Introduction

Increasing production volume was one of the main challenges in many industries for a long time. The population's rapid growth was the principal reason for that situation (Malik & Bilberg, 2019b). To face this increasing demand, industrial systems have been developed during the first and second industrial revolutions, and mass production has become a common strategy (Jepsen et al., 2021). This was followed by the introduction of robots into manufacturing systems in the third industrial revolution (Azzi et al., 2012). Their advantages convinced managers to use them widely in different operations, mainly where repetitive tasks were concerned, such as the case of automotive and electronic industries (Xu et al., 2021). In 2019, the market sales value of industrial robotics-worldwide was more than 14 billion US dollars out of which, 3.7 billion US dollars, equal to 26 percent, belonged to automotive industries, and 3.6 billion US dollars, equal to 25 percent, belonged to electronic industries (Statista, 2022a).

For decades, the worldwide usage of robots shows that traditional industrial robots are perfectly fit for mass production. However, nowadays, the challenge that industries face is not just about the production capacity in terms of throughput but is even stronger related to flexibility, customization, and ergonomics (Battini et al., 2015). At the same time, the use of robots has some disadvantages, such as the high investment and operational costs, the difficulty of integrating them in the work environment (e.g., size problems or lack of flexibility), or human-related risks (Serebrenny et al., 2019b). Eventually, the fourth industrial revolution, known as the Industry 4.0, happened not only to increase the flexibility in the production systems but also to enhance human–machine interactions (Lamon et al., 2018). Robotics is still considered as a key technology, as Boston Consulting cited this field among “the nine pillars of Industry 4.0” (Neumann et al., 2021).

In the Industry 4.0 paradigm, robots are expected to become more flexible and safer while generally being more affordable and more size efficient. Consequently, it should be easier to integrate robots into a work environment, and more opportunities are to be created for human–robot collaborations (Weckenborg & Spengler, 2019). These expectations led to the introduction of Collaborative Robots (Cobots) which help workers in their assigned tasks and improve the results and work conditions by combining workers’ skills and robots’ physical strength and endurance.

The main differences between cobots and traditional robots are: (a) cobots do not need to perform a task as quickly as robots, so it is safer for workers to be around cobots, and (b) they should not replace workers but cooperate with them, (c) they are supposed to be more flexible than robots, so they should be simple to program, locate, and relocate, (d) they should prepare a safe shared work-space for workers (de Gea Fernández et al., 2017). Besides these advantages, cobots have ergonomic benefits in the case of repetitive or dangerous tasks. In addition, there are the possibilities to use robots to help workers with physical disabilities or aging workforces (EU-OSHA, 2019).

Cobots’ advantages have led the industry to use them more and more in recent years. Based on the Statista database, the share of collaborative robots from total unit sales worldwide being 5% in 2018 is expected to reach 13% in 2022 (Statista, 2022b). Also, the market size of the collaborative robot being 700 million US dollars in 2021 is expected to reach nearly 2 billion US dollars in 2030 (NMSC, 2022). However, same as any new technology, cobotization brought some new challenges such as worker’s safety, while safety was initially a main argument for adopting cobots. For instance, due to the close interaction of humans and robots, the safety of the workers should be ensured in a collaborative station which could be a complex problem.

In the past decade, safety has been the most referenced topic in collaborative robots’ literature and designing a safe cobot was the main target of most researchers in this field (Cardoso et al., 2021, Pinheiro et al., 2022). Researchers started to further study the interaction of “safe cobots” with humans and analyse their individual and collaborative performance. In addition to safety, when cobots and humans interact to achieve a pre-defined goal, many other aspects need to be considered.

Several literature review papers on cobots have been published to assist researchers in exploring the topic. To better understand the differences between other literature review papers with regards to the main ideas, categorization, and key findings, a comprehensive analysis was conducted on literature reviews related to cobots (see Appendix Table 4). By using keywords "cobot" OR "collaborative robot" OR "human-cobot collaboration" OR "HRC" AND "review" in Scopus and Web of Science, 32 papers were found in English. These papers were studied to extract the keydata such as subjects, research field, sources, review method, classification criteria, and key findings were extracted. The number of studies in each research field indicates that these papers mainly focused on the design phase of cobots, robotics, or computer science fields (see Fig. 1), among which significant studies such as Robla-Gomez et al., 2017, El Zaatari et al., 2019, Zhang et al., 2021a, and Costa et al., 2022 can be mentioned. However, recently, some reviews that address human–robot collaboration appeared in the literature such as Simões et al., 2022, Faccio et al., 2023.

Fig. 1
figure 1

Number of published literature reviews about collaborative robots in relation to the field of study (M: Manufacturing, R: Robotics, CS: Computer Science, S: Safety, DT: Digital Twin, HC: Health Care)

Full size image

Figure 2 shows that besides the robotics and computer science fields, studying manufacturing systems has become more popular in recent years. Moreover, several studies focus on the human–machine dimension (see Appendix Table 4), for example, human–robot communication (Hjorth & Chrysostomou, 2022) or the use of computer science for analyzing human–robot collaboration (Navas-Reascos et al., 2022b) in manufacturing systems.

Fig. 2
figure 2

Number of literature review papers per year (R: Robotics, M: Manufacturing, CS: Computer Science)

Full size image

Among the 32 selected studies, 11 were systematic literature reviews out of which three were in the manufacturing research field, two focused on human factors in collaborative systems, and one focused on workstation design factors. In our systematic literature review (SLR), we expand the analysis of the contributions on cobots and human–robot collaboration according to the following criteria: the subject of study, methodology, methodology, performance criteria, and collaboration scenarios.

The rest of this article is structured as follows. Section “Basic concepts” introduces some basic concepts about cobot's research fields, performance criteria, and collaboration scenarios. Section “SLR Methodology” presents the systematic literature review approach and steps to select papers. In Sect. “Results of the systematic literature review”, a demographic analysis is reported, the categorization of the selected papers is explained, and the articles are analyzed in the defined categories. Finally, in Sect. “Research Agenda and conclusions”, open questions, and perspectives for human-cobot collaboration are summarized and discussed.

Basic concepts

This section introduces the concepts that will be used further in our literature analysis, namely: (a) subject of study, (b) performance criteria, and (c) collaboration scenarios.

Subject of study

In general, research about cobots concerns either pre-manufacturing or post-manufacturing. In the pre-manufacturing research, researchers aim to improve the design of cobots. Therefore, a cobot is typically considered an isolated entity in these studies, not an entity in a manufacturing system. On the other hand, in the post-manufacturing research, researchers focus on either (I) how to improve the human–robot interaction by changing the cobot’s program or controllers or (II) how to use or implement cobots in a production system.

Based on the previous explanations, cobot studies can be divided into three main classes: design, programming, and operation.

  • Design: The main targets of such papers are to produce a better cobot. Changing the sensors, arms angles, the material used, and arms speed are the most common ways to design new cobots with better performance.

  • Programming: The program and controller installed on the cobots act as the decision-maker. Therefore, researchers try to find a new way to develop programs/algorithms that can improve cobots’ performance.

  • Operation: Such papers discuss the integration of cobots in manufacturing systems.

Performance criteria

The selection of performance criteria is essential for designing a cobot or a collaborative system (Simões et al., 2022). Every study can consider one or more performance criteria to evaluate their design solutions. In Sect. “Results of the systematic literature review”, all performance criteria that have been employed in the cobot's literature are discussed.

Collaboration scenarios

Based on how cobots and humans collaborate to accomplish tasks (processes) in a workstation, the collaboration between humans and cobots can be classified into four different categories: independent, sequential, simultaneous, and supportive (El Zaatari, 2019) (see Fig. 3).

  • Independent: In this scenario, a human worker and a cobot process two tasks (tasks 1 and 2) separately on two different workpieces (workpieces 1 and 2). In some studies, “independent” scenarios are named “parallel” scenarios.

  • Sequential: This scenario happens when a human worker and a cobot process two different tasks (tasks 1 and 2) on the same workpiece in serial processes. This scenario is usually used to improve working conditions for workers. Delivery tasks or pick-and-place are two samples.

  • Simultaneous: Here, a human worker and a cobot process two different tasks (tasks 1 and 2) on the same workpiece at the same time. This scenario is usually used to improve ergonomics. For instance, rivets can be processed by a cobot and screws by a human worker simultaneously. Ensuring the safety of workers in this scenario is of paramount importance.

  • Supportive: When a human worker and a cobot interactively process a single task on a single workpiece, the scenario is called “supportive”. In this scenario, support each other (usually physical support) to accomplish a task. For example, a cobot holds the workpiece and a human worker fastens the screw (Zhang et al., 2021b).

Fig. 3
figure 3

Collaboration scenarios (Zhang et al, 2021b)

Full size image

SLR Methodology

In this section, all the research questions were described, then the search strategy and selection criteria were explained in detail. To find, classify, and analyse all relevant articles to the topic of interest and research questions, a Systematic Literature Review (SLR) was performed. The proposed guidelines for SLR studies from Kitchenham (2004) were followed. The SLR steps are applied in the following order:

  • Defining research questions

  • Specifying search keyworks (inclusion criteria) and search method

  • Screening titles, abstract, and conclusions to find irrelevant studies

  • Reading carefully all the selected studies

  • Using both forward and backward snowball approaches

  • Categorizing, summarizing, and reporting the results to answer the research questions.

Research questions

The first step of the SLR study is defining the research questions. The following questions were selected as the research questions of this study because answering these questions can help researchers to better understand the current situation around cobots implementation and the open questions in this field.

RQ1: What are the performance criteria used in studies related to the implementation of cobots?

RQ2: How many studies considered each collaboration scenario?

RQ3: What are the achievements of each study?

RQ4: What are the open questions about implementation of cobots?

Search and selection strategy

In order to find the papers related to the human-cobot collaboration, two online databases, Scopus and ISI Web of Science, were used.

Inclusion criteria

A combination search expression was used in Scopus and Web of Science database to conduct a systematic literature review on the implementation of cobots in manufacturing systems. The relation between search expressions is drawn in Fig. 4.

Fig. 4
figure 4

Keywords and relations for search in databases

Full size image

By using the defined search expressions and the relation between them, 237 papers were found in both databases.

Exclusion criteria

In the first step, non-English language articles were excluded from the papers. In the next step, the duplicated articles were removed. A two-step systematic approach was applied to extract the irrelevant articles: (1) examining the title, reading the abstract, and conclusions; (2) a full paper reading was considered if the title, abstract, and conclusions could not provide sufficient information about the paper's relevance. All papers not respecting at least one of the following conditions were excluded from the analysis:

  • The article does not fit the considered research questions.

  • Humans were not considered.

  • Cobot is not considered specifically.

  • Not related to the manufacturing system.

After the full reading of papers, 183 papers were selected for analysis. Additionally, a forward and backward snowball approach was realised providing 19 papers. All the search and selection processes are described in Fig. 5.

Fig. 5
figure 5

Search and selection steps

Full size image

Classification

To classify the final 201 papers, the following dimensions were considered for the analysis of each paper's research questions, results, and conclusions (see Appendix Table 5):

  1. 1.

    Subject of study: design, programming, and/or operation (as explained in Section “Subject of study”).

  2. 2.

    Methodology (key proposition): mathematical approach, mathematical modelling, simulation, framework, comparative case study, or other.

  3. 3.

    Performance criteria: safety, cost, flexibility, productivity, ergonomic, or quality.

  4. 4.

    Human-cobot collaboration type: independent, sequential, simultaneously, and supportive (as explained in Section “Collaboration scenarios”)

Descriptive analysis

In this section, some descriptive analyses related to selected articles were presented. In the next section, the main results of the systematic literature review were addressed. Based on the Scopus database, more than 95% of papers in the field of collaborative robots have been published after 2010. As a preliminary analysis of the published papers filed on cobots in the Scopus database, two networks were elaborated on the most frequent keywords used in all the articles using VOSviewer software (see Fig. 6 and Fig. 7).

Fig. 6
figure 6

Clustering of co-occurrence keywords network of the 2251 papers (Scopus Database) using VOSviewer

Full size image
Fig. 7
figure 7

Publication year demonstration of co-occurrence keywords network of the 2251 papers (Scopus Database) using VOSviewer

Full size image

Keywords classifications

In Fig. 6, the most frequent keywords are classified based on their relationship in articles. The auto classifier distinguished three different classes. As can be understood from the network and color classification, red color keywords are mostly related to the computer science and programming aspect of cobots, such as deep learning, augmented reality, and simulation. The green color keywords are mostly related to the robotics and design phase of cobots, such as physical human–robot interaction, force control, and motion planning. Additionally, the blue color keywords are mostly related to safety and risk management, such as ergonomic, safety, collision avoidance, and risk assessments.

The most frequently used keywords in cobot studies were “Collaborative robots” and “assembly” which were classified in the green class, mainly related to the robotics and design phase of cobots. This is consistent with the findings from Fig. 1 and Appendix Table 5, which show that most studies until now have been focused on the robotics and computer science aspects of cobots. As inferred from the classification shown in Fig. 6, the majority of the most-used keywords are related to the design and programming phase. However, there are some categorizations and relationships that deserve further attention. For example, "trust" was categorized in the red class, which is closely related to programming. This is because most researchers have attempted to increase trust between humans and cobots by providing better and more reliable programming. "Safety" is one of the most frequently used keywords, alongside “Collaborative robots”, "human–robot collaboration", and "industry 4.0″. It is also connected to both the design and programming phase. "Sensors” is categorized in the programming class, but it is well connected to both other classes due to its importance in designing safe human–robot collaboration.

Keywords timeline

In addition to the keyword classifications, the keyword timeline in Fig. 7 can also provide insights into the general trends in recent years in the field of cobots. The timeline shows the evolution of the keywords used over time, and it can be seen that there is a shift in the research focus from older keywords related to programming and safety, to more recent keywords related to robotics and design phase, safety, ergonomics, and the incorporation of new technologies such as digital twins, deep learning and smart factory. This indicates a growing trend towards improving the functionality, safety and efficiency of cobots, and incorporating new technologies in the design and operation of cobots. Nearly 30 percent of the top-most repeated keywords belong to papers published in the past 2 years, and more than 90 percent of the top-most repeated keywords belong to papers published in the past five years (see yellow and green nodes in Fig. 7). As seen in Fig. 7, the three oldest keywords among the top repeated keywords are "Cobot", "Haptics", and "Teleoperation". These keywords are categorized in the programming class. The next generation of keywords, represented in green in Fig. 7 and related to years between 2018 and 2020, are mostly related to the robotics or design phase of cobots and safety such as "motion planning", "force control", "impedance control", "collision avoidance", and "obstacle avoidance". "Trust", "service robots", and "ergonomics" from the safety and ergonomic class, and "digital twins", "deep learning", and "smart factory" from the programming class are the most recent keywords used frequently in studies related to cobots. This indicates that the research focus has shifted towards improving the functionality, safety and efficiency of cobots and incorporating new technologies such as digital twins and deep learning in the design and operation of cobots.

Figure 7 also highlights an important outcome. It shows a general shift in research topics related to cobots. The oldest frequent keywords were mostly related to creating a safe and usable collaboration system. A more detailed analysis shows that most of them were frameworks for designing human–robot collaboration or aimed to demonstrate the benefits of cobots for industries. As industries become more familiar with the benefits and safety of cobots, researchers began to focus on designing more efficient systems and improving the functionality of cobots. This shift in research focus reflects the growing acceptance and integration of cobots in industries and also the recent interest in human factors and ergonomics aspects.

Frequency analysis

A Pivot table based on the performance criteria and the main study's objective for all 201 selected articles clearly shows the frequency of research in each field (see Table 1). As shown in the table, productivity was the most frequently considered performance criterion in the articles to examine the production system. This factor alone is considered in more than 60 percent of articles. Safety and flexibility are the other two frequent performance criteria in literature. Quality as the performance criterion was considered just once in the previous studies.

Table 1 Frequency of using each performance criterion for each objective
Full size table

Table 1 showed that performance criteria were not used evenly across different fields. For example, papers that focus on design or programming mainly consider productivity, flexibility, and safety of the collaborative system. One reason for this could be that in these types of studies, the main goal is to enhance the performance of cobots rather than the system as a whole. On the other hand, to evaluate the cost of a collaborative system, it is necessary to consider the entire system. However, in the operation phase, except for productivity and flexibility, other performance criteria are used almost evenly. The number of studies that consider flexibility as a performance criterion is low, mainly due to the difficulty of evaluating flexibility using quantitative measures.

Results of the systematic literature review

Studies in the design and programming category usually try to improve cobots from the mechanical or programming aspect of view. However, the main target of our study is considering cobots from a management aspect of view. Therefore, this section analyses only papers belonging to the operation class (the subject of study is operation).

Since the focus of this study is on the operation phase, a more detailed analysis of the corresponding contributions is provided in Appendix Table 6 regarding this phase. In addition to the methodology and performance criteria, Appendix III summarizes the main problem addressed by each study. This provides a better understanding of the specific challenges that researchers are addressing and what they propose facing that. This facilitates a more comprehensive understanding of the field and the progress that has been made in addressing the key issues related to human–robot collaboration in manufacturing systems.

Context of performance assessment

The first aspect to consider when examining the problem definitions in different studies is the context of performance analysis. Based on the selected articles, it is possible to categorize the context of performance analysis into several classes.

Designing collaborative assembly line

As shown in Table 2, the majority of studies belong to this class. The first studies tried to provide frameworks for designing a collaborative assembly line. Matthias et al. (2011) were the first to propose a framework for designing a collaborative assembly line, and as expected, safety was the considered performance criterion. Afterwards, other studies were published with different performance criteria (Djuric et al., 2016; Schonberger et al., 2018; Cencen et al., 2018; Serebrenny et al., 2019a; Jepsen et al., 2021; Gervasi et al., 2022). After several framework studies, Gil-Vilda et al. (2017) evaluated the performance of a real collaborative assembly line in a comparative case study. Other examples include Wang et al. (2019), D'Souza et al. (2020), Inoue et al. (2021), Sordan et al. (2021), and Navas-Reascos et al., (2022a) with different performance criteria. Simulation was the next methodology used in this class (Malik et al., 2020, 2021; Thomas et al., 2018; Zhu et al., 2022). Although mathematical modeling was the last methodology used for this class, it was the most frequently used methodology for collaborative assembly line balancing. The first study in this category was conducted by Weckenborg & Spengler (2019), who considered cost and ergonomics as performance criteria. However, due to the long history and strong literature in proposing mathematical models for manual assembly line balancing problems, numerous studies have developed new mathematical models to design and optimize a collaborative assembly line (i.e.,Boschetti et al., 2021a, 2021b; Dalle Mura & Dini, 2019; Sadik et al., 2017; Weckenborg et al., 2020; Cohen et al., 2022; Zhu et al., 2022; Abdous et al., 2022).

Table 2 Performance assessment methods and used criteria for cobots operation phase
Full size table

Designing collaborative disassembly line

Early developments in collaborative assembly lines on one hand, and the advantages of using cobots in hazardous or dangerous tasks on the other hand, have led many researchers to focus on designing collaborative disassembly lines. In this field, researchers have mostly utilized existing frameworks or simulation models to design collaborative systems. For instance, Xu et al. (2021) provided a mathematical model to balance a collaborative disassembly line by considering both productivity and safety. Due to the hazardous nature of some tasks in the disassembly line, the majority of studies have considered safety as a crucial performance criterion for evaluating the collaborative system's design. For example, Lee et al. (2022), Deniz & Ozcelik (2023), and Liao et al. (2023) all attempted to assign the most hazardous tasks to cobots to increase the safety of the disassembly line for human workers. While most studies in this area used mathematical models to balance the collaborative disassembly line, Liu et al. (2022a) compared different reinforcement learning methods to explore the feasibility of designing a collaborative disassembly line.

Task allocation

On first glance, task allocation is a component of designing a collaborative assembly line problem, so it is best categorized as part of the first class. However, new approaches to human–robot communication, such as optical sensors, augmented reality (AR), and motion prediction systems, are aimed at addressing the lack of cognitive ability of cobots (Li et al., 2022). Consequently, in addition to traditional task allocation, real-time task allocation or scheduling has become more popular. Online scheduling allows for immediate task allocation to workers and cobots based on real-time data, such as production rate, task completion time, and machine status. This ensures that the right worker or cobot is assigned to the right task at the right time, thereby optimizing workflow and reducing idle time. Despite the underdeveloped infrastructure for seamless communication between humans and cobots, various researchers have investigated this topic. Petzoldt et al. (2022) and Pabolu et al. (2022) have proposed a framework for dynamic task assignment in a collaborative assembly line and have considered productivity as the performance criteria. Zhang et al., (2022a, 2022e) and Lanzoni et al. (2022) have employed AI approaches in their studies. Li et al. (2022) provides a model for calculating online fatigue in an assembly line and reducing fatigue levels by assigning tasks to cobots.

Workspace design

Workstation design is a critical component in a collaborative assembly line that affects both the safety and productivity of workers. The design must ensure that the work environment is ergonomic and safe for both workers and cobots to minimize the risk of accidents, injuries, and musculoskeletal disorders (Gualtieri et al., 2022). An effective workstation design can enhance workflow efficiency and reduce the time required to complete tasks, thereby minimizing bottlenecks, and streamlining the workflow. Moreover, an adaptable and flexible workstation design is crucial for the long-term success of a collaborative assembly line (Gervasi et al., 2021). The design should accommodate changes in product design, production volumes, and other variables that impact the assembly line's operation. This requires a modular design that enables the quick and efficient reconfiguration of workstations to meet changing needs. In summary, an optimal workstation design is essential for achieving a safe, efficient, and flexible collaborative assembly line. To design a workspace, researchers have proposed frameworks and simulation models. Malik & Bilberg (2019a) proposed a framework for designing a workspace and the positions of cobots, tools, and objects for a safe collaborative system. Malik et al. (2020) provided a new framework for designing productive, safe, and flexible collaborative workspaces and proposed a simulation model to validate it. Wojtynek & Wrede (2020) also developed a new simulation model for collaborative workspaces to design a productive workspace.

Task classification

Task classification is an important aspect of collaborative assembly lines as it helps to identify which tasks are best suited for humans and which tasks are more appropriate for collaborative robots. This ensures that each worker and cobot is performing tasks that are safe and appropriate for their respective capabilities, ultimately improving efficiency and productivity. Task classification can also optimize the allocation of tasks by assigning them based on workers' and cobots' respective strengths and abilities, enabling the assembly line to operate more smoothly. Additionally, it can aid in designing tasks that are compatible with collaborative systems. In 2018, Bruno & Antonelli used an AI classification tool (Decision Tree) to develop a task classification model, considering features such as precision and tools required (Bruno and Antonelli, 2018). In 2019, Antonelli & Bruno extended their work by adding more features to their classification models, resulting in a more complete classification (Antonelli and Bruno, 2019).

Performance criteria

These sections analyse the contributions according to the evaluation criteria used. Five performance criteria have been used in the literature of Cobots operation class: safety, cost, flexibility, productivity, and ergonomics.

Cost

Costs can be considered both in the design phase and the operational phases. Some researchers are interested in the reduction of the cost of designing or manufacturing of cobots. In the operational phase, researchers aim to reduce the cost of implementation of new cobots in a manufacturing system, cost of maintenance, cost of collaboration, or cost of production. Weckenborg & Spengler (2019) and Dalle Mura & Dini (2019) developed a new cost-oriented mathematical modeling for the collaborative assembly line. Accorsi et al. (2019) studied the economic feasibility of using cobots in the food packaging industry. Gualtieri et al. (2019) developed a new evaluation methodology for redesigning a pure manual assembly line into a collaborative assembly line. Karaulova et al. (2019) analyzed an assembly line after establishing cobots. Dalle Mura & Dini (2022), Zhu et al. (2022), Dalle Mura & Dini (2022), Abdous et al. (2022), and Belhadj et al. (2022) developed a mathematical model and considered cost as one of the objective functions. Li et al., (2021c) proposed a bi-objective mathematical model with the second objective function being the minimization of the cost of the assembly line. Fager et al. (2021) calculated the cost-effectiveness of using Cobots in a picking system. Vieira et al. (2022) developed a two-level mathematical model with a detailed discrete-event simulation model. Peron et al. (2022) proposed a Decision Support System (DSS) for implementing assistive technologies in assembly line which cost is one of the performance criteria evaluated. Lee et al. (2022), Liao et al. (2023), and Deniz & Ozcelik (2023) developed a mathematical model for collaborative disassembly line which one of the objective functions is cost. Xiang et al. (2022) developed a mathematical model for a multi-product u-shaped collaborative assembly line and tried to optimize cost of production.

Productivity

Productivity is the most used performance criterion in the 53 selected papers and it was considered in both the design and operational phases. Lamon et al. (2019) related the increased productivity with minimizing cobots’ fatigue. Wang et al. (2019) showed the improvement in productivity by using cobots in a real case study. Serebrenny et al., (2019a), Gualtieri et al., (2020b), and Gervasi et al. (2021) suggested a framework to improve productivity. Zhang and Jia (2020), Weckenborg et al. (2020), Boschetti et al., (2021a), Cohen et al. (2022), Boschetti et al., (2021b), Nourmohammadi et al. (2022), Antonelli & Aliev (2022), and Almasarwah et al. (2022) used mathematical models to maximize productivity.

Wojtynek & Wrede (2020), Malik et al. (2021), and Wang et al. (2022a) used simulation to evaluate productivity improvement by implementing cobots in assembly lines. Malik et al. (2020) elaborated a framework to simulate the designed assembly line with virtual reality to evaluate the productivity of the assembly line. Gualtieri et al., (2020a) provided a guideline for designing a product that improves the productivity of the collaborative assembly lines. Xu et al. (2021), Li et al., (2021c), Vieira et al. (2022), Li et al. (2022), and Keshvarparast et al. (2022) developed a multi-objective mathematical model in which one of the objective functions was to improve productivity. Zhang et al., (2021b) applied a new metric, a combination of productivity and ergonomics, to design a collaborative assembly line. Malik & Brem (2021) proposed a framework to use digital twins to simulate collaborative assembly lines. Rega et al. (2021) developed a knowledge-based approach to optimize the productivity of assembly lines. Sordan et al. (2022) used a case study to evaluate the workers’ idle time in an assembly line balancing problem. Gjeldum et al. (2022) proposed a 3-level decision support system for task-sharing to improve productivity. Ibanez et al. (2021) and Banziger et al. (2020) developed new simulation software to design a collaborative assembly line to evaluate the productivity of the designed assembly line. Petzoldt et al. (2022) proposed a framework for dynamic task allocation and validate the effectivity of the framework by using simulation model. Peron et al. (2022) proposed a Decision Support System (DSS) for implementing assistive technologies in assembly line which productivity is one of the performance criteria evaluated. Zhang et al. (2022b) used Reinforcement Learning for online task sequencing in a collaborative assembly line which considered productivity as a performance criterion. Pabolu et al. (2022) proposed a digital twin-based framework for evaluating a collaborative assembly line before implementation. Liu et al. (2022a) compare two different Reinforcement Learning methods and the solutions provided by each of them based on productivity evaluation.

Ergonomics

Collaborative systems such as cobots have mainly two kinds of ergonomic issues, cognitive and physical. Cognitive ergonomic issues refer to mental stress and psychological discomfort, which could be felt by operators while collaborating with robots. Papers that have considered cognitive ergonomic issues are rare. Most of the papers considered ergonomics similarly to practices for the assembly lines by considering fatigue, energy expenditure, etc. Realyvásquez-Vargas et al. (2019) tried to reduce the Occupational risk factors (e.g., awkward postures, excessive effort, and repetitive movements) in a real case study by implementing cobots. Weckenborg & Spengler (2019) used the mean work rate as an ergonomic constraint in a new cost-oriented mathematical modeling for collaborative assembly lines. Dalle Mura & Dini (2019) considered tasks' energy expenditure in cost-oriented mathematical modeling. They ensure that assigned tasks’ energy expenditure for each worker should not pass the physical limitation of the worker. Gualtieri et al., (2020a) and Gualtieri et al., (2020b) provided a guideline for designing a product that improves the ergonomics of the collaborative assembly lines. Zhang et al., (2021b) used a new metric, the combination of productivity and ergonomics, to design a collaborative assembly line. Dalle Mura & Dini (2022) developed a mathematical model and considered a combination formula of cost and ergonomics as the objective function. Banziger et al. (2020) proposed a simulation method to allocating task to human or robot. Keshvarparast et al. (2022), and Dalle Mura and Dini (2022), developed a multi-objective mathematical model that one of the objective functions was minimizing the total physical workload for each worker. Li et al. (2022) considered fatigue in a collaborative assembly line balancing. Navas-Reascos et al., (2022a) evaluate physical strain and muscular activities in a collaborative assembly line and compare it with the previous manual assembly line.

Safety

A collision-free collaboration system is a safe collaboration system (Rojas et al., 2021). Opposite to industrial robots, which are usually isolated while workers being restricted from approaching or interacting with the robots, workers and cobots can be assigned to a workstation freely. As a result, worker safety was the first issue that the researchers focused on. That is why, throughout the design process of a cobot, safety is a critical consideration, and the International Organization for Standardization (ISO) sets specific guidelines for safe, collaborative work (ISO 10218-1 and ISO 10218-2). Safety-rated monitored stop, hand guiding, speed and separation monitoring (SSM), and power and force limiting are among the four collaboration scenarios addressed by the safety standards (Costanza et al., 2021). Matthias et al. (2011) and Malik & Bilberg (2019a) provided a framework to ensure the safety of workers in a collaborative system. Gualtieri et al., (2020a) provided a guideline to design a product that is compatible with the safety procedures of the collaborative assembly lines. Malik et al. (2020) provided a framework to simulate the designed assembly line with virtual reality to evaluate the safety of the assembly line. Berger et al. (2020) introduced the “safety Bubble” concept to ensure the safety of workers. Xu et al. (2021), by referring to ISO/TS 15066, ensure the safety of workers by reducing the speed of cobots regarding the distance between worker and cobot. Rega et al. (2021) developed a knowledge-based approach to optimize the productivity and safety of assembly lines. Lee et al. (2022), Liao et al. (2023), and Deniz & Ozcelik (2023) proposed a mathematical model to safety of human worker in a hazardous disassembly environment.

Flexibility

Flexibility, like safety, is an indicator that researchers usually study in a qualitative way in the cobots design phase. In the literature, two types of flexibility are stated. First, flexible cobot which refers to how fast the cobots can reprogram or mobilize for new procedures; second type, flexible collaboration is about the number of tasks that cobots can possibly do in a given time (design of work cell). The term “flexible cobot” is usually considered in the design phase of a cobot. Cobot designers aim to design and develop a new cobot that can perform various activities by considering flexibility during the design process. “Flexible collaboration” refers to the manufacturing system design. Malik et al. (2020) provided a framework to simulate the designed assembly line with virtual reality to evaluate the flexibility of the assembly line. Jepsen et al. (2021) proposed a new framework to design a flexible assembly line. Inoue et al. (2021) used mobile cobots to transfer products in an assembly line to improve flexibility.

Papers’ methodology and adopted performance criteria

Table 2 presents an analysis of the frequency of different performance criteria and methodologies in the literature on cobot operation. The operational phase of cobots is a critical aspect that must be considered to ensure their successful implementation. The results show that most of the studies that developed mathematical models used cost and productivity as their primary performance criteria. However, the number of papers that considered cost as the performance criterion in other methodologies is very low. For example, to best of our knowledge there is not any framework which considered cost as the performance criteria. This finding indicates that cost is not a primary concern in other methodologies. In contrast, productivity is the most frequent performance criterion used in the literature. More than half of the papers considered productivity as the performance criterion, which highlights the importance of improving cobot productivity in the operational phase. The analysis also shows that ergonomics is a relatively neglected performance criterion in cobot operation studies. This result suggests that researchers need to pay more attention to ergonomics in the operational phase to ensure worker well-being. Moreover, the results indicate that safety and flexibility are rare performance criteria in mathematical modeling and mathematical approach methodologies. Instead, most papers that focused on safety and flexibility used framework approaches. Safety is also an interesting topic for review papers, suggesting that there is a need for more research on this topic. The analysis further shows that studies in design or programming classes mainly used simulations to improve the safety of cobots. Simulations were infrequent in papers that considered productivity as the performance criterion. This finding implies that researchers need to incorporate simulations into their studies to improve cobot safety.

In conclusion, the analysis of Table 2 provides valuable insights into the frequency of different performance criteria in cobot operation studies. The results highlight the need for more research on ergonomics, safety, and flexibility in cobot operation. Researchers should also consider using simulations to improve cobot safety and productivity during the operational phase.

Human-cobot collaboration scenarios

Studying cobots requires a deep understanding of the collaboration scenario, which refers to the division of tasks among team members. This is a crucial aspect of the operation phase (e.g. assembly line design) process, as the collaboration scenario chosen can have a significant impact on the planning, efficiency, and overall outcome of the production system (Antonelli & Bruno, 2019). Keshvarparast et al. (2022) have introduced a mathematical model that provides insights into how three different collaboration scenarios (sequential, simultaneous, and supportive) can impact the design and cycle time of collaborative assembly lines.

However, research that considered collaboration scenarios is limited. Table 3 shows that, more than 43% of the 58 selected papers do not consider collaboration scenarios. This is a concerning issue, as the failure to consider the collaboration scenario can lead to inefficiencies, longer cycle times, and decreased productivity. The majority of the studies that failed to consider collaboration scenarios only provided frameworks or comparative case studies. Moreover, studies that provided mathematical models considered at least one scenario, the number of studies that considered multiple scenarios is low, likely due to the complexity of considering all the different scenarios and their varying requirements and assumptions. Although all the studies that provided mathematical models considered at least one scenario, the number of studies that considered multiple scenarios is low, likely due to the complexity of considering all the different scenarios and their varying requirements and assumptions. It is worth mentioning that the specific needs of the industry, the type of tasks being performed, and the available resources, among other factors, can affect the choice of the collaboration scenario. Studying the application sector of the cobots could help practitioners in this matter.

Table 3 Frequency analysis of collaborative scenarios based on methodology
Full size table

Cobots’ application industrial sectors

The application sector of cobots is also an important aspect of research, as it helps us understand how these robots can be used in different fields, such as warehouse, automotive, and precision tasks. Cobots are designed to work alongside humans, and they have the potential to increase productivity, improve safety, and reduce costs. However, the specific requirements of different industries and tasks may vary, which is why it is important to understand the task types and industries where cobots are being used. For example, in a warehouse, cobots may be used for picking and packing, while in automotive manufacturing, they may be used for assembly or welding. In precision tasks, cobots may be used for tasks that require a high degree of accuracy, such as inspection or quality control. Understanding the specific task types and industries where cobots are being used can help identify the potential benefits and challenges of using cobots in those applications. The studies included in Appendix Table 6 explore the use of cobots in various positions and for different types of tasks, but they do not provide information on the specific task types or industries where cobots are being used. This lack of specificity highlights the need for more research to fully understand the capabilities and limitations of cobots in different applications. By identifying the specific requirements and challenges of different industries and tasks, researchers can develop cobots that are better suited to those applications, as well as identify areas where further research is needed.

In summary, the application sector of cobots is an important area of research, as it helps us understand how these robots can be used in different fields and for different types of tasks. The lack of specificity in the studies included in Appendix Table 6 highlights the need for more research to fully understand the capabilities and limitations of cobots in different applications. This research can help identify the potential benefits and challenges of using cobots in specific industries and tasks and help develop cobots that are better suited to those applications. Ultimately, understanding the application sector of cobots is crucial to realizing their potential to improve productivity, safety, and cost-effectiveness in a wide range of industries.

Digital twins for assessing the performance of cobots

The performance evaluation approach or method could be also an important topic in the study of cobots. As a complementary technology, which is largely addressed in the Industry 4.0 context, the Digital Twins could be mentioned. Digital twins are virtual replicas of physical systems, products, or processes, created using sensors, IoT devices, and other data-gathering tools to capture real-time data that is used to create a digital representation (Grieves & Vickers, 2017). They have become increasingly important in manufacturing as they enable companies to test new products, optimize production processes, and reduce downtime. By simulating the behaviour of a physical system in a virtual environment, companies can better understand the system's performance and predict how it will respond to changes in the real world (Digital Twin Consortium, 2022). Digital twins are particularly useful in manufacturing where they are used to create a virtual representation of a product or system that can be tested and optimized before the physical product is built. This helps to reduce the time and cost associated with physical testing and prototyping (Nikolakis et al., 2019). They can also be used to monitor and optimize the performance of machines and production lines by gathering data from sensors and other sources to predict when maintenance will be required, reduce downtime, and improve overall efficiency (Zhou et al., 2020).

In collaborative systems where humans and robots work together to complete tasks, digital twins should be used to create a virtual representation of the system including both the human and robot components. This enables researchers to test different configurations and control strategies in a virtual environment before deploying the system in the real world (Fuller et al., 2020; Leng et al., 2021; Perno et al., 2022). To better understand the studies, a general categorization around digital twins is provided (see Fig. 8). As it can be seen in this figure, two different categories existed for studies; first, to develop a digital twin model for using in a collaborative system (Schmidt et al., 2022; Wang et al., 2022b; Yi et al., 2022); second, to use digital twins in a collaborative system.

Fig. 8
figure 8

Digital twins in human–robot collaboration studies

Full size image

Real-time monitoring

In some studies, a digital twin based real-time monitoring proposed for collaborative systems. Ye et al. (2022) designed an interface based on digital twins to improve the real-time task allocation controller in a collaborative system. Lorenzo et al. (2022) suggested a framework to use digital twins as real-time production planning. Both the studies used productivity as the performance evaluation criteria. Franceschi et al. (2022) proposed a framework to use digital twins detecting production failure in a real-time monitoring controlling system. In this study, quality was considered as the performance evaluation criteria.

Evaluating

In manufacturing, digital twins are especially beneficial as they enable the creation of a virtual model of a product or system that can be refined and tested before its physical counterpart is constructed. This approach assists in minimizing the expenses and time associated with physical testing and prototyping (Pizoń et al., 2022). Research that uses digital twins for evaluation in collaborative systems can be divided into three classes: (1) evaluating the performance of cobot in the design phase, (2) evaluating the interaction between human and cobot, (3) evaluating the performance of collaborative systems.

Evaluating the performance of cobot in design phase

Lu et al., (2022b) discussed the development of a digital twin-based framework for human–robot collaboration in manufacturing. The framework is designed to be generic and modular, allowing for easy customization and scalability. The authors suggest that this framework can facilitate effective collaboration between humans and robots, leading to improved manufacturing processes and productivity. Sun et al. (2021) proposed a digital twin driven framework to evaluate the cognitive and improve the performance of cobots.

Evaluating the interaction between human and cobot

Pizoń et al. (2022) explored the use of digital twins to evaluate the integration of cobots into manufacturing systems and analyse potential human–robot interactions. Furthermore, they identified some of the benefits that digital twins can bring to collaborative manufacturing systems. However, human digital twins are still very rare in literature and real examples of human/cobot digital representations are urgently needed in the short future (Berti et al., 2022).

Evaluating the performance of collaboration system

Malik et al. (2020) used Augmented Reality (AR) to provide a complete simulation to evaluate productivity, flexibility, and safety for a collaborative assembly line. Pabolu et al. (2022), and Wang et al. (2022b) suggested a framework to evaluate the performance of a designed assembly line. Zhu et al. (2022) proposed a mathematical model for reconfiguration of assembly line to reduce cost of production and improve productivity then, they used digital twins for evaluating the new design before implementing. Choi et al. (2022) presented a mixed reality system that enables safe collaboration between humans and robots using deep learning and digital twin generation. The system integrates real-world and virtual environments to provide a comprehensive and immersive experience. The authors suggest that this system can improve collaboration and safety in various industrial applications.

Overall, digital twins are an important tool for improving human–robot collaboration in manufacturing and other applications. As the technology continues to evolve, digital twins are likely to become even more important in collaborative systems, enabling humans and robots to work together more effectively and efficiently (Grieves & Vickers, 2017; Zhu et al., 2022; Sun et al., 2021), inside a socio-technical ecosystem.

Research agenda and conclusions

According to the performed literature review, Cobots have attracted much attention in the previous 8 Years (from 2014 to 2022). Although the first publication in this field was published in 1994, most subsequent articles have concentrated on the mechanical aspects of enhancing Cobots, while the interaction between humans and Cobots appeared only recently (Cencen et al., 2018). One of the most important reasons for this shifting in the research focus is the safety issues that arise when Cobots are implemented in a human–machine sharing environment (Costanzo et al., 2021). Since the robots were originally isolated in the manufacturing system, the safety procedures were limited only to some specific situations. After introducing the Cobots to the industries, the safety procedure needs to change significantly (Malik & Bilberg, 2019a). With time, new Cobots have been designed in a better way and the safety, flexibility, and productivity of new Cobots are considerably higher than the old ones (Romiti et al., 2021; Lee et al., 2022; Yi et al., 2022). However, as our study has shown previously, the number of studies in this field is still insufficient to cover all the questions and to consider the human factors in a comprehensive way. To identify open questions in the field of cobots and human–robot collaboration in manufacturing systems, the authors analysed all the selected studies and summarized the opinions of the authors in the conclusion of each study.

Hereafter, our findings are discussed more in detail around the major research areas that were emphasized by the authors because of their priority or potential for further investigations.

Performance assessment criteria

Based on the discussion provided in Tables 1 and 2, a set of performance indicators have been used to evaluate cobots. Productivity is the most used performance criterion to evaluate the performance of a collaborative system in all phases of the process, including the robotic, programming, and operation phases. Other studies have measured safety and flexibility indices in the robotic or programming phase, however, there appears a lack of studies concerning the assessment of safety and flexibility performance indices in the cobot operation phase. Additionally, most studies that have considered ergonomic, safety, and flexibility have provided only qualitative frameworks, and a lack of quantitative measures can be detected. In literature, there is a strong need for new quantitative approaches able to jointly measure the ergonomic, safety, and flexibility performances of a collaborative manufacturing system in all the phases of the manufacturing/assembly process.

Collaboration scenario

The most crucial difference between robots and Cobots lies in their ability to collaborate with humans. Real collaboration occurs when the supportive collaboration scenario is in play, where the worker and the Cobot work together to complete a task, resulting in the highest level of collaboration (Zhang et al., 2021b). However, as mentioned in Section "Flexibility" and summarized in Table 3, the number of papers that consider this scenario is still very limited. To fully understand the potential of the collaboration between humans and Cobots, it is essential to study different collaboration scenarios and compare them in real-world situations. There is also a lack of studies considering more than two collaboration scenarios together. This complexity makes it difficult to understand the interplay between different collaboration scenarios (Keshvarparast et al., 2022). Moreover, most of the existing methodological frameworks for designing collaborative assembly lines do not consider collaboration scenarios. Mathematical models, on the other hand, have shown that collaboration scenarios can lead to different optimal designs for assembly lines (Keshvarparast et al., 2022). Future research is needed not only to create new models and approaches but also for comparing different human-cobot collaboration scenarios in real-world situations, in order to fully validate the findings from mathematical models and theoretical studies in real settings. Different collaboration scenarios should also be compared and assessed according to various performance metrics, such as productivity, efficiency, and worker satisfaction.

Task designing and classification

With the introduction of collaborative robots, it is necessary to design tasks that are more suitable for human–robot collaboration (Rega et al., 2021). This means considering the unique capabilities and limitations of both humans and robots, and designing tasks that can be performed safely and efficiently by both (Gualtieri et al., 2020b). By doing so, one can take advantage of the strengths of both humans and robots and create a more effective and efficient collaborative process. However, the lack of sufficient research in this area which is shown in Appendix Table 6 means that many tasks are still being designed in the same way as before, without considering the potential benefits of human–robot collaboration. Therefore, it is essential that researchers focus on studying the differences between traditional task design and task design for human–robot collaboration. By understanding these differences, more effective methods for designing tasks, that are optimized for human–robot collaboration, can be developed. Furthermore, task categorization is also important for assigning tasks to humans or cobots and selecting the best collaboration scenario. This involves considering different aspects linked to each task and to the resource who will execute the task (Bruno & Antonelli, 2018) as task time, value-add time percentage, ergonomic workload, safety level, etc. By categorizing tasks according to specific parameters, it can determine if a task is best suited for human-cobot collaboration or should be performed exclusively by humans or robots (Antonelli & Bruno, 2019). This can help to support the operation manager decision making, optimize the collaborative process, improve performance, and ensure safety.

Workforce diversity

The importance of considering human factors in cobot-human collaboration cannot be overstated (Bogataj et al., 2019, Katiraee et al., 2021; Neumann et al., 2021). The fact that workers can have differing ages, genders, skills, and physical characteristics means that the impact of these factors on the overall production process and the human-cobot relationship must be thoroughly studied. The field of human–robot interaction has only just begun to address these issues. However, as it is mentioned in several new studies, there is much room for exploration and advancement (Schonberger et al., 2018; Gualtieri et al., 2020a; Dalle Mura & Dini, 2022; Petzoldt et al., 2022; Keshvarparast et al., 2022). The effect of differences in age, gender, physical measures, and skills on Cobot acceptability by humans must be explored. This will help us better understand how to design collaborative systems that consider the unique needs and preferences of each worker (Li et al., 2021c, Mura & Dini, 2023). When workforce diversity is considered, also different learning effects need to be investigated. Research on learning curves in collaborative systems is limited, leading to an incomplete understanding of the human–robot interaction and its impact on system performance. This highlights the need for further research in this area to gain insights into the differences in learning and forgetting curves based on worker diversity (Cohen et al., 2022). Understanding the impact of worker diversity on the learning and forgetting curve in collaborative systems can lead to increased efficiency, productivity, and worker satisfaction.

Ergonomic assessment and impact on injury reduction rate

The delegation of uncomfortable and heavy tasks to a collaborative robot has been suggested to improve the ergonomic quality of work by avoiding awkward postures or tiredness caused by repetitive load (Mateus et al., 2019). This approach can solve a multi-objective job allocation problem for humans and Cobots (Zhang et al., 2021b). However, there is limited research on the true ergonomic quality level of work, when it is completed in partnership with a cobot, and further investigation is required to support the assumption that a cobot will always increase the ergonomic level of a task and reduce injuries risk. This limitation is clearly shown in Appendix Table 6. Only Realyvásquez-Vargas et al., 2019, and Karaulova et al., 2019 investigated the ergonomic indexes in a comparative case study and analysed the difference between manual assembly line and collaborative assembly line by an ergonomic point of view. Recent studies assume the ergonomic index for tasks assigned to Cobots equal to zero (Stecke & Mokhtarzadeh, 2022; Li et al., 2022; Dalle Mura & Dini, 2022). However, this assumption neglects the fact that Cobots can create a new ergonomic load on workers (Lacevic et al., 2022; Lanzoni et al., 2022; Mura and Dini, 2023). In addition, some tasks can be done in different ways in the presence of cobot in supportive mode, so ergonomic indexes are not equal to zero. This highlights the need for additional research to accurately assess the true ergonomic quality level of work in collaboration with Cobots by coupling postural, fatigue and cognitive ergonomic assessment.

ML\RL in human pose prediction and task scheduling

The use of Machine Learning (ML) algorithms as the decision-making system for cobots is an important aspect of their functionality. Cobots use motion capturing systems and motion sensors to predict the movements of their human co-workers, which helps to improve their collaboration and increase productivity. Reinforcement Learning (RL) has also been widely used in cobot controllers to optimize training and improve performance. As it can be seen in Appendix Table 6, recently some online scheduling studies used reinforcement learning such as Alessio et al., (2022). Additionally, Liu et al., (2022b), and Zhang et al., (2022b) used reinforcement learning to optimize an assembly line balancing problem. However, more detailed investigation should be conducted to verify the results. Therefore, there is a lack of research in the literature on the use of Machine Learning and Reinforcement Learning to optimize the implementation of cobots in manufacturing systems.

Real data collection

The shortage of practical case studies able to provide to the research community a real data collection and open-source databases for supporting future research, method validation and tuning is evident in the literature. To better understand the implementation of cobots in collaborative manufacturing environments, more thorough and comprehensive studies must be conducted in real manufacturing scenarios. This includes not only laboratory testing with human and cobot participants, but also real-world case studies and practical applications that provide a more realistic view of how cobots are impacting the manufacturing industry. Currently, only three studies (i.e., Gil-Vilda et al., 2017; Navas-Reascos et al., 2022a) provide data that can be used to guide future research, highlighting the need for more robust and inclusive data sets to support the growth and development of cobots in the manufacturing industry.

Sustainable human-cobot collaboration

Finally, it is important to consider the sustainability of the human-cobot collaboration in the shared working environment. If humans and robots must cooperate in the same workplace, they will mutually affect and complement each other. Here, the wellbeing of the worker and his behaviour and status parameters will be strategic to react with changes in the scheduling and balancing of the working tasks. The development of digital representation of the workers by collecting real time data regarding workers’ status, well-being, health and safety parameters (including postures and fatigue as previously discussed) will be strategic to support the operation manager decision making in the near future. Digital ergonomics tools, biosensors and wearable sensors coupled with ML techniques will be also strategic in this context to design a digital twin for human workers and create effective and sustainable collaborative working environments (Calvo & Gil, 2022; Lin & Lukodono, 2021; Berti et al, 2022). As it can be seen in Appendix Table 6, currently, a few studies investigated a sustainable collaborative system and all of them assert that more studies required to fully developed this issue, also some suggestion provided by these studies (Calvo & Gil, 2022; Gualtieri et al., 2022). As a consequence, the integration of key concepts from the human factors engineering discipline will be strategic to assess cobots use and benefits in the context of Industry 4.0 (Neumann et al., 2021). The so-called “side effects’ of the technology” needs to be investigated since there might be side effects associated with the worker comfort and trust of working with a cobot for an entire working shift of 8 h (Neumann et al, 2021).

Conclusions

This study conducted a systematic literature review to explore the integration of collaborative robots (cobots) in manufacturing systems. Accordingly, we searched for papers using the research questions' definitions and the corresponding keywords in two databases (Scopus and Web of Science), resulting in a sample of 438 articles. We then selected 202 papers based on extraction criteria, and an additional 19 papers were found through the snowball approach. The selected studies were classified based on their subject of study, methodology, performance criteria, and collaboration scenarios. Our findings revealed that productivity was the most used performance criterion, while flexibility was the least used due to the challenges in evaluating it. Moreover, collaboration scenarios were often overlooked in the selected studies, leaving gaps in our understanding of the impact of different scenarios on cobot performance. Our analysis of the literature identified several key contributions to the state of the art, including new approaches to cobot design and deployment.

Our study has practical implications for practitioners and researchers working in the field of manufacturing systems. By categorizing studies based on different industries or usage, our findings could guide the selection of the most suitable performance criteria for each cobot application. We carefully analyzed the context of analysis and categorized the problems for selected papers to better understand the frequent issues that studies previously faced. Additionally, we investigated in detail the collaborative scenarios and the importance of digital twins in collaborative systems.

Considering the elaborated research agenda, the following topics could be promising areas for future research, such as evaluating the impact of collaboration scenarios on cobot performance. Finally, it is worth mentioning that this study is not without limitations. For example, while we provided a general overview of performance criteria, future research could explore these criteria in greater depth. Additionally, further research could investigate the use of cobots in specific manufacturing contexts, such as warehouses.