Human–Computer Interaction in Smart Greenhouses: A Review of Interfaces, Technologies, and User-Centered Approaches

Abstract

Human–computer interaction (HCI) is essential for optimizing smart greenhouse management and for fostering efficient and sustainable agricultural practices. A synthesis of recent advancements in diverse interfaces, including digital twins, virtual and augmented reality, mobile applications, and sensor-based controls, alongside the integration of artificial intelligence (AI), automation, and human–robot collaboration, was examined as part of advanced automation strategies. This study highlights the importance of user-centered and context-aware design to enhance usability, address challenges like simulation sickness, and cater to varied user demographics. Emphasis is placed on responsible, adaptive, and trustworthy interaction, ensuring effective decision support and promoting human–AI synergy. This review offers an integrated perspective on current developments, identifying pathways for future sustainable interaction design in controlled-environment agriculture.

1. Introduction

The global agricultural sector is currently undergoing a transformative shift, driven by the escalating need to ensure food security for a rapidly growing world population, predicted to reach 9.7 billion by 2050 (Singh et al., 2024), alongside the imperative to enhance agricultural sustainability [1,2]. Traditional agricultural methods face immense pressures from critical challenges, including resource scarcity, global climate change, and high energy demands [3,4,5,6,7,8]. Climate change, characterized by increasingly erratic weather conditions, necessitates innovative cultivation strategies that can sustain production regardless of external climatic factors [1,7,9]. Furthermore, agriculture must drastically reduce its ecological footprint, especially considering that farming and related activities account for approximately 10–29% of all global greenhouse gas (GHG) emissions, as noted by Maraveas et al. [4].

In response to these complex environmental and production challenges, controlled-environment agriculture (CEA), particularly within greenhouses, has emerged as a fundamental solution [10,11]. Greenhouses, or protected agriculture systems, provide a reliable, controlled environment, enabling year-round crop production [1,2,5,9,12]. This controlled setting significantly enhances resource efficiency by optimizing the use of inputs like water and fertilizers [2,6,12], protecting crops from adverse weather, pests, and diseases [2,6], and yielding substantially larger biomass per unit area compared to open-field environments [13].

This necessity for optimized resource management and increased yield has accelerated the evolution of traditional greenhouses into sophisticated smart greenhouses, representing the core of precision agriculture (PA) and Agriculture 4.0 [7,14,15]. Modern greenhouses are rapidly becoming high-tech factories characterized by intensive production and heavy reliance on digital infrastructure [3,16,17,18]. This industrialization is enabled by the integration of Information and Communication Technologies (ICT), such as the Internet of Things (IoT), big data, cloud computing, and artificial intelligence (AI) [3,14,15]. These technologies utilize advanced systems and networks of sensors (Figure 1) to intensively monitor and control critical processes, including climate management, irrigation, fertigation, and crop monitoring, leading to increasingly smart and data-driven operations [3,4,5,7,14,18,19].

While the promise of smart greenhouses lies in their automation and data capacity, the transition to such technology introduces a critical dependency on human–computer interaction (HCI) [20,21]. According to Pino et al., HCI is dedicated to the design, evaluation, and implementation of computer systems tailored for human use by investigating the phenomena associated with this interaction [20]. In the agricultural context, Ibrahim’s study highlights that the user interfaces act as the essential bridge between human operators (farmers, agronomists, and technicians) and complex technological processes [21].

The effective incorporation of HCI principles is overwhelming for ensuring efficiency, usability, and widespread adoption of smart greenhouse technologies [21,22]. Modern, data-driven horticulture relies on decoupling physical flows from management, allowing growers to plan, monitor, and control operations remotely based on real-time digital information, thereby removing traditional constraints of laborious manual observation [3,16]. However, this automation mandates a profound restructuring of professional skills and cognitive processes. Because of this, farmers must integrate and interpret data outputs from sensors and control the functioning of actuators through mediated interfaces [20].

Most importantly, HCI addresses significant human barriers to technology uptake [23]. The increasing scarcity of skilled green labor-experienced employees with horticultural knowledge is also a growing challenge [3,16]. Although AI and machine learning can capture tacit knowledge through advanced analytics [3], if the resulting management systems are complex, unintuitive, or fail to align with practical user needs, adoption rates stagnate [21,23,24]. For example, studies like those conducted by Brenes et al. and Kavga et al. confirm that low technological literacy among certain user demographics, such as older farmers on small farms, makes innovation difficult [23,24]. Therefore, optimal efficiency and system effectiveness hinge on HCI solutions that prioritize usability and intuitiveness [20,21]. An effective HCI design simplifies complex communication between human users and automated greenhouse systems, as noted by Basyouny [25].

This review systematically synthesizes the state-of-the-art research regarding HCI in smart greenhouses, focusing on the sophisticated technologies used, the resulting interfaces developed, and the important role of user-centered design approaches.

The first objective is to provide a comprehensive analysis of the prevailing interfaces used for human interaction in these highly automated environments. This includes traditional mobile interfaces, such as Android applications and web-based platforms, which enable real-time remote data visualization, monitoring, and control of greenhouse parameters [6,14,26]. Advanced visualization tools were explored, particularly digital twins (DTs), which serve as digital equivalents mirroring the states and behaviors of real-life greenhouse objects for decision support [3,16,27]. Additionally, the rising trends in Extended Reality (XR), specifically augmented reality (AR) and Virtual Reality (VR) used for complex visualization, remote monitoring, simulation, and training programs were also investigated [13,16,28,29]. A key aspect of this analysis addresses the human factors associated with immersive interfaces, such as simulation sickness (cybersickness), which directly impacts user comfort and system evaluation for long-term deployment, according to Slob et al. [16].

The second objective is to consolidate the enabling technologies and design approaches to drive these interactive systems. This encompasses the foundational role of sensor networks (Wireless Sensor Network (WSN) and IoT) for data acquisition [5,18,19,30], the use of AI and automation paradigms (including Machine Learning, deep learning, fuzzy logic, and Artificial Neural Network (ANN) algorithms) for climate control, prediction, and optimization services [4,5,7,15,31] and the development of robotics and human–robot collaboration (HRC) systems for tasks like monitoring and harvesting [32,33,34]. Central to integrating these elements effectively is the role of user-centered design (UCD) methodologies, which ensure the technology is adapted to the complex and diverse requirements of agricultural users [7,20,21,23,34].

By comprehensively surveying the integration of HCI in smart greenhouses, this review offers several key contributions. First, it provides an integrated perspective linking emerging technologies (DTs, AR/VR, AI/Machine Learning (ML), and IoT) to their resultant interaction interfaces and the practical human factors issues they raise. Second, this work highlights crucial gaps in the current literature. Although there is rapid technological advancement, the critical HCI element for digital twins is still not widely studied [16]. Moreover, research often focuses narrowly on technological implementation or localized usability tests [21,23]. There is a noted gap concerning the long-term effects on users and the need to standardize metrics for assessing usability in agricultural settings [21]. Also, insufficient research connects specific HCI design choices to larger sustainability concerns, such as the relationship between user interaction and computing energy consumption (digital carbon footprint) [25,35,36].

Therefore, this review provides a foundation for directing future research towards adaptive and sustainable interaction design. Future research work must emphasize sustainable HCI (SHCI) to reduce the environmental impact of digital technologies [25], considering how technology use impacts GHG emissions [35,36]. It is imperative to develop interfaces that offer adaptive feedback and improved data visualization to support complex decision-making processes for diverse user groups [21]. This includes promoting accessibility and inclusivity, especially in rural areas often challenged by limited infrastructure and varied digital literacy levels [21]. By synthesizing existing knowledge and identifying these gaps, this review aims to guide the next generation of researchers toward developing reliable, user-friendly, and environmentally responsible interaction designs necessary for the sustainable growth of smart greenhouses in horticulture [4,21].

2. Materials and Methods

2.1. Search Strategy and Scope

The objective of the search strategy was to identify peer-reviewed research focusing specifically on the confluence of agricultural technology, system interfaces, and human factors within controlled-environment agriculture (CEA). The search was confined to publications released within the more than twenty-year span from 2000 to 2025, thereby capturing the emergence and rapid proliferation of the Internet of Things (IoT), artificial intelligence (AI), and advanced visualization technologies that characterize modern smart greenhouse systems [3,5,16].

The selection from the literature was executed across three globally recognized academic databases known for their comprehensive coverage in computer science, engineering, and agricultural technology: IEEE Xplore, Scopus, and Web of Science (WoS), as well as other relevant international databases [5,20,37]. This multi-database approach ensured broad coverage, like methodologies adopted in systematic reviews within related fields [5,10].

Although elements of systematic search protocols were employed, the review did not implement a full PRISMA systematic review. Instead, a PRISMA-like flow chart (Figure 2) was used to transparently document the identification, screening, and refinement of sources informing the synthesis. This approach aligns with structured narrative and semi-systematic reviews, which combine systematic search procedures with expert-guided thematic expansion when appropriate.

2.2. Keyword Formulation and Search Execution

A targeted set of keywords and phrases was utilized to construct search strings, ensuring coverage of the most important intersection points explored in this review. These keywords targeted the core domain and the technological interfaces:

• Domain and core technology: “smart greenhouse”, “automation”, and “digital twin”.
• Interface and interaction: “human-computer interaction”, “user interface”, and “user-centered design”.
• Advanced visualization: “AR/VR in agriculture”.

The search strings combined terms using Boolean logic (AND/OR), pairing core agricultural terms (e.g., “smart greenhouse”) with HCI and technology terms (e.g., “human-computer interaction” OR “user interface” OR “digital twin”). For instance, keyword combinations targeted articles addressing the fundamental necessity of user-centered design (UCD) in this rapidly automating environment [7,34,38], as well as emerging technologies like digital twins in horticulture [3,27] and the role of AR/VR in complex visualization and training [13,16].

Because the review employs a structured narrative approach rather than a full systematic protocol, keyword construction was optimized for conceptual breadth rather than exhaustive retrieval. Accordingly, Boolean queries were iteratively refined based on initial results and expert judgment, and additional terms were incorporated during the targeted narrative search to capture emerging technologies (e.g., “XR horticulture”, “human–robot collaboration”, and “immersive visualization”). All searches were completed in 2025, prior to the synthesis and analysis phases.

Full representative search strings used in the primary databases are documented within the methodological notes, while the PRISMA-like flow chart in Figure 2 illustrates how database results, narrative add-ons, and existing SLR-derived evidence were combined into the final evidence set (see Supplementary Materials).

2.3. Selection and Filtering Procedure

The initial extensive search yielded numerous documents, which were then subjected to a multi-stage filtering process to ensure high relevance and academic quality, consistent with rigorous methodology standards [39].

Relevance filtering was made by initial selection, which screened titles and abstracts for direct relevance to HCI principles and their implementation within greenhouse or controlled-environment agriculture settings.

Inclusion criteria were represented by peer-reviewed articles (journal papers, conference proceedings, and validated review articles) published in English, which were included to maintain academic credibility [4,20,40]. In a limited number of cases, high-relevance academic sources outside traditional peer-review channels, such as doctoral theses or institutional research reports, were also included when they offered substantial methodological, conceptual, or empirical insights not yet represented in the peer-reviewed literature.

Exclusion criteria were considered, and studies were rigorously excluded if they focused solely on general agricultural or climate science without explicit discussion of HCI, user interfaces, or user interaction design (excluding most non-HCI studies) [5]; did not focus on smart or automated agricultural systems (excluding most non-agricultural studies); or were non-peer-reviewed materials, such as white papers or non-academic reports.

To ensure comprehensive thematic coverage, additional targeted narrative searches were undertaken for emerging or underrepresented categories (e.g., AR/VR, robotics/HRC, and digital twin prototypes), and for recent years that were not comprehensively covered by existing systematic literature reviews. This complementary narrative expansion follows common practice in structured narrative and semi-systematic reviews, ensuring that relevant developments are captured, even when they fall outside the scope or time span of existing SLRs.

Therefore, the final set of included studies represents a combined dataset consisting of the following:

• Articles retrieved through database search;
• Articles identified through expert judgment and targeted narrative expansion;
• Studies extracted from existing SLRs that provided consolidated trend data.

This combined selection approach ensured that the final corpus adequately covered both established and emerging HCI-related technologies in smart greenhouse research, supporting a comprehensive and conceptually coherent synthesis [7,10,21,31,34,41].

3. Interfaces and Technologies in Smart Greenhouses

The transition of greenhouses towards high-tech, data-driven factories necessitates the convergence of Information and Communication Technologies (ICT), engineering, and human-centered design principles [3,38]. This section reviews the core technologies enabling the smart greenhouse ecosystem, focusing on their specific role in mediating HCI, along with their associated benefits, limitations, and integration potential.

The rapid digitization of CEA necessitates a quantitative understanding of the underlying HCI technologies currently dominating the research landscape. Analysis of systematic literature reviews (SLRs) published between 2015 and 2025 reveals distinct research density, maturity levels, and growth trends across foundational, intelligent, and immersive technology frameworks.

The following tables synthesize quantitative data directly derived from systematic literature reviews and focused research studies within the analyzed sources, covering the defined technology categories (IoT/WSN, AI/ML, AR/VR, digital twins, robotics/HRC, mobile/web apps, cloud/edge computing). To quantify the prevalence and evolution of major HCI-related technologies within smart greenhouse research,

Table 1. Technology frequency and growth in smart greenhouse research.

Technology Category	Count/Reference Base (n) *	Approximate Relevance	Major Growth Period (2015–2025)	Ref.
IoT/WSN	≥115 implicit IoT/WSN occurrences in DT-related studies; based on n ≈ 123 in Ariesen-Verschuur et al. [3]	Dominant infrastructural foundation.	Continuous; sharp increase from ≈2016.	[1,3,5,11,20,42,43,44,45]
AI/ML (incl. fuzzy logic)	47% ML usage in vertical farming studies (n ≈ 68) according to Siregar et al.; 19% predictive DT type (n ≈ 123) according to Ariesen-Verschuur et al. [3]	High, essential for control and prediction.	Sharp acceleration from ≈2018.	[1,3,4,15,31,46,47,48,49,50,51]
Digital Twins (DT)	8 explicit DT mentions; 6.5% of n ≈ 123 according to Ariesen-Verschuur et al. [3]	Emerging concept	Emerging, sharp increase from 2018.	[3,27,51,52,53]
Mobile/Web Apps (GUI)	88.1% GUI usage in DT studies (as reported by Slob et al. [16], based on Ariesen-Verschuur et al. [3], n ≈ 123)	Dominant HCI paradigm for remote control.	Consistently high use throughout the period.	[6,14,16,26]
Cloud/Edge Computing	Frequently incorporated in IoT architecture; no explicit quantification in source SLRs.	High (central storage/processing for DT/AI).	Consistent integration; edge emerging 2019–2021.	[3,5,15,20,50,51,54,55,56,57]
Robotics/HRC	22% robotic automation in vertical farming 2016–2022 (n ≈ 68) according to Siregar et al.	Moderate/Increasing importance for automation.	Steady rise, focus on HRC/teleoperation post-2019.	[16,20,28,31,32]
AR/VR	1 explicit VR paper in DT context; 2.4% 3D visualization (reported by Slob et al., [16], based on the Ariesen-Verschuur et al. [3] SLR, n ≈ 123).	Nascent, primary focus on simulation/training.	Early stages, emerging post-2020 (especially AR/VR for DT visualization).	[16,20,28,32]

* Count and percentage values are derived from the sample sizes reported in the source SLRs. Because many studies simultaneously employ multiple technologies, a multi-label classification rule was applied, meaning that each study was assigned to every category explicitly identified in the original reviews. As a result, category totals do not sum to 100%. A single-label classification is not applicable, as the source SLRs themselves use overlapping, non-exclusive categories.

summarizes their relative frequency, adoption rate, and growth period between 2015 and 2025. The data illustrates the progressive integration of IoT and AI systems as core infrastructural layers, alongside the emergence of digital twins, immersive interfaces, and robotic collaboration paradigms.

The chronological progression of core HCI-enabling technologies in smart greenhouse systems is summarized in

Table 2. Temporal progression and thematic shifts in HCI technologies for smart greenhouse applications (2015–2023).

Year	IoT/Cloud Technologies *	AI/ML Approaches *	DT *	AR/VR and Robotics (HRC) *	Highlights *	Ref.
2015	Steady IoT monitoring systems. High volume GT articles (n * ≈ 54).	Fuzzy logic systems established.	-	-	Low-cost control design prevalent.	[58,59,60]
2016	IoT systems show strong growth (n ≈ 63 GT articles).	Adaptive Neuro-Fuzzy Inference System (ANFIS) modeling techniques adopted.	-	-	Focus on relevant wireless monitoring systems.	[61,62]
2017	Peak GT articles (n ≈ 71). IoT is utilized for integrated management.	Fuzzy logic widely applied.	-	-	Robotics for crop monitoring continues.	[63,64,65]
2018	Sharp increase in IoT/implicit DT papers begins. Android APP development widespread.	ANN/Fuzzy models for microclimate prediction common.	First explicit DT paper published (Monteiro et al.).	VR/augmented reality (AR) conceptualized for simulation.	Robotics for harvesting (dual arm).	[26,66,67,68]
2019	Peak HCI/IoT production.	Deep learning/ML approaches prominent (47% in Vertical Farming-VF).	Predictive DT gains prominence.	Mobile Plant Health Visualizer (AR). Human–robot interaction (HRI) system Human-based collaborative intelligence (HUB-CI).	Autonomous Cloud Robotic System demonstrated.	[14,20,31,69,70,71]
2020	Wide IoT/Cloud platform adoption. Edge AI applications begin.	AI/ML models applied for energy optimization.	Imaginary DT (Building Information Model-BIM). Predictive DT research expands.	VR training simulators developed. Collaborative Intelligence (HRC) exploration.	-	[54,55,72,73,74]
2021	Peak vertical farming/AI publications (32%).	Focus on DL/Hybrid models (e.g., hybrid neuro-fuzzy).	DT Architecture/frameworks proposed for optimization.	AR in precision farming noted. Robotic 3DS system proposed.	DT acknowledged as next smart farming phase.	[53,75,76,77,78,79,80,81]
2022	Edge computing is utilized for AI inference.	ML/Deep learning (DL) confirmed superior in accuracy.	Confirmation of DT concepts in initial phase.	-	Robotics automation confirmed (22% in VF studies).	[1,3,5,31]
2023+	Narrowband Internet of Things (NB-IoT) and secure platforms.	Hybrid quantum deep learning emerging.	VR/DT interaction barriers investigated (simulation sickness).	AR/VR confirmed as visualization future. Remote HRI and VR for strawberry picking validated.	Focus on User Experience (UX) and simulation sickness in XR.	[6,16,32,82]

* Counts and percentages are derived from the sample sizes reported in the source systematic reviews or primary studies cited. Because many studies simultaneously employ multiple technologies, a multi-label classification approach was applied; therefore, category totals do not sum to 100%. A single-label breakdown is not applicable, as the original reviews report technologies in overlapping categories rather than mutually exclusive groups.

. The table provides a year-by-year overview (2015–2023) of major developments across IoT, AI/ML, digital twins, AR/VR, and human–robot collaboration (HRC), highlighting how research focus has shifted from basic sensing and automation to intelligent, immersive, and predictive control frameworks.

3.1. Sensors and IoT Networks

The foundation of the smart greenhouse is the collection and transmission of real-time environmental data, primarily achieved through sensors and Internet of Things (IoT) networks [3,5]. IoT is defined, according to Bersani et al., as a global network of intelligent, interconnected objects providing sensing, computing, and communication capabilities to support human activities [5].

Sensors act as the perceptual layer, observing critical physical variables such as temperature, humidity, light intensity, CO₂ concentration, and soil moisture, as in the example presented in Figure 3 [5,6,18,19]. This data is essential for decision-making and remote monitoring by the grower [5,6]. The use of Wireless Sensor Networks (WSNs), often leveraging protocols such as LoRa, ZigBee, or Wi-Fi [14,17], is highly effective in greenhouses as it avoids the expense and vulnerability of extensive cabling in harsh environments [30].

WSNs offer high accuracy, robustness, flexibility, fault tolerance, and cost-effectiveness compared to traditional wired systems, according to Akkaş et al.’s study [30]. They enable continuous real-time data flow, which is mandatory for automation and precise position measurements [6,30]. Low-power technologies like LoRa are particularly suitable for long-distance data transmission across large greenhouse facilities, as mentioned by Wo et. al. [14].

The reliability and responsiveness of these sensors directly influence the performance of the entire IoT system [4,19]. Sensor errors can negatively impact production and increase energy consumption due to erroneous climate control adjustments [3,19]. Also, wireless connectivity can be challenging in environments featuring heavy metal structures or tall crops [3,19].

Integration potential of sensor data forms the raw input for all higher-level intelligence (AI/ML) and visualization (DT, mobile/web) layers [3]. This layer enables remote control applications by translating physical conditions into digital information, according to Ariesen-Verschuur et al.’s study.

3.2. Mobile and Web Applications

Mobile and web applications serve as primary graphical user interfaces (GUIs) for interacting with the greenhouse system remotely [22,26].

These interfaces are fundamental for remote control, monitoring, and enabling the decoupling of greenhouse management from on-site presence [3,83]. They typically display real-time sensor readings (e.g., temperature, soil moisture), allowing the user to manage control parameters, such as fertigation scheduling (Figure 4), ventilation, and heating setpoints [6,14]. Wu et al.’s study highlights the preference for Android monitoring software, which is evident in many contemporary systems [14].

Mobile applications offer enhanced accessibility and convenience, enabling users to control and monitor operations from anywhere at any time using ubiquitous devices, like smartphones [6,30]. The web interface functions as a centralized data visualization and decision-making platform, as mentioned in Huynh et al.’s study [6]. Developers emphasize generating user interfaces (UIs) and user experience (UX) that are user-friendly to maximize efficiency and adoption [2,6,22].

A key limitation identified in current research is the frequent lack of aesthetically pleasing and functional designs. Many existing mobile applications are cited as suffering from unattractive UIs and limited features, reducing their overall effectiveness, as noted in Meirieta et al.’s study [84]. Furthermore, some systems, such as the remote measurement and control design by Xiao et al. [85], lacked a local client, requiring the user to rely entirely on a web page for monitoring.

Integration potential of mobile and web interfaces is the indispensable application layer in the IoT architecture, receiving data from cloud/server resources and translating AI-derived insights and digital twin simulations into actionable control commands for the user [3,30].

3.3. Physical Panels and Embedded Interfaces

Physical panels and embedded interfaces, as highlighted by Lu, refer to the localized control systems situated directly on or near the greenhouse equipment [83]. These often employ embedded hardware, like DIY platforms (e.g., Arduino or Raspberry Pi) [17,86].

These systems handle low-level, real-time control functions and allow direct operation of equipment such as fans, lights, and irrigation valves (actuators) [83]. They provide robust local control, which is necessary to avoid a complete reliance on internet connectivity for critical, time-sensitive interventions, according to Setiawan et al. [1]. They typically feature local graphical interfaces (GIs) or touch screens [83,87], enabling manual or automated control modes [83]. For instance, a human–computer interaction GUI based on MATLAB (The MathWorks, Inc., Natick, MA, USA) has been designed for operators to monitor real-time environmental changes and make corresponding decision controls by Shan et al. [88].

Embedded DIY platforms offer low-cost, easy-to-use capabilities and flexibility, which is also highlighted in Turk et al.’s study [86]. They provide reliable, real-time control, serving as a critical backup to remote systems [1]. Additionally, modern systems can incorporate features like local displays of measured values directly on sensor nodes, according to Sumalan et al. [89].

Traditionally, these control subsystems are often installed without designed interactions between them, leading to a lack of interoperability or centralized control [7]. Older or basic embedded systems may rely on less intuitive interfaces (e.g., serial port assistant-based host computer buttons) compared to modern mobile applications [87].

Embedded systems form the reliable bridge between the digital control commands (from the IoT/Cloud/AI layers) and the physical actuators in the greenhouse [7,90]. Advanced hardware platforms integrate the functionality of these separate control devices to achieve smart services [7].

3.4. Multimodal Interfaces

Multimodal interfaces explore interactions beyond traditional graphical user input (touch, mouse) by incorporating methods like voice, gesture, and more natural forms of human input, as highlighted in Pino et al.’s study [20,40].

The goal is to create more natural interaction techniques in the IoT environment [20,40]. While detailed implementation examples in smart greenhouses are not really described in the literature, this area aims to improve the effectiveness and ease of interaction for operators.

Research into networked telerobots, like the one conducted by Gračić, emphasizes the importance of human–robot interaction (HRI) awareness, including the robot’s knowledge of human commands [33]. The MultiHBot robotic system, designed for identifying and cutting fruits, enables human control of processes via a console and voice commands [33]. General research in related smart environments has proposed and evaluated gesture sets for smart-home control [91,92,93].

Multimodal interaction, such as combining human commands with robotic accuracy, enhances performance and allows farmers to perform tasks remotely or from safe distances (e.g., spraying) [33].

The current dominant trend in IoT applications often defaults to simple graphical interfaces [20]. Integrating reliable voice and gesture recognition in dynamic, noisy agricultural environments presents significant technical hurdles. Multimodal controls are critical for optimizing HRC, enabling seamless remote teleoperation and intuitive command execution without physical screens [33].

3.5. Virtual and Augmented Reality (AR/VR)

Immersive technologies like AR and VR offer complex visualization and interaction methods that transcend traditional 2D screen limitations, according to Slob et al. [16].

AR/VR technologies enable realistic 3D visualization, enhancing the physical world with digital content [16]. According to Slob, they are instrumental in providing a detailed overview of real-world processes, allowing users to grasp spatial patterns more clearly. Because of this, the field is rapidly becoming the next step in visualizing complex digital twin data [16].

VR simulations are valuable tools for training precision agriculture concepts and operating Electronic Control System (ECS) interfaces without the risk of damaging crops [13,16]. Examples mentioned by Slob include simulators for power tillers and systems that visualize greenhouse aerodynamics to educate farmers [16]. AR supports hands-free, context-aware interaction, enabling operators to visualize data collected by monitoring robots (e.g., camera feeds and 3D pose replication) [28,32].

A major HCI challenge, particularly for prolonged, regular use in greenhouse management, is the risk of simulation sickness (cybersickness or VR sickness), and this phenomenon, characterized by discomfort, fatigue, and disorientation [16], can deter adoption. Research confirms that simulation sickness negatively affects the overall user evaluation of the virtual environment [16]. To mitigate this, developers must optimize complex virtual models (e.g., simplifying high-detail BIMs) to ensure a playable frame rate [16].

AR and VR are the envisioned interaction layers for digital twins, offering a realistic and immersive way to interact with the virtual representation of the greenhouse and its data [16].

3.6. Digital Twins (DTs)

A digital twin (DT) is a digital equivalent (replica) of a real-life physical object or environment, mirroring its states and behaviors over its lifetime in a virtual space [3,16].

DTs signify a new era in smart horticulture (Ariesen-Verschuur et al.). From an HCI perspective, DTs are powerful tools that enable the remote planning, monitoring, and control of greenhouse operations [3]. They capture the tacit knowledge of experienced horticultural experts through advanced analytics, facilitating continuous learning and improved decision-making [3]. While dashboards are currently the most common visualization for DTs, 3D and VR visualization are emerging [3,16].

DTs substantially enhance control capabilities by allowing growers to simulate the effects of corrective and preventive actions based on real-life data before implementation [3]. DTs support various operational purposes:

• Monitoring and control: Focusing on tracking the state and behavior of real objects, often aimed at improving cost-efficiency (the majority of current DT applications) [3].
• Prediction: Projecting future states (e.g., predicting yield or quality changes based on parameter adjustments) [3].
• Prescription: Recommending optimal interventions (e.g., advising on nutrient application or optimizing production schedules considering energy costs) [3]. Examples mentioned by Ariesen-Verschuur include predictive DTs gaining prominence since 2018 [3].

The concept is still in its seminal phase, with far more implicit applications found within IoT systems (115 studies) than explicit DT frameworks [3]. Currently, the level of virtualization is often limited, focusing mainly on the greenhouse climate rather than individual plants (only about 9% virtualize plants themselves) or the company level [3]. Fully realizing the potential of advanced predictive and prescriptive DTs across the complete lifecycle is still an early stage of development [3].

DTs integrate data from the IoT layer, leverage AI/ML models for simulation and prediction, and require immersive interfaces (AR/VR) for complex visualization and interaction [3,16].

3.7. Artificial Intelligence and Machine Learning

Artificial intelligence (AI) and machine learning (ML) refer to the algorithmic and computational processes used to derive insights and automate complex decisions within the smart greenhouse system [94,95,96,97].

AI models are integrated into the control system to convert expert knowledge and real-time data into systematic, objective, and responsive practices. This fundamentally changes the cognitive process of the farmer, moving away from heuristic experience, according to Torsi et al. [38]. Therefore, AI acts as a decision support system [2,69].

AI/ML algorithms, such as fuzzy logic and Artificial Neural Networks (ANNs), are widely used for climate control, prediction, and optimization [10,98]. Some specific applications include the following:

• Predictive crop modeling and forecasting: AI algorithms predict yield, growth, and quality based on environmental factors and sensor data [2].
• Microclimate prediction: ANNs and fuzzy logic are highly effective in modeling and predicting greenhouse microclimate and managing energy expenditure [10,15,17].
• Recommendation systems: Prescriptive models (part of advanced AI) advise on necessary actions, such as optimal nutrient application or irrigation scheduling [3,4].
• Detection: AI, supported by image processing (e.g., deep learning), is used to pinpoint diseases, pests, and other issues by analyzing plant images [4,5,99,100,101].

Effective AI deployment requires high-quality, continuous data input supplied by reliable sensor networks [4,15]. Furthermore, while sophisticated algorithms like deep learning (DL) are developing, many applications still rely on simpler feedforward neural network (NN) architectures [15].

As for integration potential, AI provides the “intelligence” layer to the digital twin framework, feeding predictive and prescriptive capabilities [3,17]. It works synergistically with IoT/sensors (data input) and enables automation/robotics to act autonomously [4,7].

3.8. Automation and Robotics

Automation and robotics execute physical tasks based on predefined rules or intelligent, real-time decisions, significantly reducing manual labor and enhancing precision, as noted in Costa et al.’s study [17].

Automation systems guarantee a consistent and cost-effective microclimate by automatically managing equipment based on control algorithms [2,102]. Robotics involves sophisticated machines performing tasks like harvesting, monitoring, and precise manipulation using machine vision and AI [32]. The key HCI challenge is defining effective HRC strategies, which is highlighted in Nair et al.’s research [69].

Automation optimizes resource use and improves crop performance [2,36]. Robotics enhances efficiency, improves crop quality, and reduces high labor costs and dependence on scarce green labor [3,17,32]. HRC systems have been shown to increase detection probability (e.g., crop anomalies) compared to fully autonomous systems by combining human expertise with robot accuracy [33,69]. Remote teleoperation, often using interfaces like VR, allows human operators to guide robotic arms (e.g., Kinova Gen3) for precise manipulation and inspection tasks (e.g., capturing leaf health indicators), as described in Udekwe et al. [32].

Robotic systems, especially for complex tasks like harvesting, still face challenges related to accuracy when working in realistic conditions (e.g., obscured fruits) [33]. High initial investment costs and technical complexity remain significant barriers to widespread adoption [2].

Robotics relies on sensors/IoT for navigation and data collection, AI/ML for object recognition and decision-making [32], and specialized multimodal or VR/AR interfaces for remote human interaction and control [32,33].

3.9. Cloud and Edge Computing

These computing paradigms provide the infrastructure for storing, processing, and distributing the massive datasets generated in smart greenhouses [14,19].

Cloud computing (CC) serves as the remote server and central repository for data collected from IoT devices [14]. Cloud platforms facilitate intelligent data analysis, create digital twin abstractions, and manage the remote database [3,83]. Edge computing, considered an emerging technology [19], performs computation closer to the data source (the greenhouse floor), addressing challenges related to processing large data volumes from numerous IoT devices and enabling rapid, low-latency decision-making required for real-time control [5,17,54,55,103,104,105].

Cloud computing provides essential scalability for processing big data and remote management capabilities [17,83]. Edge computing specifically addresses the need for low-latency decision-making, which is mandatory for fast environmental anomaly detection and control systems [5,54,55,103,104,105]. Cloud platforms are used to store data, provide raw material for data mining, and for predictive event models [86].

The efficacy of cloud-based systems is highly dependent on a stable internet connection and sufficient bandwidth, which is particularly challenging in rural agricultural areas [17,19]. Technical issues, like system integration and data analytics capabilities, are common problems in high-tech agricultural settings [106].

Cloud computing (CC) and edge computing are the necessary backbone, linking the IoT sensors to the AI algorithms and distributing results back to the mobile/web interfaces for human viewing and control [22,83]. To provide a clearer and more operational understanding of how computation is distributed across smart greenhouse infrastructures,

Table 3. Typical cloud vs. edge deployment patterns in smart greenhouse systems.

Component/Function	Typical Location (Cloud/Edge)	Rationale/Constraints	Typical Latency Target *	Ref.
Low-Level Environmental Control Loops (e.g., PID for Fertigation, On/Off Actuation)	Edge (Local Controller/Gateway: Raspberry Pi, PLC, Microcontroller)	Hardware must be low-power for remote/off-grid locations, and resilient to harsh environments.	Milliseconds to Seconds (e.g., Solenoid valve closing ≤0.3 s; PID control stability/response time ≈ 150 s for EC/pH regulation.	[6,14,21,86]
Real-Time Sensor Data Processing, Filtering and Critical Event Detection/Inference	Edge (Gateway/Fog Node)	Reduces high bandwidth demand and load on the central cloud server by preprocessing, aggregating and transferring only essential data. Supports low-latency, real-time interactions for event detection.	Low Latency (Near Real-Time, <1 s to minimize delays in control actions).	[19,107,108]
Historical Data Storage, Visualization and Remote Monitoring	Cloud (Public/Private Platform)	Provides highly scalable and reliable storage for large volumes of historical and real-time data for retrospective analysis. Enables centralized remote access and control via web/mobile interfaces for users globally.	High Latency Tolerated (Near Real-Time for display, usually seconds or minutes for monitoring updates).	[2,14,26,86]

* Latency values, deployment patterns and architectural distinctions summarized in this Table are synthesized from previously published experimental studies and review articles, as cited in the Reference column. Because the reported systems differ in sensor types, network protocols, controller hardware, and evaluation setups, the numerical ranges should be interpreted as representative values rather than strict benchmarks. The Table reflects consolidated findings extracted from secondary literature, not measurements conducted directly in this review.

summarizes the typical deployment patterns observed in the reviewed literature, highlighting what components run at the edge versus in the cloud, along with representative latency expectations.

3.10. Dominant Trends, Integration, Gaps, and Sustainability

The descriptive trend analysis (Figure 5) illustrates the broad temporal trajectory of HCI-related technologies in smart greenhouses since 2015. The chart provides an indicative visualization of publication activity, with the most intense activity observed between 2018 and 2022, corresponding to the expansion of IoT- and AI-driven approaches.

The trend representations below are constructed from patterns reported in prior systematic literature reviews, complemented by qualitative and numerical indicators extracted from individual studies. These figures serve as illustrative syntheses rather than reconstructions of raw datasets.

This synthesis draws on trend indications reported in SLRs focusing on greenhouse IoT/DT and AI/fuzzy logic research between 2015 and 2022 (Ariesen-Verschuur et al.; Vanegas-Ayala et al.). A multi-label interpretive approach is used where appropriate, consistent with how the source SLRs classify overlapping technological layers. Recent studies (2023–2025) and emerging categories, such as AR/VR and robotics, were incorporated qualitatively to reflect developments not covered by the main SLRs. The synthesized patterns consistently highlight the centrality of IoT/WSN across the period. To illustrate the relative emphasis on different technological domains across the literature, a proportional visualization was generated from the synthesized secondary evidence (Figure 6). This Figure highlights how the thematic weight of each category evolves over time, based on patterns reported in the reviewed SLRs and individual studies.

This proportional visualization suggests shifts in research emphasis. IoT/WSN and mobile/web interfaces remain central throughout the period, while AI/ML and DT show the steepest growth, with a marked shift around 2018 coinciding with one of the first explicit DT frameworks in horticulture. The illustrative trend index (Figure 7) highlights the growing innovation intensity in AI/ML and digital twins during 2019–2021. Emerging fields, such as AR/VR and robotics/HRC, show an upward trajectory post-2020, transitioning from conceptual proposals to early prototypes.

Despite the technological maturity in IoT and basic AI, several adoption gaps persist. The reliance on ad hoc studies, the difficulty of data generalization, and the consequent lack of standardized datasets and architectures remain a significant barrier to scaling predictive models across diverse greenhouse contexts [11,15]. Furthermore, high initial investment costs and the technical complexity of implementing advanced DT or predictive AI models limit their adoption among small- and medium-sized enterprises (SMEs) [3,31].

From a sustainability perspective, the shift to AI and DT holds substantial promise by enabling precise resource optimization, enhancing productivity, and sustainability [3,4]. Intelligent algorithms assist in precise recommendations for climate control, irrigation, and energy management, which translates to efficiency gains, such as achieving energy efficiency goals [3]. However, the analysis also necessitates addressing the environmental cost of the underlying technology. The proliferation of powerful AI models, particularly deep learning, requires massive energy consumption for training and deployment [35,109]. Future research must, therefore, be guided by sustainable HCI (SHCI) principles, ensuring that the substantial computational costs and energy demand of these high-tech systems are consciously balanced against the resource savings they generate in the physical world [20]. This tension between computational energy and agricultural efficiency defines a key ethical and technological challenge for the future of HCI in CEA.

3.11. Comparative Performance Analysis of HCI-Related Technologies in Smart Greenhouse Systems

The performance analysis of HCI-related technologies in smart greenhouses requires evaluating both traditional technical metrics (e.g., accuracy, latency) and user experience (UX) outcomes (e.g., System Usability Scale (SUS) scores, cognitive load). The nine core technologies discussed here span a wide spectrum of maturity, impacting their quantifiable performance data and adoption viability in CEA.

Table 4. Key technical and HCI metrics per technology.

Technology Category	Key Metrics *	Typical Performance Values *	HCI Limitations and Technical Barriers	Ref.
IoT/WSN	Accuracy, Data Loss Rate, Latency, Energy Consumption.	Sensor accuracy highly variable: 2–25% to >90%. Data loss rate (gateway to server): 0.4%. Successful data storage rate: 99.76%. Wireless transmission reliability: No data packet loss (using Low-Power Listening-LPL technology).	Variable sensor quality; cost constraints; reliance on stable network connectivity.	[3,5,15,19,30]
Mobile/Web interfacce	Usability (System Usability Scale-SUS Score), User Experience (UX/User Experience Questionnaire-UEQ).	SUS Score: 82.75 (Categorized as Excellent). UEQ Score: Excellent for attractiveness, perspicuity, efficiency, stimulation, novelty; good for dependability. Prototype User Interface (UI) average score: 49 (design A superior to design B, 43.4).	Limited features and visual constraint compared to 3D systems; usability dependent on UCD rigor.	[23,34,84,90]
Embedded panels	Classification Accuracy, Root Mean Square Error (RMSE).	Overall microclimate classification accuracy: 97.33%. RMSE (Temperature): 2.398; RMSE (Humidity): 1.483; RMSE (Light): 392.225.	Limited computational power for complex AI; data compression required for cloud transfer.	[10,15,82,107,108]
AI/ML	Prediction Accuracy, Coefficient of Determination (R2), Energy Savings, Error Reduction.	DL disease detection accuracy: 99%. CO2 prediction R2: 0.97. Internal temperature prediction error: <1 °C (Nonlinear AutoRegressive with eXogenous Inputs Neural Network-NNARX model). Energy prediction accuracy: 95%. Recurrent Neural Network-RNN control error (Ec): 344.12 (compared to Ec =533.31 for simple NN).	Difficulty in generalizing models to different greenhouse contexts; dependence on massive datasets for DL.	[15,82,107,108]
DT	Visualization Preference, Predictive Capability (% of applications).	DT purpose: Monitoring (73%), Prediction (19%). Representation: 88.1% via dashboard; 2.4% via 3D visualization.	Conceptual phase; current focus limited to monitoring state visualization; advanced prescriptive capabilities are still emerging.	[3,16,52,53]
AR/VR	Simulator Sickness Questionnaire-SSQ, User Evaluation Score, Convenience/Quality correlation.	Increase in SSQ score leads to lower overall evaluation score. Convenience (realism/playability) has a positive statistical influence on evaluation; visual quality does not.	Risk of simulation sickness/cybersickness; specialized hardware required; low overall application rate (<3% of DT studies use 3D/VR).	[13,16]
Robotics/HRC	Detection Accuracy Gain, Error Rate, Operational Efficiency (Cycle Completion Time-CCT).	Detection increase using HRC (HUB-CI): 4% (vs. human alone); 14% (vs. fully autonomous system). Collaboration yields significantly fewer errors. Yield monitoring CCT (eg., strawberry): Measured across two trials for eight plant groups.	High complexity in managing physical collaboration tasks requires robust workflow protocols to minimize conflicts and errors.	[32,69]
Cloud/Edge Computing	Latency, Cost Efficiency (Levelized Cost of Energy-LCOE).	Data loss rate (gateway to server): 0.4%. LCOE (Photovoltaic-PV system with Latent Thermal Energy Storage-LTES storage): 0.068 USD/kWh (70% reduction vs. diesel system: 0.230 USD/kWh).	Dependence on network stability; data security concerns; high energy consumption for Generative AI models.	[3,4,15,82,109]
Multimodal (voice/gesture)	Cognitive Load (Fixation Frequency Correlation).	Transition from web to Mixed Reality impacts fixation frequency (correlation with cognitive load).	Lack of widespread application data; technical difficulty in accurate gesture/voice recognition in noisy greenhouse settings.	[110,111,112,113,114]

* Reported metrics and values are taken directly from the source studies and systematic reviews cited. Because the included works use different experimental designs, sample sizes, and evaluation metrics, the values presented here reflect those reported in the original publications and are not derived from a unified dataset. Some technologies (e.g., DT, AR/VR, and robotics) inherently overlap with multiple layers of smart greenhouse systems, and, therefore, multi-label classification applies conceptually; a single-label separation is not meaningful for this typology.

synthesizes the main technical and HCI metrics reported across the analyzed literature.

The analysis of performance metrics across nine core HCI-related technologies highlights a clear demarcation between mature, dependable foundational systems and emerging, high-potential yet high-risk paradigms.

The mature technologies, namely IoT/WSN, mobile/web interfaces, and embedded panels, demonstrate fundamental technical reliability and high measured usability. IoT/WSN technologies serve as the backbone, achieving high data fidelity, with average data loss rates between gateway and server as low as 0.4% [15], supporting real-time data flow that is essential for remote management. Similarly, mobile/web interfaces provide the most successful user front-end, consistently achieving high user satisfaction metrics, exemplified by measured System Usability Scale (SUS) scores of 82.75 (categorized as Excellent) [84]. These systems are preferred for their low complexity and high accessibility [26,38].

In contrast, emerging paradigms like AR/VR, DT for prescription, and robotics/HRC promise transformative functionality but face significant performance hurdles centered on user experience (UX) and complexity integration. VR environments, designed to enhance spatial understanding, are undermined by poor user comfort, where an increase in SSQ scores has a statistically significant impact on the overall evaluation, resulting in lower user ratings [16]. Robotics systems demonstrate superior technical efficiency in tasks like anomaly detection, achieving a 14% increase in detection probability compared to fully autonomous systems when human collaboration is integrated [69]. However, the measured performance relies heavily on complex integration models like HUB-CI to ensure efficiency by managing coordination and minimizing errors [69].

The central trade-off observed is the balance between reliability and immersion. Conventional graphical interfaces provide reliable access to real-time IoT data but require the grower to synthesize information and rely on heuristic knowledge [38]. Conversely, immersive technologies like AR/VR offer a 3D visualization platform that provides a higher sense of realism [16], but this added spatial complexity and novel interaction mode introduce significant UX risks, such as cybersickness [16], reducing the system’s long-term reliability for prolonged professional use. Critically, among the analyzed literature, it was noted that while the convenience (playability and realism) of the VR DT positively influences evaluation, the visual quality does not have a measurable statistical impact [16].

A secondary trade-off exists between automation and control transparency. AI/ML systems provide phenomenal performance gains, such as a 99% accuracy rate in deep learning disease detection models [108] or achieving prediction accuracy of 95% for energy consumption. This high level of automation reduces the farmer’s cognitive load, moving them from heuristic reasoning to systematic, scientific practice [15]. However, while AI expert systems successfully demonstrate operational performance in prototype form, historical research suggests that high cost and integration difficulty often impede actual deployment in practice, implying a resistance to complex, opaque automation [38]. Furthermore, AI accuracy is often domain-specific; while one ANN achieved superior temperature and humidity prediction accuracy over classic methods (81% and 62%, respectively) [15], the models are limited by the difficulty of generalizing results across different greenhouse installations [15].

The established integration pattern is predominantly AI + IoT + Cloud/Edge Computing. IoT/WSN collects real-time data [5], which is transmitted via protocols like LoRa or ZigBee [5] to Cloud platforms (e.g., MySQL or Firebase) for storage [5,14]. AI/ML models then utilize this aggregated data to generate intelligent insights, achieving, for instance, a reliable air temperature forecast with an R² score of 0.965 [5]. This pattern underpins the shift toward DT development, leveraging AI prediction models to simulate future states [3].

The emerging pattern is the integration of AR/VR + DT + Robotics/HRC. The digital twin serves as the central virtual representation, visualized via AR/VR technology for immersive interaction [16]. This allows human experts to remotely supervise or collaborate with complex robotic platforms [32], leveraging the robots’ physical capability and the human’s specialized intelligence to improve overall system detection effectiveness by 14% over full autonomy [69].

Current research suffers from several critical gaps. First, DT applications remain largely conceptual and monitoring-focused [3], with only a few articles explicitly addressing DTs in greenhouse horticulture in one review (Ariesen-Verschuur et al.), indicating a low maturity level for fully prescriptive and autonomous applications. Second, there is a fundamental limitation in performance assessment standardization. While sophisticated metrics are used in individual technical studies (e.g., RMSE, R² for AI models; CCT for robotics), there is a lack of consistent, comparative HCI evaluation frameworks specific to agriculture. This is particularly relevant when evaluating the sustainability cost of advanced AI/ML, as the energy consumption of large generative models is rarely reported using HCI-focused metrics, complicating the analysis of their environmental impact [109]. The reliance on ad hoc studies and the limited size and diversity of user groups (e.g., one VR study used only 30 participants from a single university) [16] further necessitate the development of robust, standardized HCI evaluation protocols for smart agriculture.

4. User-Centered Design for Smart Greenhouse Technology

The sophisticated technologies deployed in modern agriculture, such as IoT systems, AI, and DTs, necessitate a deliberate focus on HCI and UCD [21]. UCD principles are vital for enhancing the efficiency, accessibility and overall usability of technology in farming practices, as highlighted in Ibrahim et al.’s study [21]. By integrating the user perspective, UCD actively seeks to maximize system intelligence and ensure that the solutions align precisely with the specific needs and preferences of agricultural stakeholders [21,22]. Consequently, incorporating UCD is essential for overcoming barriers to adoption of technology, which often include the complexity of use, lack of suitability to farmer needs, and potential increased workload [21,23].

4.1. Participatory Design

Participatory design emphasizes the active involvement of greenhouse growers and operators in the design and iterative development cycle of new technological systems [21].

This approach is mandatory to acquiring accurate contextual information, as existing research confirms that the introduction of modern Information and Communication Technologies (ICT) requires a significant shift in the paradigms and cognitive processes of farmers [38,69]. Expert farmers rely heavily on professional visions and heuristic reasoning developed over years of manual experience (particularly for tasks like irrigation) [38,115]. By employing UCD, designers can incorporate this tacit knowledge and operational heritage into greenhouse innovation processes, according to Torsi et al. [38]. For instance, collaborative intelligence algorithms designed for networked telerobots involve managing and coordinating interactions between software, hardware, and human agents to reduce errors and conflicts [38,116,117]. A clear example of effective participatory design involves the joint effort of agronomists and ICT technicians defining new services within automated systems, ensuring that the resulting hardware–software platforms yield improved efficiency and operational services adapted to real installation requirements [7,118].

While participatory approaches enhance usability [21,119], a recognized limitation in technology assessment is the tendency to focus studies primarily on evaluating the technical effectiveness of solutions, frequently overlooking the important user perspective and intrinsic perceptions [23,120]. Comprehensive user feedback sessions are necessary to assess usability, functionality, and overall user satisfaction in real-world agricultural contexts [21].

4.2. Context-Aware Interfaces

Context-aware interfaces are designed to dynamically adjust their operation, display, and features based on the user’s identity, location, expertise, current task demands, and surrounding environmental conditions [21].

This principle is critical due to the harsh and variable conditions characteristic of agricultural settings, requiring tailored HCI solutions to address factors such as fluctuating weather, diverse user demographics, and limited connectivity [21]. Key design requirements include promoting task efficiency and minimizing the cognitive load on users during complex activities [21,120,121]. A core aspect of context awareness is resilience to environmental factors, necessitating interfaces that are reliable and durable under varying circumstances [21]. Also, AI and DT technologies embody advanced context awareness by leveraging huge datasets from reliable sensor networks to convert the grower’s heuristic experience into systematic, objective, and responsive practices [38,122,123,124,125,126]. This computational process captures the tacit knowledge of experienced horticultural experts through advanced analytics [2,127,128].

Designing genuinely context-aware systems demands a deep understanding of the agricultural work environment, implying that developers must consistently employ contextual inquiry techniques to accurately identify environment-specific challenges and ensure seamless integration with users’ daily tasks [21,129,130,131].

4.3. Ergonomics and User Experience (UX)

Ergonomics and UX focus on designing interactions that ensure usability, efficiency, accessibility, and inclusion across a diverse user base, which is important given the wide variation in technical literacy among greenhouse operators [21,22,132].

Effective HCI design significantly improves user experience and the quality of life associated with managing smart systems, according to Zhao et al.’s analysis [22]. Researchers advocate that successful agricultural interfaces must accommodate diverse demographics by considering differences in age, education, and technological proficiency [22,133]. Accessibility is paramount, particularly for the elderly and people with disabilities [22,134]. Systems should feature variable accessibility, allowing the user interface to be configured based on individual capabilities, potentially switching from touch controls to voice-controlled interfaces for physically disabled users to manipulate smart devices [22,135,136]. Moreover, the use of visualized models over simple button pressing is essential for improving the user experience [22].

In terms of UX design specifically for ubiquitous devices, development teams focus on applying UI/UX design principles to address drawbacks common in existing mobile applications, such as unattractive interfaces and limited features [84,136,137,138]. A practical example of this UCD application, shown in Meirieta’s study, involved developing a smart mobile greenhouse prototype using the design thinking methodology, which, following user testing, achieved a SUS score of 82.75, categorized as excellent usability [84,139]. In this study, the UCD process was operationalized through a full design thinking cycle—empathizing, defining, ideating, prototyping and testing, supported by concrete design artifacts such as user interviews and questionnaires, empathy maps, affinity mapping, creation of sitemaps, low-fidelity wireframes, and a high-fidelity Figma prototype; usability was then evaluated with N = 10 prospective users, all greenhouse practitioners, or agriculture-technology users [84]. This focus on intuitiveness and efficient navigation facilitates task completion and increases user satisfaction [21].

4.4. Specific Challenges

The adoption of sophisticated interfaces in smart greenhouses faces several specific challenges related to technology acceptance, user demographics, and the limitations inherent in immersive systems.

For interfaces utilizing VR, such as those used for 3D visualization of DTs, simulation sickness (cybersickness or VR sickness) presents a significant HCI challenge for prolonged and regular use [16]. This phenomenon causes symptoms like discomfort, fatigue, and disorientation [16,140]. Studies have confirmed that users who experience simulation sickness tend to give a lower overall evaluation score of the virtual environment [16]. Developers must mitigate this risk by optimizing complex virtual environments (e.g., simplifying highly detailed BIMs) to maintain a playable frame rate [16,141,142].

The rapid introduction of ICT is creating a significant generational, cultural, and practice gap in cultivation [20,40]. Research indicates that most greenhouse farmers are elderly and often possess a low level of education, which contributes to a lack of innovation culture and makes it difficult for them to manage modern control systems (e.g., climate control or fertigation systems) [24]. Therefore, their strategy often focuses on survival and strict cost control, as highlighted by Kavga et al.’s study [24]. This reluctance necessitates that technology developers focus on the usability and intuitiveness of interfaces to minimize the alterations farmers must make to their existing routines, according to Zeidler et al. [28]. Furthermore, the investigation of emerging interfaces (e.g., haptics or speech) could offer potential strategies to overcome language and literacy barriers that might impede technology adoption [37].

Adherence to UCD principles is of the utmost importance as it directly correlates with technology acceptance, trust, and operational efficiency in smart greenhouses. Interfaces that are intuitive, easy to navigate, and responsive to user needs facilitate the efficient performance of tasks, ultimately leading to greater uptake and utilization of digital tools by farmers [21]. Successful UCD implementation, particularly through features like user-friendly interfaces, makes complex, automated systems more accessible to a wider range of users, thereby helping address the increasing scarcity of experienced green labor [2,3].

Decisively, HCI principles applied through UCD contribute significantly to environmental sustainability. By integrating decision support systems and visualization tools that focus on resource efficiency (e.g., smart irrigation data visualization or DT simulations), UCD enables precision agriculture techniques that minimize input waste, leading to a reduction in environmental impacts and resource consumption [6,21]. Thus, the success of technological innovation in smart greenhouses relies not just on optimizing control algorithms but fundamentally on designing interactions that respect and augment human capabilities, thereby ensuring both business profitability (yield quality and quantity) and long-term ecological sustainability [3,11,21,102].

4.5. Quantitative and Comparative Analysis of User-Centered Design Approaches in Smart Greenhouse Systems

The integration of advanced technologies, such as IoT, DT, and AI, in CEA has intensified the focus on HCI [5,14,21,83]. HCI is important for designing, evaluating, and implementing systems, aiming to enhance the efficiency, accessibility, and overall usability of technology in farming practices [20,21]. The success of intelligent greenhouse systems depends not only on advanced hardware but also on excellent HCI design [22]. Next, we analyzed the application frequency and evolution of core UCD paradigms by including UCD, participatory design (PD), experience-centered design (ECD), cognitive ergonomics, and human–robot interaction (HRI), applied within smart greenhouse HCI research from 2015 to 2025.

Research confirms that user-centered methods, particularly those focusing on usability and UX, are integral to the successful adoption of agricultural technology [21,23]. Several studies provide implicit and explicit quantitative data regarding the adoption and evaluation of user-centered features, demonstrating an emerging but uneven focus across subdomains (

Table 5. Quantitative synthesis of user-centered evaluation in smart greenhouse studies.

Evaluation/ Design Type	Context	Frequency/Year *	Ref.
UX/Usability	DT Visualization	Dashboard interface dominant (88.1% of 123 reviewed articles).	[16]
		3D Visualization (2.4%, or 2 articles).
	VR/DT Interface	Only one paper (out of 123) implicitly addressed VR
Usability (SUS)	Mobile App Design	Achieved an excellent usability score of 82.75 (evaluated by 10 potential users).	[84]
Participatory/Qualitative	VR/DT Experiment	Limited sample size of 30 participants (all from one university/WUR) noted as a limitation.	[16,38]
	Grower Cognition	Qualitative work based on 11 expert interviews to isolate constants for improving technology usability.
Participatory (UCD)	Microclimate GUI	Study involved 10 respondents (three hydroponic practitioners, three agricultural technology users and others) for needs analysis.	[34]
Collaborative HRI	Target Recognition	Collaboration increased detection probability by 4% compared to a human alone and 14% compared to a fully autonomous system	[33,69]
UX/Cognitive Load	Alert System Assessment	NASA Task Load Index (NASA-TLX) and UEQ employed found that alert delivery mechanisms generated a similar mental workload.	[23]

* Percentages, counts, and participant numbers are reported as provided in the original studies. Because several studies involve multiple evaluation or design methods simultaneously, a multi-label classification was applied; therefore, totals may not add up to 100%.

).

The analysis reveals a dominant focus on traditional interface paradigms like dashboards in early DT applications (88.1% of articles reviewed up to 2022) [16], contrasted with the nascent adoption of immersive technologies like VR (only one article implicitly addressing VR in DT before 2022) [16]. However, recent studies (2024–2025) demonstrate a commitment to rigorous usability testing, exemplified by the SUS resulting in an excellent score of 82.75 for a mobile application prototype designed using UCD and design thinking [84].

The application of UCD in smart greenhouse HCI is highly contextual, addressing the complexity of automation systems, the variability of the agricultural environment, and the need to support expert decision-making [10,84]. The primary paradigms identified are compared in

Table 6. Comparative summary of user-centered design paradigms in smart greenhouse systems.

Paradigm	Focus Area	Evaluation	Interface Types	Ref.
User-Centered Design (UCD)	System development, ensuring usability, functionality and utility for the end-user (e.g., microclimate monitoring).	Usability Testing, Surveys/Questionnaires, System Usability Scale (SUS).	Graphical User Interfaces (GUIs), Mobile Applications (Android), Web Interfaces.	[6,7,14,34,84]
Participatory Design (PD)/Design Thinking	Collaborative creation, identifying user problems and needs, incorporating stakeholder knowledge (agronomists, technicians, growers).	Interviews (semi-structured), Qualitative Data Analysis (affinity mapping, empathy maps), Persona/Journey Mapping.	Conceptual Models, Prototypes (low and high-fidelity), Mobile App Mockups (Figma).	[7,38,84,106]
Experience-Centered Design (ECD)/UX	Holistic user perception, focusing on the quality of interaction, emotional inclusion and avoiding negative consequences like simulator sickness.	Simulation Sickness Questionnaire (SSQ), User Experience Questionnaire (UEQ), Qualitative feedback.	Virtual Reality environments, 3D visualizations, Interactive prototypes.	[16,23,143]
Cognitive Ergonomics	Optimizing system interaction to reduce cognitive load, match technology to existing professional heuristics, and minimize errors in complex tasks.	NASA Task Load Index (NASA-TLX), Error rates, Qualitative analysis of professional routines/practices.	Alert Systems (multi-modal delivery: voice, sound, text), Process control interfaces (fuzzy logic, AI models for prediction).	[16,23,143]
Human–Robot Interaction (HRI)	Enabling remote control and collaborative intelligence between human operators and robotic platforms for tasks like monitoring or harvesting.	Detection accuracy, Cycle completion times, Error reduction quantification, Performance metrics.	Virtual Reality teleoperation interfaces, Wi-Fi consoles, robotic vision feeds, digital twin replicas of robot status.	[32,33,69]

.

The foundation of HCI in greenhouse technology relies heavily on UCD principles, which mandate placing users, such as greenhouse growers, agricultural specialists, and analysts, at the core of the system development process [34,107]. This approach ensures repeatability, transferability, and ease of modification, treating the system as a tool to further knowledge rather than a hindrance [144]. Relatedly, PD, often integrated through design thinking phases (empathizing, defining, ideating, prototyping, and testing), is essential for gathering insights into user needs, complaints, and preferences early in the design phase [84]. For instance, combining design thinking with the socio-technical-ecological systems (STES) perspective helps analyze the grower’s entire journey in adopting energy management solutions, providing a comprehensive view of the problem space [106].

Cognitive ergonomics is highly relevant, especially for managing the shift in professional skills required by automation [38]. Experienced growers historically rely on heuristics and trained perception to manage the ecosystem; automation necessitates changing cognitive processes to incorporate information from sensors and actuators [38]. Studies focusing on minimizing conflicts, optimizing collaborative routines, and evaluating mental workload using metrics like NASA-TLX address these cognitive challenges [23,69].

The HRI paradigm is important for teleoperated or autonomous systems performing complex tasks like yield monitoring [32,33,69]. VR interfaces are increasingly used for remote HRI, allowing operators to oversee and control robotic operations while visualizing the greenhouse environment from the robot’s perspective [32]. Finally, while not explicitly named, ECD encompasses research focusing on the perceived utility and overall UX of high-fidelity interfaces, such as evaluating simulation sickness (cybersickness) in VR-based digital twins [16]. The core benefit of using VR in combination with DT is the UX derived from the realistic representation of the environment [16].

5. Comparative Analysis of Human–Computer Interaction Paradigms in Smart Greenhouse Systems

Research shows a dominant focus on monitoring and control systems at the greenhouse level, addressing climate and energy management [3]. To better illustrate the conceptual and functional differences among existing HCI paradigms,

Table 7. Comparative features of HCI paradigms in smart greenhouse systems.

Feature	Conventional GUI-Based Interaction	Immersive and Multimodal Interaction	AI-Driven Adaptive and Predictive Interfaces (DT/DSS)	Ref.
Interaction mode	2D graphics, direct manipulation, menu navigation, remote manual control.	3D immersive (VR/AR), gesture, voice, telerobotics/HRI, simulated physical action.	System suggestions, predictive visualization, autonomous execution, prescription of optimal strategies.	[3,22,26,28,32]
Cognitive load	Moderate/High (requires manual synthesis of data, heuristic decision-making).	Variable (can be reduced by visualization/AR; increased by cybersickness or complex controls).	Low (automates decision-making; predictive alerts minimize surprises).	[3,16,21,38,83]
Automation level	Low/Medium (automated data acquisition; manual setpoint control).	Medium (automated tasks via HRI/teleoperation; human-in-the-loop control).	High/Autonomous (self-adjusting algorithms, predictive control, optimization models).	[3,7,32,69,83]
Adoption potential	High (low cost, high familiarity, mobile device accessibility).	Medium/Emerging (limited by simulation sickness, specialized hardware costs/complexity).	Medium (high initial cost, requirement for skilled staff, generalization issues).	[13,15,16,26,84,89,106]
Sustainability impact	Generally neutral/positive (enables monitoring resource use, UCD focus).	Positive (enables efficient HRI for precision tasks, reducing waste).	Variable (high positive impact via resource optimization; negative impact from AI/GenAI energy consumption).	[4,5,6,15,21,33,109]

compares interaction modes, cognitive load, automation levels, adoption potential, and sustainability impact.

Predictive DTs for growth simulation and yield prediction have been gaining prominence since around 2018 [3]. AI/ML models (ANNs, FL) are widely studied for microclimate prediction (e.g., temperature, humidity), energy optimization, and specific controls, like CO₂ concentration [5,15]. For instance, fuzzy logic is utilized in systems for regulating humidity and temperature [6,15]. The HUB-CI model demonstrates a system based on collaborative intelligence, integrating AI and workflow protocols to manage physical collaboration in telerobotic agricultural systems for anomaly detection [69].

While Table 6 addressed the conceptual frameworks of HCI paradigms,

Table 8. Comparative summary of representative HCI systems and findings in smart greenhouse horticulture.

System	Technology/Core Function	Interface Type	HCI Role	Ref.
Android remote control	IoT, Raspberry Pi, MySQL database for remote control of fertigation (Subsystem D).	2D mobile GUI (Android)	Remote monitoring and control	[14,26]
VR greenhouse training	Virtual Reality (VR), Electronic Control System (ECS) simulation	3D immersive (HMD, controllers)	Training/simulation	[13,16]
Hub-CI (telerobotics)	Collaborative Intelligence (CI), Telerobotics (spectral imaging, workflow protocols).	Cyber-augmented collaboration (web-based/remote)	Human–robot supervision	[69]
Fuzzy logic control	AI (Fuzzy Inference Systems, ANFIS), IoT sensing for climate control.	Embedded/2D dashboards (often remote access)	Adaptive automation	[5,10,89]
Mobile Plant Health Visualizer (MPHV)	Augmented reality (AR), SI-NDVI imaging.	AR (HMD/smartphone display)	Context-aware visualization	[28]

focuses on practical implementations, identifying representative systems that exemplify the convergence between IoT, AI, and immersive technologies in smart greenhouse operations.

The evolution of HCI in smart greenhouses reflects a continuous effort to manage complexity, optimize resources, and enhance the efficiency of controlled-environment agriculture. The future direction of smart greenhouse management necessitates the seamless integration of adaptive, sustainable, and HCI principles. To further synthesize the comparative findings,

Table 9. Evaluation of HCI solutions in smart greenhouse management: advantages and limitations.

System	Advantages	Limitations	Ref.
Android remote control	Real-time monitoring, visualization, remote adjustment of fertigation/settings.	Requires network status for access, limited functionality compared to advanced systems.	[14,26]
VR greenhouse training	Risk-free training environment, replicating real-world ECS interfaces, fosters engagement.	Challenges in mimicking physical actions, risk of simulation sickness, hardware dependent.	[13,16]
Hub-CI, telerobotics	Improves detection accuracy (14% over autonomous systems), optimizes collaboration tasks, decision support.	Requires complex management/coordination protocols to prevent errors	[69]
Fuzzy logic control	High interpretability, better performance for nonlinear systems, potential resource savings (22% energy, 33% water)	Reluctance in adoption due to complexity/non-standardization compared to PID.	[5,10,89]
Mobile Plant Health Visualizer (MPHV)	Enables quick visualization of plant health (PHM) status in cultivation area; remote monitoring.	Complexity, cost, dependence on wireless signals, and limited optical resolution (future concern).	[28]

evaluates representative HCI solutions in terms of their operational advantages and limitations within smart greenhouse management.

The current trend shows a clear move towards integrating AI-driven predictive capabilities with increasingly immersive and intuitive interfaces [21,83]. Digital twins are emerging as the core technological enabler, promising to completely decouple physical management from information access, allowing growers to transition fully into remote, supervisory roles [3]. However, achieving true predictive and prescriptive autonomy remains a development focus [3,5,15]. This advanced intelligence must be paired with multimodal interaction techniques (gesture, voice) to provide efficient and intuitive access to complex control systems, especially in environments where hands-free operation is desirable [21].

The widespread adoption of these sophisticated systems hinges on rigorous human-centered design [21,23,38]. Interfaces must be designed to mitigate the cognitive friction associated with the shift from experiential, heuristic reasoning to systematic, scientific practices mandated by AI/sensor data [38]. This requires simplifying the complexity of control algorithms and integrating DSS suggestions intuitively, perhaps through visually rich immersive interfaces that minimize cognitive load [21]. Furthermore, accessibility and inclusivity must be prioritized to address diverse user demographics and technological literacy within the agricultural workforce [21,23].

Finally, the sustainability imperative mandates that HCI address not only the optimization of energy and water consumption within the greenhouse environment (Figure 8), which intelligent control already facilitates efficiently [4,5,102], but also the environmental impact of the computing infrastructure itself. The increasing reliance on powerful AI models also generates a substantial carbon footprint [109].

Future research must adopt a sustainable HCI (SHCI) or post-growth HCI orientation, focusing on designing resource-conscious algorithms and quantifying the carbon cost of computational choices to ensure that technological advancement in smart greenhouses genuinely contributes to global environmental goals while advancing productivity [25,45,109,145,146]. Thus, the next generation of smart greenhouse HCI will be characterized by resilient, adaptable, and ethically scaled systems that harmonize automation capabilities with human ecological responsibilities.

6. Challenges and Future Directions

The integration of advanced technologies and human-centered design principles into smart greenhouses presents a roadmap toward enhanced productivity and sustainability. However, realizing the full potential of these Cyber-Physical Systems (CPS) requires proactively addressing several critical challenges and dedicating research efforts toward specific future directions that prioritize the human operator, system trustworthiness, and ecological responsibility.

6.1. Technological and Infrastructure Challenges

6.1.1. Interoperability Issues Between Heterogeneous Systems

A primary technical hurdle is ensuring interoperability between heterogeneous systems [21]. Current automated greenhouses often rely on multiple subsystems (e.g., climate control, irrigation, and automatic vent openers) that are installed without designing interactions between them [7]. This lack of integration prevents the effective combination of control systems to create the perfect combination of control [7]. Research conducted by Andrianto shows that greenhouses face common IT issues related to systems integration and data analytics capabilities [106]. Future research must prioritize integration models to convert these disparate scenarios into intelligent systems capable of predicting automatic actions and making all subsystems interoperable [7]. Overcoming the complexity of integrating multiple technologies and ensuring system compatibility remains an urgent issue [21].

6.1.2. Data Security and Privacy in IoT-Based Greenhouses

As smart greenhouses rely on ubiquitous sensor networks (IoT), data security and privacy protection are critical limitations. Agricultural activities involve sensitive data, such as crop yields and soil conditions [21]. Robust measures are required to protect farmers’ information [21]. Current efforts include using automatic security system architectures, such as face recognition, to safeguard greenhouse assets and employee privacy [147]. However, gaps remain, as data security and privacy protection in the cloud are still incomplete [83]. Major efforts are needed to guarantee secure data transmission and collection, addressing constraints identified as barriers to IoT-enabled smart farming adoption [5,19].

6.1.3. High Costs, Digital Literacy, and Infrastructure Gaps

Widespread adoption is hampered by significant economic and social barriers:

• High Costs: High initial investment costs and technical complexity are major challenges to the adoption of greenhouse technology [2]. The cost of IoT applications, particularly for highly functional sensors (which can exceed 1000 USD per sensor), impedes commercialization by smallholder farmers and limits widespread adoption [19,83].
• Digital Literacy: Many farmers, especially older generations, often have low levels of education and lack the necessary knowledge to manage modern control systems (e.g., climate control or fertigation systems), contributing to a lack of innovation culture [24]. Designing interfaces that cater to the broad spectrum of technical literacy within the agricultural workforce requires careful consideration [21].
• Infrastructure Gaps: The digital divide is critical. Rural areas, where agriculture is concentrated, often suffer from limited technological infrastructure, insufficient internet connectivity, and unreliable power supply, hindering the seamless integration of sophisticated technologies [19,21]. The efficacy of IoT systems is limited by internet instability, particularly when managing many devices [17]. Future research must validate solutions like Low-Earth Orbit (LEO) constellations, which promise global broadband internet coverage with low latency, potentially eliminating infrastructural barriers to mass IoT adoption in remote areas [19].

6.2. Human-Centered and Interface Development Directions

6.2.1. Lack of HCI Standards Specific to Agriculture

Currently, there is an absence of standardized metrics for evaluating HCI specifically in agricultural settings, which complicates the assessment of usability and user satisfaction [21]. Also, the inherent heterogeneity of agricultural practices, which vary significantly across different regions and crop types, makes designing universal HCI solutions challenging [21]. Future research must aim to inform policy insights to establish technology standards and guidelines that recognize the role of user-centered design [21].

6.2.2. Adaptive and Predictive Interfaces

The future trajectory of HCI in smart greenhouses lies in building interfaces that are both adaptive and predictive, deeply integrating AI and UX principles. AI models capture tacit expert knowledge, translating heuristic experience into systematic practices [38]. Current research literature shows the emergence of predictive DTs that allow the simulation of interventions and the projection of future states [2]. However, more advanced applications focusing on prescriptive capabilities are still in an early stage of development [2]. Future interfaces should incorporate adaptive feedback mechanisms that provide guidance and suggestions based on the current state of the agricultural system [21]. Furthermore, while most AI applications currently use basic feedforward neural network architectures, future research should explore more sophisticated models, such as Recurrent Neural Networks (RNNs) and hybrid deep learning (DL) models, to enhance accuracy in complex tasks like microclimate prediction [15].

6.2.3. Personalization According to User Role and Expertise

To maximize adoption and efficiency, interfaces must move beyond one-size-fits-all designs towards deep personalization. This involves catering to the diverse user demographics and their varying technological proficiencies [21]. Designing for variable accessibility, perhaps switching between graphical and voice-controlled interfaces, is essential for inclusion [22]. The core requirement is that interfaces must align with the farmer’s established routines and expertise (professional skills) to minimize the alterations they need to make to their practices [38].

6.2.4. Reliable Multimodal Systems

Research needs to focus on creating reliable multimodal interaction systems that combine multiple input/output modalities (e.g., AR visualizations augmented by voice control). Emerging interfaces such as gestures and speech can help overcome traditional language and literacy barriers [37]. AR is particularly promising as it can enhance the physical world with digital content and is considered the next step for visualizing complex DT data [16,28]. The development of multimodal systems is critical for specialized applications like remote HRC [33]. However, specialized challenges, such as simulation sickness (cybersickness or VR sickness) in immersive VR environments, must be studied and mitigated by optimizing virtual environments for comfort during prolonged use [16].

6.3. Sustainability and Evaluation Gaps

6.3.1. Energy Efficiency and Digital Inclusion

HCI is increasingly acknowledging its role in addressing sustainability predicaments, moving beyond a focus solely on technological growth [145,148,149]. The future calls for sustainable design, integrating energy efficiency and digital inclusion. Automated control systems have been proven to reduce energy demand [102], and IoT infrastructure plays a key role in optimizing greenhouse energy management [4,5]. Future research should explore integrating carbon capture technologies and advanced smart glass systems within greenhouse structures to offset the digital carbon footprint [4]. Furthermore, adopting a post-growth HCI perspective promotes designs rooted in solidarity, cooperation, and social justice [10,145], ensuring that technology advancement supports sustainable development goals, including digital inclusion [148].

6.3.2. Standardization and Longitudinal Evaluations in Real-World Environments

A significant gap remains in the long-term assessment and standardization of new systems. Most current applications are classified as prototypes [31]. Longitudinal studies are necessary to assess the real-world, long-term impact and usability of agricultural interfaces, enabling iterative refinement based on user feedback and changing agricultural practices [150]. Standardization is complicated by the myriad different technical standards employed across networks, data rates, and power requirements [19]. Future research methodologies, informed by UCD, should involve full, iterative cycles of interface development, accompanied by extensive data collection and systematic reflection on the findings obtained in real-world work contexts [151].

7. Conclusions

The analysis confirms that human–computer interaction serves as the critical nexus for achieving efficiency, sustainability, and widespread adoption of complex smart greenhouse technologies. The review maps a clear technological trajectory from foundational IoT infrastructure and dependable GUIs toward intelligent, highly integrated systems leveraging AI, DTs, and XR. While established paradigms, like mobile and web interfaces, that demonstrate high operational usability often achieve excellent SUS scores of 82.75 [84], the integration of emerging technologies fundamentally necessitates a profound restructuring of the grower’s cognitive role, moving decision-making away from heuristic expertise toward systematic, data-driven practices.

Future research must therefore prioritize the theoretical integration of human-centered paradigms to manage increasing system complexity. The adoption of cognitive ergonomics is essential to minimize the cognitive load associated with supervising high levels of automation and interpreting sensor outputs. For the nascent DT framework to evolve from its current monitoring-centric state to truly prescriptive systems, robust models for human–AI collaboration (HAC) must be formalized, demanding transparent, trustworthy, and explainable decision support logic. Concurrently, the viability of immersive interfaces (AR/VR) is envisioned as the next step for complex 3D DT visualization and hinges on dedicated research on human factors to mitigate the critical barrier of simulation sickness, which is proven to negatively impact overall user evaluation and comfort during prolonged professional use.

Despite the technological maturity in core components, systemic and methodological gaps constrain scalability and equitable adoption. A significant limitation is the absence of standardized HCI metrics specific to agricultural settings, which impedes the rigorous comparative assessment and longitudinal evaluation necessary for iterative refinement. Furthermore, while high-performance AI promises substantial resource optimization, a critical ethical gap exists in quantifying the digital carbon footprint associated with the massive energy consumption required by DL models. Practical adoption is further restricted by high initial investment costs and critical infrastructure challenges, including the lack of internet stability and the low technological literacy prevalent in many rural agricultural areas. Addressing these barriers requires focusing future research on deep interface personalization and reliable multimodal systems to cater to diverse user demographics.

To strengthen the practical implementation of SHCI within AI-enabled agricultural systems, future studies should adopt a minimal, standardized reporting checklist that ensures transparency regarding the environmental costs of computation. At a minimum, researchers should specify the following: (1) the model type used (e.g., CNN, ANN) and dataset scale; (2) the hardware configuration for training and inference (GPU/CPU model, memory); (3) the total energy consumed during training and inference, reported in kWh; and (4) the carbon intensity factor associated with the cloud or edge location where computation was executed. Including these elements allows future work to quantify digital energy demand accurately and relate it to potential efficiency gains in greenhouse operations, thereby supporting verifiable, ethically aligned SHCI practices.

Consequently, the future direction of HCI in CEA must be guided by the SHCI imperative, ensuring that efficiency gains in the physical world are not offset by unsustainable computational costs. Moving forward, researchers must focus on developing adaptive, resilient, and ethically scaled systems. This requires establishing open interoperability architectures to unify disparate greenhouse subsystems, validating resilient infrastructure solutions (e.g., edge computing), and committing to longitudinal evaluations that validate usability in real-world agricultural contexts. The ultimate success of smart greenhouse innovation depends not merely on optimizing algorithms but fundamentally on designing interactions that augment human capabilities and harmonize automation with long-term ecological and social responsibilities.