Companion Robots Supporting the Emotional Needs of the Elderly: Research Trends and Future Directions

Abstract

The accelerating global population aging has brought increasing attention to the loneliness and emotional needs experienced by older adults due to shrinking social networks and the loss of relatives and friends, which significantly impair their quality of life and psychological well-being. In this context, companion robots powered by artificial intelligence are increasingly regarded as a scalable and sustainable form of emotional intervention that can address older people’s affective and social requirements. This study systematically reviews research trends in this field, analyzing the structure of emotional needs among older users and their acceptance mechanisms toward robot functionalities. First, a keyword co-occurrence analysis was conducted using VOSviewer on relevant literature published between 2000 and 2025 from the Web of Science database, revealing focal research topics and emerging trends. Subsequently, questionnaire surveys and in-depth interviews were carried out to identify emotional needs and functional preferences among elderly users. Findings indicate that the field is characterized by increasing interdisciplinary integration, with affective computing and naturalistic interaction becoming central concerns. Empirical results reveal significant differences in need structures across age groups: the oldest-old prioritize safety monitoring and daily assistance, whereas the young-old emphasize social interaction and developmental activities. Regarding emotional interaction, older adults generally prefer natural and non-intrusive expressive styles and exhibit reserved attitudes toward highly anthropomorphic designs. Key factors influencing acceptance include practicality, ease of use, privacy protection, and emotional warmth. The study concludes that effective companion robot design should be grounded in a nuanced understanding of the heterogeneous needs of the aging population, integrating functionality, interaction, and emotional value. Future development should emphasize adaptive and customizable capabilities, adopt natural yet restrained interaction strategies, and strengthen real-world cross-cultural and long-term evaluations.

1. Introduction

The global demographic landscape is undergoing a profound and irreversible shift toward population aging. According to projections by the United Nations Department of Economic and Social Affairs, the population aged 65 years or older worldwide will double by 2050, from the current 761 million to approximately 1.6 billion. Concurrently, the share of this age group in the global population is projected to rise from 9.8% in 2022 to 16.3% in 2050. This significant demographic evolution is not only fundamentally reshaping societal structures but also posing complex social and public health challenges [1]. Among these, loneliness among older adults, driven by factors such as shrinking social networks, the loss of loved ones, and geographical isolation, has emerged as a critical and increasingly urgent issue, underscoring a growing need for emotional support. Loneliness is not only closely associated with psychological distress at the subjective level but is also linked to various adverse health outcomes, significantly impairing older adults’ quality of life and overall well-being [2]. Substantial research indicates that loneliness not only acts as a key trigger for various forms of psychological distress, such as depression and anxiety, but is also closely linked to serious physical health issues, including cognitive decline, an increased risk of cardiovascular diseases, and even elevated all-cause mortality [3]. Notably, both chronic loneliness and depression are recognized as modifiable risk factors for Alzheimer’s disease and other dementias. The underlying mechanisms may involve chronic neuroinflammatory responses and persistently elevated stress hormone levels, thereby accelerating the process of neurodegenerative pathology [4].

The concurrent progression of population aging and the shifting spectrum of diseases is marked by a substantial increase in the proportion of older adults who live with multiple chronic conditions while retaining functional independence. As this growing constituency, which constitutes the core of the older population, navigates life with illness, their essential needs have transcended conventional care, pointing toward the necessity of a more comprehensive support system [5]. The distinct challenges they encounter include the daily cognitive load and safety risks associated with complex medication regimens; persistent psychological stress, often masked by an outwardly functional appearance, stemming from health-related anxieties and fear of emergencies; and loneliness exacerbated by disease-driven withdrawal from social engagement. This reality necessitates a fundamental shift in the paradigm of care, moving from one centered primarily on physical assistance to an integrated model that synthesizes systematic health management, safety monitoring, and psychosocial support. However, existing non-continuous family support and traditional periodic home-based services prove inadequate in addressing such persistent and dynamically evolving needs, thereby revealing a critical gap in care provision. Against this backdrop, there is a pressing need to explore scalable and sustainable emotional support infrastructures. Driven by rapid advancements in artificial intelligence and robotics, companion robots are emerging as a valuable non-pharmacological intervention. By responding to the emotional and social needs of older adults through personalized, interactive, and constant engagement, they hold promise for alleviating loneliness, slowing cognitive decline, and enhancing their sense of purpose and overall quality of life [6]. Therefore, conducting systematic research on how companion robots can effectively support the emotional needs of older adults carries significant theoretical value and far-reaching practical implications.

To systematically map the intellectual foundation and research evolution in this field, this study employs bibliometric methods, integrating the VOSviewer (v1.6.20) visualization tool, to conduct a multi-layered analysis of existing literature. First, from the perspective of “Elderly Companion Robot Design Research,” we analyze research foci and evolutionary trends in technological implementation pathways, interaction modalities, and functional configurations. Second, focusing on “Research on Emotional Design for the Elderly,” we explore how affective computing, user perception, and experience design principles can be incorporated into robotic systems to enhance emotional interactivity and social presence. Finally, returning to the fundamental issue of “Research on the Emotional Needs of the Elderly,” we identify differences and commonalities in the types, intensity, and expressive modes of emotional needs across diverse cultural and social contexts. Through this three-tier bibliometric analysis, the study aims to identify gaps and understudied areas in the current literature, thereby laying a solid foundation for subsequent theoretical development and empirical inquiry.

Building on a systematic understanding of existing academic progress, this study further employs field investigations and needs assessment to achieve the following research objectives: (1) systematically identify and quantify the need structures and core challenges of older adults across different age groups in home-based care settings, with particular attention to intra-group differences in daily activities, health monitoring, and social connectedness; (2) delve into the specific expectations and desires of older adults in different age segments regarding emotional companionship, including their preferences for functional attributes such as communication style, interaction frequency, and empathic capability; (3) assess the acceptance of and potential barriers to emotional companion robots among older adults of different ages, and systematically gather their concrete expectations regarding product appearance, functionality, and interaction patterns. This study contributes not only to bridging the gap between theoretical research and practical needs but also to providing a robust empirical foundation for the personalized design and promotion of companion robots tailored for older adults—particularly those managing chronic conditions and maintaining functional independence.

To achieve the aforementioned research objectives, this study follows a systematic research trajectory. Section 2 delineates the core concepts, including companion robots for older adults, characteristics of aging users, and human–robot interaction design, thereby establishing the theoretical foundation and analytical framework. Subsequently, Section 3 elaborates the methodological approach, detailing the data sources and processing procedures for bibliometric analysis alongside the mixed-methods design employed in the empirical investigation of older users. Building on this foundation, Section 4 presents both the domain knowledge mapping revealed through bibliometric analysis and the user needs data obtained from empirical investigations, followed by a multidimensional analysis of the findings. Finally, Section 5 synthesizes and critically discusses the research findings, deriving theoretical contributions and practical implications while reflecting on study limitations to suggest directions for future research.

2. Related Concepts

2.1. Elderly Companion Robot

Companion robots for older adults represent a key technological response to the global challenge of population aging. At their core, these robots are autonomous or semi-autonomous systems equipped with perception, cognition, decision-making, and actuation capabilities, designed to provide multidimensional and integrated support services for the elderly [7]. To systematically review companion robot design for older adults, a clear delineation of its functional scope is essential. As illustrated in Figure 1, this study proposes a five-tier functional model that evolves from fundamental to advanced capabilities. This model outlines a progressive pathway: starting with media interaction as an intelligent device, advancing to basic robotic functions like environmental perception and mobility, then incorporating eldercare-specific features such as safety monitoring and health management, progressing further to smart home-integrated residential services, and ultimately achieving advanced emotional companionship through multimodal affective computing and adaptive dialogue systems.

(1)

The first tier comprises basic device functions, including auxiliary services such as audiovisual media interaction, information query, and voice reporting [8].

(2)

The second tier entails basic robotic functions, encompassing modules such as voice interaction, environmental perception, and basic autonomous mobility, providing the robot with fundamental responsive and movement abilities [9].

(3)

The third tier consists of elderly care functions, focusing on core safety and health needs by providing limited-scope health management services such as fall detection, emergency alerts, physiological monitoring, and medication reminders [10].

(4)

The fourth tier represents smart home elderly care functions, demonstrated through coordinated operation with smart home systems to enable automated management of environmental elements, including lighting, security, and door or window status, thereby enhancing both home safety and daily convenience [11].

(5)

The highest level, the fifth tier, constitutes companion robot functions. This tier is what fundamentally distinguishes companion robots from conventional medical or assistive robots. The core technology of this functionality is built upon multimodal affective computing, long-term user modeling, and adaptive dialogue systems. In terms of emotion recognition, to address the limitations of single-modality perception, the system employs a multimodal information fusion strategy that coordinately analyzes vocal features, including tone and rhythm, visual signals encompassing facial muscle action units and body posture, and physiological data such as heart rate variability and electrodermal activity acquired with user authorization through non-contact sensors or wearable devices [12]. This integrated mechanism effectively accommodates the distinctive emotional expression characteristics of older adults, such as subtle facial expression variations due to reduced muscle tone or vocal characteristic changes resulting from laryngeal muscle aging. Through cross-validation of multi-source information, it significantly enhances the robustness and accuracy of affective state inference.

To ensure the applicability of emotion recognition algorithms for older adults, particularly those with unique expressive habits, it is essential to address the performance degradation and limited generalization capability observed when models trained on general datasets are applied to this demographic. Viable technical approaches primarily include domain adaptation and personalized online learning [13]. Domain adaptation techniques, particularly transfer learning-based methods, aim to minimize distributional discrepancies between data from the general population and elderly cohorts. Specific implementations may involve feature alignment across domains using measures such as maximum mean discrepancy, or incorporating adversarial training to enable feature extractors to learn domain-invariant representations. These enhancements improve model adaptation to characteristic expressive patterns in older adults, such as weakened activation of specific facial action units or altered fundamental frequency ranges [14]. Research demonstrates that such methods can effectively improve model adaptability even with limited annotated data.

Following initial domain adaptation, personalized online learning mechanisms can be introduced to address individual-level heterogeneity. Operating with users’ informed consent and data security protocols, this approach utilizes weakly supervised signals collected during long-term interaction, including direct feedback such as emotion response accuracy ratings and indirect behavioral cues like sustained engagement willingness and dialogue turn length, for continuous model fine-tuning [15]. This process facilitates gradual model evolution from generic performance toward individualized capability. To evaluate model generalization, cross validation strategies such as leave one user out or leave one expression pattern out can be implemented to specifically test performance with new users or novel expressive patterns. By monitoring the degree of degradation in key metrics like the macro-averaged F1 score and comparing with baseline models, generalization error can be quantitatively assessed [16]. Simultaneously, analyzing model performance on samples with high predictive uncertainty, combined with active learning strategies to select informative samples for user clarification, can progressively expand the model decision boundaries and enhance recognition robustness for heterogeneous expressions [17]. This integrated framework that combines domain knowledge with incremental individual learning provides crucial support for transitioning emotion recognition technology from laboratory research to practical applications.

Building upon this precise perception, the system dynamically constructs and continuously updates personalized profiles that reflect user personality, habits, and emotional needs through long-term user modeling. This model then drives an adaptive dialogue system employing a hybrid architecture: it utilizes a rule-based finite state machine to ensure deterministic delivery of critical information while integrating a large language model-based generative dialogue component to maintain conversational openness and response flexibility. This balanced approach achieves equilibrium between natural fluency and operational reliability. Ultimately, the system aims to deliver profound socioemotional support through high-accuracy emotion recognition with empathetic responses, personalized social interaction, family contact assistance, and adaptive entertainment companionship. The fundamental objective is to dynamically alleviate loneliness among older adults while promoting their psychological well-being and cognitive engagement [18].

To deliver these functions, companion robots for the elderly typically integrate high-precision environmental sensors, AI-based semantic understanding and affective computing modules, multimodal human–robot interaction interfaces, and contextual response strategy mechanisms. Their social value lies in offering intelligent and integrated services that effectively mitigate risks of social isolation and mental health pressures, help older adults maintain independence and cognitive function, and thereby reflect a deep integration of technology and social well-being. As such, these robots emerge as critical technological enablers in advancing the strategies of active aging and healthy aging.

To concretely elucidate the integration of the aforementioned key technologies, Figure 2 illustrates the comprehensive architecture of the affective interaction system for companion robots designed for older adults. This framework adopts a layered design, which progresses upward through the multimodal perception layer, the feature extraction and fusion layer, the emotion recognition and user modeling layer, the adaptive decision-making layer, and the multimodal output layer.

The multimodal perception layer is tasked with synchronously acquiring vocal, visual, and physiological signals from older users. Vocal signals are first preprocessed to extract key acoustic parameters, such as fundamental frequency, energy, and spectral features. Visual signals, captured as facial image sequences by cameras, are used to extract the activation intensities of key action units based on the Facial Action Coding System. Upon obtaining explicit user authorization, the physiological signal monitoring module acquires indicators like heart rate variability and electrodermal activity via non-contact sensors or wearable devices [19].

The feature extraction and fusion layer integrates multi-source information effectively. It employs an early fusion strategy based on an attention mechanism, which begins by temporally aligning asynchronous data from different modalities. A cross-modal attention network then dynamically learns the complementary relationships and importance weights across these modalities. This process yields a unified fused feature representation, forming the foundation for subsequent emotion recognition [20].

The emotion recognition and user modeling layer features a hierarchical, progressive structure. At its foundation, an emotion classifier enhances a generic model with domain adaptation techniques to mitigate distributional discrepancies between general population data and elderly cohort data, thereby improving the recognition of age-specific expressive patterns. The intermediate layer constructs and dynamically updates a long-term user profile that records emotional history, interaction preferences, and personality traits. The uppermost layer employs an online learning mechanism, which continuously refines the model using real-time user feedback to progressively evolve its performance from generic to individualized [21].

The adaptive decision-making layer utilizes a hybrid architecture that integrates rule-driven and data-driven methodologies. Its dialogue management system merges the deterministic processes of finite-state machines with the generative capabilities of large language models (LLMs). The former guarantees the reliable delivery of critical information, such as health reminders and safety alerts, whereas the latter ensures natural and fluent open-domain conversation, enriched by the incorporation of user profiles for personalization. Furthermore, the response strategy generation module dynamically adapts empathy depth, linguistic style, and interaction pace based on the user’s recognized emotional state and stored preferences [22].

The multimodal output layer converts abstract decisions into concrete affective responses. This is achieved by synthesizing expressive speech through neural techniques, animating corresponding facial expressions and body movements on virtual avatars or physical robots, and coordinating with smart home systems to adjust the ambient environment in sync with the interaction context [23].

2.2. Elderly User Characteristics

A thorough understanding of the characteristics of older user groups is a fundamental prerequisite for successful age-friendly interaction design [24]. This user profile is notably complex, manifested through the interplay of declining physiological functions, changes in cognitive processing capabilities, and dynamically evolving socioemotional needs (see Figure 3).

At the physiological level, older adults commonly experience declined sensory organ functions, typified by reduced visual acuity, a narrowed auditory range, diminished tactile sensitivity, as well as decreased motor coordination and slower response times [25]. In response to these changes, designs should aim to enhance the perceptibility of information, improve operational fault tolerance, and ensure ease of use and accessibility in physical interactions.

At the cognitive level, the process of healthy aging is accompanied by structural transformations in cognitive resources. While fluid intelligence, which relies on neural processing efficiency and manifests in domains such as information processing speed, working memory capacity, and attentional allocation, typically undergoes normative age-related decline, crystallized intelligence, grounded in knowledge accumulation and reflected in vocabulary, experiential wisdom, and contextual judgment, can generally be maintained or even further developed [26]. Specifically, although individuals may experience declines in executive functions and the efficiency of episodic memory retrieval, procedural memory and semantic memory typically remain relatively stable.

Critically, cognition and emotion exhibit a dynamic, reciprocal relationship throughout this process. On one hand, the decline in cognitive resources directly compromises the capacity for emotion regulation. For instance, diminished working memory and inhibitory control may impair an older adult’s ability to disengage from negative emotional states, thereby exacerbating feelings of loneliness or anxiety [27]. On the other hand, emotional states significantly influence cognitive performance. Positive emotions, such as joy and a sense of security, can broaden cognitive perspectives, enhance problem-solving flexibility, and increase motivation to learn. Conversely, negative emotions like anxiety and depression consume already limited cognitive resources, which can lead to a further decline in cognitive performance, establishing a self-reinforcing negative cycle [28].

Consequently, cognitive-level design should aim not merely to reduce extrinsic cognitive load through measures such as providing clear information architecture and minimizing memory demands, but should also proactively foster intrinsic cognitive gains by crafting positive interactive experiences. Through explicit task guidance, timely positive feedback, and predictable system behavior, design can enhance users’ sense of control and accomplishment. Such approaches not only directly alleviate cognitive pressure but also indirectly support emotion regulation, thereby promoting a virtuous cycle of cognitive-emotional interaction.

At the psychosocial level, older adults experience changes in social roles and shrinking relational networks, which lead to increased needs for emotional connection, social belonging, and self-actualization, alongside heightened risks of loneliness. At the same time, they may exhibit varying degrees of technology-related anxiety or fluctuating motivation to learn [29]. Given the previously established close relationship between cognition and emotion, the central design challenge lies in creating interactions that compensate for diminishing cognitive resources while simultaneously addressing higher-level emotional needs. Designs should prioritize creating positive, low-anxiety interactive experiences that enhance the sense of supportive connection and maintain a sense of personal agency. It is particularly important to avoid categorizing older users simplistically as a homogeneous vulnerable group and instead recognize their intrinsic diversity and intra-group variability; for instance, significant differences may exist between healthy active older adults and those with mild to moderate cognitive impairments.

2.3. Human–Computer Interaction Design

Human–computer interaction (HCI) design is an interdisciplinary field concerned with the dynamic interplay of information and collaborative engagement between humans and computational systems. It aims to design interaction modalities that are not only efficient and usable but also enjoyable and aligned with human cognitive and behavioral patterns [30]. When it comes to intelligent devices with physical and social presence, such as companion robots for older adults, the design challenges extend far beyond those of traditional graphical user interfaces.

The design process must be approached holistically, addressing multiple dimensions that include the robot’s form factor, motion and behavior planning, multimodal input and output integration, natural dialogue logic, context-aware response mechanisms, and strategies for maintaining long-term engagement [31]. Core design principles should encompass the following: maintaining a focus on users’ authentic needs and real-world scenarios; ensuring the system is easy to learn and remember; establishing inclusive and intelligible error tolerance and recovery mechanisms; and naturally incorporating social cues and emotional feedback within interactions.

Compared to general human–computer systems, the interaction design of companion robots for older adults requires a deeper understanding of the characteristics of elderly users and a systematic application of emotional design theory. Design efforts must simultaneously address physical ease of use and accessibility, cognitive clarity, and low burden, as well as socioemotional identification and connectedness [32]. This process advances the design philosophy from “user-centered” to “experience-centered,” emphasizing the seamless integration of functionality and emotionality. The ultimate goal is to deliver warm, sustained, and meaningful companionship through technology, thereby enhancing older adults’ psychological well-being and quality of life.

3. Materials and Methods

Compared to traditional literature review methods, scientific knowledge mapping takes the scientific literature of a domain as its object and employs visualization techniques to reveal the intellectual structure, evolution, and underlying relationships within a discipline [33]. Unlike systematic reviews, which emphasize comprehensive coverage of research questions and assessment of evidence quality, scientific knowledge mapping focuses more on delineating the macro-level knowledge structure. It excels at identifying research hotspots, analyzing development trends, and uncovering the intellectual foundations of a field [34]. Compared to scoping reviews, it offers more intuitive advantages in revealing the breadth and structural relationships within research domains [35], and in contrast to narrative reviews that rely on authors’ subjective narration and critical analysis, it provides a relatively objective, data-driven representation of the field’s architecture [36]. Leveraging the method’s strengths in macro-structural analysis, this study employs a hybrid research design integrating scientific knowledge mapping [37] with qualitative content analysis [38] to systematically examine the research landscape of companion robots for older adults while effectively bridging macro-level perspectives with micro-level user needs.

Specifically, the scientific knowledge mapping analysis aims to capture the field’s overall intellectual structure and research frontiers at a macro level. This method utilizes the VOSviewer tool to visualize and quantitatively analyze keyword co-occurrence networks derived from relevant literature. Meanwhile, the qualitative content analysis seeks to delve into users’ contextualized deep needs and experiences at a micro level. Grounded in questionnaire surveys and in-depth interview data collected from older users, this approach employs systematic coding and thematic extraction techniques to identify their core requirements and usage barriers across emotional companionship, functional needs, and human–robot interaction. The integration of these two methodologies achieves a cross-level synthesis from macro-domain trends to micro-user insights, thereby providing methodological support for the mutual reinforcement between theoretical construction and empirical investigation.

3.1. Literature Data Collection and Screening

Research on interaction design for companion robots for the elderly is inherently a highly interdisciplinary and synthetic field. Its theoretical foundation and methodological approach draw upon and integrate multiple disciplines, including human–computer interaction, gerontology, sociology, design studies, robotics, computer science, and informatics [39]. To ensure the representativeness and academic authority of the literature sample, this study selected the Web of Science Core Collection as the data source, covering major sub-databases such as SCI-Expanded, SSCI, A&HCI, and EI.

The core search topics were “companion robot design for the elderly,” “emotional design for the elderly,” and “emotional needs of the elderly,” with a time span from 2000 to 2025. The search was conducted on 1 September 2025. All initially retrieved records were exported and saved in BibTeX format for reference management, including full records and cited references. Subsequently, a systematic literature screening was conducted: an initial screening based on bibliographic information excluded entries clearly irrelevant to the research topic; the retained literature then underwent an in-depth review of abstracts and full texts to eliminate irrelevant records, such as conference announcements, book reviews, duplicate publications, and studies irrelevant to the research topic. This process yielded final valid literature samples of 155, 218, and 293 documents, respectively, forming the foundational dataset for subsequent scientific knowledge mapping analysis.

This study utilized VOSviewer as the core tool for bibliometric analysis. Developed by the van Eck and Waltman team at Leiden University in the Netherlands, this software is widely used for the construction and visualization of scientific knowledge maps [40]. VOSviewer demonstrates high efficiency and strong visualization capabilities in processing large-scale bibliographic data, making it well-suited for identifying keyword co-occurrence patterns and generating cluster maps [41]. Leveraging its similarity-based network construction algorithm, this study conducted a cluster analysis of keywords from the literature. Using visualization methods such as density views and network views, the study systematically revealed research hotspots and intellectual structures in this domain, thereby identifying the major research themes and development trends.

3.2. Sampling and Research Methods

This study employed a mixed-methods approach, integrating both quantitative and qualitative strategies to capture, in a comprehensive manner, both the explicit characteristics and the underlying motivations behind the demand for companion robots among home-dwelling older adults across different age groups. The research first utilized a questionnaire survey to systematically obtain data on the distribution of needs and the statistical relationships between variables. This was followed by in-depth interviews designed to uncover the emotional drivers, contextual experiences, and nuanced attitudes toward, as well as any potential concerns regarding, human–robot emotional interaction.

The survey employed a stratified random sampling method to enhance the representativeness of the sample across key dimensions, including age, gender, living arrangement, and self-care ability [42]. Wuxi, a city in China’s Jiangsu Province, was selected as the research site for the following reasons: First, Wuxi’s aging rate is significantly higher than the national average. As of late 2023, the proportion of its population aged 60 or above has exceeded 26.9%. Its earlier and more profound demographic aging offers high typicality and practical urgency for in-depth research [43]. Second, as an economically developed city in the Yangtze River Delta with mature community infrastructure, Wuxi has been at the forefront of implementing smart-eldercare and technology-assisted aging initiatives. The generally higher acceptance of emerging technologies among older adults in Wuxi provides a favorable environment for investigating the feasibility and acceptance of companion robots [44]. Third, the research team’s institution is located in Wuxi, which facilitates sustained and in-depth field investigations. This proximity helps ensure high-quality questionnaire distribution and collection, and provides significant logistical and institutional advantages for subsequent in-depth interviews.

It must be acknowledged that although the study sample was drawn from a single city selected for its typical demographic characteristics, it demonstrates relative homogeneity in cultural background, economic status, and eldercare policy environment. This limitation may constrain the direct generalizability of findings to older populations in different socioeconomic and cultural regions. However, the primary objective of this research is to conduct an in-depth, contextualized exploration of “older users’ demand structure and acceptance mechanisms regarding companion robots” while establishing a preliminary theoretical model, rather than pursuing broad statistical representativeness. In this context, Wuxi serves as an appropriate research setting as a representative case of an economically developed city experiencing advanced population aging, providing suitable conditions for examining intrinsic relationships between variables and constructing fundamental theoretical frameworks. The value of this study lies in its systematic revelation of potential demand dimensions, key influencing factors, and their interactive mechanisms, while the extent of its external validity and generalizability requires further verification through subsequent replications and extensions across diverse populations.

In terms of questionnaire design, this study integrated multiple established scales to ensure the content validity of the measurement instrument. As summarized in

Table 1. Questionnaire structure planning.

Structure	Core Content and Objectives	Key Variables/Example Questions
Part 1: Introduction and Informed Consent	Building Trust and Clarifying Rights and Responsibilities	Research Background, Research Purpose, Emphasis on Response Value, Participant Rights, Estimated Duration
Part 2: Basic Profile of Interviewed Elderly Users	Collected for Basic Demographic and Background Information for Subsequent Subgroup Analysis	Age, Gender, Education Level, Health Status, Living Arrangements, Proficiency with Smart Devices, Routine Social Activities
Part 3: Current Status of Emotional and Companionship Needs Among Elderly Users	Assessment of Current Emotional State and Primary Companionship Needs	Loneliness (UCLA Loneliness Scale), Life Satisfaction (Life Satisfaction Index-A), Primary Emotional Needs, Desired Companionship Content
Part 4: Experiences and Evaluations of Existing Companionship Products	To Assess the Strengths and Weaknesses of Existing Solutions and Identify Unmet Needs	Primary Sources of Companionship, Satisfaction Level, Pain Points
Part 5: Perceptions, Expectations, and Concerns Regarding Elderly Companion Robots	To Investigate the Acceptance, Functional Expectations, and Potential Concerns Regarding Emerging Technologies	Perceptions and Impressions, Functional Expectations, Interaction Mode Preferences, Appearance Form Preferences, Usage Concerns
Part 6: Preferences and Acceptance of Emotional Design Elements	Investigation of Specific Emotional Design Elements (Core Section)	Anthropomorphism, Voice and Dialogue Style, Emotional Feedback Mechanisms, Proactivity and Personalization
Part 7: Open-ended Questions and Suggestions	To Collect Personalized, In-depth Information Beyond the Reach of Quantitative Data	“What would your ideal elderly companion robot be like?” “Do you have any additional thoughts, suggestions, or concerns regarding the design?” “Any suggestions for improving this questionnaire?”

, the final questionnaire structure comprises the following key components: the first section collects basic demographic information; the second section assesses current emotional and companionship needs using the UCLA Loneliness Scale [45] and the Life Satisfaction Index A (LSI-A) [46]; the third section explores perceptions, expectations, and concerns regarding companion robots; the fourth section serves as the core measurement component, focusing on preferences and acceptance levels concerning emotional design elements. This segment incorporates items from several validated instruments: the Assessment of Emotional Design Scale (AEDS) [47], a combined scale integrating the technology acceptance model (TAM) with the Seniors Technology Acceptance and Adoption Test (STAAT) [48,49], and an interaction experience assessment scale adapted for older users based on the core dimensions of the User Experience Questionnaire (UEQ) [50].

To validate the psychometric properties of the research instruments, a pilot test was conducted following standard procedures prior to the formal survey. The first phase involved three domain experts specializing in gerontology and human–computer interaction, who evaluated the questionnaire’s content validity and item appropriateness for older users. The instrument’s wording was subsequently refined based on their feedback. The second phase implemented a small-scale test with 35 participants from the target population, followed by reliability and validity assessments of the core measurement scales. Reliability analysis demonstrated acceptable internal consistency across all core constructs: specifically, the combined scale integrating the technology acceptance model and the Seniors Technology Acceptance and Adoption Test achieved a Cronbach’s α of 0.86, the Assessment of Emotional Design Scale reached 0.89, and the interaction experience scale adapted from the User Experience Questionnaire obtained 0.91. For validity testing, exploratory factor analysis confirmed satisfactory construct validity, with all items loading above 0.6 on their primary factors and cumulative variance explanation exceeding 60%, indicating a clear and effective factor structure. Incorporating both quantitative and qualitative feedback from the pilot test, final revisions were made to several ambiguously phrased items, resulting in the finalized questionnaire for the formal investigation.

A total of 550 questionnaires were distributed in this study, with 503 valid responses collected, achieving a valid response rate of 91.5%. In the qualitative phase, a purposive sampling strategy was employed to select 32 older adults representing diverse age groups, health statuses, and social backgrounds for semi-structured interviews. This sampling strategy ensured the diversity and representativeness of the sample across key characteristics, thereby enhancing the interpretability and generalizability of the findings (The questionnaire structure layout is shown in Table 1).

To investigate the underlying structure and emotional drivers of older adults’ needs for companion robots, this study conducted systematic content analysis of qualitative data obtained through in-depth interviews. Employing a methodological approach that integrates thematic extraction with coding consistency verification, the research ensures analytical rigor and enhances the reproducibility of its findings. Specific procedures are as follows:

The specific procedures were as follows. First, the interview audio recordings were transcribed verbatim, and the text was thoroughly reviewed to familiarize the research team with the content. Next, initial open coding was conducted. Two researchers with backgrounds in human–computer interaction and gerontology independently performed a line-by-line analysis of the transcripts. They systematically extracted statements pertaining to emotional needs, functional preferences, interaction barriers, and factors influencing acceptance, assigning concrete initial codes such as “preference for a gentle voice,” “concern about privacy leakage,” and “preference for simple light feedback.” The process then advanced to theme refinement and axial coding. Through multiple iterative discussions, the researchers merged related initial codes to form broader, higher-level themes, such as “preference for natural interaction,” “privacy and security concerns,” and “expectation for emotional warmth,” ensuring these themes remained grounded in the data while achieving an appropriate level of abstraction.

To ensure coding objectivity, a rigorous inter-coder reliability assessment was conducted. Following independent coding by two researchers, Cohen’s Kappa coefficient [51] was calculated to evaluate consistency for the primary themes. The resulting Kappa value of 0.82 was statistically significant, indicating substantial agreement. Any coding discrepancies were resolved through discussion facilitated by a third researcher until a consensus was achieved. Finally, the thematic framework was validated via triangulation with quantitative survey data. For instance, the frequent preference against highly anthropomorphic robots expressed in interviews aligned with the lower acceptance ratings for such designs in the survey, thereby reinforcing the validity of the moderate anthropomorphism design principle.

The systematic coding and thematic analysis not only transformed raw interview data into a structured framework but also meaningfully integrated qualitative insights with quantitative findings. This integration validates the mixed-methods research design and demonstrates the coherence of the synthesized conclusions.

4. Results

This section presents a systematic overview of the international research landscape in the field of companion robots for older adults by examining research hotspots and evolutionary trends through three interconnected dimensions: technological implementation, emotional integration, and fundamental human needs, as revealed through bibliometric knowledge mapping. First, the analysis focuses on the core research of companion robot design for older adults, tracing the evolution of technical approaches and functional configurations. Subsequently, it expands the perspective to include emotional design studies for the elderly, investigating how affective computing and experiential design principles are integrated into systems to improve interaction quality. Finally, the examination returns to the fundamental driver of the field’s development, namely research on the emotional needs of the aging population, identifying its core dimensions and key influencing factors. This progressively structured analytical framework, moving from surface manifestations to underlying principles and from technological considerations to user-centric perspectives, aims to systematically reveal the field’s knowledge architecture and developmental logic.

4.1. Bibliometric Analysis

4.1.1. Elderly Companion Robot Design Research

Internationally, research into the design of companion robots for older adults originated in the field of rehabilitation robotics in the latter half of the 20th century. Beginning in the 1980s, alongside the emerging trend of population aging and preliminary advances in intelligent technology, scholars began to systematically explore the potential of robotic technologies in addressing the diverse daily needs of older individuals. The research focus has evolved markedly: from an initial core emphasis on rehabilitation assistance, it expanded to encompass daily life care and subsequently deepened toward emotional support and social interaction [52]. Currently, research on companion robots for the elderly has entered an intelligent developmental phase, driven by artificial intelligence, affective computing, and social robot theory. It is now characterized by growing interdisciplinary integration, continuous theoretical innovation, increasingly sophisticated technological implementation, and steadily expanding application scenarios.

To delineate the research landscape of companion robots for older adults, this study generated a keyword co-occurrence clustering map using VOSviewer (Figure 4). In the map, node size represents keyword frequency, link strength indicates co-occurrence relationships, and distinct color clusters denote thematic groups. The visualization reveals several primary research hotspots: The red cluster focuses on human–robot interaction and affective computing, featuring keywords like “social robot” and “emotion recognition,” which highlight the pursuit of social attributes and natural interaction. The green cluster represents health monitoring and smart home integration, including “health monitoring” and “fall detection,” emphasizing functional roles in safety and environmental coordination. The blue cluster reflects studies on user acceptance and care needs, with keywords such as “elderly care” and “user acceptance,” addressing the alignment with user characteristics and care scenarios. Finally, the yellow cluster points to technological implementation pathways, containing core concepts like “artificial intelligence” and “sensor,” underscoring the supporting role of key enabling technologies. The knowledge map analysis reveals that research hotspots in the field of companion robots for older adults within the international academic community primarily cluster around: application development, integration of sensor technologies, differentiated analysis of user groups, needs assessment, behavioral and cognitive characteristics of elderly users, implementation of caregiving functions, emotional companionship mechanisms, product form, and interface design, as well as material innovations. Together, these themes depict a multifaceted research landscape that equally prioritizes technological implementation and user studies, balancing functional and emotional dimensions.

Recent years have witnessed a marked year-on-year increase in research outputs within this field. Concurrently, a growing number of scholars from diverse disciplines, including nursing science, psychology, computer science, and industrial design, are actively engaging in this area, which continues to expand and deepen the scope and substance of related research. Meanwhile, cutting-edge research is increasingly focused on innovations in design theory and methodology, experimental interventions and effectiveness evaluation, specificity modeling of user behavior, theoretical construction of the necessity of companionship, as well as real-time interaction and adaptive systems. This indicates that the field is gradually shifting from exploring technical feasibility toward deepening user experience, enhancing system intelligence, and developing theoretical systematization. This evolution underscores the field’s maturation into a more robust and multidimensional academic framework.

Among core research themes, social assistance and emotional support have consistently remained primary research foci within the international academic community. Huschilt and Clune [53] focused on the potential role and efficacy of socially assistive robots (SARs) in providing emotional solace, facilitating social interaction, and alleviating loneliness and depressive symptoms among older adults. They systematically examined the feasibility of robots as companions, their practical effectiveness, and their long-term impact on psychological well-being, thus laying a crucial theoretical and empirical foundation for the field. Human–robot interaction (HRI) and user acceptance represent another critical research dimension supporting the development of this domain. Culley and Madhavan [54] not only addressed the technical implementation of natural interaction and multimodal interfaces but also delved into the dynamic evolution of human–robot relationships during long-term use, the mechanisms of trust-building, and patterns of emotional attachment.

From a theoretical perspective, scholars have employed multiple models to analyze key factors influencing technology acceptance behavior among older users. Frameworks such as the behavioral reasoning theory (BRT) and the extended unified theory of acceptance and use of technology (UTAUT2) have been widely applied to examine the mechanisms through which various variables shape user attitudes and behavioral intentions. These efforts are facilitating a shift from a purely technology-driven approach to a user-centered paradigm [55]. Furthermore, user participation and iterative design are considered by Fischer et al. [56] to be essential strategies for enhancing product usability and user experience, demonstrating the comprehensive integration of user-centered design principles. Functional diversity and service integration underscore the highly application-oriented nature of the field. Robotic functionalities have evolved from early single-task execution toward integrated multidimensional services encompassing assistance with activities of daily living, health monitoring and management, cognitive training, remote monitoring, and recreational activities.

Ethics, privacy, and social impact represent critical and unavoidable considerations in international research within this field. As robotic functions continue to advance and their role in the daily lives of older adults becomes increasingly prominent, Wirtz et al. [57] conducted systematic explorations into related ethical dilemmas, data privacy protection mechanisms, risks of social role replacement, and potential user overreliance. Their work also aims to establish corresponding ethical guidelines and design standards to promote the synergistic development of technological applications and ethical practice. At the theoretical and methodological level, international research exhibits a high degree of rigor and methodological sophistication. Multiple theoretical perspectives are incorporated into the research design and interpretation. Beyond the widely applied technology acceptance models, research has employed Rodgers’ evolutionary concept analysis to clarify the conceptual connotations, prerequisites, attribute dimensions, and practical consequences of companion robots [58]. Similarly, the concept of inclusive design aims to enhance the accessibility and suitability of products and services for older adults with varying abilities and needs [59]. Meanwhile, the Living Lab approach, which emphasizes user behavior observation and co-creation in real-world contexts, has been extensively used for iterative validation of service models and user experience optimization. This approach significantly strengthens the practical applicability and social value of research outcomes [60].

Regarding research design, mixed-methods approaches have been widely adopted for their capacity to generate complementary insights. Longitudinal studies are employed to examine the long-term effects of technology use and the evolution of user attitudes, whereas cross-cultural research helps uncover demand variations across different cultural contexts. Participatory design and user experience evaluation are integral throughout the product development cycle [61]. In technological innovation, cutting-edge technologies such as artificial intelligence, machine learning, affective computing, natural language processing, and computer vision are increasingly integrated into companion robots for older adults. These advancements aim to enhance their perceptual, interactive, and decision-making capabilities, thereby enabling more precise responses to user needs. Strohmann et al. [62] combined both quantitative and qualitative methodologies to systematically investigate design principles and frameworks for virtual companionship (see Figure 5), thereby providing valuable theoretical and practical guidance for the development of emotionally interactive robotic dialogue agents.

4.1.2. Research on Emotional Design for the Elderly

Globally, research on emotional design for older adults has evolved into a significant interdisciplinary field, synthesizing theories and methods from design studies, gerontology, human–computer interaction, and other disciplines. Its emergence arises from a critique of and a move beyond the traditional functionalist design paradigm, reflecting a sustained commitment to enhancing the quality of life and psychological well-being of older adults [63]. To elucidate the intellectual structure of “emotional design for older adults,” Figure 6 presents a keyword co-occurrence clustering map. The visualization reveals several core, interconnected research directions: The blue cluster, defined by keywords like health, care, and medicine, represents the theme of health and medical care, focusing on applications in health management, disease care, and assistive products. The green cluster, characterized by loneliness, quality of life, and well-being, centers on addressing loneliness and enhancing psychological welfare, exploring how design can improve social connectedness and subjective well-being. The red cluster, featuring product, design, and technology, pertains to product development and technological implementation, dealing with the creation of emotionally valuable products and smart technologies. Finally, the yellow cluster, with keywords such as age, population, and impact, reflects a demographic and societal perspective, examining the aging phenomenon, user diversity, and the broader social impact of design. Based on the results of the keyword cluster analysis from the knowledge map, research hotspots in this domain can be broadly categorized into five main directions. High-frequency keywords include health, care, medical treatment, loneliness, and product, indicating concentrated research efforts across multiple dimensions, including functional support, healthcare provision, emotional care, and physical interaction.

Affective design theory, particularly Donald Norman’s three-level model of emotion, provides a critical theoretical framework for understanding how products evoke emotional responses from users and offers systematic guidance for design practice [64]. Its application in research related to older adults focuses on creating products, services, and environmental systems that not only meet functional needs but also evoke positive emotional experiences.

The core research themes and theoretical perspectives in this domain exhibit notable diversity. A major research direction involves enhancing the subjective well-being and quality of life of older adults. Talamo et al. [65] explored how affective design strategies can foster positive emotions, life satisfaction, social connectedness, and autonomy among the elderly, closely aligning with the core principles of healthy aging and active aging. The emotional interaction mechanisms of technological products represent another current research hotspot. Amidst the rapid development of smart devices and assistive technologies, particularly social and companion robots, scholars aim to design system interfaces and experiences capable of emotional interaction. Jo and Hong [66] focused on how a robot‘s appearance, vocal expressions, behavioral patterns, and its mechanisms for recognizing and responding to user emotions can facilitate the establishment of emotional bonds, build trust, and enhance long-term acceptance.

Emotionally supportive environments have also garnered significant scholarly attention, with related research covering residential settings, medical care facilities, and urban public spaces. Chen et al. [67] analyzed how methods such as color psychology, lighting design, integration of natural elements, material strategies, and spatial layout can be leveraged to create environments that are safe, comfortable, familiar, and capable of evoking positive emotions, thereby supporting the physical and mental states of older users. Furthermore, inclusive design and cross-cultural perspectives have gained prominence as significant topics in international research. Mohd et al. [68] emphasized the considerable differences in emotional needs among older adults from diverse cultural backgrounds, health statuses, and socioeconomic positions. They advocate using inclusive design principles to enhance the applicability of design solutions and stress the importance of cultural sensitivity in emotional expression and interpretation.

Regarding theoretical frameworks and research methodologies, this field integrates a wide range of theoretical perspectives and methodological tools beyond Norman’s theory of emotional design. Kansei Engineering, a methodology for systematically translating users’ affective and cognitive responses into concrete design elements, has been extensively studied and applied in regions such as Japan, significantly enhancing the alignment between products and users’ emotional needs [69]. User experience (UX) design theory provides a comprehensive framework for understanding and optimizing the holistic experience of older adults during their interaction with products and services [70]. The core concepts of positive psychology, such as focusing on individual strengths, cultivating positive emotions, and enhancing a sense of meaning in life, have also provided important theoretical impetus for emotional design.

Participatory design [71] and empathic design [72] involve older users actively in the design process to ensure that outcomes genuinely reflect their actual needs and emotional expectations. Experience prototyping [73] employs interactive prototypes to gather users’ emotional feedback on conceptual solutions early in the design process, effectively supporting iterative refinement. Biofeedback measurement techniques, including the monitoring of physiological indicators such as facial expression analysis, electrodermal activity (EDA), and heart rate variability (HRV), provide quantifiable data for objectively assessing older adults’ emotional responses to design stimuli [74].

Technological applications and empirical research represent fundamental components in the development of this field. With continuous advances in affective computing and artificial intelligence, it is increasingly feasible to develop intelligent systems capable of recognizing, understanding, expressing, and adapting to human emotions, thereby providing a solid technical foundation for emotional design. These technologies are being explored in the design and development of more emotionally responsive companion robots, personalized health management systems, and emotional support applications, significantly enhancing the adaptability and humanization of interactive experiences. As part of smart aging initiatives, Torta et al. [75] have worked to systematically integrate emotional design principles into urban planning and public service facilities to improve social participation and environmental belonging among older adults. Other studies have focused on the design of therapeutic environments for specific elderly groups, exploring how to elicit positive emotions and facilitate the retrieval of situational memory through sensory stimulation and nostalgic atmosphere creation [76]. Within empirical research, outcome evaluation has received increasing emphasis. Curumsing et al. [77] have adopted experimental designs and longitudinal approaches to systematically assess the potential impacts of emotional design interventions on the emotional states, behavioral patterns, and quality of life of older adults, thereby providing an empirical basis for optimizing design strategies.

4.1.3. Research on the Emotional Needs of the Elderly

Research on the emotional needs of older adults has an established international and long-standing academic tradition, with its origins and evolution rooted in a systematic concern for psychosocial needs across the human lifespan, particularly in later stages of life. Figure 7 presents a keyword co-occurrence clustering map for the research domain of “emotional needs of older adults,” revealing its multifaceted and interconnected frontiers. The primary thematic clusters and their significance are as follows: The red cluster, centered on emotional companionship, social support, and interpersonal relationships, represents research on emotional support networks. It focuses on establishing intimate bonds, strengthening social ties, and identifying pathways to alleviate loneliness. The blue cluster, encompassing older patients, dementia, and therapeutic interventions, concentrates on specific populations and therapies. It examines the emotional needs of groups like those with cognitive impairments or chronic diseases, and explores the value of non-pharmacological interventions for emotional adjustment. The green cluster, characterized by emotional needs, mechanisms of action, and influencing factors, delves into the internal structure and formation mechanisms of these needs. It aims to theoretically deconstruct their components, developmental patterns, and determinants at both individual and societal levels. The yellow cluster, containing emotional design, intelligent technology, and home environment, demonstrates research on technology-enabled solutions and design responses. It reflects academic efforts to address emotional needs through design methodologies, intelligent technologies, and smart wellness systems. Based on the results of the knowledge mapping analysis, research hotspots in this field are predominantly clustered around keywords such as emotional companionship, geriatric patients, therapeutic activities, and the intrinsic mechanisms of emotional needs, reflecting its multidisciplinary and practice-oriented nature. Current cutting-edge research is increasingly focused on emotional design, emotional support strategies, technology-supported home-based care, and the development of smart wellness systems, demonstrating a distinct shift from theoretical inquiry to applied practice.

Multiple theoretical frameworks provide diversified perspectives for understanding the emotional needs of older adults. Socioemotional selectivity theory (SST) posits that when individuals perceive their future time as limited, they tend to prioritize goals that generate positive emotional experiences and actively optimize their social networks to enhance the quality of emotional connections [78]. Attachment theory focuses on how attachment patterns formed in early life stages influence the expression and fulfillment of emotional needs in old age, particularly emphasizing the role of security and intimate relationships in psychological adjustment [79]. Meanwhile, theoretical models such as “successful aging” [80] and “active aging” [81] regard positive emotional states, sustained social engagement, and life satisfaction as core indicators of quality of life in later years, thereby guiding the development of relevant policies and design practices at a macro level.

Regarding the core dimensions and assessment of emotional needs, international research has primarily focused on older adults’desire for social connectedness and belonging, the mechanisms through which intimate relationships serve as sources of emotional support, the relationship between different types of emotional support and psychological well-being, and their sense of purpose, need for autonomy, and emotional regulation strategies for coping with stress and loss. To enhance research consistency and comparability, Zaharia et al. [82] developed and widely implemented a set of standardized assessment tools to systematically measure emotional states, levels of loneliness, social support network structures, and overall quality of life among older adults. The investigation of factors influencing the emotional needs of older adults constitutes another significant research direction in this field. At the individual level, numerous psychological and socio-demographic variables are closely associated with emotional experiences. The role of social networks has attracted significant research attention; Sun et al. [83] conducted a systematic examination of the impact of family structure, intergenerational relationship quality, characteristics of friendship networks, and degree of social participation on the emotional health of older adults. Furthermore, broader socio-cultural contexts also shape modes of emotional expression, support-seeking behaviors, and the perception of loneliness.

In the realm of theoretical models and research paradigms, the social ecological model is frequently employed to comprehensively analyze multi-level factors spanning individual, interpersonal, organizational, community, and policy dimensions [84]. The Life Course Perspective facilitates a dynamic understanding of the emotional needs of older adults by taking into account early life experiences, key life transitions, and broader socio-historical contexts [85]. Research designs demonstrate considerable diversity: longitudinal studies provide data to explore trajectories of emotional needs, influencing factors, and long-term effects; cross-national comparative research helps identify commonalities and differences in needs across diverse cultural and social settings; mixed methods research combines quantitative breadth and qualitative depth to achieve a more holistic understanding; interventional experimental studies, particularly randomized controlled trials (RCTs), are applied to evaluate the effectiveness of strategies aimed at improving emotional well-being [86].

Empirical research on emotional support and intervention strategies continues to develop. Sun et al. [87] reported that community-based programs aimed at enhancing social participation may be associated with reduced loneliness and increased perceived social support. Psychological interventions such as cognitive behavioral therapy (CBT), problem-solving therapy, and reminiscence therapy have been linked to improvements in depressive and anxiety symptoms in a number of studies [88]. The potential of technology-enabled social interventions is currently being explored, though their applicability and possible limitations warrant further attention. There is a growing focus on the needs of specific populations. Current research frontiers encompass multiple directions: studies on psychological resilience from a positive psychology perspective, examining the mechanisms through which older adults maintain positive affect in the face of adversity [89]; investigations into the spiritual needs of the elderly and their potential impact on sense of meaning and emotional solace; and analysis of the dual role technology can play in emotional support.

4.2. Analysis of Survey Results

Based on a systematic bibliometric analysis of international literature in the field of companion robots for older adults, combined with a mixed-methods study conducted in Wuxi City (including questionnaire surveys and in-depth interviews with 503 community-dwelling older adults), this study preliminarily obtained the basic demographic characteristics of the sample (see

Table 2. Demographic characteristics and group difference tests of the survey sample (N = 503).

Characteristics	60–69 Years (N = 195)	70–79 Years (N = 188)	80+ Years (N = 120)	Total (N = 503)	Test Statistic	p-Value
Gender (Male/Female)	48.2%/51.8%	45.7%/54.3%	42.5%/57.5%	46.1%/53.9%	χ2 = 1.24	0.538
Mean Age ± SD (years)	64.7 ± 2.8	74.3 ± 2.9	83.9 ± 3.5	73.1 ± 7.6	F = 1250.67	<0.001
Education Level (Primary school and below)	25.6%	38.8%	55.0%	37.8%	χ2 = 38.52	<0.001
Marital Status (Married)	78.5%	65.4%	40.8%	64.6%	χ2 = 62.31	<0.001
Living Arrangements (Living alone)	15.4%	25.0%	35.8%	23.9%	χ2 = 20.15	<0.001
Monthly Income (<¥3000)	30.3%	40.1%	48.2%	38.6%	χ2 = 15.84	<0.01
Self-rated Health (Mean ± SD, 1–5 points)	3.8 ± 0.9	3.2 ± 1.0	2.7 ± 1.1	3.3 ± 1.1	F = 65.34	<0.001
Number of Chronic Diseases (≥2 types)	45.1%	60.6%	75.8%	58.4%	χ2 = 36.77	<0.001
Smartphone Proficiency (Proficient/Relatively proficient)	75.9%	55.3%	28.3%	57.1%	χ2 = 85.90	<0.001

) and their acceptance preferences regarding different emotional expressions of robots (see

Table 3. Older adults’ acceptance of different robotic affective expression modalities (mean ± SD).

Affective Expression Modalities	60–69 Years (N = 195)	70–79 Years (N = 188)	80+ Years (N = 120)	Total (N = 503)	F-Value (p-Value)
Varied Vocal Tones	4.0 ± 0.9	3.8 ± 1.0	3.5 ± 1.1	3.8 ± 1.0	9.87 (<0.001)
Dynamic Facial Expressions on Screen	3.7 ± 1.1	3.4 ± 1.2	3.0 ± 1.2	3.4 ± 1.2	15.23 (<0.001)
Simple Light/Sound Signals	4.2 ± 0.8	4.1 ± 0.8	4.0 ± 0.9	4.1 ± 0.8	2.15 (0.117)
Bionic Movements	3.3 ± 1.2	3.0 ± 1.3	2.6 ± 1.3	3.0 ± 1.3	13.56 (<0.001)

). The findings not only reflect the social composition characteristics of the older user group but also provide an important data foundation for a deeper understanding of their emotional interaction needs. Preliminary analysis suggests that factors such as age, health status, and technology acceptance may significantly influence older adults’ preferences and acceptance levels toward emotional feedback patterns in companion robots. To investigate the significance of differences in needs and preferences across various age groups of older adults, this study employed inferential statistical methods, including one-way ANOVA and chi-square tests for analysis, following the characterization of basic sample features through descriptive statistics [90].

First, chi-square tests were employed to analyze the distributional differences in demographic characteristics across age groups. The results, presented in Table 2, revealed significant differences among older adults of different age groups in educational attainment, marital status, living arrangements, monthly income level, number of chronic conditions, and smartphone proficiency. The respective chi-square values were χ² = 38.52, p < 0.001; χ² = 62.31, p < 0.001; χ² = 20.15, p < 0.001; χ² = 15.84, p < 0.01; χ² = 36.77, p < 0.001; and χ² = 85.90, p < 0.001. Furthermore, analysis of variance indicated a statistically significant difference in self-rated health scores across age groups (F(2, 500) = 65.34, p < 0.001). Collectively, these findings demonstrate substantial group heterogeneity within the study sample, thereby justifying subsequent comparative analyses by age group.

Second, to investigate potential differences in the acceptance of emotional expression modalities across various age groups of older adults, a one-way analysis of variance (ANOVA) was performed. The results, as shown in Table 3, indicated statistically significant differences in acceptance levels across age groups for all four expression modalities. A subsequent Tukey HSD post hoc analysis further delineated the specific inter-group differences. Regarding the modalities of “dynamic on-screen facial expressions” and “bionic movements,” the acceptance level in the 60–69 age group was significantly higher than that in both the 70–79 age group and the 80 and above age group. Furthermore, the acceptance level in the 70–79 age group was also significantly higher than that in the 80 and above age group. For the modality of “varying vocal tones and speech patterns,” the acceptance level in the 60–69 age group was significantly higher than that in the 80 and above age group. In contrast, for the modality of “simple light/sound signals,” no statistically significant differences in acceptance were observed among the three age groups. Moreover, this particular modality received the highest acceptance ratings across all groups. These findings statistically corroborate that younger older adults demonstrate a relatively higher acceptance of complex, anthropomorphic interactions, whereas simple and explicit feedback mechanisms transcend age barriers, garnering universal and consistent preference.

This study, through a mixed-methods investigation, reveals several key findings regarding the needs of older adults for aging in place and their design preferences for companion robots, as outlined below:

First, the results indicate significant differences in the structure and urgency of home-based care needs among older adults of different age groups. Adults aged 80 and above, often classified as the “oldest-old,” demonstrate substantially higher needs for safety monitoring and assistance with activities of daily living, due to physiological decline and reduced mobility, highlighting their greater dependency and vulnerability. In contrast, the “young-old” (aged 60–69), who generally enjoy better physical health and a stronger willingness for social adaptation, express needs centered more around social interaction, cultural and recreational activities, learning new skills, and community participation, reflecting psychologically and socially motivated desires aligned with active aging.

Second, regarding emotional interaction patterns, older users exhibit diverse yet clearly oriented preferences. Questionnaire data show that over 70% of respondents believe robots should possess proactive greeting and caring functions, about 65% expect them to listen patiently and offer responses, and more than 60% approve of the provision of useful information and reminders. In-depth interviews further reveal a widespread preference among older adults for natural, gentle, and non-intrusive emotional expressions, such as soothing voice tones and timely yet restrained screen-based facial feedback. Statistical tests confirm that aversion toward highly anthropomorphic or complex emotional simulations increases significantly with advancing age. This suggests that emotional interaction design should strike a balance between human-like qualities and user comfort, avoiding excessive anthropomorphism that may lead to negative experiences.

Third, key factors influencing older adults’ acceptance of companion robots span multiple dimensions. Correlational analysis revealed a significant positive relationship between perceived ease of use and acceptance (r = 0.48, p < 0.01), whereas privacy and security concerns demonstrated a significant negative correlation with acceptance (r = −0.35, p < 0.01). In descending order of average importance ratings, these include practicality, ease of use, privacy and security assurance, affordability, and perceived warmth and affinity. Among these, the perception of warmth emphasizes the need for robots to avoid appearing cold or mechanical, instead conveying respect, empathy, and a trustworthy social presence. Privacy and security emerged as major concerns for most older users, particularly in contexts involving data collection and in-home monitoring. Transparent and controllable mechanisms for information processing are fundamental to building trust.

Finally, in terms of functional expectations, older adults widely desire companion robots with integrated, multi-tasking core capabilities. These primarily include safety-related functions such as emergency and health monitoring, reminders for daily tasks to mitigate the effects of cognitive decline, as well as meaningful, context-aware conversation and emotional support features. To genuinely alleviate loneliness and enhance a sense of purpose in life, these functions must be supported by a certain degree of emotional intelligence and memory continuity.

5. Conclusions

5.1. Comparison of Research Findings with Existing Literature

Based on an integrated examination of bibliometric analysis and mixed-methods empirical data, this study systematically reveals the research evolution, technological pathways, and user acceptance mechanisms of companion robots in supporting the emotional needs of older adults, offering multi-layered implications for theoretical development and practical application. Literature analysis indicates that the field has progressively expanded from functional assistance to emotional companionship, demonstrating a characteristic trend of interdisciplinary integration. Empirical investigation further delineates both commonalities and intra-group differences among older users in terms of need structures, interaction behaviors, and conditions for acceptance. These two methodological approaches mutually reinforce each other, collectively indicating that effective companion robot design must be grounded in a thorough understanding of the heterogeneous needs of the elderly, thereby achieving an organic integration of technical functionality, interactive experience, and emotional value.

First, regarding the heterogeneity of needs among older users, this study provides more nuanced empirical support for the theories of “active aging” and “healthy aging.” While existing research notes the shift in elderly needs from physical care toward comprehensive support and emphasizes the importance of emotional goals within the socioemotional selectivity theory [6,7,12], our empirical data not only confirm this macro trend but also reveal its underlying dynamic structure. The needs of younger older adults (aged 60–69) align closely with socioemotional selectivity theory, showing a stronger inclination toward social engagement and developmental activities. In contrast, the oldest-old adults (aged 80 and above), due to physiological decline, recenter their needs more prominently around safety and health monitoring, reflecting an evolving need structure influenced by age and health status. This finding contributes to moving beyond the research paradigm that treats “older adults” as a homogeneous group and provides a basis for developing segmented and adaptive design theories for companion robots.

Second, concerning emotional interaction patterns, this study’s findings critically refine the prevailing trend toward high anthropomorphism. Whereas existing research often emphasizes developing robots with rich facial expressions and biomimetic movements to enhance social presence [33,36,41], our integrated data from questionnaires and interviews reveal that older users consistently prefer natural, gentle, and unobtrusive interactions. They exhibit a measured skepticism toward highly anthropomorphic designs, a tendency that strengthens with age. This observation resonates with scholarly cautions about the potential risks of anthropomorphism in human–robot interaction, suggesting that effective emotional engagement may depend more fundamentally on “humanized” qualities like respect and empathy than on external “anthropomorphic” representations. The key contribution of this study is its clarification of how to implement the “moderate anthropomorphism” principle for older adults through low-cognitive-load feedback mechanisms, such as variations in vocal inflection and simple light signals. This provides new, age-appropriate reference points for interaction design in affective computing and social robotics.

Third, this study extends the theoretical framework for understanding technology acceptance among older adults. While traditional models and their extensions focus primarily on perceived usefulness and ease of use [34,35], our findings identify privacy, security, and emotional warmth as equally critical factors [37]. This insight aligns with ongoing discussions on ethics in service robotics and resonates with international advocacy for inclusive design [39,40]. By translating these relatively abstract constructs into concrete, empirically grounded variables that influence user decision-making, this research contributes to a more comprehensive and explanatory acceptance model for companion robots, thereby laying the foundation for subsequent scale development and empirical testing.

Finally, regarding functional integration, this study refines the documented trend toward “multifunctional service integration.” Its primary contribution is the systematic prioritization of these functions, established through comprehensive user research: safety monitoring forms the foundational cornerstone, cognitive assistance provides daily support, and meaningful emotional dialogue serves as the higher-order objective. This qualitative insight aligns with bibliometric findings that identify “emotional companionship mechanisms” as an emerging research frontier. Together, they indicate that future companion robots must embody an organic unity of function and affect. Emotional interaction divorced from practical utility risks becoming hollow, while pragmatic functions lacking emotional warmth will fail to achieve profound companionship.

5.2. Technical Approach and Solutions

To address the specific technical implementation challenges in emotion recognition and interactive systems, this study proposes architectural-level solutions. For emotion recognition, the system can employ a multimodal information fusion strategy. This involves the coordinated analysis of vocal features, including tone and rhythm, visual signals encompassing facial muscle movements and body posture, and authorized physiological data such as heart rate variability and electrodermal activity. This approach enables robust perception of emotional states in older users. The fusion mechanism effectively adapts to the unique emotional expression characteristics of the elderly population, such as subtle facial expression variations due to reduced muscle tone and vocal characteristic changes resulting from laryngeal aging. It enhances inference accuracy through cross-validation from multiple information sources. To mitigate potential performance degradation of general models when applied to the elderly demographic, we recommend implementing domain adaptation and transfer learning techniques. This process involves pre-training models on general datasets, followed by fine-tuning using smaller-scale annotated data from older adults. Concurrently, a personalized online learning mechanism can be introduced where, upon obtaining users’ informed consent, the model undergoes continuous calibration based on satisfaction feedback and interaction behavior patterns. This facilitates a progressive evolution from generic models toward individualized, specialized models.

In the architectural design of adaptive dialogue systems, this study proposes a hybrid framework to balance natural conversational flow with safety and reliability. This system integrates a rule-based finite state machine with a large language-model-powered generative dialogue component. The former ensures deterministic delivery and controllability of critical information such as health reminders and safety alerts. The latter maintains conversational openness and response flexibility to support more natural social interaction. However, when deploying large language models in elderly companion scenarios, careful consideration must be given to potential hallucinations, information overload issues, and context window limitations. To address these challenges, several technical strategies can be implemented: guiding generated content toward safe and health-related topics through carefully engineered prompts and domain-specific knowledge bases; establishing a layered information output mechanism that prioritizes concise presentations with optional detail expansion; and maintaining essential context dynamically through long-term user modeling to compensate for the limited memory window of large models. This hybrid architecture aims to achieve synergy between rule-guided precision and generative capabilities, thereby establishing a stable equilibrium between natural interaction and safe companionship.

In terms of designing emotional interaction modes, the findings of this study are consistent with advances identified in the literature. International research emphasizes the importance of natural interaction, non-intrusive design, and moderate anthropomorphism, which aligns well with the widespread preference among older adults in our study for gentle, simple interaction styles that avoid excessive human-likeness. For example, older users showed higher acceptance of natural variations in vocal tone and simple light feedback, while expressing reservations toward complex biomimetic movements. This reflects their desire for technological “humanization” rather than “anthropomorphism.” These findings invite reflection on previous design tendencies that emphasized high degrees of mimicry, suggesting that the effectiveness of emotional interaction should be based on compatibility with older adults’ cognitive habits and cultural background, rather than on the extent of morphological realism.

Furthermore, both the literature and empirical data suggest a multi-factor model influencing acceptance among older adults. Beyond classic constructs such as perceived usefulness and ease of use, this study identifies privacy security, emotional warmth, and affordability as equally critical decision variables. This aligns with recent international scholarly focus on ethical issues concerning privacy and principles of inclusive design. Notably, older users place considerable importance on the sense of emotional warmth conveyed by technological products, indicating that they expect robots to function not only as practical tools but also as socially present companions. Therefore, design must balance technical reliability with emotional affinity, ensuring that emphasis on intelligent functionality does not come at the expense of the emotional dimension inherent in human interaction.

Affective computing and adaptive interaction technologies hold significant potential in the field of elderly companionship, yet their practical implementation continues to face multiple challenges. The hybrid architecture proposed in this study represents merely one possible solution among several. Future technological development should transcend the singular pursuit of cutting-edge innovation and instead adopt a core focus on addressing genuine problems and user needs. Specifically, for the emotion recognition module, a multimodal solution leveraging Wav2Vec 2.0 and the Facial Action Coding System (FACS) can be adopted to enhance adaptability to elderly expressive characteristics. The vocal modality utilizes the Wav2Vec 2.0 model pre-trained on large-scale general corpora for feature extraction, followed by domain-adaptive fine-tuning on elderly inclusive emotional datasets such as ElderReact. This process captures acoustic feature variations like fundamental frequency and formant shifts resulting from age-related laryngeal changes. The visual modality employs a pre-trained model based on the Facial Action Coding System to analyze activation intensities of facial muscle action units, with particular attention to subtle expression variations attenuated by reduced muscle elasticity in older adults, such as eyebrow lowering and lip corner pulling. For multimodal fusion, the extracted vocal and visual feature vectors are combined with physiological features like heart rate variability acquired via non-contact sensors under user authorization. These integrated features are fed into a fusion network based on cross-modal attention mechanisms for joint modeling. This network dynamically learns contextual contribution weights across different modalities to generate robust emotional state embeddings. To construct a dedicated multimodal corpus supporting this system, synchronized vocal, facial video, and physiological signal data should be systematically collected from elderly participants in living lab environments following ethical approval and informed consent. Data collection protocols may include guided situational recall, exposure to emotion-eliciting materials, and open-ended dialogues with robots. The collected data subsequently undergoes multiple rounds of refined annotation by consistently trained annotators according to both discrete and dimensional emotion models.

5.3. Research Limitations and Future Work

This study has several limitations. Although the sample possesses a certain degree of representativeness, it was drawn from a single city; caution is therefore warranted when generalizing the findings to rural populations or groups from different cultural backgrounds. Furthermore, while the literature analysis covered mainstream international research, it included limited non-English publications, which may have omitted regionally specific innovations. Future research could pursue cross-cultural comparative studies and integrate longitudinal data to explore the evolution of human–robot relationships and the mechanisms of emotional attachment over time. This study proposes a multimodal affective fusion and adaptive interaction architecture, establishing a theoretical and technical foundation for the core affective computing capabilities of companion robots for older adults. However, a key limitation is that the architecture’s performance with real elderly users remains to be validated with rigorous real-world data. Therefore, follow-up research is planned to construct a living lab platform that simulates authentic home environments, where representative older users will participate in medium- to long-term longitudinal studies, pending ethical approval and informed consent. This study will systematically collect and annotate multimodal interaction data to evaluate key performance metrics. These include the accuracy, recall, and macro F1 score of cross-modal emotion recognition, with special attention to the characteristically subtle expressivity of the elderly; the false positive and negative rates in critical scenarios; the convergence and performance of personalized online learning algorithms; and the system’s robustness against environmental interference like lighting and noise changes. These quantitative experiments aim not only to validate the proposed architecture but also to establish empirically grounded benchmarks for affective computing performance tailored to older users.

In summary, as a technological approach to addressing the challenges of population aging, the effectiveness of companion robot design hinges on the integration of technological capabilities with humanistic considerations. Through dual validation via literature and empirical evidence, this study suggests that future research should focus on developing mechanisms responsive to the diverse needs of older adults across different age groups, health conditions, and cultural contexts; constructing natural, gentle, and reliable emotional interaction paradigms; refining multidimensional acceptance models that incorporate emotional, ethical, and socio-cultural variables; and advancing long-term real-world evaluation methods to enhance the sustainability and social acceptability of technological solutions. It is paramount that future work focuses on overcoming key technical bottlenecks in affective computing and adaptive interaction, such as enhancing the robustness of multimodal emotion recognition, optimizing hybrid dialogue system architectures for elderly users, and enabling continuous evolution of personalized models during extended interactions. When selecting technological pathways, the central guiding principle must be the enhancement of older users’ sense of well-being, security, and companionship experience. Through these pathways, companion robots hold the potential to evolve from technological products into partners capable of providing emotional support, thereby promoting the psychological well-being and quality of life of older adults.