Presentation
A Taxonomic Framework for Human-Robot Interaction in Healthcare - Research Gaps and Future Research Directions
SessionPoster Session 2
DescriptionHuman-robot interaction (HRI) in healthcare has evolved significantly in recent years, with numerous applications across various healthcare domains. These advancements often represent HRI applications in various ways, making it challenging to systematically categorize targeted populations, types of robots, interaction elements, and context in healthcare settings. This study addresses a long-standing need to establish a taxonomic framework that can accurately represent the diversity of HRI in healthcare. To accomplish this, we analyzed research articles, review studies, and reports to identify existing taxonomic classifications for HRI in healthcare. Archival analysis was conducted, examining original research articles, peer-reviewed studies, data and records from 2020 to 2025, as our primary source materials. This research study addresses two fundamental questions: Does our current taxonomy of HRI in healthcare adequately capture the various facets of user categories, implementation contexts, and interaction types? What taxonomic refinements need to be made to various aspects of HRI in healthcare to advance research and practice? The overall goal is to simplify the taxonomy to make it easier to categorize the target population, identify potential usability gaps, and build a roadmap to wider acceptance and longer-term use.
Method
This study performed a literature review using PRISMA (Preferred Reporting Items for Systematic Reviews and Meta analysis) guidelines to propose a taxonomic framework for HRI in healthcare. We collected 60 articles by searching keywords: HRI in healthcare, healthcare robots, and socially assistive robots from the various platforms, including Google Scholar, IEEE Xplore, ACM Digital Library, and The University of Tennessee, Knoxville Library between 2020 and 2025. After reading the titles and summaries, 10 articles were excluded from the collections for not meeting criteria of HRI in healthcare. In the words frequency analysis, the keywords that appeared most frequently include robot, care, services, interaction, and patient, which confirms relevance of processed articles. After duplicate removal, we had a total of 49 articles for the taxonomy analysis. These articles were processed using QDA Miner and WordStat software for analysis. Our study employed a four-dimensional taxonomic framework, analyzing existing literature to identify the gaps and quantify the research distributions. The dimensions of the framework include user categories, robot types, interaction elements, and context.
Results and Discussion
The user type taxonomy showed a critical imbalance across various user types. At the parent code level, taxonomy revealed direct healthcare staff (84%), patients (72%), and older adults (76%) as prominent research areas, while understudied areas include support staff with only 12% attention. There is a substantial research gap to understand how administrative and technical user types interact within healthcare. In subcategorized user types, the distribution seems more imbalanced. Caregivers seem more prominent with 63%, diagnostic staff second with 60%, and nurses attained 57% attention. Healthcare professionals, including doctors (38%) and dementia patients (36%), have received moderate attention; however, psychologists (3%), technicians (5%), and administrators (18%) are the least studied areas. The caregiving shows HRI research in healthcare is more focused on supporting both formal and informal caregiving activities. Patient subcategories also revealed disparities, as dementia (36%), autism (29%), and disabilities (28%) received moderate attention; however, depression (16%), stroke (20%), and other diseases (23%) received little attention. Older adults overall received very high attention; however, their subcategories received little attention: independent (17%), dependent (8%), and frail (6%). In clinical and medical staff subcategories, though doctors and nurses are well- studied, therapists (14%), specialists (3%), and psychologists (3%) are least studied areas. Overall, prominent areas of user types indicate a lot of research is going on involving these users with HRI in healthcare; however, less prominent areas indicate that a huge research gap exists to explore the least studied areas.
The robot types taxonomy revealed research imbalance. Dominating research areas include physically assistive robots (68%), humanoid robots (64%), and socially assistive robots (54%). Surgical robots and telepresence robots received moderate attention with 34% and 40% respectively. Less studied areas include routine task robots (16%), automated dispensing robots (10%), admin robots (10%), monitoring robots (8%), mechanoid robots (8%), diagnostic robots (6%), and non-biomimetic robots (2%). Subcategories analysis shows that Pepper is well-studied with 34%, Paro second with 32%, and rehabilitation robot with 28%. Physical assistive robots, including rehabilitation robots (28%) and walking aids (22%), received moderate attention compared to Exoskeletons (18%) and prosthetics (16%). Surgical robots, including da Vinci (20%), are moderately studied compared to Cyberknife (6%) and ROBODOC (4%). Other robots received very little attention including routine task robots like Tug (10%), RELAY (8%), HelpMate (4%); and disinfecting robots, including TASKI (2%), and UVD (10%). Among socially assistive robots, Paro (32%) is well-studied, while NAO (16%), CuDDler (2%), Kasper (4%), and Kiwi (2%) are less studied. This analysis revealed that humanoid shape robots, physically assistive and socially assistive robots, show promising research; however, other areas are filled with potential gaps to explore, including task-based robots, monitoring robots, animalistic robots, and administration robots.
The interaction types taxonomy revealed three major elements: task-based interactions, social interactions, and therapeutic interactions. All types appeared well-studied: therapeutic with 98%, task-based interaction with 92%, and social interaction with 82%. However, subclassification showed imbalance. Prominent areas include monitoring (50%), rehabilitation (46%), and delivery (42%). Therapeutic subcategories showed rehabilitation (46%), and mental health support (18%) received moderate attention, while memory training (16%), exercise guidance (12%),and cognitive therapy (2%) received low attention. In task-based interaction, monitoring, delivery, and cleaning (24%) are moderately studied; however, decision support (2%), data analysis (6%), and supply management (8%) are least studied. In social interactions, only companionship (36%) is moderately studied, while Entertainment (22%), conversational (18%), emotional support (18%), and personal counseling (2%) are least studied. The interaction types of taxonomy revealed the least studied research areas for future research directions.
The interaction context revealed significant concentration in healthcare settings. Prominent areas include telehealth (64%), and moderate categories include research centers and labs (48%), and acute care hospitals (46%). Rest of categories are less focused: community-based centers (26%), specialized health care units (24%), and institutional care units (32%). At subcategory level, acute care hospitals are less studied: emergency departments (16%) and intensive care units (8%). In telehealth, telemedicine (46%) is moderately studied, while remote monitoring (14%) is less studied. Clinical research (24%), and pilot programs (30%) showed moderate representation. In institutional care units, nursing homes (32%) are moderately covered, while university research labs (4%), living centers and memory care units are covered very little (1%). In community-based centers, patient homes (34%) are moderately studied, while outpatient clinics (26%), mobile healthcare units (22%), and community health units (1%) are less focused. Moreover, all subcategories of special healthcare units are less focused on average, 4% of the cases. So, these least studied areas have potential for inclusion of HRI research in these healthcare settings.
Conclusion
In answer to our research questions - This study provides a gap analysis which presents the dominating current research areas and understudied areas as per taxonomy. Our current taxonomy of HRI in healthcare does not adequately capture the various facets of healthcare field. This analysis therefore helps identify understudied populations, robots, interaction types, and contexts. This is important to guide research, design and implementation efforts targeted to meet the needs of the field. The refined taxonomic framework is essential for clearly and accurately defining the target populations, outlining their needs, identifying the types of interactions that could potentially meet those needs, and also addressing the interaction challenges. The proposed taxonomy redefines human-robot interaction in healthcare research into an organized, systematic framework, which facilitates identifying gaps, charting viable research directions, and building a roadmap to a wider-reaching impact of HRI in healthcare.
Method
This study performed a literature review using PRISMA (Preferred Reporting Items for Systematic Reviews and Meta analysis) guidelines to propose a taxonomic framework for HRI in healthcare. We collected 60 articles by searching keywords: HRI in healthcare, healthcare robots, and socially assistive robots from the various platforms, including Google Scholar, IEEE Xplore, ACM Digital Library, and The University of Tennessee, Knoxville Library between 2020 and 2025. After reading the titles and summaries, 10 articles were excluded from the collections for not meeting criteria of HRI in healthcare. In the words frequency analysis, the keywords that appeared most frequently include robot, care, services, interaction, and patient, which confirms relevance of processed articles. After duplicate removal, we had a total of 49 articles for the taxonomy analysis. These articles were processed using QDA Miner and WordStat software for analysis. Our study employed a four-dimensional taxonomic framework, analyzing existing literature to identify the gaps and quantify the research distributions. The dimensions of the framework include user categories, robot types, interaction elements, and context.
Results and Discussion
The user type taxonomy showed a critical imbalance across various user types. At the parent code level, taxonomy revealed direct healthcare staff (84%), patients (72%), and older adults (76%) as prominent research areas, while understudied areas include support staff with only 12% attention. There is a substantial research gap to understand how administrative and technical user types interact within healthcare. In subcategorized user types, the distribution seems more imbalanced. Caregivers seem more prominent with 63%, diagnostic staff second with 60%, and nurses attained 57% attention. Healthcare professionals, including doctors (38%) and dementia patients (36%), have received moderate attention; however, psychologists (3%), technicians (5%), and administrators (18%) are the least studied areas. The caregiving shows HRI research in healthcare is more focused on supporting both formal and informal caregiving activities. Patient subcategories also revealed disparities, as dementia (36%), autism (29%), and disabilities (28%) received moderate attention; however, depression (16%), stroke (20%), and other diseases (23%) received little attention. Older adults overall received very high attention; however, their subcategories received little attention: independent (17%), dependent (8%), and frail (6%). In clinical and medical staff subcategories, though doctors and nurses are well- studied, therapists (14%), specialists (3%), and psychologists (3%) are least studied areas. Overall, prominent areas of user types indicate a lot of research is going on involving these users with HRI in healthcare; however, less prominent areas indicate that a huge research gap exists to explore the least studied areas.
The robot types taxonomy revealed research imbalance. Dominating research areas include physically assistive robots (68%), humanoid robots (64%), and socially assistive robots (54%). Surgical robots and telepresence robots received moderate attention with 34% and 40% respectively. Less studied areas include routine task robots (16%), automated dispensing robots (10%), admin robots (10%), monitoring robots (8%), mechanoid robots (8%), diagnostic robots (6%), and non-biomimetic robots (2%). Subcategories analysis shows that Pepper is well-studied with 34%, Paro second with 32%, and rehabilitation robot with 28%. Physical assistive robots, including rehabilitation robots (28%) and walking aids (22%), received moderate attention compared to Exoskeletons (18%) and prosthetics (16%). Surgical robots, including da Vinci (20%), are moderately studied compared to Cyberknife (6%) and ROBODOC (4%). Other robots received very little attention including routine task robots like Tug (10%), RELAY (8%), HelpMate (4%); and disinfecting robots, including TASKI (2%), and UVD (10%). Among socially assistive robots, Paro (32%) is well-studied, while NAO (16%), CuDDler (2%), Kasper (4%), and Kiwi (2%) are less studied. This analysis revealed that humanoid shape robots, physically assistive and socially assistive robots, show promising research; however, other areas are filled with potential gaps to explore, including task-based robots, monitoring robots, animalistic robots, and administration robots.
The interaction types taxonomy revealed three major elements: task-based interactions, social interactions, and therapeutic interactions. All types appeared well-studied: therapeutic with 98%, task-based interaction with 92%, and social interaction with 82%. However, subclassification showed imbalance. Prominent areas include monitoring (50%), rehabilitation (46%), and delivery (42%). Therapeutic subcategories showed rehabilitation (46%), and mental health support (18%) received moderate attention, while memory training (16%), exercise guidance (12%),and cognitive therapy (2%) received low attention. In task-based interaction, monitoring, delivery, and cleaning (24%) are moderately studied; however, decision support (2%), data analysis (6%), and supply management (8%) are least studied. In social interactions, only companionship (36%) is moderately studied, while Entertainment (22%), conversational (18%), emotional support (18%), and personal counseling (2%) are least studied. The interaction types of taxonomy revealed the least studied research areas for future research directions.
The interaction context revealed significant concentration in healthcare settings. Prominent areas include telehealth (64%), and moderate categories include research centers and labs (48%), and acute care hospitals (46%). Rest of categories are less focused: community-based centers (26%), specialized health care units (24%), and institutional care units (32%). At subcategory level, acute care hospitals are less studied: emergency departments (16%) and intensive care units (8%). In telehealth, telemedicine (46%) is moderately studied, while remote monitoring (14%) is less studied. Clinical research (24%), and pilot programs (30%) showed moderate representation. In institutional care units, nursing homes (32%) are moderately covered, while university research labs (4%), living centers and memory care units are covered very little (1%). In community-based centers, patient homes (34%) are moderately studied, while outpatient clinics (26%), mobile healthcare units (22%), and community health units (1%) are less focused. Moreover, all subcategories of special healthcare units are less focused on average, 4% of the cases. So, these least studied areas have potential for inclusion of HRI research in these healthcare settings.
Conclusion
In answer to our research questions - This study provides a gap analysis which presents the dominating current research areas and understudied areas as per taxonomy. Our current taxonomy of HRI in healthcare does not adequately capture the various facets of healthcare field. This analysis therefore helps identify understudied populations, robots, interaction types, and contexts. This is important to guide research, design and implementation efforts targeted to meet the needs of the field. The refined taxonomic framework is essential for clearly and accurately defining the target populations, outlining their needs, identifying the types of interactions that could potentially meet those needs, and also addressing the interaction challenges. The proposed taxonomy redefines human-robot interaction in healthcare research into an organized, systematic framework, which facilitates identifying gaps, charting viable research directions, and building a roadmap to a wider-reaching impact of HRI in healthcare.
Event Type
Poster Presentation
TimeTuesday, March 244:45pm - 6:15pm EDT
LocationRhinelander Gallery
Digital Health