Socially Assistive Robots (SARs) are a type of robot which assist humans through social interaction (Bedaf et al., 2015). They can serve as companions, pets, or service robots (Matarić and Scassellati, 2016). According to Broekens et al. (2009), SARs are understood as social entities that can communicate with users. Based on the type of assistance they provide, SARs can be categorized as either contact assistive robots or social interactive robots. While contact assistive robots provide physical assistance, social interactive robots provide assistance through social interaction Feil-Seifer and Mataric (2005). A similar categorization is suggested by Fracasso et al. (2022) and Heerink et al. (2010), who categorize SARs as either service robots or companion robots. Service robots help in assisting with a variety of physical activities, such as carrying heavy loads or walking assistance. RIBA robot (2023), with its human-type arms is an example of a service robot which is designed to help patients with lifting and moving heavy objects. RIBA is also capable of moving patients between a bed and a wheelchair (Joseph et al., 2018). On the other hand, Companion robots provide social interaction for emotional, social, or psychological support (Fracasso et al., 2022).

Studies have shown that companion robots are particularly useful for older people as they can reduce stress (Saito et al., 2003), depression (Wada et al., 2005), regulate blood pressure (Robinson et al., 2015), and improve people’s mood (Wada et al., 2003). Companion robots are available in different forms and shapes such as pet-like (e.g., pet robots) and human-like (e.g., humanoids). PARO (2023), a robotic baby seal pet, is a popular pet robot which carries various sensors to sense touch, sounds, and visual objects (Vitanza et al., 2019). Similarly, Pepper robot (2024) is a popular humanoid companion robot. Pearl (Pineau et al., 2003) is another popular companion robot which assists older patients by helping with ADL such as giving reminders about medication and appointments and using motion sensors to detect falls and physical inactivity (BUDDY, 2023).

In this section, we first discuss the importance of trust and the factors that influence trust in human-human interactions (HHI) and human-robot interactions (HRI). We then consider trust of older adults on robots and discuss the different factors that are uniquely important in such interactions.

In this work, we used the Kitchenham and Charters’ Systematic Literature Review (SLR) methodology (Kitchenham and Charters, 2007) in which the research questions and search strategy are defined first. The research questions were structured using the Population, Intervention, Comparison, Outcome, and Context (PICOC) method (Eriksen and Frandsen, 2018), as shown in Figure 1.

We were particularly interested in reviewing the methodologies used in the literature for exploring trust in SARs with older people. Therefore, our population consisted of old people, old people with disabilities, elderly, caregiver, carer, unpaid carer, geriatric, and older adults (the inclusion of caregivers, carers and unpaid carers as keywords was motivated by the aim of ensuring comprehensive coverage of relevant literature associated with older adults). By incorporating these additional keywords, we aimed to capture any papers that might be directly or indirectly related to older adults and their trust in SARs). Similarly, as an intervention, we looked at the methods, techniques, or strategies which measured the level of trust in SARs. As, we were interested in determining the methodologies, rather than comparing them, the comparison criteria was excluded. Our output, included the identification of methods for finding trust levels in SARs. Finally, we used a generalized context as we did not want to restrict our research questions to any particular context.

Our search strategy comprised of three phases: identification, screening, and finalization.

The papers included in our review were from 11 different conferences and 21 different journals. The names of these conferences and journals are listed in Table 3. Nearly 70% of these journals and conferences were multidisciplinary in nature (Medical Sciences, Technology, Social Sciences, Computer Science, Engineering, Arts and Humanities and Mathematics) and 30% were from the Computer Science category. This diverse representation underscores the collaborative and inclusive approach taken in exploring various facets of trust. The mix of these contributions gives us a wider view and shows how different areas of study collaborate to better understand the relationship between older adults and SARs. The inclusion of such a broad spectrum of disciplines contributes to a comprehensive understanding of the complex dynamics surrounding the topic of trust, shedding light on the diverse perspectives that shape our insights into the multidimensional aspects of trust in SARs. From Table 3, the most number of papers selected from any year was 8 (from 2024), while the least number of papers selected from any year was 1.

All of the selected papers focused on evaluating trust on SARs in the context of older adults. Therefore, the participants included in these studies were either older adults or were responsible for taking care of older adults. The nature of interactions with the robots differed across various studies. In certain studies, participants directly engaged with the robot. In contrast, in some instances, the interaction was more indirect, involving participants viewing pictures/videos of robots, while in other studies, participants relied on their perceptions of robots. Classifications used in the SLR were derived from our research questions outlined in Section 1 to capture the key themes found across the literature. Figure 3 presents an overview of our SLR classification. The complete information gathered from the studies included in our SLR is presented in the Supplementary Appendix 1 .

In the dynamic domain of SARs, measuring trust involves a multifaceted approach (Park, 2020). Evaluating the trust of SARs encompasses diverse methodologies, including structured questionnaires to gauge user perceptions, and qualitative insights derived from open-ended questions, interviews, and discussions. Additionally, the integration of advanced techniques like Machine Learning (ML) facilitates objective assessments of system performance. This convergence of methods creates a holistic framework for comprehensively measuring trust in the use of SARs. In our SLR encompassing 47 papers, an inquiry revolved around the methodologies employed to measure trust in SARs when interacting with older adults. The importance of answering this question lies in the critical role trust plays in the successful integration of SARs into the lives of older adults. By comprehensively examining these studies, we aimed to not only identify the diverse methods in use but also determine if there is a consensus on the most effective approach for measuring trust in this context. The significance of uncovering such consensus or trends in methodology preferences is two-fold: it informs best practices for researchers and developers, and it contributes to the establishment of standardized approaches that enhance the reliability and validity of trust measurement in SARs designed for older adults. From our thorough exploration, we identified that the studies included in the SLR employed Validated questionnaires that have undergone rigorous testing and validation processes, establishing their reliability and validity across various contexts; Study specific questionnaires tailored to the unique objectives and context of a particular research study, and machine derived trust assessment using machine learning algorithms.

In our second question, we looked at different robots used in studies about trust in SARs. We wanted to understand what types of robots researchers used to study, how older adults trust and interact with them. This exploration gave us insights into the technology used, showing us different applications that play a role in building trust between older adults and their robotic companions. We found that different robots have been used in studies and categorized them into three types based on interaction/exposure with the participants, i.e., Direct interaction (robots), indirect interaction (computer simulators, pictures, or videos of robots) or no interaction (no robots). A similar categorization was used by Naneva et al. (2020).

Among the 47 selected research papers, almost 49% (23 out of 47) utilized robots as part of their study. In contrast, almost 28% (13 out of 47) of the studies used computer simulators, pictures, or videos of robots, while in the remaining 23% (11 studies), no robots were used (in these studies, the perception of trust in robots was assessed through qualitative approaches, such as conducting interviews or administering questionnaires). Next, we discuss the studies in which each type of robot interaction was used.

In our exploration of the third question, we aimed to look into the contextual dimension of the studies in terms of research setting, the ratio of humans to robots in these research settings as well as factors influencing trust that the researchers explored. We sought to understand the settings in which these studies were conducted, specifically differentiating between laboratory (lab) and wild (e.g., home environments). The choice between these environments holds significant implications for the validity and applicability of the findings, especially given the unique needs and behaviors of older individuals (Molina-Mula et al., 2020). As laboratory setting provides controlled conditions, enabling precise measurements and controlled variables. While this offers experimental rigor, it may not fully capture the real-world intricacies and challenges that older adults might encounter when interacting with robots in their homes.

In addition to investigating the research environments, we also analyzed the types of interaction ratios commonly used in these studies. To structure this analysis, we adapted the interaction framework proposed by Sørensen et al. (2014), originally developed for human-artifact interactions. This framework categorizes interactions into four basic structures: 1. many users interacting with many artifacts, 2. one user interacting with many artifacts, 3. many users interacting with one artifact, and 4. one user interacting with one artifact. While Sørensen et al. (2014) applied this framework to digital artifacts, we tailored it to human-robot interactions, categorizing them as dyadic (1:1) or non-dyadic scenarios. Non-dyadic interactions include 1:Many (a robot engaging with multiple humans), Many:1 (multiple robots assisting a single human), and Many:Many (group-based interactions involving multiple humans and robots). This allowed us to assess which interaction ratios were most commonly employed in laboratory and in-the-wild studies, providing a clearer understanding of how SARs are typically evaluated across different contexts.

In terms of factors influencing trust, encompassed a comprehensive examination of whether the focus of trust assessment was directed towards the robot itself, the human involved, or the environmental aspects surrounding the interaction. For instance, did studies predominantly measure trust in the robot’s capabilities, reliability, and behavior? Or did the assessment pivot towards the human factors, considering aspects such as user expectations, perceptions, and preferences? Furthermore, we explored whether environmental factors, such as the physical surroundings and contextual scenarios, played a pivotal role in shaping trust dynamics.

Our selected studies were conducted in various environments such as lab, home, care homes or nursing homes. Table 6 shows the distribution of the contexts with respect to their interaction type. Lab environment was the most common context in which these studies were conducted as it accounted for almost 49% of the studies. The interaction with the robot in lab was either direct or indirect. The following studies were conducted in lab environment that have direct interaction with the robot: Yan et al. (2013), Cavallo et al. (2014), Ono et al. (2015), Mann et al. (2015), Correia et al. (2016), Ting et al. (2017), Piasek and Wieczorowska-Tobis (2018), Newaz and Saplacan (2018), Rossi et al. (2018), Loghmani et al. (2019), Fitter et al. (2020), Kumar et al. (2022), Giorgi et al. (2023), Rahman (2023), Lorusso et al. (2023), Gul et al. (2024), Wald et al. (2024), and Aharony et al. (2024) and the following 5 studies were conducted in lab but interaction was indirect: Fakhrhosseini et al. (2020), Fracasso et al. (2022), Zafrani et al. (2023), Rahman (2023), and Aly et al. (2024). The “Other” category in Table 6 includes contexts such as a library, a quiet space, and an office. The following studies were conducted in a home environment: Torta et al. (2014), Begum et al. (2015), Piasek and Wieczorowska-Tobis (2018), Sorrentino et al. (2021), and Fiorini et al. (2023). Two studies (Pascher et al., 2022; Harris and Rogers, 2021) were conducted remotely via telephone. Ejdys (2022) used Computer-Assisted Web Interview (CAWI) survey technique while the study conducted by Marin and Lee (2013) and Mo et al. (2017) was conducted in a controlled environment (i.e., a space in one of the coffee areas in an elderly center). Three studies (Correia et al., 2016; Piasek and Wieczorowska-Tobis, 2018; Harris and Rogers, 2021) used more than one context for conducting their experiments (i.e., experiments were partly conducted in a lab and partly conducted in a home environment or online via email) while the study conducted by Giorgi et al. (2023) in a lab environment called Robot Home, was designed to resemble a real living room/home. An experiment conducted by Cavallo et al. (2014) used three service robots in three different contexts, naming the contexts as domestic (DomoCasa Lab, a domotic house developed and managed by the BioRobotics Institute of Scuola Superiore Sant’Anna in Peccioli (Italy)), condominium (common areas, such as the entrance hall, corridors and elevator, of the building where the DomoCasa Lab is located) and urban (the surrounding outdoor pedestrian area) (Cavallo et al., 2014). A study conducted by Huang (2022), interviewees were surveyed in the bustling commercial street in Zhanjiang City in western Guangdong Province, southern China.

Upon reviewing 47 studies, a noticeable pattern emerges where the majority of studies into direct interactions with robots involving older adults were carried out in laboratory settings. Surprisingly, only five studies extended their examination to till. This discrepancy underscores a prevalent inclination toward controlled experimental conditions, likely driven by factors such as regulated variables and experimental control. The limited exploration of direct interactions with robots in home settings specifically for older adults suggests a potential gap in understanding how these interactions unfold in real-world home environments. There appears to be a pertinent need for increased research focusing on older adults and direct robot interactions within home settings to enhance the applicability of findings to their everyday lives. In terms of robot human ratio, Table 6 highlights a significant reliance on the dyadic interaction approach in existing studies, where trust is predominantly measured in 1:1 engagements between a robot and a participant. Only two studies explored the 1:Many interaction approach, primarily through videos of robots presented to participants, indicating limited exploration of non-dyadic interaction dynamics. These limited explorations of non-dyadic dynamics underscore the need for more comprehensive research on how trust operates in interactions involving more than two parties.

As part of the context, we also identified the purpose of evaluating trust within these studies. We found that studies were focused on evaluating different factors of trust. Some of the articles were focused on evaluating acceptance, with trust measurements included as part of acceptance models (i.e., Almere, UTAUT). Other articles focused on understanding specific features of robots and how that features like behaviour, reliability of the robot, etc. relates to trust. To understand better how multiple factors that impact trust have been explored within the scope of SARs and older adults, we categorised factors influencing trust according to the revised human-trust model proposed by Hancock et al. (2021), that offers a com-prehensive exploration of factors influencing trust in human-robot interactions presented in the Figure 4. Each tier of the Figure 4, from Robot - Related Factors to Human -Related Factors and Environmental, provides a breakdown of dimensions of trust in different studies. We have found that studies (Marin and Lee, 2013; Mann et al., 2015; Branyon and Pak, 2015; Correia et al., 2016; Fakhrhosseini et al., 2020; Newaz and Saplacan, 2018; Torta et al., 2014; Loghmani et al., 2019; Poulsen and Burmeister, 2019; Do et al., 2021; Erebak and Turgut, 2019; Fracasso et al., 2022; Ejdys, 2022; Huang, 2022; Kumar et al., 2022; Lorusso et al., 2023) on evaluating robot factors, particularly performance-based, including reliability, communication method, behaviour and failures. Studies Pascher et al. (2022), Branyon and Pak (2015), and Stuck and Rogers (2017) were also focused on robots’ factors, however, they explored trust from the angle of robots’ attributes that includes anthropomorphism and physical appearance. The following studies Cavallo et al. (2014), Piasek and Wieczorowska-Tobis (2018), Mann et al. (2015), and Marin and Lee (2013) explored trust according to a mix of robot factors, evaluating both aspects of performance and attributes (i.e., performance, reliability, and appearance combined).

Regarding trust evaluated accordingly to human - related factors, we found that studies Begum et al. (2015), Mo et al. (2017), Ting et al. (2017), Daniele et al. (2019), Pak et al. (2020), Fitter et al. (2020), Rossi et al. (2018), and Sorrentino et al. (2021) focused on characteristic-based factors such as users’ personality traits, users’ comfort with robots, attitudes towards robots and their expectancy and ability based focused on factors like situational awareness. As Environmental Factors, studies Yan et al. (2013), Zhang et al. (2022), Camilleri et al. (2022), andIshak and Nathan-Roberts (2015) were oriented to explore factors affecting trust from an angle of team collaboration, with elements such as role interdependence and interaction frequency. Finally, 5 studies Harris and Rogers (2021), Erebak and Turgut (2019), Hoppe et al. (2022), Correia et al. (2016), and Fakhrhosseini et al. (2020) did explore a combination of factors that impact trust in the scope of human - related factors, environmental factors, and robot - related factors (i.e., ability base, performance, reliability). In order to understand which contexts have been under explored, we represent factors using heatmap shown in the Figure 5. Each factor is represented along the y-axis, while the x-axis indicates the count of studies addressing that factor. The heat intensity increases with higher counts, as shown by the accompanying color bar.

Analyzing the model represented in the Figure 5, it’s evident that certain factors have been extensively explored in academic research on trust in SAR by older adults, while others may need more attention.

In the Robot Related Factors category, Reliability under Performance based stands out as a heavily researched aspect, with references spanning multiple studies, including works by Cavallo et al. (2014), Piasek and Wieczorowska-Tobis (2018), and Mann et al. (2015) and many more. This indicates a consistent scholarly interest in understanding how the reliability of robots impacts the establishment of trust in SARs by older adults. The second most explored dimension is Behavior.

The analysis of the context in which studies on trust in SARs were conducted shows the significant prevalence of laboratory-based experiments. Nearly half of the studies almost 49%) were carried out in controlled laboratory environments, emphasizing the controlled conditions and precise measurements available in such settings. However, it is noteworthy that only 11% of the studies explored the home environment as a setting for their experiments. The home environment is particularly important when investigating trust among older adults, offering valuable insights into their interactions with robots in daily life (Bajones et al., 2019). This proportion indicates a potential gap in research, highlighting the need for more in-depth exploration of trust dynamics in home settings. Further investigations in this direction can provide a richer understanding of the challenges and opportunities associated with implementing SARs in real-world scenarios, especially among older populations. In our selected studies, much of the existing research predominantly relies on a dyadic interaction, where trust is measured in 1:1 engagements between a robot and a participant. In real-world applications, particularly in caregiving environments, the dynamics of trust can be significantly more complex. SARs, which show promise for use in the care of older adults, are often envisioned for scenarios involving not just the older adult, but also formal or unpaid carers. While robots may directly interact with older adults, the presence of a carer could influence how trust is established and maintained between the robot and the individual receiving care. This introduces a dynamic not typically accounted for in current research, where the carer’s role and their influence on the relationship between the older adult and the robot are not fully explored. The lack of research into non-dyadic interactions, particularly those involving multiple human actors or multiple robots, highlights a critical gap. Trust dynamics in caregiving scenarios, where a robot, carer, and older adult interact require further investigation (Gul et al., 2025). Understanding how trust functions in this triadic context is crucial, as the presence of a carer could shape both the older adult’s trust in the robot and the overall trust dynamics within the caregiving relationship. This gap in research presents an opportunity to better understand and design robots that can function effectively in multiuser caregiving environments.

Upon examining the factors influencing trust measurement methods, a clear distinction emerges between human factors and robotic factors, particularly concerning whether the measurement is subjective or objective. Upon analyzing the data regarding measurement methods and influencing factors, most studies employing subjective measurement methods predominantly focus on robotic factors, such as reliability and robot’s behavior, to gauge trust. The emphasis on these aspects of the robot’s performance suggests that subjective measurements often capture the qualitative, experiential aspects of trust influenced by the robot’s observable characteristics. Moreover, within subjective measurement methods, communication emerges as a primary environmental factor. The significance placed on communication underscores the importance of interactive and engaging features in shaping trust perceptions. These studies likely employ surveys, interviews, or observational techniques to capture the subjective experiences. Conversely, when examining into objective measurement methods, there is a shift towards human factors, specifically personality traits and propensity to trust. The utilization of these human-centric factors in objective measurements implies a more quantitative and systematic approach, likely involving computational models like machine learning to analyze patterns and behaviors objectively.

In this question, we directed our attention to the diverse demographics that researchers had explored in the context of measuring trust in SARs. In our investigation into population size and age range across each study, a key objective was to discern the health status of participants. Our inquiry aimed to determine whether individuals involved in these studies had age-related disabilities or if the participant pool primarily comprised healthy older adults. The demographics were characterized by analyzing the distribution of papers based on the countries to which the authors were affiliated. This approach allowed us to understand the representation of different nations in the body of literature on trust in SARs. By focusing on the author’s country of affiliation, we gained insights into the geographic diversity of research efforts, providing a lens through which to explore how trust in SARs has been studied and understood within various international contexts. Simultaneously, we sought to uncover any underrepresented demographics, thereby revealing potential gaps in our understanding. It should be noted that while our analysis identified the geographical distribution of studies, we did not perform a direct comparison of trust levels between countries, as the included studies varied widely in their contexts, aims, and trust measurement methods. Furthermore, we delved into the population sizes that featured in these studies, examining the scale at which trust in SARs had been scrutinized across different contexts.

The distribution of papers per country (of author affiliation) are shown in Figure 6. The map in figure reveals varying levels of research activity on SARs across different countries. The United States stands out with the highest number of papers, totaling 12. Following closely is Italy, contributing significantly with 7. Countries such as the United Kingdom, China, Germany, Israel and Korea have also shown notable research engagement, each having either 2 or 3 papers on SARs. On the other hand, numerous countries, including Australia, Austria, Canada, France, Japan, Netherlands, Malaysia, New Zealand, Norway, Portugal, Singapore, and Turkey, each have one paper. It’s noteworthy that several regions, particularly in Africa and South America, appear to be underrepresented in the provided data, as they do not have any recorded papers on trust of older adults in SARs. This indicates a potential area for future research growth and collaboration to ensure a more globally inclusive perspective SARs studies. Table 7 shows the population sample sizes used in each of our selected studies. The largest population size was 1,149 (in Ejdys (2022)) while the smallest population size was 4 (in Piasek and Wieczorowska-Tobis (2018)).

Among the studies that used a service robot, the largest population size was 67 (in Huang (2022). On the other hand, among the studies that used a companion robot, the largest population size was 65 (in Mann et al. (2015)). In studies that used computer simulators pictures, or videos of robots are categorised as Other and the largest population size was 384. In our collection of 47 papers, only 7 studies did not involve any participants. For instance, the research by Zhang et al. (2022) did not use real robots and, as a result, did not include any participants in their experiments. This might be because they were focused on developing a model or method rather than testing it with people. Similarly, Camilleri et al. (2022) and Branyon and Pak (2015) created a method or model but did not deploy or test it with participants. The decision could be attributed to the early stages of development or a focus on theoretical aspects before engaging in practical testing. Additionally, in the study by Ono et al. (2015), the number of participants involved was not specified, which could be due to oversight in reporting or a deliberate omission in their research methodology. The ages ranges of the participants in each of the selected studies are shown in Table 7. 25 studies have included older adults (aged 60, or over) as their participants. The eldest person included in these studies was 95 years old (in Torta et al. (2014). The mean age in these studies ranged from 30 to 90.21 years. Six studies (Correia et al., 2016; Mo et al., 2017; Poulsen et al., 2018; Zhang et al., 2022; Camilleri et al., 2022; Ono et al., 2015) did not mention the age ranges of their population while one study (Ishak and Nathan-Roberts, 2015) did not include any participants. All the studies that used service robots included older adults as their participants. In these studies, the mean population age ranged from 73.8 years to 78.5 years. 7 out of 8 studies which used companion robots, included older adults as their participants. In these studies, the mean population age ranged from 24.31 to 90.21. In two studies (Mo et al., 2017; Correia et al., 2016), only the participants’ average age information is available while information about the age ranges of the participants is not available.

Our comprehensive analysis reveals a concentration of research efforts on SARs predominantly in the United States and Italy, with these two countries contributing significantly to the field. The map of research activity illustrates a global landscape with notable research engagement in these regions. However, it’s crucial to acknowledge the under representation of several areas, particularly in Africa and South America, pointing to potential opportunities for research growth and collaboration to ensure a more balanced and globally inclusive perspective on SARs.

Looking into participant demographics, our findings indicate a predominant focus on healthy older adults in the selected studies. This observation raises awareness of potential biases in the participant selection process and underscores the importance of expanding research to include a more diverse representation of older adults, encompassing those with varying health conditions and backgrounds. As the field of SARs continues to evolve, addressing these geographical and demographic imbalances will contribute to a more comprehensive understanding of the dynamics surrounding trust and interaction with robots, particularly among older populations.

In investigating RQ1, how trust is measured in SARs for older adults, our review exposes a lack of standardized methods for finding trust in SARs. One of the most popular methods used was questionnaires. 14 out of 47 papers, researchers used proxy examination of trust and used questionnaires like ATAQ (Eriksen and Frandsen, 2018) or and UTAUT (Ahmad, 2015). Apart from questionnaires, study specific surveys/questionnaires were also popular tools for measuring the level of trust in SARs for finding trust of older adults. Additionally, we observed a lack of consensus regarding the preferred method and its suitability for specific types of experiments. Our second finding in terms of methodology is that existing methods used for measuring trust of old people in SARs were based on subjective evaluation of trust. Subjective evaluation or self-report measures is typically based on personal assessment of the environment. Self-report measures are intrusive, however these methods are not viable in applied setting. In old age, capability to correctly assess the environment may be affected due to cognitive limitations (Erickson et al., 2022). According to Edelstein et al. (2010), the accuracy, reliability, and validity of older adult self-reports is mixed, suggesting that one should be cautious when using the self-report method and, when possible, utilize multiple methods.

Due to advancement in automation and technology, different standards are being developed (e.g., Standard for Clinical Internet of Things (IoT) Data and Device Interoperability with TIPPSS–Trust, Identity, Privacy, Protection, Safety, Security (IEEE, 2019)) have been developed to ensure consistency and interoperability across solutions and technologies. Trust in SARs needs to have some standard framework, as standard framework provide many benefits like consistency in methods, technology, terminology and work-processes. Another important consideration is the use of objective evaluation along with subjective evaluation. Due to limitations of subjective evaluation methods of trust in old age, it is important to also use objective evaluation methods for measuring trust in SARs. Given the identified gaps in standardized trust measurement methods and the limitations of subjective evaluations alone, we propose a conceptual framework for measuring trust in SARs. This framework integrates both subjective self-reports (validated questionnaires and study-specific surveys) and objective evaluation methods (behavioral and physiological measures) to provide a more reliable and holistic assessment of trust.

From our analysis of the RQ2 and RQ3, where we examined 47 selected research papers, we found that approximately half (23 out of 47) utilized robots in their studies. Notably, our findings suggest a predominant use of robots in controlled environments, such as labs. This controlled setting, while providing a scenario conducive to research, may influence the level of trust observed in older adults, as they might feel more at ease with expert assistance in a lab environment, potentially impacting the generalizability to home environments where trust dynamics may differ. Trust is crucial for sustainable interaction with assistive technology, especially in sensitive contexts like homes and intimate spaces (Schwaninger, 2020). Hence, it is expected that the level of trust in SARs shown by the older adults in a lab environment may not be a true reflection of their level of trust in a home environment. Similar findings were reported in Whelan et al. (2018). It needs to be investigated further to know if the level of trust increases or decreases with time and if living in homes have a positive or a negative impact on the level of trust in SARs. Similar observations were made by (Bajones et al., 2019) who conducted field trials with a mobile service robot in a private home environment and found that trials should be moved to homes in order to better understand real world challenges. Bemelmans et al. also indicated that further investigation is required to evaluate the effects of SARs within real elderly care settings (Bemelmans et al., 2012). In examining the factors influencing trust in SARs in RQ3, the studies analyzed were categorized into three primary groups: robot-related factors, human-related factors, and environmental factors. Robot-related factors, particularly those related to performance, received substantial attention, with a focus on reliability, behavior, and the handling of failures by SARs (Marin and Lee, 2013; Mann et al., 2015; Correia et al., 2016; Fakhrhosseini et al., 2020; Branyon and Pak, 2015; Ejdys, 2022; Giorgi et al., 2023; Zafrani et al., 2023; Piasek and Wieczorowska-Tobis, 2018; Poulin and Haase, 2015; Do et al., 2021; Lee et al., 2017; Kumar et al., 2022; Wonseok et al., 2021; Huang, 2022; Fracasso et al., 2022). Similar findings were reported by Brule et al. (2014) that performance based factors have a large influence in perceived trust in HRI. Another significant subset of studies delved into the impact of robot attributes, such as anthropomorphism and physical appearance, highlighting the importance of considering emotional and psychological aspects in designing SARs for HRI, especially in the context of older adults. While some studies adopted a comprehensive perspective by considering both performance and attributes in evaluating trust, there was no consensus on the dominant factor. Human-related factors, encompassing user characteristics and traits, were explored in a separate cluster of studies, emphasizing the role of individual and psychological aspects in trust formation. Additionally, 13 studies demonstrated the influence of environmental factors, particularly those related to team collaboration, suggesting that trust is shaped not only by the robot or the individual user but also by the broader context in which SARs are utilized. This broad understanding is essential for the effective integration of SARs into the lives of older adults, fostering trust and acceptance (Langer et al., 2019). While a subset of studies recognized the interplay between human, environmental, and robot factors, acknowledging the complexity of trust formation, there was no consensus on which factor exerted greater influence. Moreover, certain factors, such as level of automation, false alarm, predictability, dependability, operator overload, demographics, self-confidence, attentional capacity, group membership, multi-tasking requirements, and physical environment were notably under explored. This highlights gaps in the current research landscape, suggesting avenues for future exploration to comprehensively understand the broad nature of trust in SARs.

From our analysis of RQ4, we examined demographics according to author affiliation in which study on SARs was conducted. The distribution of studies across these countries reveals a varying level of research activity in this field. Notably, the majority of studies have been conducted in countries with strong research and development ecosystems, including the United States and Italy, which have 8 and 6 studies, respectively. Furthermore, several other nations in Europe and Asia have contributed to the research landscape, each with one to two studies. It is essential to acknowledge the varying levels of research activity in different regions, as this may reflect disparities in technological exposure, cultural contexts, and research priorities when exploring trust within the area of SARs and older adults as different cultures have different levels of trust and effective interaction with robots. The study highlighted in Papadopoulos et al. (2018) emphasized that older adults’ acceptance of healthcare robots is shaped by individual factors such as cultural background and suggested that culturally competent assistive robots should be employed ethically, serving as valuable tools for human caregivers and that these robots are designed to complement rather than replace human caregivers, in accordance with the principles outlined in the BSI 2016 guidelines (BSI, 2024). Different demographics exhibit varying levels of trust and effectiveness in engaging with robotic technology (Seaborn et al., 2023). Recognizing and addressing these regional variations in technological familiarity is essential for achieving a more comprehensive understanding of how trust in SARs is influenced by human factors when interacting with older adults. Although our analysis did not compare trust levels across countries, prior literature indicates that such differences, such as the higher trust in robots often reported in Japan compared to many European contexts, may reflect deeper cultural orientations toward technology, historical exposure to robotics, and differing institutional frameworks (Ikari et al., 2023; Castelo and Sarvary, 2022). Future research incorporating cross-cultural designs and standardized trust measures could provide valuable insights into how these factors shape trust in SARs among older adults. We also found that participants in the selected studies were healthy older adults with no age-related disabilities. To foster a more comprehensive understanding of trust in SARs, there is a clear need for increased involvement of older adults with age-related disabilities in future studies. As age-related disabilities can impact an individual’s physical, cognitive, and social capabilities, which in turn may influence their confidence and trust in using technology (Birkhäuer et al., 2017). By intentionally including individuals facing age-related challenges, researchers can explore and develop methods specifically tailored to address the unique trust dynamics that may arise in this demographic. More involvement of old adults with age-related disabilities is required to investigate methods for finding trust in SARs.

There are some limitations of our work we would like to acknowledge. For example, we only included papers written in the English language and limited our research for 11 years, i.e., from 2013 until 2024, this was because it represented the largest proportion of results; we excluded articles that were not published in peer-reviewed journals or conferences, and thus we may have missed out on some research and commercial solutions as a result. Additionally, while all included studies underwent independent screening by two authors to assess their methodological appropriateness and relevance, we did not apply a formalised risk-of-bias assessment tool such as AXIS (Downes et al., 2016), MIXED METHODS APPRAISAL TOOL (MMAT) (Hong et al., 2018), or Risk of Bias (RoB) (Whiting et al., 2016). The use of such a tool could provide a quantitative appraisal of study quality and comparability, and we recommend its incorporation in future reviews to further strengthen methodological rigour. Building upon the current findings, our future work aims to undertake longitudinal studies with SARs involving older adults within their home environments to explore whether initial trust in SARs is sustained or evolves, while also providing a detailed analysis of participant demographics, including age ranges, gender balance, cultural contexts, and the inclusion of older adults with disabilities, to better understand how these factors influence interactions with SARs and influence trust. Additionally, future research could explore advanced methods for measuring trust as psychophysiological assessment, such as data-driven fuzzy logic approaches, in the broader context of HRI research.