The Impact of Real-World Interaction on the Perception of a Humanoid Social Robot in Care for Institutionalised Older Adults

Abstract

(1) Background: For the residents of long-term care (LTC) units, a humanoid social robot (HSR) may be not only a caregiver but also a companion. The aim of our study was to analyse changes in its perception following a real-world interaction; (2) Methods: One hundred LTC residents were assessed twice with the Godspeed Questionnaire Series (GQS): after viewing a photograph of HSR TIAGo only and after interacting with it in a practical manner. The perception parameters were evaluated on a scale of 1–5 in five series: I-Anthropomorphism, II-Animation, III-Likeability, IV-Perceived intelligence, and V-Perceived safety. (3) Results: In the post-interaction assessment of the TIAGo robot, no lower scores were observed relative to the first (photo-based) scoring. Positive changes were observed in III (p < 0.001), I (p < 0.01), II (p < 0.05), and IV (p < 0.05). In multivariable analysis, high levels of loneliness constituted a correlate for improvement after interaction in I (p < 0.05); computer skills—in III (p < 0.01), and GDS score corresponding to depression—in IV (p < 0.01). (4) Conclusions: Our study reveals a positive change in older people’s perception of an HSR after interacting with it. Interaction is thus an indispensable element in the development process. Developers and implementers should pay particular attention to the robot’s smart functions, movements, and responsiveness.

1. Introduction

For various reasons, many people face the necessity of relocating to a care institution at some point [1]. In most instances, this procedure cannot be described as easy to accept. Upon moving in, older individuals begin the process of establishing their homes within the new environment. Four factors associated with this transition have been identified as particularly crucial: continuity, the preservation of personal identity, a sense of belonging, and the ability to remain active and engaged in work [2]. Meeting these needs can prove complex, as the institutional structure of long-term care (LTC) settings has been shown to influence the functioning of older individuals and their decision-making processes [3]. The restriction of autonomy perpetuates the stereotype of LTC residents as passive [4]. Their perceived lack of control often leads to a sense of powerlessness [5], which is especially evident among those with dementia [6], many of whom report experiencing a poor quality of life [7]. These challenges underscore the inherent issues within LTC units that require attention. Furthermore, the growing populations of older adults and the simultaneous decline in the number of younger individuals in contemporary societies—not limited to developed countries—contribute to an expanding care gap (a growing insufficiency in the availability of both formal and informal caregivers). LTC systems worldwide, especially in the aftermath of the COVID-19 pandemic, are struggling with inadequate staffing levels [8]. Consequently, it is imperative to explore and implement novel solutions that, among other goals, can contribute to optimising the time budgets of LTC employees. While staff members regard non-pharmacological interventions as vital for enhancing resident care, there has been relatively little discourse on the implementation of innovative interventions or technologies, highlighting the need to raise awareness in this area [9]. This calls for novel approaches to efficient and ethical LTC. The application of new technologies [10], including social robots [11], in care for older adults presents a promising solution, particularly when these technologies are perceived as reasonable, appropriate, and easy to use by prospective users [12].

The barriers, facilitators, and intentions associated with the use of humanoid social robots (HSRs) in care have been extensively studied, including by our team [13,14,15,16,17,18,19] (e.g., consistently worse attitudes of formal caregivers than those of older adults). A systematic review conducted by Papadopoulos et al. [20] found that many of the enablers (enjoyment, personalisation, and familiarisation), as well as barriers (limited capabilities and negative preconceptions), are related to the perception of the robot. Among the conclusions drawn, key insights touch on the acceptance of proposed solutions. We have demonstrated that assessing potential users’ needs and requirements and providing opportunities for tangible interaction with the robot is essential, as this interaction influences factors related to acceptance [21]. We have also highlighted that the perception of the robot is of central importance when considering its various potential roles and functions [22]. Therefore, the aim of this study was to analyse changes in the perception of an HSR resulting from the opportunity to experience its operation, investigate the correlates of these changes, and discuss their implications for the implementation of robotic solutions in LTC. The first research question was thus: does interaction with an HSR influence the perception of its usefulness, acceptability, and implementation potential in LTC facilities? The second research question was: what are the correlates of change of perception of the robot? Based on these questions, we formulated six hypotheses.

In contrast to the Pepper robot, which features articulated arms and a skirt-shaped base [23], our TIAGo robot did not possess a human-like body or extremities. Consequently, we hypothesise that this absence of anthropomorphic features will negatively influence anthropomorphism-related parameters (H1). Animacy is typically characterised by an agent’s capacity to act autonomously, as opposed to being passively controlled [24]. This distinction may account for the lower animacy ratings observed for static robots such as Furhat in comparison to mobile robots like Pepper or NAO [25]. Therefore, as the TIAGo robot is capable of autonomous activity, we expect that interaction with TIAGo will result in an increased animacy score (H2). The robot’s appearance is recognised as an important factor of its likeability [26]; however, older adults may prefer its functionality over appearance [27]. Accordingly, we anticipate that direct interaction with the TIAGo robot will similarly enhance likeability ratings among older adults in our study, particularly due to the hands-on availability of its various functions (H3). Perceived intelligence in robots is contingent upon their demonstrated competence. Given the technological constraints of the TIAGo platform, we postulate that perceived intelligence will remain unchanged following interaction with the robot (H4). Finally, perceived safety is influenced by the robot’s physical size; larger robots may be viewed as less safe [25]. Therefore, we hypothesise that the substantial size of the TIAGo robot will not positively affect perceived safety following interaction (H5). Furthermore, as we showed previously that attitudes toward the robot were not correlated with sociodemographic factors, self-assessment scores or selected items of the Comprehensive Geriatric Assessment [28], we expect that there will be no universal correlates of the perception of the robot (H6).

2. Materials and Methods

2.1. Research Methodology

The study model allowed for a comparison of perceptions of the HSR TIAGo (PAL Robotics, Spain) after viewing a photograph and following direct interaction. Ethical approval for this study was granted by the Bioethics Committee of Poznan University of Medical Sciences under Protocol No. 711/18.

2.2. Study Group

This study involved older individuals (aged 60 years and above) from six LTC homes in the Greater Poland region. Initially, informational sessions were held to explain the aims of the project. All residents who expressed interest were screened for functional capacity, which included assessments of basic activities of daily living (ADL) using the Barthel Index (BI) [29], cognitive functions using the Mini-Mental State Examination (MMSE) [30], and mood using the Geriatric Depression Scale (GDS) [31]. Additionally, participants completed a questionnaire containing demographic data (age, gender assigned at birth, education) and declarative self-assessment items rated on a 1–5 scale. These items included the following:

• Computer skills (1—no skills, 5—excellent skills);
• Ease of use of technology (1—extreme difficulty/lack of ability, 5—no issues with new technologies);
• Health status (1—the worst possible, 5—the best health condition);
• Physical fitness (1—the worst possible, 5—the highest possible independence);
• Loneliness (1—no loneliness, 5—extreme loneliness).

2.3. Ethical Considerations

For ethical reasons, all interested individuals participated in the subsequent stages of the meeting; however, the analysis of results included only those participants who scored at least 15 points on the MMSE, which is considered to indicate the ability to engage in logical communication, understand questions, and provide appropriate responses [32,33]. The subjects were also verbally asked whether they would like to express any ethical reservations about this study.

2.4. The TIAGo Robot

We used a customised version of the TIAGo robot that incorporated several sensors (an RGB-D camera, a thermal camera, RFID equipment, a laser scanner, various environment sensors, and radar distance sensors), a microphone, a loudspeaker, and a tablet for communication with the user, mounted on the robot’s chest section. The robot was wirelessly coupled with a remote computer and connected to the Internet. It was able to navigate in a semi-autonomous manner or follow the user based on the thermal sensor. The options available during the interaction included reminders, safety measures, dietary recommendations, physical exercises, cognitive games, video connectivity (e.g., for contacting friends), and the provision of news and weather. Presentation of environmental values based on local and networked sensors (temperature, humidity, air pressure, air quality, etc.) was also accessible.

2.5. Procedure

After participants viewed a photograph of the TIAGo, they expressed their opinions on its perception for the first time. They were then given the opportunity to interact with the robot in person for as long as they deemed necessary (Figure 1). A detailed description of the presentation can be found in a previous publication [20]. The interaction lasted from approximately 90 to 150 min (the duration was determined by the number of participating subjects and their interest in operating the robot). The final stage of the meeting involved a re-assessment of the participants’ perceptions of the robot, taking into account their interaction experience.

The Godspeed Questionnaire Series (GQS) was employed to assess robot perceptions. The Polish-language version of the questionnaire was validated in a separate study [22]. The GQS consists of five series of terms related to the robot: 1—Anthropomorphism, 2—Animacy, 3—Likeability, 4—Perceived intelligence, and 5—Perceived safety. Each item is rated using five-point semantic differentials with opposing terms. Respondents are asked to rate each category on a scale from 1–5 (e.g., Dead (1) vs. Alive (5) from the Animacy series).

In this study, we also analysed factors that could potentially influence perception, including the following:

• Demographic data (age, gender, education);
• Self-assessment (subjective) items;
• Objective data on functional capacity (cognitive functions, presence of depressive symptoms, and ADL performance).

2.6. Statistical Analysis

Statistical analysis was conducted using STATISTICA 13.3 software (StatSoft, Poland). The results were presented as means, standard deviations, and medians due to the non-normal distribution of some variables, verified using the Shapiro–Wilk test.

The perceptions of the robot, as assessed after viewing the photograph and after interaction (paired data), were compared using the Wilcoxon test. Differences for individual parameters were analysed using the χ² test, considering the following:

• Demographic data: age (60–79 years vs. 80+), gender (F vs. M), and education (less than secondary vs. at least secondary);
• Self-assessment items: scores below the central value vs. scores equal to or greater than the central value (i.e., 1–2 vs. 3–5, except for loneliness, where 4–5 vs. 1–3 was used, as 5 represents the most negative score);
• Objective data on functional capacity:

  ○

  Cognitive function assessment via the MMSE scale: participants with at least 24 points (indicating no more than mild cognitive impairment) vs. those with a lower score;

  ○

  Depression symptom assessment via the GDS scale: participants scoring up to 5 points vs. those with a higher score;

  ○

  ADL performance on the Barthel scale: participants with scores below 80 vs. those scoring at least 80.

A multiple regression model (logistic regression) was employed to assess simultaneous interdependence between variables, yielding the odds ratio (OR) and the confidence interval (CI) with a 95% confidence limit. All variables found to be significant in the univariable analysis were included in the multiple linear regression analysis. A p-value of <0.05 was considered statistically significant, while p-values between 0.05 and 0.10 were interpreted as indicative of non-significant trends.

3. Results

This study involved 100 older adults (aged 60 years or more), with a mean age of 76.6 ± 8.9 years (median 75.5 years; the oldest participant was 94 years old). Of these, 47 respondents were aged 75 years or older. There were 53 women among the participants. The largest group (41 participants) had completed secondary education, while only nine held a university degree.

The mean declared computer skills score was 1.6 ± 1.3 (median 1.0), with 79 participants reporting no skills and only eight indicating excellent skills. The mean score for ease of use of technology was considerably higher at 3.0 ± 1.6 (median 3.0), as only 31 participants reported a lack of abilities, while 25 indicated no problems with technology.

The mean value of the respondents’ self-assessment of health was 3.0 ± 1.1 (median 3.0), with 13 participants giving the lowest and eight people—the highest rating. The mean score for physical fitness was 3.2 ± 1.2 (median 3.0), with the lowest and highest ratings provided by nine and 15 participants, respectively. Regarding declared loneliness, the mean score was 2.6 ± 1.6 (median 2.5), with 18 subjects indicating extreme loneliness and 41 reporting no loneliness.

The mean BI score was 77.0 ± 23.4 (median 85.0). No participant scored between 0 and 20 points (indicating total dependency), and 21 subjects achieved the maximum score of 100 points, indicating no limitations in independence. The mean score on the MMSE was 23.3 ± 4.1 (median 23.5). Fifty participants scored below 24, suggesting advanced cognitive impairment, with 21 scoring between 15 and 19 points. Thirty-one subjects scored within the normal range (27–30 points). The mean GDS score was 3.9 ± 2.9 (median 3.0), with 75 participants scoring within the range indicating no depressive symptoms (1–5 points). The remaining participants had higher scores, with three individuals presenting symptoms of severe depression (GDS score of 11 points or above).

3.1. The Impact of Interaction with the TIAGo Robot on GQS Results

No perception scores were lower on any parameter or GQS series in the post-interaction assessments compared to those based on the robot’s photograph. Mean perception scores for all series were higher after the interaction, except for the Perceived Safety series (mean ± SD; median: 3.7 ± 0.9; 3.7 vs. 3.9 ± 0.9; 4.0)—see Figure 2.

The most substantial changes were observed in the Likeability series (4.1 ± 0.9; 4.2 vs. 4.4 ± 0.8; 4.8; p < 0.001). These changes were related to changes in ratings for all components of the parameters in this series, with the exception of the Dislike–Like item, for which an insignificant trend was observed (3.9 ± 1.2; 4 vs. 4.2 ± 1.3; 5; p = 0.058). Specifically, significant increases were noted in the following parameters: Unfriendly–Friendly (4.1 ± 1.2; 4.5 vs. 4.5 ± 0.9; 5.0; p < 0.01), Unkind–Kind (4.1 ± 1.0; 4.6 vs. 4.5 ± 0.8; 5.0; p < 0.01), Unpleasant–Pleasant (4.1 ± 1.1; 5.0 vs. 4.5 ± 0.9; 5.0; p < 0.01), and Awful–Nice (4.1 ± 1.0; 4.0 vs. 4.6 ± 0.8; 5.0; p < 0.001).

In the Anthropomorphism series (2.8 ± 1.0; 2.8 vs. 3.0 ± 1.0; 3.0; p < 0.01), the higher perception was primarily linked to a change in the perception of the robot’s movement parameter (Moving rigidly vs. Moving elegantly: 3.4 ± 1.3; 3.0 vs. 4.0 ± 1.4; 5.0; p < 0.0001) as no significant changes were noted for other parameters (Machinelike–Humanlike: 2.0 ± 1.4; 1.0 vs. 2.2 ± 1.5; 1.0, Unconscious–Conscious: 3.0 ± 1.6; 3.0 vs. 3.2 ± 1.5; 3.0, Artificial–Lifelike: 2.5 ± 1.6; 3.0 vs. 2.5 ± 1.5; 3.0). For the Fake–Natural item, an insignificant trend was noted (3.0 ± 1.5; 3.0 vs. 3.3 ± 1.0; 3.0; p = 0.0896).

Significantly higher perception scores were also recorded for the Animacy series (3.0 ± 0.9; 2.8 vs. 3.3 ± 1.0; 3.3; p < 0.01), primarily driven by a change in the Dead–Alive item (2.7 ± 1.6; 3 vs. 3.3 ± 1.6; 4; p < 0.001), and the Inert–Interactive parameter (3.3 ± 1.6; 3 vs. 3.6 ± 1.5; 4; p < 0.05). No changes were observed for the remaining parameters.

The mean score of the Perceived Intelligence series was also higher (4.2 ± 0.9; 4.6 vs. 4.4 ± 0.7; 4.8; p < 0.05), primarily due to an increase in the Unintelligent–Intelligent item (4.3 ± 1.0; 5 vs. 4.6 ± 0.7; 5; p < 0.01). For other items, no significant changes were noted.

3.2. Correlates of Change in Perception of the TIAGo Robot After Interacting with It

The frequency of higher perception of the TIAGo robot following interaction, compared to perception based on the photograph, with a distinction between the analysed parameters, is presented in

Table 1. The characteristics of study subjects whose perception improved after interaction with HSR TIAGo.

Parameter		Anthropo Morphism	Animacy	Likeability	Perceived Intelligence	Perceived Safety
Age (years)	60–74 (n = 43)	28 (65.1%)	25 (58.1%)	28 (65.1%)	22 (51.2%)	21 (48.8%)
	≥75 (n = 57)	30 (52.6%)	34 (59.6%)	24 (42.1%) p = 0.0272	20 (35.1%)	22 (38.6%)
Gender	Females (n = 53)	31 (58.5%)	29 (54.7%)	21 (39.6%)	17 (32.0%)	27 (50.9%)
	Males (n = 47)	27 (57.4%)	30 (63.8%)	31 (65.9%) p = 0.0099	25 (53.2%) p = 0.0428	14 (29.8%) p = 0.0419
Education	Below secondary (n = 44)	28 (63.6%)	29 (65.9%)	24 (54,5%)	22 (50.0%)	17 (38.6%)
	At least secondary (n = 56)	30 (53.6%)	30 (53.6%)	28 (50.0%)	20 (35.7%)	26 (46.4%)
Ease of use of technology	1–2 (n = 40)	27 (67.5%)	24 (60.0%)	22 (55.5%)	15 (37.5%)	18 (45.0%)
	3–5 (n = 60)	31 (51.7%)	35 (58.3%)	30 (50.0%)	27 (45.0%)	25 (41.7%)
Computer skills	1–2 (n = 81)	46 (56.8%)	46 (56.8%)	36 (44.4%)	30 (37.0%)	33 (40.7%)
	3–5 (n = 19)	12 (63.2%)	13 (68.4%)	16 (84.2%) p = 0.0020	12 (63.2%) p = 0.0435	10 (52.6%)
BI	≤80 (n = 48)	28 (58.3%)	32 (66.7%)	24 (50.0%)	20 (41.7%)	24 (50.0%)
	>80 (n = 52)	30 (57.7%)	27 (51.9%)	28 (53.8%)	22 (42.3%)	19 (36.5%)
MMSE	≤23 (n = 50)	33 (66.0%)	32 (64.0%)	25 (50.0%)	20 (40.0%)	22 (44.0%)
	>23 (n = 50)	25 (50.0%)	27 (54.0%)	27 (54.0%)	22 (44.0%)	21 (42.0%)
GDS	<6 (n = 75)	40 (53.3%)	42 (56.0%)	39 (52.0%)	31 (41.3%)	25 (33.3%)
	≥6 (n = 25)	18 (72.0%)	17 (68.0%)	13 (52.0%)	11 (44.0%)	18 (72.0%) p = 0.0010
Physical fitness	1–2 (n = 26)	16 (61.5%)	17 (65.4%)	11 (42.3%)	9 (34.6%)	10 (38.5%)
	3–5 (n = 74)	26 (35.1%) p = 0.0226	42 (56.8%)	41 (55.4%)	33 (44.6%)	46 (62.2%) p = 0.0418
Health status	1–2 (n = 18)	18 (64.3%)	18 (64.3%)	14 (50.0%)	15 (53.6%)	17 (60.7%)
	3–5 (n = 72)	40 (55.6%)	41 (56.9%)	38 (52.8%)	27 (37.5%)	26 (36.1%) p = 0.0419
Loneliness *	1–3 (n = 68)	33 (48.5%)	40 (58.8%)	37 (54.4%)	29 (42.6%)	25 (36.8%)
	4–5 (n = 30)	23 (76.7%) p = 0.0142	17 (56.7%)	13 (43.3%)	12 (40.0%)	16 (53.3%) p = 0.0817

* Two participants provided no values.

.

In the multivariable analysis, which assessed all variables being significant in the univariable analysis, we found that the odds ratios for a better perception of the robot after interaction were higher for the Anthropomorphism series: individuals who declared high levels of loneliness had an odds ratio of 3.34 (95% confidence interval: 1.26–8.87; p < 0.05). For the Likeability series, subjects with declared computer literacy had an OR of 6.74 (95% CI: 1.74–26.09; p < 0.01). For the Perceived safety series, participants with a GDS score indicative of depression had an OR of 4.75 (95% CI: 1.63–13.87; p < 0.01).

No correlates were identified for the remaining series. The results of the multi-parameter analysis are presented in detail in

Table 2. Multivariable analysis of correlates of improvement after interaction with HSR TIAGo.

Series I: Anthropomorphism
		OR	95% CI	p
Physical fitness	1–2 vs. 3–5	1.665	0.618–4.487	0.314
Loneliness	4–5 vs. 1–3	3.340	1.258–8.872	0.016
Series III: Likeability
		OR	95% CI	p
Age (years)	60–74 vs. at least 75	2.126	0.860–5.257	0.102
Gender	Males vs. females	2.332	0.953–5.706	0.064
Computer skills	3–5 vs. 1–2	6.735	1.738–26.093	0.006
Series IV: Perceived intelligence
		OR	95% CI	p
Gender	Males vs. females	2.283	0.997–5.228	0.051
Computer skills	3–5 vs. 1–2	2.719	0.945–7.821	0.064
Series V: Perceived safety
		OR	95% CI	p
Gender	Males vs. females	0.421	0.168–1.055	0.065
Physical fitness	3–5 vs. 1–2	0.763	0.238–2.441	0.648
Health status	3–5 vs. 1–2	0.511	0.164–1.597	0.248
Loneliness	4–5 vs. 1–3	1.381	0.524–3.642	0.514
GDS	6 and above vs. less than 6	4.753	1.628–13.871	0.004

.

4. Discussion

Based on the literature and our previous studies, attitudes toward the use of HSRs in care settings are deeply intertwined with various aspects of their perception [22,28,34]; we thus pose the question of whether a real-world interaction influences the perception of the robot. Understanding how LTC residents perceive a robot is essential for the successful implementation of robotic solutions [35], which could ultimately reduce staff workload. Staffing in LTC institutions is a critical issue, affecting not only the overall quality of life for residents but also their mental health, as demonstrated by Chappell et al. [36].

Our study is one of the first to analyse the impact of institutionalised subjects’ real-world interaction with an HSR on their perception of the robot. While numerous studies on robot perception have been published, few incorporate actual interaction with the technology, particularly in the context of its use in care. D’Onofrio et al. assessed post-exposure perceptions of a robot’s usefulness and found positive feedback, regardless of the participants’ familiarity with new technologies [37], similar to our findings. Beer et al. applied a 2.5-h exposure to a robot and observed an improvement in perceived ease of use [38]—a parameter generally related to perception (this aspect of technology acceptance has been recently strengthened in an analysis using a Large Language Model [39]). Moreover, Ke et al. noted an improvement in LTC residents with dementia after 32 weeks of interaction with a robot [40], which complements the results we obtained. Our study assessed the change in perception of the HSR TIAGo between viewing it in a photograph and interacting with it in full contact. We previously demonstrated that a tangible interaction with an HSR changes attitudes towards the robot, with perception being a key factor [21]. This current study also examined potential correlates of the robot’s perception, including demographic data, self-assessment information, and selected items of the Comprehensive Geriatric Assessment.

Our results show a generally positive change in older people’s perceptions of an HSR. The greatest difference, congruently with our H3, was observed in the Likeability series, with almost all items being rated higher after the interaction. It is consistent with the study by Robinson et al. using the Pepper HSR [41]. This observation is significant because the Likeability series captures elements related to the human–robot relationship, such as friendliness and kindness. Reciprocal responses are known to play an important role in shaping these relationships [42].

The Anthropomorphism series also showed improvement (matching our H1), primarily due to the movement parameter (e.g., Moving rigidly–Moving elegantly). Notably, Schulz et al. analysed the characteristics of the Fetch robot and found that participants preferred to engage in interaction with the robot rather than watch it move [43]. The robot’s movement quality seems important for its perception; however, the uncanny valley effect (UVE) should also be considered in this context. Tu et al. demonstrated differences in UVE perception between younger and middle-aged adults and older individuals, suggesting that the robot’s motor functions should be designed with the target group in mind [44].

Scores for the Animacy series improved significantly, primarily due to the Dead–Alive and Inert–Interactive items, consistent with our H2. This suggests that future robot users may place a high value on how quickly the robot reacts and prefer more dynamic behaviour.

The Perceived intelligence series was also rated higher post-interaction (confirming our H4), with the Unintelligent–Intelligent item driving this change. Hence, prospective users seem to expect robots to exhibit intelligence, and interaction with the TIAGo reinforced the perception of an HSR as such. Similar results were obtained by Sturgeon et al. using the relatively small NAO HSR (presented in a video sequence), though no change was observed in our study for the Perceived safety series, reflecting our H5. This finding warrants further investigation because comfort, experience, familiarity, predictability, sense of control, transparency, and trust are key factors influencing perceived safety in human–robot interaction [45].

In the multivariable analysis, no common correlates for change in all series were found, as suggested in our H6. Loneliness was observed to correlate with changes in Anthropomorphism scores. Single-assessment studies by Chen et al. [46], Li and Sung [47], and Dang and Liu, particularly for Western cultures [48], have also noted a relationship between loneliness and anthropomorphism, which suggests that special attention should be paid to human-like features when designing robotic interventions for people experiencing loneliness. This is a critical consideration, given that older adults transitioning to LTC facilities are often lonely. For changes in the Likeability series, computer skills were found to be a correlate, a finding not previously demonstrated. Hence, computer literacy appears to play a positive role in moderating the acceptance of HSRs in care. Gender yielded an insignificant trend here. The literature data related to the impact of gender on the acceptance of robots are inconsistent: from no [49] to substantial influence [50]. Our univariable analysis indicated that males scored Likeability and Perceived intelligence better, whereas females scored Perceived safety better. For Perceived safety, a GDS score indicating depression was a correlate (a trend also observed in a pilot study by Harrington et al. [51]). This finding is important because safety features are a priority for prospective HSR users in care [52]. GDS, commonly used as a screening tool for mood and depression, could serve as an indicator of the significance of safety functions in robot design. In univariate analysis, physical fitness and loneliness were also significant factors influencing changes in Perceived safety. While our participants consistently interacted in close proximity to the machine, Rugabotti et al. observed that spatial aspects may also affect the perceived safety of the robot [53]. All this suggests that the perception of safety in an HSR is a multifaceted phenomenon, as highlighted in previous studies [53,54], and should be carefully examined during the pre-implementation phase of robotic interventions. Indicated factors may shape user expectations and acceptance, ultimately affecting the success of robotic integration not only in long-term care settings. From a clinical perspective, the findings underscore the importance of tailoring robotic solutions to their older users. Integrating robots that are perceived as responsive, friendly, and intuitive may contribute to improved mood, greater social stimulation, and potentially reduce caregiver burden.

The results of this study support the hypothesis that interaction with the robot led to a positive change in its perception. Participants demonstrated significantly improved evaluations across multiple dimensions, particularly in terms of likeability. These observations indicate that real-world interaction plays a critical role in shaping user attitudes towards robotic solutions, reinforcing the value of experiential exposure in promoting acceptance and trust in assistive technologies.

Limitations

The cross-sectional nature of this study means that the results suggest important relationships but cannot establish causality. Furthermore, the first interaction with a robot may be influenced by the novelty effect; however, participants had sufficient time to operate the robot as they saw fit. Despite these limitations, the study group can be considered relatively large for this type of analysis.

5. Conclusions

Real-world interaction with an HSR results in both qualitative and quantitative changes in its perception. To obtain reliable results, studies on HSR perception and acceptance should incorporate actual interaction with the robot. Developers and staff responsible for implementing these systems should pay particular attention to the robot’s smart functions, movements, and responsiveness. It is essential to thoroughly understand the target group’s characteristics before designing the robot, particularly in terms of loneliness, mood (including symptoms of depression), and computer literacy. Our findings may also benefit the LTC management and staff when preparing interventions using HSRs. Their deployment should be accompanied by staff training and user education to maximise therapeutic potential and ensure effective interaction between users and robotic systems.