Articles that made it past the initial screening were listed in a spreadsheet which included pertinent details for review as well as references to the articles. First, raters individually reviewed the titles of each article and excluded articles that were easily identifiable to the researchers as irrelevant to the purpose of this review. Criteria used to remove irrelevant titles included, (1) non-peer-reviewed (e.g., dissertations), and not original research (e.g., other literature reviews) articles, (2) duplicate articles (3) articles before 2018 (4) articles not written in English. Beyond this, raters removed articles which were obviously not related to healthcare and which did not feature a robot intended for human interaction. Raters were intentionally conservative in ratings such that articles which were deemed possibly irrelevant were still included for abstract review to reduce the possibility of inadvertently removing relevant articles.

After titles were reviewed, n = 949 articles were excluded and n = 656 eligible articles were reviewed further by their abstracts.

The abstracts were retrieved by hand and sorted in a shared sheet for review by the group. During the abstract review, a set of exclusion criteria was used to screen whether the content of the article was relevant for the purpose of this systematic review. Three independent raters manually reviewed the abstracts of the articles of the screened articles. A table of the abstract exclusion/inclusion criteria is shown below in Table 2 alongside their respective codes. Articles were excluded if (1) the robot was used in the wrong context, including environments outside of the hospital or for purposes outside of healthcare entirely, (2) if the study featured a technology which did not meet our definition of a robot. We defined a robot as a physically embodied mechanical device with autonomous features. We also excluded surgical robots and autonomous wheelchairs because they were not relevant to our research goals. We also excluded (3) articles which did not involve actual human-robot interactions, either because the robot was not intended for direct human interaction, or because the study only involved hypothetical interactions with a robot, such as an opinion survey. The human interaction also needed to be with a nurse or with a patient as a means of reducing the nurses’ workload (e.g., delivering medications to patients). Finally, we excluded (4) articles which were ineligible using the title review criteria.

The raters independently examined the abstracts from about 4.6% of the papers (n = 30 articles) to gauge consensus. The first agreement on the articles to include and exclude was 67%. After this initial review, the raters reviewed a portion of the articles that had been coded together and discussed the rationale behind each coding choice in order to reach a better consensus. Following this exercise, each rater independently reevaluated the initial 30 article abstracts using their own and their co-rater’s codes which yielded an 80% consensus (agreement) on articles to include. In cases in which raters disagreed, the senior author (STJ) examined abstracts to make a final determination of inclusion. The remaining articles (n = 626) were randomly divided between the raters to review independently. N = 487 articles were excluded for not meeting inclusion criteria (in which articles could be excluded for more than one reason), and n = 169 articles were taken into consideration for a complete article evaluation.

The remaining n = 169 articles were divided between the three raters during the full article review process, and eligibility was decided based on the same inclusion/exclusion criteria listed in Table 2 as well as additional criteria which include, excluding studies wherein (1) the usage of collaborative robots is in other healthcare settings (e.g., rehabilitation centers, nursing homes, etc.) other than a hospital, and (2) studies that do not include physical robots like survey papers. After a full review of the 169 articles, a total of n = 141 articles were eliminated, leaving a total of n = 28 articles that were included.

The studies in the included literature described various applications of cobots in healthcare settings. Of the 28 included articles, the most common applications were for service and logistical tasks comprising twenty-three articles majorly for aiding nurse’s work (e.g., Anthony et al., 2020; Oishi et al., 2021; Sun et al., 2022), and assistance during the COVID-19 pandemic (e.g., Prabhakar et al., 2020; Dangi et al., 2021; Virgolin et al., 2022), then social cobots (n = 8) with three overlaps wherein the cobot performs both services and social functions (Boumans et al., 2018; Boumans et al., 2019; Miyake et al., 2020). The social cobots identified in the included articles were used to provide emotional support to patients especially children (e.g., Farrier et al., 2019; Ali et al., 2020; Rossi et al., 2022) and for providing information or taking clinical screening interviews (Do et al., 2021; Yoo et al., 2021; see “Cobot application” columns in Table 3). The literature review categorizes and examines these various applications, shedding light on the multifaceted roles cobots play in healthcare.

The locations of the cobots were categorized into either “Near patients” or “Roaming healthcare facility” and “Other” for studies that did not report the location of the cobot. The most common locations were near patients as reported by about sixteen studies (Hossain et al., 2020; Yoo et al., 2021; Rossi et al., 2022). The majority of the cobots located near the patients were aiding nurses (n = 10) in performing tasks such as medication reminders (Hossain et al., 2020), or recording patients’ data (Miyake et al., 2020). All social support cobots (n = 6), and 75% of the elder care and/or memory support cobot (n = 3) were near the patients. Cobots that are roaming freely in the healthcare facilities (n = 13; Thamrongaphichartkul et al., 2020; Oishi et al., 2021) are majorly assisting nurses work (n = 9) such as a cobotic cart (Konara et al., 2020) for medication delivery, Carver-Cap cobot for logistics (Thamrongaphichartkul et al., 2020), mobile personal robot 2 (PR2), and adaptive cobotic nursing assistant (ARNA) platform for fetching objects for nurses and measuring patients’ temperatures (Dalal et al., 2018; Abubakar et al., 2020; Lundberg et al., 2022) to mention a few, while three studies did not report the location of the cobots (He et al., 2019; Toney et al., 2022; Sun et al., 2023; see “Cobot location” columns in Table 3).

A couple of studies utilized existing commercial cobots like “Pepper” (n = 2), “NAO” (n = 1), or “MEDi” (n = 2), a modified version of NAO (Herald, 2015; Farrier et al., 2019). Beyond this, there were quite many studies that provided details about other cobot platforms such as “ISOLDE” (Virgolin et al., 2021) and “PR2” (Dala et al., 2018; Lundberg et al., 2022). The majority of the cobots were of “Unique design” (n = 20) developed by the design team such as “Lio”, a personal cobot assistant with a mobile base and a rechargeable battery that disinfects and detects elevated body temperature remotely (Mišeikis et al., 2020), “CARVER”, a cobotic cart with a mobile base and storage function used for medication delivery (Thamrongaphichartkul et al., 2020; Vongbunyong et al., 2021), or an “active monitoring bedside agent” used to prevent falls for older adults (Miyake et al., 2019; see “Cobot product used” columns in Table 3).

Regarding common features, most of these cobots were described as having a mobile base (n = 20). Half of the platforms had a functional robotic arm for manipulation (n = 14), and a natural language interface (n = 14). Some platforms had storage functionality (n = 4) or could record survey data (n = 3). A number of platforms could perform some nursing tasks such as medication reminder and delivery (n = 8), taking patient vitals (n = 5), or recording other patient data aside surveys or vitals data (n = 6). A number of studies reported the capability of performing “other” nursing tasks such as collecting oropharyngeal samples (Sun et al., 2023) or providing information to patients for a more self-directed hospital stay (Yoo et al., 2021). Other cobot features identified in the literature include “JAKA 141 ZU3” cobot with a DH-Robotics AG-95 gripper, single shot-multibox detector (SSD) that is used for sample collection (Sun et al., 2023), “Aido-Bot” used for vital monitoring with a specification combination of IR sensors, pulse oximeter sensor, and a mobile base (Hossain et al., 2020; see “Cobot features” columns in Table 3).

In terms of morphology, many cobots had relatively mechanistic designs (n = 16) in which the designers did not try to create a humanlike or animal-like appearance (i.e., zoomorphic) and/or the design focused on utilitarian function instead of form but may choose to adopt a different appearance for future versions of the cobot. Some applications showed exposed electronics and wires which were likely due to the stage of the prototype and not the intention for the finalized product such as an autonomous mobile cobot (AMR) called “D-Bot” (Guan et al., 2021), or a cobotic cart called “CARVER” (Thamrongaphichartkul et al., 2020; Vongbunyong et al., 2021). The rest of the cobots in the included articles featured anthropomorphic designs (n = 12) such as the humanoid cobot “Pepper” from United Group Robotics (Boumans et al., 2018; Boumans et al., 2019), “NAO” and “MEDi” developed by Aldebaran from United Robotics Group (Farrier et al., 2019; Ali et al., 2020; Rossi et al., 2022), “PR2” (Dalal et al., 2018; Lundberg et al., 2022), or a multimodal interactive mobile cobot called “ISOLDE” (Virgolin et al., 2021; see “Morphology” columns in Table 3).

The technology readiness level (TRL) reported for the cobots ranged from one to 9 (Manning, 2023), indicating prototypes to fully commercialized products. Most fell in the middle of the TRL range of 4 (n = 6) to 5 (n = 8), suggesting these cobots have moved from proof-of-concept to testing/validating components and/or breadboards in relevant environments, or simulation of environments that are realistic as possible. Five studies featured cobots at TRL 7, which means the system’ prototype is being demonstrated in a space environment, while only one study featured a cobot at TRL 9, suggesting the cobot has been implemented into the system and is being tagged “flight-qualified” and “flight proven” (Mišeikis et al., 2020), one study featured a cobot at TRL 2, indicating the cobot is at the early stage with little to no experimental proof-of-concept (Konara et al., 2020), and three studies featured cobots at TRL 3. The remaining studies (n = 4) featured cobots at either TRL 6 (n = 2) or TRL 8 (n = 2) that were in the early prototype to demonstration systems. This indicates most healthcare cobots are still at the testing/validation level (see Figures 5A, B and “Technology Readiness Level (TRL)” columns in Table 3).

In summary, these studies indicate a significant interest in the use of collaborative cobots to aid nurses in various healthcare settings, with a wide range of cobot types and applications. The Technology Readiness Level (TRL) varied across the studies, suggesting varying stages of development and implementation of these cobotic solutions.

Table 3 shows the resulting literature included after stages of the systematic review process using PRISMA guidelines (Page et al., 2021).

The literature shows that most healthcare cobots are patient-centered, with very few designed specifically to assist nurses. Additionally, most reported applications are based in India, with limited evidence from the U.S. In addition, a large proportion of COVID-related cobots emerged from India, representing a surge in development prompted by the pandemic. In terms of target populations, most existing cobots focus on patients in isolation or the elderly, delivering medications and other essentials, and providing emotional and companionship support, with minimal cobots designed for pediatric populations. Finally, the review highlights that the majority of market-ready technologies are based on off-the-shelf platforms like Pepper, NAO, MEDi, and PR2, rather than bespoke nurse-assistive designs.

This review reveals several trends that provide insight into the future landscape of collaborative cobots in healthcare. India accounted for the majority (21.4%) of published cobotics research wherein the majority (83.3%) were developed and deployed as COVID-specific cobots, illustrating significant investments in this area and the power of urgent demands driving innovation. However, evidence of nurse-assistive cobots in U.S. hospitals was generally strikingly lacking, despite the known adoption of general-purpose cobots like TUG Automated Robotic Delivery System by Aethon Inc. (Summerfield et al., 2011; Teng et al., 2022) and Moxi service cobot by Diligent Robotics (Bernardo et al., 2022; Tietze and Mcbride, 2020; Christensen et al., 2023). This indicates a gap between commercially-driven cobot purchases and academically-published scientific research, thereby highlighting a need for more research and reporting on real-world implementations rather than conceptual proposals as evidenced in Morgan et al. (2022).

The emphasis on care cobots is patient-centric rather than nurse-centric when the self-described purpose of most designs is to “aid nurses”, but instead they are replacing them. Cost analyses, maintenance planning, and hardware/software adaptability received little attention, though these factors are critical for long-term viability. Cybersecurity is also a major concern given the confidential patient data that healthcare cobots access (Guo, 2022; Whittaker, 2022). Without deliberation of these issues, seemingly promising cobots may fail prematurely.

With few exceptions, cobots appear to augment rather than automate nursing work. But truly nurse-centered designs remain rare. Overall, realizing the full potential of nurse-assistive cobots will require committed research addressing multi-year product lifecycles, versatile applications, and localized human-cobot collaboration. Rather than replace nurses, the ideal future sees collaborative robots efficiently handling rote tasks like materials transport, and documentation to free up nurses for high-value caregiving. This future will only emerge through deep partnerships between industry providers, healthcare organizations, and nurse stakeholders, acknowledging that successful adoption depends on far more than technological capabilities alone.

Overall, this review demonstrates significant gaps in the literature and market availability of collaborative robots tailored to aid nursing practice, particularly in the U.S. The findings suggest opportunities to apply human-centered design approaches engaging frontline nursing staff, to develop innovative cobots that can address nurses’ unique needs and workflows. Increased research commitments and industry partnerships focusing on nurse-assistive cobotics could help overcome existing barriers to adoption and implementation.