Introduction

Meeting increasing global food demands, while sustaining agri-ecosystem services, protecting biodiversity and mitigating climate change, requires an extraordinary transformation of agriculture production. Agri-innovations, including new and novel technologies, products, services, practices, organizational and business models will be the driving force in making progress in the transformation of the agrifood system [1]. Agri-innovations are credited with enhancing resource efficiency, reducing labor costs, contributing to climate adaptation and mitigation, enhancing environmental sustainability, improving crop yields, preserving ecosystem services and biodiversity, increasing land productivity, driving economic growth and competitiveness, and contributing to food security. With recent promising advances in agriculture science and technology, the importance of driving adoption of agri-innovations has garnered renewed interest within academic, policy and extension communities. The research presented in this paper was informed and guided by the research question: What are the determinants of adoption of automation, robotics and precision agriculture technology in the agriculture industry?

Adoption is not merely a simple decision to acquire a new technology or implement a new production practice. It involves a complex interplay of factors across multiple temporal and spatial and organization scales, such as economic and market constraints, technological compatibility and complexity, access to information and advisory services, policy and governance support, and infrastructure availability, social and cultural influences, and demographic factors. Numerous studies have highlighted the role of economic feasibility in shaping the adoption of innovative technologies or practices. This includes the cost of technology, the availability of financial support or incentives, and the potential for increased profitability or efficiency [2,3,4,5,6]. Government policies and regulations significantly influence adoption. Policies that support sustainable practices, provide incentives, or promote certain technologies can drive adoption [7,8,9]. Social norms, cultural practices, and peer influence are important determinants of adoption [4]. Farmers are more likely to adopt new practices if they see them being successfully implemented by their peers or if there is a cultural shift towards more sustainable or technologically advanced practices [3,10]. Farmers’ attitudes towards climate change or sustainability influence their willingness to adopt eco-friendly practices and technologies [3,6,11]. Access to information and advisory services are linked to adoption [2,7,11,12]. The compatibility of new technologies with existing farming practices and the complexity of these technologies are key factors in adoption; as are farmers’ beliefs about the benefits and drawbacks of these technologies and practices [3,6,10,13]. Adequate infrastructure, including reliable utilities and communication networks, influences the adoption of new agricultural technologies [2,4,5,9]. Demographic factors like age, education level, and farm size can influence the adoption of innovative technologies and practices [2,3,5,9,10,11,12,13] found that religious beliefs influenced the adoption of digital technologies in the beef industry. These factors interact in complex ways to shape farmers’ attitudes and intentions towards adopting innovative practices and technologies in agriculture.

Robotics and automation technologies stand as pivotal agri-innovations, garnering increasing attention due to the promise for transforming agricultural practices. These technologies include precision agriculture (e.g. variable rate application, GPS, remote sensing technology), digital technologies, information and communications technologies, automation equipment, artificial intelligence and machine learning, cloud computing and data analytics technologies, as well as tracking and monitoring technologies, data management and decision support systems. These technologies are integral to enhancing agricultural sustainability, efficiency and productivity by optimizing resource use, and addressing the challenges of global food security, environmental sustainability, agri-ecosystem services, biodiversity and climate change [2,3,4,14]. Additionally, automation and robotics enable more accurate and data-driven decisions and targeted interventions [4,5,10]. This can lead to improved crop yields, reduced use of inputs such as water, fertilizer, and pesticides and better overall farm management [14,15].

Despite the recognized potential, the actual adoption of robotics and automation technology in agriculture is relatively low [3,14]. For example, adoption of unmanned aerial vehicles and robots among farmers in Germany, United States, and Australia ranges from 4% - 22% [3]. This disparity reveals a notable gap in the literature related to understanding the determinants influencing the adoption of robotics and automation in agriculture. Questions remain about what drives or hinders the adoption of these technologies and how these factors interact in diverse agricultural contexts. Two critical gaps persist: 1) The lack of consensus in the literature on the determinants of adoption, which leaves a fragmented understanding of what drives or impedes the process [13,15]; 2). A particular scarcity of research focusing on understanding the determinants of adoption of nascent technologies such as robotics and automation [3,10]. Understanding these determinants is critical to harnessing the full potential of these innovations.

This paper addresses these gaps by employing a mixed methods, narrative, interpretive knowledge synthesis with thematic analysis to curate and synthesize the existing research on the determinants of adoption for robotics and automation technology in agriculture [16,17]. This method of knowledge synthesis systematically integrates both qualitative and quantitative studies, using a narrative inductive method of analysis. It involves qualitative description, interpretation and synthesis to identify key and recurring themes within the dataset [17,18]. This approach was chosen for its ability to comprehensively integrate diverse studies and identify a coherent set of determinants, thus providing clearer guidance for future research, policy-making, and practical application in this rapidly evolving field [19]. By determinants, we consider factors that influence and shape the adoption of automation and robotics technologies, including barriers and drivers.

Mixed methods, narrative, interpretive knowledge syntheses are valuable tools for identifying and mapping what is already known about the determinants of adoption and where gaps in knowledge exist [16,18,19]. Such knowledge syntheses are critical not only for developing theories and conceptual frameworks for adoption, but for developing policies, strategies and programs that mitigate risk and accelerate technology adoption [19,20]. The research was informed and guided by this research question: What are the determinants of adoption of automation, robotics and precision agriculture technology in the agriculture industry?

We analyzed a dataset of 72 articles published between 2017–2021 that considered the determinants of adoption of robotics and automation technologies in developed agricultural industries. Based on this analysis, we identified 13 determinants of adoption.

The paper is organized as follows. The next section describes the method used for the narrative synthesis, including characterization and thematic analysis. The descriptive results of the characterization and 13 determinants of adoption identified through the thematic analysis are described in the Findings section. We highlight key insights and implications of our findings in the Discussion section. The findings, insights and implications presented in this paper will be of interest to agricultural researchers, agribusiness, general farm and commodity organizations, agricultural service providers and agricultural policy makers.

Methods

We used a mixed methods, narrative, interpretive knowledge synthesis [16,18,19] with thematic analysis [17] approach to search for and identify the relevant literature and identify determinants of adoption of robotics and automation technology in agriculture. A mixed methods, narrative, interpretive knowledge synthesis is a qualitative approach to a systematic review that produces an account of a body of knowledge based on a formal analytical process to generate new insights or concepts [16,18,20]. The method follows a rigorous process that includes systematic searching, article selection, data extraction, coding and synthesis [21]. The key characteristics of the method are rigour, objectivity, repeatability and transparency [21]. The design and methods were adapted from the Collaboration for Environmental Evidence’s (CEE) guidelines for evidence synthesis in environmental management [21,22,23].

Search strategy

The search string included four components which were combined with the Boolean operator AND: 1) agriculture (agricult*); 2) technology (robot*, automa*, precision agriculture; 3) adoption (use of, using, use, adopt*); 4) determinant (barrier*, challenge*, driver*, constrain*, outlook, factor*). Seven databases were searched: Web of Science, Proquest, Ebscohost, CAB Direct, Agricola, IEEE Explore and AGRIS. The search string was adapted to the format of each database. Boolean operators (AND, OR) and wildcards (*) were used in the databases where available. The searches were restricted to those studies published in English from 2017–2021. Table 1 provides the final search parameters and their corresponding database.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Search parameters used in each database.

https://doi.org/10.1371/journal.pstr.0000110.t001

Article relevance screening, characterization & thematic analysis

Search results were uploaded to Endnote and then exported to Covidence. The searches identified 6,548 studies. Of these, 1,882 were removed as duplicates. Two independent researchers reviewed the articles through a two-stage process to select relevant articles for the final dataset. The first screening process involved reviewing the article title and abstract using predetermined inclusion criteria (Table 2). The initial screening included 4,666 studies, of which 4,367 were deemed irrelevant based on the inclusion criteria (Table 2). All articles meeting the inclusion criteria were included in the second-stage screening, where the full text articles were reviewed using the same inclusion criteria. The remaining 299 studies were included in the full text screening, of which 219 were deemed ineligible based on the inclusion criteria (Table 2) and eight for which full text was not available.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Inclusion and exclusion criteria used to screen literature search results to identify eligible articles for the characterization dataset.

https://doi.org/10.1371/journal.pstr.0000110.t002

Our data generally included two broad groups of articles: empirical case studies and literature reviews. Case studies either collected and analyzed primary data using surveys, observation, interviews or focus groups OR analyzed secondary data. Case studies had to use data from countries or regions with institutional structures and agricultural production systems similar to Canada, including Australia, Europe, New Zealand, the United Kingdom, Canada or the United States (Table 2). Literature reviews were included if it drew most of its data from the target countries (Table 2). We did not use authors’ institutions as a criterion for inclusion.

The final dataset included 72 studies (S1 Table). The complete process is summarized in Fig 1. Conflicts between the two reviewers were resolved by consensus.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow diagram of synthesis process.

https://doi.org/10.1371/journal.pstr.0000110.g001

Articles in the final dataset were characterized, which involved extracting descriptive information and metadata (including key words, type of study, type, name and date of publication, study location, study design/methods, theoretical/conceptual framework, commodity, technology), and identifying and assigning potential codes for thematic analysis. Articles were assigned a unique identification number in the Covidence platform. We use these identification numbers for in-text citations surrounded by round brackets (). The dataset with the identification numbers is provided in S1 Table.

The thematic analysis involved coding each full text article in the final dataset using both deductive and inductive codes. The deductive codes were drawn from a prior reading of the literature on adoption of agricultural innovations, as well as our prior research [24,25] and were largely used for characterizing and prioritizing the final dataset. The inductive codes were generated through an iterative and interpretive process of reading and re-reading each articles in the prioritization dataset to identify potential determinants of adoption. The codes generated through the inductive process were reviewed, combined and integrated through an interactive, iterative and interpretive process to construct a set of 13 determinants of adoption[17]. The codes associated with each of the determinants are provided in S2 Table.

Findings

Descriptive results

This section summarizes the results of the characterization process.

Most of the 72 articles in the final dataset were empirical studies (67%), 27% were review articles and 6% were opinion or editorial (Fig 2). Most of the articles were published in peer-reviewed academic journals (79%), 10% were conference papers, 8% were from the grey literature and 4% were book chapters (Fig 3). Of the articles published in peer reviewed journals, more than half (51%) were published in six journals: ten were published in the journal Precision Agriculture, five were published in the Journal of Rural Studies, four each were published in Sustainability, Animal: the international journal of animal biosciences and the Journal of Dairy Science and two papers were published in Agronomy. Most of the articles were published in 2020 (29%) and 2021 (25%) (Fig 4). In terms of the commodities of interest, 25% of the articles considered agriculture in general, 53% focused on crop production with some looking at specific crop such as fruit, viticulture, field crops, corn, rice, cotton, apples, soybeans, wheat, and potatoes. The remaining 22% considered adoption in livestock production, which included dairy and sheep.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Types of articles in the dataset.

https://doi.org/10.1371/journal.pstr.0000110.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Types of publications in the dataset.

https://doi.org/10.1371/journal.pstr.0000110.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Per cent of articles published by year.

https://doi.org/10.1371/journal.pstr.0000110.g004

There was a wide range of technologies represented under the rubric of automation and robotics, which were not defined consistently across the dataset (Table 3). This was especially obvious in terms of proposed typologies. For instance, while some typologies included AI or drones, others did not. While this might reflect disciplinary differences, we still contend that a more uniform technology typology and definitions would make it easier for researchers and users to better understand determinants of and trends in adoption. Using ChatGPT 4.0, we make an initial attempt to organize the technologies listed in Table 3 in a typology that we call data-driven agricultural technologies (Table 4.). We propose seven classes of technology within the typology: Automation and Robotics, Sensing and Monitoring Technologies, Information and Communication Technology (ICT), Artificial Intelligence and Machine Learning, Precision Agriculture Tools, Remote Sensing and Aerial Technologies, Digital Decision Support Systems. The prompts used and dialogue with ChatGPT4.0 that generated the classification and typology are provided in S1 Appendix.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Technologies of interest identified in final dataset.

https://doi.org/10.1371/journal.pstr.0000110.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Classification framework of technologies identified in final dataset, which represent a typology best defined as data-driven agricultural technologies generated with the use of ChatGPT 4.0.

https://doi.org/10.1371/journal.pstr.0000110.t004

Farm characteristics and surrounding physical environment

Farm characteristics include features of the farm and its physical environment. Frequently mentioned farm features included: size (e.g., acreage, output, profit, number of animals) (283; 672, 973, 2367)); type (e.g., crops, livestock) (1352); organizational form (e.g., own or lease land) (3664); practices (e.g., traditional or organic) (2171); and built environment (e.g., dairy parlours, RFID fence systems) (794, 2970, 4595). The physical environment split along local and regional features. Local features included soil type, soil heterogeneity, field slope, elevation, drainage patterns, and terrain uniformity (2836, 3855, but cf. 2839). It extended to local climactic features, principally precipitation and temperature patterns, but sometimes prevailing winds (2880). Regional features encompass climate (i.e., seasonal regularities in wind, precipitation and temperature patterns) (2367) and less frequently, surface conditions (e.g., soil types) (2171). Within these seasonal and climactic parameters farmers then specialize (5874, 6141), often leading to the rise of regional specialization in particular forms of livestock or crops.

The reviewed literature in this determinant suggests that larger farms are more prone to adopt automation and robotics technology (2367). In this literature, common measures of farm size included annual revenue, profit margins, land under cultivation, head of livestock and number of workers (3252; 4838). Explicitly and implicitly, these explanations seemed to hinge in part on the importance of scale economies (3855, 4792, 6015 but cf. 3664 and 4595). However, size did not always matter (4595). Some agricultural commodities exhibit few opportunities for scale economies such as organic production of crops and livestock. More generally, the findings were heterogenous, making it difficult to ascribe causality between farm characteristics in general and the decision to adopt. A clearer pattern emerged when examining a particular kind of agricultural operation, with dairy cow milking seeming to provide the clearest patterns (4595). However, extrapolations from those findings beyond that industry should only be done cautiously.

Farmer characteristics

In the reviewed literature, farmers were often implicitly conceived of as entrepreneurs, managers and stewards. In turn, farmers were described in terms of common sociodemographic variables such sex/gender (4595), age, years of experience as farmer, their generation as farmer, formal education, and full- or part-time status as farmer (2171). Sometimes farmers were explicitly characterized by other adoption specific variables when they were asked about, for instance, their reasons for not adopting (6015), openness to adoption (2880, 3664, 5665), sources of information on adoption (3252, 5694), and frequency with which they sought out this information (3252).

Compared to farm characteristics, the reviewed literature offered fewer certainties in determinants of adoption. Some found that younger farmers were more likely to adopt (973, 2367, 3252, 6015), but not necessarily to have previously adopted (4595). Education levels tended to be a positive determinant (2880, 3252, 6015). Status as full-time farmer, when explicitly measured, was the most uniformly mentioned positive determinant (896, 2171, 4595). Given full-time status would seem to be both easy to measure and important to know, we wonder why more studies did not measure (6141) full-time status or if they did, report its association with adoption (4595). Familiarity with computers was often positively associated with adoption (896).

Policy and regulation

Policy and regulations play a significant role in the adoption of robotics, automation, digital farming, smart farming, and precision agriculture technologies (5092; 6015). As one manuscript notes, “Agriculture is expected to benefit from robust regulatory and institutional policies that promote both a national and international agenda for PA adoption" (2360 p.1033). Policy incentives, both financial and non-financial, play a crucial role in influencing farmers’ intentions to adopt digital agricultural technologies (608, 3375, 6015). Reductions in technology costs, government interventions, training support, and technical assistance through policy can encourage adoption. Agricultural policies can create incentives or disincentives for adopting digital agricultural technologies. Regulations can influence the adoption of agricultural technologies by providing incentives for adoption, implementing penalties for pollution, and using digital technologies to verify compliance with environmental protection programs (2189, 2360). Regulatory demands can also act as barriers to the adoption of agricultural technologies. There is a need for targeted policies and performance-based approaches to optimize technology adoption, particularly in areas where they can provide significant benefits at minimal cost (6015). Policy choices are also necessary to meet the ethical challenges likely to arise as agricultural robots become used more widely, and to maximise the social, environmental, and economic benefits of robotics in agriculture (1157).

Labour’s absorptive capacity

Labour’s absorptive capacity corresponds to human and institutional factors that enable successful adoption and continued operation of agricultural automation and robotics. Frequently these factors involve the larger context in which farmers and farms find themselves (2726, 2836, 5091, 5335). Many sources cited labor shortages (both in terms of number of workers and their skillset) as a problem that robotics and automation could solve (2970, 4838, 6476). Usually, worker availability is a precondition for skills availability; farmers will not adopt a technology if they cannot find someone to operate it (2839, 3375, 5130, 6477). Both specific (e.g., basic computer literacy, familiarity with poultry) (5268) and general (e.g., problem-solving combining mechanical, electronics and farming skills) (4714) skills were identified. Most sources recognized that learning to use these new technologies was difficult and took a long time (3257, 6061). Cost-reimbursement programs (6105), availability of outside help (e.g., sales staff, consultants, extension agents, other farmers, university technical assistance) (1932, 2171, 3252, 5694, 6015) and alternate acquisition models (6141) were frequently mentioned as institutional determinants of absorptive capacity. However, farmers’ own confidence in the technology (2887), familiarity with related technologies (1641, 2880), and norms favoring adoption (5130) were individual determinants of absorptive capacity. Scale-neutral benefits (i.e., the benefit did not scale with output, meaning smaller farms could also adopt the tech) (3855), speed of technological change, and benefit of incremental adoption (i.e., significant benefits arise from acquiring a given technology without need to restructure or upgrade technology in up and downstream tasks) (6061) were technology-specific determinants of absorptive capacity.

Social elements

The adoption of agriculture technologies is influenced by a complex interplay of cultural, societal, and ethical factors that shape farmers’ attitudes, their willingness to adopt and their perspectives on the values and risks associated with robotics and automation technology (1157, 3855). Attitudes towards environmental stewardship, social approval, and work-life balance can be influenced by cultural and societal norms. These attitudes, in turn, impact the adoption of technologies (3252). Adoption decisions can be influenced by the adoption decisions of other farmers and other stakeholders who support adoption (3855, 608). This suggests that social normative pressures act as determinants of adoption (608).

Societal imperatives such as food security, climate change, biodiversity loss, and sustainability are driving the demand for sustainable and resilient agricultural systems (5091). Digital technologies are seen as essential tools to address these social imperatives (1352, 5091). These technologies not only offer private benefits to farmers but also contribute to the public good.

Ethical considerations also play a role in the adoption of agricultural technologies (1148). The use of robots in agriculture raises questions about environmental sustainability, economic decisions, and social impacts (1157). Farmers must grapple with the ethical implications of their technology choices, such as the potential environmental consequences of certain practices and the social impacts on labor dynamics (1157).

Interoperability

Interoperability plays a significant role in the adoption of agricultural technologies. Interoperability refers to the ability of different systems, devices, and applications to access, exchange, integrate, and cooperatively use data in a coordinated manner, within and across organizational boundaries, to provide timely and seamless portability of information, leading to improved agricultural practices and outcomes (6476). In the context of agriculture, interoperability is crucial for the effective sharing and utilization of data across various platforms (1349, 1352, 3855). For instance, machine data may be stored in an equipment company’s cloud, weather data may come from a public source, genetic information may reside in the seed company’s service database, and chemical application data may come from a service provider (6476, 3855). The ability of these disparate sources to share and integrate data effectively is a key aspect in the efficient sharing of data among multiple stakeholders, which is crucial for generating useful information for decision-making in the agricultural sector (6476, 1403, 1349, 1352, 3855). The urgency of interoperability is highlighted by (6476): "Regardless of whether there is ’public’ sharing of data, there is always a need for interoperability. This is currently a dire need in both cropping and livestock systems.” (p. 13). Interoperability creates value by linking data within the farm and across the value chain (5091, 3855). This suggests that interoperability is not just about the technical ability of systems to communicate, but also about the meaningful exchange and interpretation of data (734, 3855).

However, achieving interoperability in agriculture is a challenge due to the existence of assorted data formats, naming conventions, and equipment compatibilities (6476, 283, 2360, 3855). To overcome challenges, concerted efforts are required to establish standardized APIs, promote open-source initiatives, and explore innovative data models (2078, 1403, 6476).

Standards

Standards serve as a pivotal determinant in the adoption of robotics and automation technologies. They play a role in building trust and fostering accountability, particularly when addressing concerns related to data privacy, security and interoperability (1403, 2078, 2365, 5092). By providing clear guidelines and delineating responsibilities, standards can facilitate trust and collaboration among farmers, technology-solutions developers, other industry stakeholders and consumers (734, 1403, 3375, 5092). Harmonized data standards promote transparency, and enhance data quality, facilitating more valuable insights and knowledge extraction from agricultural datasets (734, 1403, 2078, 3375). Standards address liability concerns, ensure consistent performance and quality, and facilitate seamless integration with existing production systems, ultimately contributing to safer, more efficient, and sustainable agricultural practices (1403, 6476). Furthermore, standards can play a role in regulating stakeholder behavior, accelerating product design and development and facilitating investments in technology development 6476, 3375). A key takeaway from the literature is that the current lack of standards acts as barriers to adopting automation and robotics technologies (734, 2078, 3375, 5092).

Access to information

Access to information plays a pivotal role in technology adoption, as it affects farmers’ decision-making processes and their ability to understand and implement new agricultural practices (3252, 3855, 5694). Farmers rely on various information sources, such as agricultural media, other farmers, extension services, training courses, internet/social media, exhibitions, and discussions with peers, consultants, trade organizations, relatives, agricultural service providers and commercial vendors, to access knowledge about new technologies (794, 3252, 3375, 3855, 5092, 5694). Access to information encompasses various elements, such as awareness, learning, trust, and the availability of specialized advisory services (1352, 3252, 4792, 5092, 5694). Public and private research, extension and advisory organizations play crucial roles in supporting and promoting technology adoption through knowledge sharing, training, and standardization efforts (794, 1641, 3252, 4792, 5092, 5694). Collaboration between industry, academia, and extension programs is essential for technology adoption (4792, 5092, 5694). Lack of sufficient technical support can hinder the adoption of new technologies (1641, 3855).

The time taken to learn about how to use a technology also influences adoption levels (3855). Information acquisition happens gradually, and time to learn about how to use a technology also augments adoption levels. Awareness is a necessary condition for adoption; farmers who are unaware of a technology will not adopt it (5092, 5694).

Neutral extension services are important for building trust in the technology, lowering learning costs, and protecting farmer interests (3855). Trust in extension, agriculture consultants, and technology solution providers significantly influences the adoption of agricultural technologies (3855, 4792). This trust is necessary for information exchange, learning, risk management, and ultimately, the successful implementation of innovative technologies (3855, 4792, 5092). Growers trust the expertise of researchers, extension specialists, agriculture service companies or consultants to share information that facilitates the adoption of new technologies (3855, 4792, 5092). Farmers need assurance that solutions developed by technology providers will solve actual problems, perform as claimed, improve the profitability of their farms and reduce risks (1641, 4792, 5092, 5694).

Operational benefits

Farmers consider several operational costs and benefits when deciding to adopt robotic and automation technologies. On the benefit side, farmers are looking for practical benefits from robotics and automation technology that are linked to their operations and production (1349, 1403, 1641, 2374, 2892, 4792). This includes tangible economic benefits such as cost savings, time savings, workload reduction, increased yields, efficiency, and productivity (794, 1403, 1789, 4588, 5665, 5694, 6476). They are also looking for more subjective benefits, such as usefulness, ease of use, reliability, quality, improved health and welfare of livestock, health and safety of staff and environmental sustainability (734, 794, 1789, 2078, 2374, 2970, 4588, 5665, 6476). Farmers want technology that solves real problems and has proven performance (3252, 4792, 6015). Ambiguous evidence for the value of automation and robotics technologies is a barrier to adoption (1352, 1403, 3855, 4792, 6015).

On the cost side, adoption often involves upfront costs, such as the purchase of new equipment or learning how to use the technology (2374, 2892, 5268). These technologies are capital intensive (1148, 2892, 3855, 4588). High startup costs with a risk of insufficient return on investment can make it challenging for producers to afford these technologies (672, 2374, 5091). The high upfront investments needed can come with the risk of locking a farmer in with a technology provider or vendor as they pay off the debts incurred (3375, 5091). Further, these technologies can increase overhead costs due to installation charges, and the time and effort spent learning how to use and maintain them (672, 794, 1403, 1789, 5335). If a farmer chooses to outsource technology services to a custom service provider, it also imposes costs (2078, 2360, 3855, 4792).

Frequently costs and benefits are assessed over multi-year time horizons, for instance, related to expected depreciation of other investments (e.g., milking barns) and farmer lifespans (e.g., number of years to retirement) (794, 1349). Overall, the decision to adopt agricultural technologies involves a careful consideration of these operational costs and benefits, with the aim of maximizing profits, minimizing expected loss and reducing risks and uncertainties (734, 1349, 1352, 1641, 2892, 3694, 5335, 6015).

Public infrastructure

Public infrastructure refers to all off-farm physical infrastructure. This includes reliable access to Internet, irrigation systems, electricity and roads. However, Internet infrastructure was most frequently mentioned, since it underpins many novel agricultural technologies. Internet here means the network of copper-wire, broadband, cellular towers and satellite systems needed to reliably provide sufficient bandwidth to use recent advances in agricultural automation and robotics. Internet was often characterized as unreliable (606, 2360, 3694), of insufficient bandwidth (606, 1007) or unavailable in rural agricultural locations (734, 5335), especially in mountains (2374). These problems extend even to basic cellular telephone coverage necessary to enable real-time farming applications on farmers’ phone (2880). This is a problem because manufacturers of novel automation and robotics technologies (e.g., real-time GPS enabled weed-pulling robots or self-steering tractors) frequently assume that end-users have sufficient and reliable Internet access. This public infrastructure was implicitly characterized as collective infrastructure, provided by some combination of public (e.g., irrigation authorities) and regulated private entities (e.g., telephone companies). Regardless of kind, collective infrastructure is a necessary but not sufficient condition for the successful adoption of many kinds of automation and robotics in agriculture. However, once that hurdle was cleared, other determinants might still then come into play to shape farmers’ decision to adopt.

Technological characteristics

The characteristics of a technology strongly influence the adoption of robotic and automation technologies (2200, 3252). Technical maturity or readiness, market availability, need-driven development, ease of use, perceived usefulness, technical complexity, adaptability and complementary technologies are technical characteristics that have been linked to adoption (608, 1148, 1352, 1641, 2360, 2367, 3252, 4792, 5091, 5092, 5665, 5694, 6015). These characteristics interact with farmer objectives and resources to influence adoption decisions (794, 2200, 2367, 5665). The adaptability of technology to individual farmer objectives and existing practices and technology is crucial for successful adoption (1148, 1352, 1641, 1789, 3252, 5091, 6015).

The maturity or level of readiness of technology was cited as a factor that influenced adoption (1148, 1641, 2892, 4792, 5091). This was related to the efficacy of the technology and whether it had been validated under operational conditions (3252, 4792, 5091). The unproven nature of technologies can be a barrier (1352, 4792, 5091). Technologies that are developed to address specific needs and challenges faced by farmers tend to be more readily adopted (1352, 1789). Farmers’ trust in the technology is essential for its adoption (794, 3855, 4792). They need to perceive a relative advantage and anticipate increased profitability from its use. Farmers’ perception of the technology’s usefulness and ease of use significantly impacts their intention to adopt it (608, 1403, 2892, 5665, 5694). Whether the technology is divisible (can be adopted incrementally) or indivisible (must be adopted all at once) has a significant effect (3252). Divisible technologies may have higher adoption rates due to their incremental adoption approach (3252). The complex nature of precision agricultural technologies is a barrier to adoption and may require farmers to seek specialized advisory and consultancy services (1403, 3252, 3855, 5092).

Uncertainty and risk

While both risk and uncertainty influence the decision to adopt agricultural technologies, they do so in separate ways. Uncertainty refers to the lack of clarity or predictability about the outcomes or benefits of a technology, which can lead to delays in adoption until uncertainties are resolved (1642, 6015). The role of risk in the adoption of agricultural technologies is complex and depends on a range of factors, including the nature of the technology, the specific risks involved, and the risk preferences of the farmer (3855, 2367, 2374). Risk is generally associated with the variability of outcomes and is often accounted for in decision-making models (2367, 3855, 5335). While risk and uncertainty are distinct concepts, they are conflated in our dataset, with risk being used as a synonym for uncertainty, which plays a significant role in the adoption of automation and robotics technologies.

Uncertainty stems from a range of factors, such as a lack of information about how to implement technologies profitably, insufficient technical support, or unpredictability about the future, making it difficult to quantify (734, 1641, 3855, 4342, 6015). Uncertainty can act as a barrier by increasing the perceived risks and costs of adopting new technologies (734, 1641, 2374, 4342, 6015). It can also act as a driver by leading farmers to adopt technologies that can help manage and reduce these risks and uncertainties (2367, 3855, 4588). Uncertainty about the benefits and/or costs of adopting a technology that involves large sunk costs can affect decisions about the timing of adoption (1641, 2367, 3855, 4342). The real option approach was used by 1641 to show adoption as a dynamic investment decision not only about whether to adopt a technology but when to adopt. If there is uncertainty about the technologies in the future, there may be gains (option-value) from waiting until uncertainties are clarified (1641). Uncertainty affects the learning process, with some technologies involving significant learning and information costs (3855, 5092, 5335).

Discussion

Considering our findings, we share seven key insights and conclusions. Furthermore, as the nature of qualitative narrative syntheses tends to be hypothesis generating, we propose a series hypotheses that could inform and guide future research on the determinants of adoption, which could have application beyond automation and robotics technologies to agri-innovations, in general.

First, automation and robotics technologies will not be suitable across all farm types, production systems and commodities. Several determinants are at play in this conclusion. Farmers and their farms, even of the same commodity in the same region, are not homogenous. Farmers vary in terms of comfort with risk, experience and adaptability. Farms vary in terms of local climate, soil, and already-existing infrastructure. This especially struck us considering the observation that this literature noted that larger and better capitalized farms are more likely to adopt automation and robotics. Therefore, we hypothesize that smaller and less capitalized farms will instead tend to adopt other agri-innovations better aligned with business realities. Likewise, we hypothesize that part-time and older farmers will tend to innovate in ways that do not involve automation and robotics. A surprising number of empirical papers did not assess the importance of a farmer’s full-time or part-time status for adoption of automation and robotics. This is perplexing because, depending on the commodity and context, the importance of full-time versus part-time status varies. Regardless of the circumstances, we recommend that future research measure or otherwise control for full-time and part-time status whenever possible. Furthermore, a farmer’s decision to adopt is a business decision based on the information they have at hand. While there are market pressures for contemporary farms to change their production techniques, even for producers of the same commodity, we hypothesize that uptake of automation and robotics will still be uneven due to farmers making rational business decisions and choosing to adopt the agri-innovations best suited to their operations.

Second, recognizing the inherent complexity of agriculture production systems, taking a reductive approach to understanding adoption of agri-innovations may hinder the transformation to sustainable agriculture production systems. Adoption is not determined on the basis of a single determinant. The determinants interact in complex, and still unknown, ways. Future research should focus on advancing our knowledge and understanding of the dynamic interplay between and among the determinants and how that shapes adoption decisions.

Third, public infrastructure is more than the Internet. Thus, it is curious that the papers we reviewed largely failed to mention other supporting public infrastructure like electric grid, roads, or irrigation systems as determinants of adoption. Obliquely, education institutions and technology transfer were mentioned, but generally not as a kind of shared-cost infrastructure. At the same time, this lacuna makes sense given that the papers were drawn from countries in the global North. Internet access might vary by location, but by-and-large, roads, irrigation networks and electric grids are well-established. Future research, especially on technology transfer of automation and robotics to countries outside the Global North, will want to bear this in mind.

Fourth, many papers highlighted the need for collective policy and governance solutions to drive automation and robotics adoption in agriculture. These collective solutions need to alter the incentive structure. However, few papers provided concrete examples of how to do this. In general, this is a problem with much scholarship on technology adoption, and in our case also reflects the article selection criteria. One-size-fits-all policies make sense in the conclusion of a scholarly paper, but the mechanics of an effective policy can rarely be compressed into a single paragraph. If the adoption of automation and robotics technology is desirable, more concrete examples of how to alter incentives to adopt would be useful for policy makers. At the same time, we recognize the trade-off between commodity-specific or technology-specific policies and agriculture-wide policies. What works for a farm producing a specific commodity will probably not be applicable to farms producing significantly different commodities (e.g., row crops vs. livestock). At the same time, agricultural policymakers are under pressure to create transparent, uniform and simple-to-monitor policies. This creates pressure for one-size-fits-all policies. In practice, we see these tensions leading to two levels of agricultural policy: those dealing with the entire agriculture sector; and those dealing with specific commodities or specific technologies. Nonetheless, we hypothesize that incentive systems tailored to specific commodities will be more effective in driving adoption of automation and robotics technology.

Fifth, trust and transparency are central to adoption. Technology developers need to take seriously farmers’ concerns about data privacy, ownership and security and technology performance and reliability. In light of this, we hypothesize that technologies that protect farmers’ privacy will be adopted at higher rates than those that do not. Furthermore, we hypothesize the same for technologies that have demonstrable, tangible benefits and are perceived to be more reliable and meet performance specifications.

Sixth, automation and robotics technologies must represent solutions to real agriculture production problems and must directly address farmers’ needs for more efficient resource use, increased productivity (i.e., yields), and enhanced environmental sustainability. Technology developers who lack an understanding of their end-users’ problems and actual needs will find it difficult to succeed in the agricultural automation and robotics market. The diversity of agriculture (farm sizes, production systems, commodities and environmental conditions) poses challenges to identifying actual problems with a scale and scope that represent a viable market opportunity worth pursuing.

Finally, we make an initial attempt to organize the diverse array technologies included in the final dataset as a typology that we call data-driven agricultural technologies (Table 4). This typology can be tested, validated and adapted through future research.

Identifying and understanding the determinants of adoption of data-driven agriculture technologies is critical for developing effective strategies and approaches that promote and support widespread uptake, implementation and long-term use of data-driven agriculture technologies. These emerging and nascent technologies play a critical role in the transition to more sustainable agriculture production systems. By taking the determinants of adoption into account, stakeholders will ensure that the promise of these technologies for building resilient food systems, maintaining valuable ecosystem services and improving resource efficiency and operational productivity will realized.

Supporting information

Acknowledgments

We would like to thank Olubunmi Okuwa for her assistance in finding and curating the full text documents of the dataset.