for an adaptation, whereas perceived usefulness in TAM and performance expectancy in UTAUT and UTAUT2 describe the expected benefits associated with using a technology more generally. The adaptation capability threshold within ARAM concerns whether a particular adaptation is perceived as sufficient to address the recognized need under the current conditions. Perceived usefulness and perceived ease of use in TAM instead concern whether using a technology is expected to improve performance and the effort involved in using it, respectively (Davis, 1989). An architectural robot may therefore be perceived as useful and easy to use even when a specific adaptation is considered incapable of addressing a particular need. Moreover, because adaptation capability operates as an upper threshold, an adaptation that is initially perceived as capable may become insufficient as the perceived stress it aims to address becomes more severe. The social justification threshold concerns whether the consequences of the adaptation can be justified to co-located occupants who may also be affected by the resulting spatial change. Social influence in UTAUT and UTAUT2 instead concerns the extent to which important others believe that a technology should be used (Venkatesh et al., 2003, 2012). ARAM further relates these three considerations through their relative threshold positions along a common continuum of perceived stress severity, whereas the established models discussed above generally relate multiple constructs to outcomes such as behavioral intention and subsequent technology use.

These differences also influence how the models may be applied over time. Established acceptance models generally examine intention to adopt or use a technology and its subsequent use over a broader technology-use trajectory. ARAM instead focuses on repeated appraisals of individual adaptations as occupant needs, expectations, and social situations change. ARAM can therefore support retrospective explanations of why a particular adaptation was received in a certain way, as well as prospective estimates of how an occupant may respond to a forthcoming adaptation when the relevant threshold positions can be established. The models also address different acceptance consequences. Established acceptance models primarily focus on behavioral intention and technology use, whereas ARAM distinguishes how an occupant responds to a particular adaptation by endorsing, refraining from, rejecting, or repurposing it. These responses can also have consequences for co-located occupants because architectural robot adaptations may physically alter a shared environment. In this sense, ARAM complements established acceptance models by describing how related acceptance considerations operate as situated thresholds that shape responses to individual adaptations over time.

5.3 Representing the acceptance thresholds

To represent the acceptance thresholds qualitatively, we position them along a common continuum of perceived stress severity and discretize this continuum into three qualitative ordinal levels: low, medium, and high. These levels do not represent measured numerical values or equal intervals, but serve as interpretive references for comparing the relative positions of the three acceptance thresholds in a given situation. Accordingly, the ordinal positions indicate when each appraisal changes state relative to the severity of the same focal need, rather than treating personal need, adaptation capability, and social justification as measurements on a common psychological scale. This approach is consistent with (a) qualitative magnitude coding, where ordered labels can express the relative intensity of a phenomenon without treating it as a quantitative measure (Miles et al., 2020; Saldaña, 2021), and (b) qualitative data displays that use matrices or visual summaries to compare interpretive findings across cases (Gale et al., 2013).

For retrospective qualitative representation, these ordinal levels can be coded from occupants' accounts of how consequential an unmet or anticipated need becomes in relation to a specific adaptation. A low level indicates that the need is anticipated or only mildly consequential; a medium level indicates that it has begun to affect comfort or activity while remaining tolerable; and a high level indicates that it is experienced as urgent or severe. The position of each threshold is then identified at the point where its corresponding appraisal changes state. For the two lower thresholds, the ordinal level indicates the perceived stress severity at which the appraisal changes from unrecognized to recognized personal need, or from socially unjustifiable to justifiable. For the adaptation capability threshold, the logic is reversed: the ordinal level indicates the point beyond which the adaptation is no longer perceived as sufficient, for example because the architectural robot is too small, too slow, or introduces new disruptions.

The illustrative configuration in Figure 4a is based on the acceptance response of S3.O05 to a robotic partition adaptation aimed at blocking morning glare. In this case, S3.O05 recognized a personal need for the adaptation at a relatively low level of perceived stress severity, even before the glare fully disrupted her work. She generally perceived the partition as capable of blocking glare, but only up to the point where the sun became too high for the partition to block. At the same time, she considered the adaptation socially justifiable only at a medium level of perceived stress severity, because she avoided disturbing her colleagues when the glare was still tolerable. Consequently, she endorsed the adaptation only when her perceived stress from glare reached a medium to high level.

Our results further show that the relative positions of the three acceptance thresholds vary not only between occupants, but also for the same occupant across contexts. For example, an occupant who prioritizes social harmony in an unfamiliar context may consider an adaptation that affects others socially justifiable only at a high level of perceived stress severity, causing them to refrain from it at lower levels even if it addresses their own need. In a familiar context, however, the same occupant may consider the adaptation socially justifiable at a lower level of perceived stress severity, trusting that co-located others will understand its purpose in addressing their personal need. As such, ARAM does not aim to assign fixed thresholds to particular occupants or adaptations, but to explain how these thresholds, and consequently the acceptance responses, can shift across specific occupant-adaptation-need-context relations.

5.4 Propositions for testability

To enable future empirical testing of the model, we formulate two propositions that can be evaluated using data collected independently of the present model development dataset. The first proposition concerns whether independently assessed combinations of the three appraisal states correspond systematically to the acceptance responses identified in our qualitative analysis: recognized need + capable + justifiable should correspond to endorsement; recognized need + capable + unjustifiable to refraining; recognized need + incapable + justifiable to repurposing; and either an unrecognized need or recognized need + incapable + unjustifiable to rejection. To test this proposition, the three appraisal states must be measured independently of the acceptance response used as the outcome.

The second proposition concerns whether increasing perceived stress severity corresponds to the ordered sequence of acceptance responses implied by an independently established threshold order. For example, when personal need lies below social justification, and both lie below adaptation capability, the expected sequence is rejecting, refraining, endorsing, and finally repurposing. This sequence, however, can only be tested when the threshold positions remain stable while perceived stress severity changes. If the thresholds also shift, the resulting sequence cannot be attributed to changes in perceived stress severity alone. Repeated observations that contradict either proposition under these independently established conditions would therefore count against ARAM. For example, repeated endorsement when personal need is independently assessed as unrecognized would contradict the first proposition, while a different response sequence under a stable, independently established threshold order would contradict the second.

6 Discussion

We first discuss how to interpret and scope the Architectural Robot Acceptance Model (ARAM). We then reflect on how the distinctive affordances of architectural robots may shape occupant acceptance of their adaptations. Lastly, we propose how the three acceptance thresholds could be operationalised, and discuss how ARAM may inform future applications in human-building interaction and human-robot interaction.

6.1 Interpreting the Architectural Robot Acceptance Model (ARAM)

We propose that ARAM should be interpreted as a situated model with dynamic thresholds that vary across particular occupants, adaptations, and contexts. Its unit of interpretation is a specific occupant-adaptation-need-context relation at a particular moment. Once the relevant appraisal states and threshold positions have been independently established for that relation, ARAM supports conditional estimates of the likely acceptance response.

Within an occupant-adaptation-need-context relation, the continuum of perceived stress severity represents how consequential an unmet or anticipated need becomes for an occupant in a given situation. Such personal needs may involve mitigating an existing disturbance such as environmental discomfort (Onishi et al., 2022; Nguyen and Vande Moere, 2024) or difficulty completing an intended action (Takashima et al., 2016; Brown et al., 2024), or enabling a desired future state, such as social interaction (Grönvall et al., 2014; Sadka et al., 2022) or ergonomic behavior (Lee et al., 2019; Shin et al., 2019).

As such, adaptation capability should be interpreted through the occupant's expectation of how far the architectural robot can address that need through its architectural and robotic affordances (Gibson, 1979; Maier et al., 2009). This expectation may depend on whether the occupant perceives a clear connection between the adaptation and its intended benefit. Adaptations that react to current needs may be more readily perceived as capable because their effects are easier to recognize. In contrast, proactive adaptations aimed at future benefits may be perceived as incapable when occupants do not yet recognize how the adaptation relates to their current needs.

Lastly, social justification should be interpreted through how occupants evaluate the adaptation in relation to co-located others. This evaluation depends on whether an occupant believes that co-located others can understand the need that the adaptation aims to address. When an occupant expects others to understand the need behind an adaptation, they may begin to weigh the personal benefit of the adaptation against its social impact, such as whether it disturbs others, violates social norms, or requires shared agreement. This interpretation aligns with theories that emphasize subjective norms, perceived behavioral control, and responsibility in shared environmental actions (Ajzen, 1991; Xu et al., 2020).

6.2 Scoping the Architectural Robot Acceptance Model (ARAM)

We propose that ARAM is applicable when adaptations performed by architectural robots enter occupant experience. This means that an adaptation must be perceptible or consequential enough for occupants to appraise it in relation to their own needs, expectations, or social context. The boundary of ARAM is therefore the presence of occupant appraisal rather than a specific robotic typology.

This scope includes adaptations that directly support occupants, as well as those primarily driven by building-level goals, such as improving energy performance through adaptive building facades (Loonen et al., 2013; Attia et al., 2018). In smart buildings, such adaptations may pursue collective benefits that differ in perceived value across individual occupants. ARAM applies when these adaptations enter occupant experience through perceptible or consequential changes, allowing their purpose and effects to be appraised in relation to personal need, adaptation capability, and social justification. For example, an occupant may notice an adaptation intended to reduce building energy use but see no personal relevance in that goal, leaving the personal need criterion unrecognized. Alternatively, an occupant may consider energy saving personally important but fail to understand how the adaptation contributes to that goal, leading them to perceive the adaptation as incapable of addressing that goal. These cases remain within the scope of ARAM because the adaptation is still being appraised, even when its purpose or benefit is not recognized. Conversely, adaptations that remain outside occupants' awareness or produce no perceivable or consequential effect that allows appraisal fall outside this scope.

Within this scope, ARAM concerns how occupants perceive and appraise adaptations performed by architectural robots, rather than the objective technical performance of those robots alone. An adaptation may perform well in measurable terms but still be perceived as incapable if occupants do not recognize its effect, such as when acoustic or air-quality benefits remain too subtle. Conversely, occupants may perceive an adaptation as capable even when its measurable performance is limited, if they believe it addresses their need or provides a meaningful architectural configuration.

Future research should therefore test ARAM across different occupant needs, social contexts, and technological typologies to examine where its components remain sufficient, and where additional criteria, thresholds, or responses may emerge.

6.3 Explaining the acceptance mechanism

While ARAM resonates with established knowledge from human-building interaction, human-robot interaction, and environmental and social psychology, it also highlights an acceptance mechanism shaped by the distinctive combination of affordances of architectural robots: they are both robotic systems and spatial elements architecturally embedded in the built environment.

From the perspective of human-robot interaction, ARAM aligns with prior findings that people are more likely to accept robots when they adapt to personal preferences (de Graaf and Ben Allouch, 2013), provide support in daily life (Heerink et al., 2009), and behave in socially appropriate ways (Niemelä et al., 2019). Architectural robots may benefit from these expectations when their robotic affordances are legible and their adaptations appear useful, predictable, and appropriate. However, unlike many social robots, architectural robots are not encountered only as interaction partners. They are also spatial elements that occupants use, inhabit, and expect to remain functionally reliable. This dual role may explain why occupants in our studies scrutinized each adaptation closely: an adaptation is evaluated not only as robotic behavior, but also as a change to the spatial context that occupants inhabit.

From the perspective of human-building interaction, this scrutiny may also reflect occupants' tendency to resist building automation when it reduces perceived control (O'Brien and Gunay, 2014). Since occupants often prefer to retain control over their surroundings (Lechner et al., 2021; Schweiker and Wagner, 2016), adaptations performed by architectural robots can feel intrusive when they appear unexpected, poorly timed, or difficult to override. We therefore suggest that occupant acceptance may depend partly on how the architectural robot is initially perceived: as a robot with agency, as a building automation system, or as a static spatial element. Architectural robots with visibly robotic morphologies may set clearer expectations for their mobility and agency (Onishi et al., 2022; Brown et al., 2024), whereas those that resemble static building elements may provoke stronger discomfort when they suddenly adapt (Grönvall et al., 2014; Sirkin et al., 2015). These expectations and associated social norms, however, may change over time as robots become more common and occupants grow more familiar with them. Adaptations that are currently perceived as socially inappropriate may therefore become more acceptable in the future.

From the perspective of environmental and social psychology, adaptations performed by architectural robots may also be interpreted as environmental adjustments that affect multiple occupants at once. In our studies, occupants sometimes treated an adaptation as if it reflected their own intention, and therefore considered whether it could be justified to co-located others. This observation aligns with theories suggesting that environmental adjustments in shared spaces are shaped by social norms, perceived responsibility, and perceived behavioral control (Ajzen, 1991; Xu et al., 2020). Unlike interactions with humanoid robots, where social embarrassment can influence occupants' evaluations (Bartneck et al., 2010), architectural robots may elicit social responsibility because their adaptations physically reconfigure the shared space.

Taken together, these perspectives suggest that the situated appraisal mechanism proposed by ARAM may be particularly salient for architectural robots because their adaptations can simultaneously be experienced as robotic behaviors, building automation, and socially situated environmental adjustments.

6.4 Operationalising acceptance thresholds

Operationalising ARAM serves two related purposes: estimating how an occupant may respond to an architectural robot adaptation, and empirically evaluating whether the proposed relationships are supported. For future estimation during interaction, environmental conditions, interaction histories, occupant feedback, and observable behaviors may indicate personal need, adaptation capability, and social justification. For empirical evaluation, researchers should assess these appraisal states independently of the occupant's subsequent acceptance response, then compare the ARAM-based estimate with the response actually observed. Repeated agreement or disagreement would provide evidence for or against the proposed relationships. The following sections outline factors that may inform each threshold and how these estimates may eventually guide architectural robot adaptations.

6.4.1 Personal need

Our findings indicate that occupants adopted different acceptance responses to the same adaptation depending on how strongly they perceived the personal need it aimed to address in a particular situation.

When personal need concerns the surrounding environment, this interpretation aligns with human comfort research. Multi-domain comfort theory shows that occupants perceive environmental conditions differently depending on personal, contextual, and environmental factors (Schweiker et al., 2020). These differences influence whether and when they adapt their surroundings (Hong et al., 2020). Although ARAM represents the personal need threshold along a single continuum of perceived stress severity, this perceived stress may be shaped by physiological characteristics, such as age, sex, and body mass index (Rewitz and Müller, 2022; Schweiker et al., 2018), as well as psychological and contextual factors, such as cultural background (Jowkar et al., 2020) and location-specific norms (Vittori et al., 2021). It may also emerge from interactions between multiple environmental domains. For example, lighting color can influence thermal comfort (Chinazzo et al., 2021), while productivity can be shaped by overall environmental comfort (Sadick et al., 2020).

Building on these complexities, we propose that the personal need threshold may be estimated indirectly by sensing environmental and spatial conditions that are likely to generate perceived stress. Architectural robots could monitor indoor environmental conditions including the visual domain with sub-dimensions such as glare and luminance; the acoustic domain with sound pressure level and speech activity; the thermal domain including air temperature, humidity, and air velocity; and the air quality domain accounting for relevant gases or particulate concentration. Additional monitoring could include spatial conditions such as occupancy density, interpersonal proximity, visual exposure, or movement obstruction. These data should be understood as proxies for potentially stressful conditions, rather than direct measurements of personal needs. They could be triangulated with occupant feedback, interaction histories, override behavior, or Personal Comfort Models (PCM) to estimate when a personal need is likely to become recognized (Kim et al., 2018; Bogatu et al., 2023). Physiological sensing, such as Heart Rate Variability (HRV), may provide a supplementary signal of physiological arousal, but should not be treated as a proxy for perceived stress severity because such arousal can have many causes. Continuous environmental mapping could further reveal whether this threshold shifts over time, as comfort perception changes with context, prior exposure, and adaptation (Vellei et al., 2023). Lastly, physics-based or data-driven simulations may help architectural robots anticipate conditions under which unmet needs are likely to emerge, supporting more informed and timely adaptations (Bavaresco et al., 2024).

6.4.2 Adaptation capability

Our findings indicate that occupants adopted different acceptance responses to the same adaptation depending on whether they expected it to be capable of addressing their needs.

This emphasis on expectation aligns with human-robot interaction research, where perceived capability can shape acceptance as strongly as, or even more strongly than, actual technical performance (Naneva et al., 2020). Such expectations are influenced by occupant characteristics, including age, gender, cultural background, and prior exposure to robots (Bröhl et al., 2019; Nikolovska et al., 2024), as well as by whether the robot's apparent affordances align with what it can actually do (Kwak et al., 2017). When expectations are unmet, occupants may overestimate or underestimate the robot's capability (Cha et al., 2015), which can undermine trust (Malle et al., 2020). Establishing appropriate expectations before and during interaction is therefore important (Paepcke and Takayama, 2010), particularly through design features that make the robot's capabilities legible (Kunold et al., 2023).

For architectural robots, our findings suggest that perceived adaptation capability is shaped by four interrelated dimensions. First, occupants assess whether the element is recognizable as capable of robotic action, rather than as an ordinary static building element. Second, they evaluate whether its architectural affordances can address their need. Third, they consider whether its robotic affordances, including speed, timing, or trajectory, support or hinder the adaptation. Fourth, they evaluate whether the available interaction modalities allow them to control, adjust, or override the adaptation sufficiently to address their need.

We therefore propose that the adaptation capability threshold could be estimated by examining how occupants perceive the architectural and robotic affordances of an architectural robot before and after use. This could be done through qualitative surveys, onboarding studies, interaction logs, and post-adaptation feedback that assess whether adaptations confirm or challenge occupants' expectations. Familiarization, demonstrations, and training may help occupants form more accurate expectations of what the architectural robot can and cannot do, reducing misperception and improving alignment between expected and actual adaptation capability. Once estimated, the adaptation capability threshold could potentially inform not only the physical design of the architectural robot, but also its behavior, such as by providing additional explanation, requesting confirmation from occupants, or offering additional control before performing an adaptation when its perceived capability might be uncertain.

6.4.3 Social justification

Our findings indicate that occupants adopted different acceptance responses to the same adaptation depending on whether they felt able to justify the adaptation to co-located others.

This interpretation aligns with environmental and social psychology, which shows that decisions in shared settings are shaped not only by individual preferences, but also by social norms, perceived responsibility, and perceived behavioral control (Ajzen, 1991; Xu et al., 2020). Because individuals tend to take fewer social risks in group contexts (Rabow et al., 1966), occupants may hesitate to accept an adaptation if it could disturb others, appear selfish, or violate implicit norms of the space. However, this hesitation is not uniform. Social hierarchy, familiarity, personality, and perceived control may all influence how much responsibility occupants feel for justifying an adaptation. More influential or confident occupants may be more willing to accept adaptations that affect others, whereas more socially cautious occupants may refrain even when the adaptation addresses their own need. At the same time, occupants may still prioritize their own needs when perceived control over the environment is important for their satisfaction (Lechner et al., 2021). Since this sense of control tends to decrease as the number of people in a shared space increases (Schweiker and Wagner, 2016), architectural robots need to support individual agency while remaining sensitive to social norms.

We propose that the social justification threshold could be estimated by examining occupants' familiarity with the social context and co-located others, their perceived social norms and hierarchy, their perceived control over the shared environment, their sense of responsibility, and relevant personality traits. These factors could then be examined as predictors of occupants' independently reported social justification, while subsequent acceptance responses remain a separate outcome for evaluating ARAM's estimates. Once estimated, the social justification threshold could inform how architectural robots negotiate adaptations in shared spaces. For instance, an architectural robot could make the reason for an adaptation more legible to all co-located occupants, or request confirmation when its social impact is uncertain. It could also provide subtle 'nudges' that help occupants recognize persistent unmet needs, without forcing them to take responsibility for the adaptation alone. Given the importance of social norms, educational sessions could accompany the introduction of architectural robots, clarifying that adaptations are not only for personal benefit, but may also address unmet needs among co-located occupants.

6.5 Applying the Architectural Robot Acceptance Model (ARAM)

We propose that ARAM can be applied in three ways: to explain how occupants form acceptance responses to adaptations performed by architectural robots, to estimate how adaptations may be received, and to inform how architectural robots might adapt in more acceptable ways over time.

First, we propose that the three acceptance criteria introduced in ARAM—personal need, adaptation capability, and social justification—can guide the design of questionnaires, interview protocols, and observational studies that aim to explain why occupants endorse, reject, refrain from, or repurpose specific adaptations. ARAM also provides a common analytical lens for published HBI studies, allowing occupant responses to be categorized through the three criteria, as illustrated in Figure 5. This categorization may reveal recurring acceptance patterns across studies and contexts, while independent validation requires future studies that prospectively assess the three criteria separately from the resulting acceptance response.

To provide such validation, future studies should isolate individual criteria to examine how each one shapes acceptance responses across different occupants, adaptations, and social conditions. Controlled studies may vary the severity of an occupant need, the capability of an adaptation to address it, or its consequences for co-located others, while assessing the relevant appraisal before observing the acceptance response. Comparing these independently assessed appraisals with observed responses would evaluate whether the configurations specified by ARAM are supported. Future work should also examine how these adaptation-level responses relate to broader measures of technology acceptance, such as TAM and UTAUT, while recognizing that they address different levels of acceptance.

Beyond empirical validation, once the three acceptance thresholds are operationalised, as discussed in Section operationalising acceptance thresholds, they may help to anticipate how an occupant is likely to respond to a future adaptation. Such anticipation should remain a situated conditional estimation, based on whether an occupant currently recognizes a personal need, perceives the adaptation as capable, and considers it socially justifiable.

Because ARAM is designed around dynamic acceptance thresholds, the resulting estimates may shift over time, even for the same occupant and adaptation. We hypothesize that these shifts may reflect changes in physiological state, psychological state, familiarity, trust, social relations, or recent experiences, as illustrated in Figure 6. Shifts in threshold positions should therefore be assessed through measures independent of the acceptance response and, where appropriate, through repeated longitudinal observations, rather than inferred after observing the response itself. Anticipating acceptance over time would consequently require repeated estimation of these threshold positions and the conditions that shape them. Real-time estimation may require continuous sensing of behavioral, physiological, environmental, or social information, which introduces technical challenges as well as questions of privacy, consent, and the appropriateness of collecting such data.

Where reliable estimates are available, they may inform the behavior of architectural robots over time. For example, an architectural robot could prioritize adaptations that occupants are likely to endorse, avoid or delay adaptations that may be rejected, request confirmation when occupants may refrain, or offer alternatives when occupants may repurpose the adaptation. In this way, ARAM can inform not only the analysis of occupant acceptance, but also the design of more adaptive and appropriate architectural robot behaviors.

However, designing architectural robots to always act in predictably acceptable ways may not be the only goal. Occupants who prefer full control over their surroundings may value predictability, but excessive predictability could also reduce the curiosity, engagement, or serendipity that can arise when robotic systems behave in unexpected ways (Brown et al., 2024; Agnihotri and Knight, 2019; Hoggenmueller et al., 2020). Rather than assuming that architectural robots should always maximize acceptance, future work should examine how they can balance different levels of robotic autonomy (Beer et al., 2014): allowing occupants to take direct control, performing adaptations independently within clearly acceptable boundaries, or occasionally proposing more exploratory adaptations that may invite engagement or suggest new capabilities while still preserving occupant agency.

7 Limitations and future work

This research is based on a qualitative re-analysis of three previous studies on robotic partitions, which operate at the “space plan” layer (Brand, 1994). While ARAM may be transferable to other architectural robots, its broader applicability remains to be evaluated. Future research should therefore test the model across a wider range of occupant-perceptible robotic, automated, intelligent, and smart technologies embedded in the built environment, particularly those that differ in materiality, autonomy, and social visibility.

Our qualitative methodology relied primarily on occupants' self-reported experiences, which may be affected by recall and did not provide independent measurements of the three proposed thresholds. Because inclusion required sufficiently documented occupant reasoning, tacit, ambiguous, or weakly articulated acceptance responses may be under-represented in our analyzed cases. In addition, reflexive thematic analysis recognizes that qualitative findings are necessarily shaped by researchers' interpretive positions rather than produced independently of them (Braun and Clarke, 2019, 2021). This is particularly relevant because the first and last authors designed and conducted the original studies. Their contextual familiarity provided an important interpretive resource, while the involvement of co-authors who had not participated in those studies introduced differently situated perspectives through which prior assumptions could be challenged. We therefore critically interrogated prior interpretations through a differently situated coding perspective from the second author, interdisciplinary discussion, examination of cases that did not readily fit the developing analysis, and analytical memo-writing. Consistent with reflexive thematic analysis, these procedures were intended to challenge prior assumptions and broaden interpretation across differently situated perspectives. Nevertheless, ARAM has not yet been independently validated using data collected specifically to test its propositions. Future research should therefore test these propositions using independently assessed appraisal states and acceptance responses.

ARAM also primarily captures the situated relationship between the three acceptance thresholds in response to specific adaptations. However, these thresholds may change over time and may not vary independently. For example, repeated exposure to an architectural robot may increase perceived adaptation capability, while changing social familiarity may lower or raise the social justification threshold. Future studies should therefore examine how repeated adaptations, changing environmental conditions, or evolving social relations shift acceptance thresholds over time.

Lastly, ARAM currently explains acceptance in relation to a focal occupant need and a specific adaptation. In real-world environments, however, multiple occupants may have diverse, simultaneous, or competing needs. Future research should investigate how overlapping needs are negotiated, how thresholds differ between co-located occupants, and how architectural robots might prioritize or mediate adaptations that affect several needs at once.

8 Conclusion

Through a qualitative analysis of occupant experiences from three previous empirical studies on robotic partitions, we introduced the Architectural Robot Acceptance Model (ARAM), which explains the acceptance responses that occupants may adopt to adaptations performed by architectural robots. Within ARAM, occupants appraise an adaptation through three criteria: whether they recognize a personal need, whether they perceive the adaptation as capable of addressing that need, and whether they consider the adaptation socially justifiable to co-located others. By representing these criteria as thresholds along a continuum of perceived stress severity, ARAM distinguishes four acceptance responses: endorsing, rejecting, refraining from, and repurposing. As perceived stress severity changes relative to these thresholds, an occupant may shift between these responses in relation to a specific adaptation and context. We then discussed how ARAM should be interpreted as a situated model, including its scope and potential transferability. We also reflected on how architectural robot adaptations may be simultaneously experienced as robotic behaviors, building automation, and socially situated environmental adjustments, and how this combination may shape occupant acceptance. Finally, by outlining how the three acceptance thresholds may be operationalised using insights from human-building interaction, human-robot interaction, environmental and social psychology, and human comfort research, we identified how their appraisal states may be estimated and empirically evaluated. We proposed testable propositions for examining the relationship between these estimates and observed occupant responses to architectural robot adaptations, with the goal of informing adaptations that occupants perceive as capable of addressing recognized needs and socially justifiable to others. Independent empirical validation is required to establish whether these relationships hold reliably, and whether ARAM transfers beyond robotic partitions.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Author Disclaimer

Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or Health and Digital Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.

Footnotes

Edited by: Henriette Bier, Delft University of Technology, Netherlands

Reviewed by: Felix Höpfl, University of Applied Sciences Rosenheim, Germany

Zvinodashe Revesai, University of KwaZulu-Natal, South Africa

Footnotes

The photographs in this subsection were retrieved from the corresponding cited publications.

Each occupant quote in this section is annotated with the study number (e.g., S1), the occupant number within that study (e.g., O01), the need that the adaptation aimed to address (e.g., improving privacy) where relevant, and the resulting acceptance response (e.g., endorsing). For example, a quote may be annotated as follows: (S1.O01, improving privacy, endorsing).