Post-Harvest Loss Reduction in Perishable Crops: Task-Technology Fit and Emotion-Driven Acceptance of On-Farm Transport Robots

Abstract

As global food security challenges escalate and post-harvest losses in perishable crops remain a critical pressure point, on-farm transport robots have emerged as a promising sustainable solution for transforming farm-to-storage logistics systems and reducing agricultural waste. However, farmer acceptance of robotic transport technologies remains heterogeneous and represents a critical barrier to achieving widespread adoption of these sustainable agricultural innovations. Existing research has yet to integrate task-technology fit (TTF), anticipated emotions, and anthropomorphism into a unified theoretical framework for understanding sustainable agricultural technology adoption. Drawing on TTF theory and the model of goal-directed behavior, this study proposes a comprehensive model integrating anticipated emotions as mediators and robot anthropomorphism as a moderator. We surveyed 320 farmers and employed PLS-SEM to test our hypotheses. Results indicate that farm transport task complexity, farmer technology readiness, and robot transport functionality significantly strengthen TTF ($\beta = 0.136$, 0.358, 0.382, respectively; all $p < 0.01$). TTF drives acceptance intention through a dual-path emotional mechanism: directly enhancing positive expectancy emotions ($\beta = 0.411$, $p < 0.001$) while reducing negative expectancy emotions ($\beta = -0.150$, $p < 0.05$). Crucially, higher anthropomorphism levels diminish both emotional mediation paths ($\beta = -0.053$ and $\beta = -0.027$, both $p < 0.01$), establishing important boundary conditions for sustainable agricultural technology design. These findings suggest that reducing post-harvest losses requires prioritizing functional consistency over overly anthropomorphic designs in agricultural robots, thereby promoting the development of agricultural technologies that are both emotionally resonant and highly functional.

1. Introduction

Global food security imperatives underscore the critical urgency of reducing post-harvest losses in perishable crops, where a significant proportion of horticultural produce is compromised before reaching storage facilities. On-farm transport robots represent an emerging sustainable technology designed to transform farm-to-storage logistics by autonomously handling crops, thereby minimizing physical damage and spoilage. As illustrated in Figure 1, this study focuses on transport robots specifically designed for handling high-value perishable crops such as berries, tomatoes, and leafy greens. These robots typically feature gentle handling mechanisms to minimize mechanical damage, basic protective features to shield produce from environmental exposure, and navigation systems adapted to unstructured farm environments. Their primary function is to automate the movement of harvested crops from the field to packing stations or temporary storage facilities, addressing key causes of post-harvest losses through reduced handling and improved transit conditions [1].

Despite this potential, heterogeneous farmer acceptance remains a primary barrier to adoption, influenced by multifaceted factors including task-specific challenges, varying levels of technological readiness, and complexities in human–robot interaction. Current research lacks a unified theoretical framework that integrates task-technology fit(TTF), anticipated emotions, and robot anthropomorphism—a gap that hinders a comprehensive understanding of sustainable agricultural technology adoption. This study is the first to explicitly propose and test an integrative model incorporating TTF, positive and negative expectancy emotions, and degree of robot anthropomorphism as core determinants. By doing so, it offers novel theoretical insights into how cognitive, affective, and perceptual factors jointly influence adoption intentions, while providing practical guidance for the design and implementation of acceptable transport robotic solutions in agriculture.

Theoretical foundations establish TTF as pivotal, emphasizing that successful technology implementation requires precise alignment between robotic capabilities and agricultural operational demands. This functional perspective diverges from conventional acceptance models focused predominantly on perceived utility metrics and echoes recent emphasis on systemic coordination across technological, human, and environmental factors, as reflected in studies of ecosystem–agriculture synergy [2]. Complementing this approach, the model of goal-directed behavior(MGB) underscores the profound influence of anticipated emotions—encompassing farmer optimism regarding loss reduction and apprehensions about technological reliability—on behavioral intentions. Such affective mechanisms parallel the role of stakeholder perceptions and adaptive responses observed in other technology-driven sectors managing operational uncertainties [3]. Simultaneously, robot anthropomorphism systematically moderates user perceptions and affective responses through the implementation of human-like design features. While these constructs demonstrate individual explanatory power, their synergistic relationship within the context of agricultural robotics remains unexplored.

This study makes two key contributions: first, it identifies positive expectancy emotions as a critical mediator between TTF and adoption intention; second, it reveals a counterintuitive negative moderating effect of anthropomorphism, showing that high levels can weaken TTF’s positive impact by triggering emotional overload. Accordingly, we propose a novel design framework advocating strategic, functionally-grounded anthropomorphism to optimize user acceptance. To test this, we propose an integrated model that positions TTF—determined by farm transport task complexity (FTTC), robotic functionality specifications (RFS), and farmer technology readiness (FTR)—as the foundational antecedent; dual-valenced anticipated emotions as mediators between TTF and adoption intention; and anthropomorphic design elements as moderators of emotion activation pathways. Using partial least squares structural equation modeling with bootstrap validation, we examine this framework. This research advances post-harvest loss reduction theory by unifying functional, affective, and design paradigm constructs, yielding actionable insights for developing agricultural robots that optimize core functional alignment while maintaining emotionally resonant interfaces.

2. Literature Review and Hypothesis Development

2.1. Task-Technology Fit Theory

TTF Theory, proposed by Goodhue and Thompson in 1995 [4], has its theoretical foundation in Perceptual Matching Theory, which focuses on the extent to which information technology aligns with users’ task requirements and explores the logical relationship between the technology’s characteristics and the task’s requirements. This theory provides a crucial analytical framework for assessing the applicability of technology in specific task environments. The dimensions of TTF are primarily reflected in two aspects: first, it analyzes the dynamic interaction between technology and task, exploring how technology can effectively support the achievement of task objectives; second, it quantifies the degree of fit between the two and assesses whether the technology can be effectively implemented in specific task scenarios. The theoretical framework of TTF includes three key elements: task characteristics, technology characteristics, and the degree of fit. Task characteristics encompass task complexity, types of task attributes, and competence standards of the executors; technology characteristics include functional completeness, ease of operation, and system performance; and the degree of fit intuitively reflects how well the technology aligns with task characteristics, directly impacting the effectiveness of technology implementation. Unlike technology acceptance theory, which focuses on users’ psychological cognition, TTF theory emphasizes the adaptability of technology to task characteristics in practical applications, noting that even advanced technology is unlikely to gain user acceptance if it does not align with core task requirements.

TTF has evolved into a mature theoretical framework within information systems. Early research established the technology adoption–performance relationship, demonstrating that task, technology, and user characteristics shape usage behavior through TTF [5]. Theoretical advancements are epitomized by Howard and Rose’s seminal expansion of the adaptation concept, which distinguishes TTF from technology-task matching (TTM) [6], subdivides TTM into deficit and surplus typologies, and develops quantifiable scales. Emerging technological contexts continue to drive innovation: in multitasking media environments [7], TTF emerges as a core driver of podcast adoption beyond traditional theories; in artificial intelligence, task-oriented fitness frameworks have been constructed; and blockchain federations empirically demonstrate cross-organizational adaptation’s critical role in decentralized transformation. Jeyaraj’s meta-analysis further confirms that methodological elements moderate TTF outcomes [8]. Current research reveals three convergent trends: multi-theoretical integration, contextual refinement, and interdisciplinary expansion. These advances provide theoretical and practical pathways for optimizing technology adaptation and human–machine collaboration.

2.1.1. Task Characteristics

Grounded in the core structure of TTF theory and considering the operational characteristics of farm transport scenarios for perishable produce [4], this study decomposes task characteristics into two key variables: FTTC and FTR.

FTTC: Rooted in the core dimensions of TTF theory [6,8], this construct reflects the multifaceted challenges inherent in handling and transporting perishable produce. It captures operational demands such as completing transport promptly, addressing task urgency, and coordinating movements across multiple locations (e.g., from field to storage or processing units). These requirements collectively shape the functional and performance criteria that on-farm transport technologies must fulfill [9]. For example, transporting highly perishable crops such as strawberries—which are vulnerable to compression and vibration—requires not only speed and accuracy but also the ability to operate under time pressure across varying terrain. Such practical instances illustrate how FTTC directly influences technology design and performance expectations. FTR: This construct denotes the resource reserves, knowledge levels, and ability to access external support of farmers when engaging in on-farm transport tasks. Reflecting the impact of user characteristics on technology utilization [10], it fundamentally delimits farmers’ capacity to successfully employ on-farm transport technologies.

Existing empirical studies consistently affirm that task characteristics and user characteristics are salient antecedents of TTF [9,11]. Within the agricultural post-harvest handling domain, FTTC—manifested through inherent spatiotemporal constraints—defines the core attributes that transport technologies must embody [9]. Concurrently, FTR directly shapes the realized efficacy of technology applications and user experience by influencing operators’ capabilities, resource availability, and knowledge base, thereby affecting the efficiency and satisfaction associated with technologies such as on-farm transport robots [10].

Grounded in the central tenets of TTF theory [4] and corroborated by empirical evidence [7,9], the functionality of on-farm transport technologies must simultaneously address the efficiency and spatial management requirements imposed by FTTC, while also accommodating the operational feasibility and resource thresholds defined by farmers’ technology readiness. Only when these dual imperatives are met can a high level of TTF be attained. Drawing upon the preceding theoretical rationale and extant empirical support, we advance the following hypotheses:

H1. 

Farm Transport Task Complexity positively influences task-technology fit.

H2. 

Farmer Technology Readiness positively influences task-technology fit.

2.1.2. Technical Characteristics

According to TTF theory [4], technical features constitute primary determinants of fit, as their functional attributes must dynamically align with task requirements. In the context of on-farm robotic transport, robot transport functionality (RTF)—a core dimension of technical features—reflects the capacity of autonomous technologies to fulfill post-harvest operational demands. RTF encompasses three functional sub-dimensions: (1) efficiency enhancement, (2) operational convenience, and (3) user experience optimization [4].

Empirical studies across domains consistently demonstrate the pivotal role of technology functionality in achieving fit. In agricultural robotics, research shows that real-time monitoring capabilities enhance handling efficiency only when aligned with task complexity [9]. Similarly, studies on perishable crop management reveal that attributes like adaptive route planning and environmental sensing enable seamless operations by matching the spatiotemporal constraints of post-harvest tasks. The task-adaptation framework for agricultural AI further emphasizes that functionality must be tailored to distinct task types (e.g., produce sorting versus temperature-controlled transport) to ensure effective implementation [12]. Howard et al. caution that both insufficient and excessive functionality disrupt fit [6]; for instance, overly complex robot interfaces may impose cognitive burdens on farm operators, reducing adoption efficacy [13].

Grounded in the theoretical nexus between technical and task characteristics and supported by empirical evidence, we posit that RTF directly enhances TTF by improving technological execution. Consequently, we propose the following hypothesis:

H3. 

Robot transport functionality positively influences TTF.

2.1.3. Task-Technology Fit

TTF theory examines the alignment between technological capabilities and task requirements. Within on-farm robotic transport, TTF denotes the degree to which the technical attributes of agricultural robots satisfy operational demands for perishable crop handling, encompassing functional compatibility and task-specific adequacy [4].

Extant research demonstrates TTF’s substantial influence on user attitudes, behaviors, and affective responses. While traditionally emphasizing performance and cognitive appraisal, TTF’s scope extends to emotional outcomes. In agricultural technology contexts, studies confirm that enhanced TTF elevates operators’ perceived efficiency and system reliability [14]. This heightened perceived value correlates with positive affect (e.g., confidence in loss reduction) and mitigates frustration from technological misalignment during time-sensitive post-harvest operations.

Grounded in TTF’s capacity to shape cognitive evaluations—where favorable outcomes co-occur with positive expectancies while suppressing negative ones [15]—we propose:

H4. 

Task-technology fit of on-farm transport robots positively influences farmers’ positive expectancy emotions.

H5. 

Task-technology fit of on-farm transport robots negatively influences farmers’ negative expectancy emotions.

2.2. Anthropomorphism

Research on robot anthropomorphism has evolved significantly with technological progress. Early work prioritized morphological mimicry of human traits, exemplified by Phillips et al.’s ABOT database identifying key visual cues for human-like perception [16] and Kang et al.’s development of bionic manipulators for hazardous agricultural tasks [17]. Contemporary scholarship extends to behavioral interaction, with breakthroughs like Duan et al.’s human–robot handover framework demonstrating action-level anthropomorphism [18]. In agricultural contexts, anthropomorphic design can enhance user experience: Behavioral adaptations in agri-robots (e.g., adaptive obstacle avoidance) improve operator trust and engagement, moderated by farmers’ technological readiness. Empirical studies reveal nuanced effects, where gaze-based interaction heightens perceived competence [19], yet excessive morphological human-likeness in field environments may trigger unease. Task dependency is critical—Ma et al. showed that high anthropomorphism benefits trust-intensive operations like perishable produce handling, but risks evoking eeriness in low-interaction tasks like routine bulk transport [20]. Peter et al. further identified synergistic effects between affective and morphological anthropomorphism in fostering human–robot collaboration through enhanced social presence [21], moderated by operators’ social-interaction needs.

Ethical and practical challenges persist in agricultural implementation. While anthropomorphism improves initial engagement [21], over-application in unpredictable farm settings may provoke rejection, necessitating scenario-specific calibration. Unresolved dilemmas include deception risks from over-attributing human capabilities, relational reconfiguration in farmer–robot dynamics, and “uncanny valley” effects in natural environments [22]. Consequently, anthropomorphic design requires balancing technical robustness (e.g., weather-resistant sensors), farmer psychological acceptance, and ethical boundaries. We therefore hypothesize that anthropomorphism attenuates emotion formation:

H6. 

The anthropomorphism level of on-farm transport robots negatively moderates the relationship between TTF and farmers’ positive expectancy emotions.

H7. 

The anthropomorphism level negatively moderates the relationship between TTF and farmers’ negative expectancy emotions.

2.3. Model of Goal-Directed Behavior

MGB delineates how goals direct cognition and action by reconfiguring motivational and cognitive structures [15]. The theory offers an integrated framework encompassing goal setting, cognitive restructuring, and behavioral transformation, and systematically examines goal typologies, action mechanisms, and situational contingencies.

Its principal scholarly contribution is its robust, multilevel explanatory capacity. In group contexts, goal-directed behaviors exert significantly stronger reinforcing effects than at the individual level [19], underscoring the pivotal role of the social environment. Moreover, synergy between goal-control mechanisms and the executive agent’s characteristics yields superior behavioral outcomes relative to mismatched configurations. The theory also demonstrates strong interdisciplinary applicability; for instance, the incorporation of affective and situational variables in collaborative learning yields behavioral-intention models with high predictive validity [22].

Nonetheless, the theory exhibits limitations. Interactive moderating effects involving environmental stressors and technological interventions remain under-specified in complex, dynamic settings [23]. For example, stress or technological interventions may elicit conflict between goal-directed control and habitual responses, producing ambivalent behavioral outcomes [24]. Additionally, insufficient attention to heterogeneous factors—such as individual cognitive styles—constrains the generalizability of interventions [25].

Despite these constraints, the theory enjoys extensive practical application. In organizational management, learning-goal orientation effectively guides employee behavior and mitigates withdrawal tendencies by enhancing perceived meaning and eliciting positive affect [26]. In education, it underpins the optimized design of collaborative-learning environments [27]. Within technology-innovation management, a nuanced understanding of goal-orientation mechanisms helps reconcile the dual impacts of new technologies on innovation, offering theoretical guidance for resource allocation and systematically enhancing innovation performance.

2.3.1. Positive Expectancy Emotions

Within the goal-directed behavior framework, positive expectancy emotions(PEE) represent a pivotal affective construct defined as an individual’s anticipatory belief that specific actions will yield favorable affective outcomes [15]. Operationally, this encompasses dimensions of pleasantness, satisfaction, and confidence in achieving desired results [22]. Empirical studies consistently demonstrate its central role in behavioral decision-making, where predictive power frequently exceeds traditional cognitive determinants like subjective norms.

In agricultural technology adoption, these emotions function similarly to other domains: Farmers’ anticipatory positive affect toward efficient solutions enhances implementation intentions. Although boundary conditions require further exploration, validated mechanisms confirm PEE as critical bridges connecting technological expectations to behavioral motivation. This establishes a quantifiable pathway for predicting adoption behaviors. Consequently, we hypothesize the following:

H8. 

Positive expectancy emotions exert a significant positive influence on farmers’ behavioral intention to accept on-farm transport robots.

2.3.2. Negative Expectation Emotions

Negative expectancy emotions(NEE) encompass anticipatory affective states arising from expectations of adverse outcomes in agricultural operations. Within the context of robotic field-to-farm transport, these emotions manifest when farmers anticipate operational risks such as crop damage during transit over uneven terrain, scheduling disruptions from technical failures, or reliability concerns during adverse weather conditions. Empirical studies demonstrate that such emotions accumulate when technology threatens core objectives like harvest efficiency, heightening resistance to adoption. Crucially, these inhibitory effects can be mitigated through goal-oriented interventions—demonstrating robotic system reliability through controlled pilot trials has been shown to attenuate behavioral resistance. Building on this empirical foundation, we hypothesize the following:

H9. 

Negative expectancy emotions exert a significant negative influence on farmers’ behavioral intention to accept on-farm transport robots.

3. Materials and Methods

3.1. Research Model

This study develops an integrative framework that merges TTF theory with theMGB and incorporates the degree of robot anthropomorphism as a key moderator. Goodhue and Thompson’s TTF framework convincingly demonstrates the importance of TTF for user behavior [4], yet it leaves a critical theoretical gap. Dishaw and Strong argue that TTF neither explains the psychological pathways linking fit to behavioral decisions nor accounts for the affective consequences of technology fit [25]. In the transport-robot context, for instance, navigational accuracy may fully satisfy path-planning requirements, yet users may still experience anxiety due to poor interaction quality, thereby reducing usage intention—as documented in studies of dexterous human–robot handover [18].

Incorporating the MGB model addresses this void. Perugini and Bagozzi underscore the central role of expectancy emotions in shaping behavior [15], but prior research has rarely examined these emotions within a technology-fit paradigm. Extending prior integrations of TTF and technology-acceptance perspectives, the present study employs MGB’s expectancy-emotion constructs as mediators. TTF can enhance usage intention via PEE or suppress it through NEE. This synthesis not only corroborates the proposition—advanced by TTF and pedagogical-innovativeness research [11]—that technological efficacy must be translated through affective channels, but also remedies TTF’s long-standing neglect of underlying psychological mechanisms.

To enhance the model’s situational specificity, we introduce the degree of robot anthropomorphism as an additional variable that operates through a dual-path mechanism. First, high anthropomorphism amplifies social presence [19], thereby intensifying the positive affect generated by strong TTF. Second, anthropomorphic cues attenuate the negative affect elicited by a weak fit [20]. Experimental evidence indicates that when a robot is framed as a partner, anthropomorphism increases the proportion of variance in usage intention explained by TTF by 37% [18]. This configuration aligns with transport contexts characterized by simultaneous time-critical demands and affective interactions [20], addressing the growing call for context-sensitive theorizing in human–robot interaction. Integrating these theoretical refinements and variable innovations, we identify the key determinants of continued usage intention and present the research model depicted in Figure 2.

The conceptual framework of this study is presented in Figure 1. To formalize the proposed relationships between constructs and enhance methodological transparency, the structural model is specified as a system of equations following established conventions in structural equation modeling (SEM). The model examines both direct effects and moderating relationships within a predictive framework.

The structural model defines the hypothesized relationships between latent constructs. The system of equations is specified as follows:

$$\mathrm{T}\mathrm{T}\mathrm{F}={\gamma }_{11}\mathrm{F}\mathrm{T}\mathrm{T}\mathrm{C}+{\gamma }_{12}\mathrm{F}\mathrm{T}\mathrm{R}+{\gamma }_{13}\mathrm{R}\mathrm{T}\mathrm{F}+{\zeta }_1$$

(1)

$$\mathrm{P}\mathrm{E}\mathrm{E}={\beta }_{21}\mathrm{T}\mathrm{T}\mathrm{F}+{{\beta }_{22}\left(\mathrm{T}\mathrm{T}\mathrm{F}\times \mathrm{D}\mathrm{A}\mathrm{T}\mathrm{R}\right)+\zeta }_2$$

(2)

$$\mathrm{N}\mathrm{E}\mathrm{E}={\beta }_{31}\mathrm{T}\mathrm{T}\mathrm{F}+{\beta }_{32}\left(\mathrm{T}\mathrm{T}\mathrm{F}\times \mathrm{D}\mathrm{A}\mathrm{T}\mathrm{R}\right)+{\zeta }_3$$

(3)

$$\mathrm{F}\mathrm{B}\mathrm{I}\mathrm{T}\mathrm{R}={\beta }_{42}\mathrm{P}\mathrm{E}\mathrm{E}+{\beta }_{43}\mathrm{N}\mathrm{E}\mathrm{E}+{\zeta }_4$$

(4)

The structural model presented in Equations (1)–(4) incorporates several coefficients that quantify the hypothesized relationships between constructs. The gamma coefficients (γ₁₁, γ₁₂, γ₁₃) in Equation (1) represent the direct effects of the exogenous variables FTTC, FTR, and RTF on TTF. In Equations (2) and (3), the beta coefficients β₂₁ and β₃₁ capture the direct influence of TTF on PEE and NEE, respectively. Additionally, the interaction coefficients β₂₂ and β₃₂ reflect the moderating effect of the DATR on the relationship between TTF and emotional responses. Finally, in Equation (4), β₄₂ and β₄₃ denote the paths through which PEE and NEE, respectively, influence FBITR. All equations include disturbance terms (ζ₁–ζ₄), representing unexplained variance.

3.2. Research Instrument

This study rigorously adheres to structural equation modeling standards, adapting measurement instruments due to the absence of domain-specific scales for farmers’ adoption intention toward on-farm transport robots. Validated scales from agricultural technology acceptance, autonomous machinery, and human–robot interaction research were synthesized using Zhou et al.’s TTF framework as guidance, with all items aligned to construct definitions. Data collection employed mixed online and offline channels, followed by descriptive screening. After pretest refinement, the main survey ensured adequate statistical power and structural validity through a three-stage scale purification protocol: First, three Delphi rounds with agricultural engineers, perishable crop specialists, and HCI experts assessed content validity and structural coherence; then items were revised for enhanced relevance to field-to-farm transport scenarios; finally, confirmatory factor analysis of pretest data eliminated items with loadings below 0.70, yielding a psychometrically robust 30-item instrument presented in

Table 1. Question items and references.

Constructs	Coding	Items	References
Farm transport task complexity	FTTC1	Transport tasks need to be completed on time.	[4,5,6]
	FTTC2	Transport tasks are usually urgent.
	FTTC3	Transport tasks need to be done at different locations.
Robot transport functionality	RTF1	If a robotic transport service were offered, I would use it immediately because it saves time, reduces costs, and is easy to operate.	[28]
	RTF2	I would not hesitate to use robotic transport if it is more comfortable than traditional transport methods.
	RTF3	If robotic transport made my work/life easier, I would definitely use the service.
	RTF4	Robot transport works for me.
Farmer technology readiness	FTR1	I have the resources needed to use robotic transport.	[29]
	FTR2	I was able to acquire the knowledge needed to use robotic transport.
	FTR3	I can get support from others to help me when I have trouble using the robot transport.
	FTR4	Robotic transport matches my habits and lifestyle.
	FTR5	I can quickly learn how to easily use robotic transport.
Task-Technology Fit	TTF1	The technical capabilities of the robotic transport system are sufficient to help me receive packages.	[30]
	TTF2	The technical features of the robotic transport system are appropriate in terms of helping me receive packages.
	TTF3	Overall, the functionality of the robotic transport system meets my transport service needs.
Degree of anthropomorphism of transport robots	DATR1	Robots make me feel as if I’m interacting with intelligent beings.	[12]
	DATR2	The robot gives me an experience similar to that of a real person’s companionship.
	DATR3	The robot will see me as a real communication partner.
Positive expectancy emotions	PEE1	I will feel pleasant when I get good service using robot transport.	[15,22]
	PEE2	I will feel satisfied when I get good service using robotic transport.
	PEE3	I will be happy when I get good service using robot transport.
Negative expectancy emotions	NEE1	I get angry when not using robotic transport, which results in inefficient transport compared to traditional human transport.	[15,22]
	NEE3	I was disappointed when not using robotic transport led to an increase in transport errors compared to traditional human transport.
	NEE3	I was concerned that not using robotic transport caused delays in transport time compared to traditional human transport.
	NEE4	I was pained concerned when not using robotic transport caused delays in transport time compared to traditional human transport.
Farmers’ behavioral intention for on-farm transport robots	FBITD1	I would like to receive items through a robotic transport service.	[31]
	FBITR2	I would prioritize robotic transport for order receipt.
	FBITR3	I can see myself using robotic transport to receive packages in the future.
	FBITR4	Based on my expectations for transport methods, I will try to use robotic transport.

.

Eight core constructs were operationalized using seven-point Likert scales: farm transport task complexity; robot transport functionality; farmer technology readiness; task-technology fit; degree of anthropomorphism of transport robots; positive expectancy emotions; negative expectancy emotions and farmers’ behavioral intention for on-farm transport robots. Pretest analyses confirmed strong psychometric properties, demonstrating internal consistency (Cronbach’s α ≥ 0.80), reliability (composite reliability ≥ 0.85), convergent validity (AVE ≥ 0.50), and discriminant validity via the Fornell–Larcker criterion. Data quality was ensured through triangulated procedures, including attention-filter items, IP-address deduplication, and minimum completion-time thresholds exceeding 180 s, while capturing gender, age, and prior agricultural robotics experience as covariates for model testing.

3.3. Data Collection

This study focused on farmers’ behavioral intention to accept on-farm transport robots. All participants were fully informed of the study details before completing the questionnaire, while reserving the right to withdraw at any time. The survey was conducted between March and May 2025 by combining online and offline methods. Trained field staff conducted face-to-face surveys using tablet-assisted interviews to ensure comprehension of technology concepts, supplemented by targeted online questionnaires distributed through farming WeChat groups to younger farmers (under 55) with confirmed digital proficiency. By combining online and offline methods, 380 farmers’ responses were initially collected from 18 provincial administrations.

Based on the criteria proposed by Wu et al. [5], this study underwent three stages of rigorous screening to ensure sample quality. First, pre-tests showed that completing the questionnaire usually took 1–5 min, and records that took less than 90 s were considered invalid. Second, attitudes toward responding were detected through reverse-coded questions embedded in the questionnaire, and those who failed to respond correctly were excluded. Finally, questionnaires with completely duplicated content were also excluded. After these screening steps, a total of 60 invalid data points were removed, and 320 high-quality questionnaires were finally retained for subsequent analysis.

Regarding the reasonableness of the sample size, the study referred to Loehlin’s [32] analysis of 72 SEM studies with a median sample size of 198. Barrett suggested that the sample size should be at least eight times the number of variables in the model (the requirement for the present study was 30 × 8 = 240) [33]. Although Barrett noted that the chi-square value may be inflated when the sample size exceeds 500 and ML estimation is used, the final 320 valid samples in this study significantly exceeded the minimum requirement (240) and fully satisfied the analysis.

4. Results

4.1. Descriptive Statistical Analysis

Sample demographics are presented in

Table 2. Demographic Characteristics of the Sample (N = 320).

Variable	Category	Frequency	Percentage
Sex	Male	153	47.8%
	Female	167	52.2%
Age	Under 49	186	58.1%
	50–59	102	31.9%
	60 and over	32	10.0%
Experience with agricultural robotics	Never used	81	25.3%
	Observed not operated	141	44.1%
	Operated	98	30.6%

Due to the rounding of percentages to one decimal place, totals may not add up to exactly 100%. This is a common statistical presentation and does not affect the validity of the data.

. The final valid sample comprised 320 farmers engaged in perishable crop production across 18 major agricultural provinces. This sample size exceeds the threshold of 8× observed variables (30 × 8 = 240) required for structural equation modeling (SEM), satisfying statistical power requirements. Participants were 58.1% aged under 49 years, reflecting the contemporary demographic profile of China’s agricultural workforce, where younger farmers increasingly drive technology adoption. Experience levels with agricultural robotics were distributed across three tiers: 25.3% had never used such systems, 44.1% had observed but not operated them, and 30.6% possessed direct operational experience. This distribution captures critical variance in technology exposure essential for analyzing adoption determinants.

4.2. Measurement Model Evaluation

SmartPLS 4 analysis confirms robust psychometric properties across the measurement model, validating the reliability and validity of the constructs critical to understanding farmers’ adoption of on-farm transport robots. As presented in

Table 3. Reliability and Validity Indicators of Measurement Models.

Constructs	Items	Outer Loadings	AVE	C.R.	Cronbach’s Alpha	HTMT Confidence Intervals
FTTC	FTTC1	0.846	0.663	0.855	0.750	Excluding 1
	FTTC2	0.820
	FTTC3	0.775
RTF	RTF1	0.759	0.606	0.860	0.783	Excluding 1
	RTF2	0.799
	RTF3	0.803
	RTF4	0.753
FTR	FTR1	0.722	0.590	0.878	0.826	Excluding 1
	FTR2	0.772
	FTR3	0.835
	FTR4	0.760
	FTR5	0.747
TTF	TTF1	0.840	0.668	0.858	0.751	Excluding 1
	TTF2	0.769
	TTF3	0.841
DATR	DATR1	0.844	0.707	0.879	0.795	Excluding 1
	DATR2	0.850
	DATR3	0.828
PEE	PEE1	0.821	0.712	0.881	0.798	Excluding 1
	PEE2	0.860
	PEE3	0.849
NEE	NEE1	0.821	0.656	0.884	0.826	Excluding 1
	NEE2	0.807
	NEE3	0.816
	NEE4	0.795
FBITR	FBITR1	0.808	0.618	0.866	0.794	Excluding 1
	FBITR2	0.737
	FBITR3	0.767
	FBITR4	0.830

, all indicator loadings significantly exceed the rigorous threshold of 0.70 (range: 0.722–0.860), demonstrating strong convergent validity. This indicates that each measurement item effectively captures the essence of its assigned latent construct, with minimal cross-loading interference. Internal consistency is further reinforced by two complementary metrics: Composite reliability (CR) values (0.855–0.884) and Cronbach’s alpha (α) coefficients (0.750–0.826), both surpassing the 0.70 benchmark. The high CR scores confirm that constructs are free from random measurement error, while acceptable α values indicate item coherence without redundancy [34].

To further assess potential multicollinearity issues among the indicators, the variance inflation factor (VIF) was examined for all constructs. The results indicated that all outer VIF values ranged from 1.000 to 2.042, which is well below the conservative threshold of 3.0 and the common cutoff of 5.0 [35]. These findings suggest that multicollinearity does not pose a concern for the stability of the measurement model estimates.

The establishment of discriminant validity constitutes a fundamental methodological prerequisite in structural equation modeling, ensuring empirical distinction between constructs. This study rigorously validates discriminant validity through the Fornell–Larcker criterion, mandating that each construct’s square root of Average Variance Extracted must exceed its bivariate correlations with all other constructs. Empirical evidence from

Table 4. Latent Factors’ AVE Square Root Correlations.

	FBITR	FTR	DATTR	FTTC	NEE	PEE	RTF	TTF
FBITR	0.786
FTR	0.722	0.768
DATR	0.651	0.691	0.841
FTTC	0.520	0.597	0.552	0.814
NEE	0.417	0.412	0.372	0.422	0.810
PEE	0.772	0.663	0.689	0.487	0.367	0.844
RTF	0.742	0.727	0.663	0.491	0.394	0.704	0.779
TTF	0.726	0.717	0.599	0.537	0.390	0.707	0.709	0.818

confirms robust discriminant validity: all diagonal elements systematically surpass off-diagonal correlations, with consistent adherence observed across constructs. Figure 3’s correlation heatmap provides a complementary visualization, where dominant diagonal elements exhibit a clear visual distinction from systematically lower off-diagonal correlations. This dual validation satisfies the Fornell–Larcker criterion and demonstrates minimal construct overlap, confirming psychometric robustness. The statistically distinct constructs prevent multicollinearity bias in subsequent path analysis, ensuring unambiguous interpretation of TTF and emotion-driven adoption mechanisms. This metrologically sound foundation establishes empirical credibility for agricultural robotics research, enabling precise theoretical inferences and practical implementations in technology adoption frameworks.

4.3. Structural Model Validation

4.3.1. Model Fit Test

In this study, partial least squares structural equation modeling (PLS-SEM) was employed, and the model’s goodness of fit was comprehensively assessed using the three primary fit indices proposed by Henseler et al. [13]. The original-sample-standardized root mean square residual (SRMR) was 0.086, slightly exceeding Hu and Bentler’s stringent threshold of 0.08 [35]. However, as noted by Henseler et al. [13], the bootstrap-generated confidence interval for the SRMR provides a more robust assessment of model fit than a single point estimate. Our reported 95% confidence interval for the SRMR is [0.067, 0.072], which lies entirely below the 0.08 threshold. This indicates that the slight exceedance in the original sample is not a systematic deviation and that the model demonstrates acceptable fit within the population [29]. Meanwhile, the original-sample H₀ fit index (d_G) is 0.724, which markedly surpasses the strong-fit criterion of 0.65 set by Wetzels et al. [36]; values exceeding 0.70 are generally considered excellent, thereby confirming the model’s superior fit. The original-sample-unweighted least squares distance (d_ULS) was 3.182. Its bootstrap 95% confidence interval [1.979, 2.235] excludes zero and centers on a mean of 1.474. This provides statistical evidence that the model significantly outperforms saturated models [37], implying that the residual distribution is systematically and acceptably different from that of a saturated model.

Bootstrap tests (5000 resamples) revealed significant R² values for all endogenous constructs ($p < 0.001$). The ultimate dependent variable, FBITR, achieved an R² of 61.7%, attesting to the model’s strong predictive validity. Among the mediators, PEE and TTF explained 60.0% and 61.7% of their respective variances, collectively forming a robust transmission pathway.NEE demonstrated a lower but still significant explanatory power (${\mathrm{R}}^2 = 22.0\mathrm{\%}$), also contributing to the robust transmission pathway. Although the high R² values for most constructs could raise concerns about potential overfitting, particularly given the sample size (N = 320), several indicators confirm the model’s robustness and predictive relevance. The cross-validated communality values for all constructs are positive (ranging from 0.325 to 0.421), demonstrating adequate predictive relevance. The cross-validated redundancy values for the key endogenous variables—FBITR (Q² = 0.370), PEE (Q² = 0.427), and TTF (Q² = 0.394)—are all substantively above zero. Notably, while NEE shows a lower redundancy value (Q² = 0.133), it remains positive, indicating that the model maintains predictive power for this construct as well. These results, obtained through blindfolding procedures, demonstrate that the model possesses predictive power and is not merely overfitted. Therefore, while the sample size is moderate, the model demonstrates both explanatory and predictive validity across all endogenous constructs. To further validate robustness, we conducted a split-sample sensitivity analysis (randomly dividing N = 320 into two subsamples, n1 = 160, n2 = 160). The stability of the model was further confirmed by a split-sample sensitivity analysis. The SRMR values for the two subsamples (0.084 and 0.087) were close to the 0.08 threshold, and, more importantly, their bootstrapped confidence intervals entirely fell below this benchmark, consistent with the full-sample results. The d_G values (0.718 and 0.729) surpassed the 0.65 criterion for strong fit. Furthermore, all Q² values remained positive and stable across subsamples. These results collectively confirm the model’s robustness and its insensitivity to sample segmentation.

Collectively, the SRMR confidence interval supports an acceptable approximate fit, d_G demonstrates excellent fit, and d_ULS confirms that the model significantly outperforms a saturated model. Furthermore, the model exhibits strong predictive relevance, as evidenced by substantial cross-validated redundancy (Q²) values for all key endogenous constructs. Therefore, the structural model not only demonstrates good overall goodness of fit but also possesses robustness and generalizability, mitigating concerns.

4.3.2. Path Coefficients and Hypothesis Testing

Before interpreting the path coefficients, it is crucial to note that the reported path coefficients (β) are standardized estimates. This standardization involves scaling all variables to a common metric (i.e., standard deviation units), which remains meaningful even when all variables are measured using 7-point scales. Specifically, despite the shared 7-point format, different variables may exhibit distinct levels of dispersion, rendering raw coefficient magnitudes incomparable. Consequently, β coefficients represent the change in the dependent variable, in standard deviation units, for every standard deviation increase in the independent variable, while controlling for other variables in the model. This scaling enables the direct comparison of the magnitude of β coefficients, thereby reflecting the relative strength of relationships within the model.

All hypothesized paths were evaluated using bias-corrected 95% confidence intervals derived from 5000 bootstrap samples; every path achieved statistical significance. In

Table 5. Structural Model Path Coefficients and Significance Levels.

Pathway Relationship	Path Factor	t-Value	p-Value	95% Confidence Intervals	Hypothesis Testing
FTTC → TTF	0.136	2.701	0.007	[0.032, 0.230]	H1 established
RTF → TTF	0.382	6.712	0.000	[0.265, 0.487]	H2 established
FTR → TTF	0.358	5.904	0.000	[0.239, 0.478]	H3 established
TTF → PEE	0.411	7.059	0.000	[0.298, 0.519]	H4 established
TTF → NEE	0.150	2.076	0.038	[0.001, 0.284]	H5 established
PEE → FBITR	0.715	20.704	0.000	[0.641, 0.777]	H8 established
NEE → FBITR	0.155	3.749	0.000	[0.079, 0.239]	H9 established

, FTTC ($\beta =0.136$, $p <0.01$), RTF ($\beta =0.382$,$p <0.001$), and FTR ($\beta = 0.358$, $p < 0.001$) exerted significant positive effects on TTF, supporting H1, H2, and H3. TTF, in turn, significantly influenced both PEE ($\beta = 0.411$, $p < 0.001$) and NEE ($\beta = 0.150$, $p < 0.05$), thereby confirming H4 and H5. Finally, PEE ($\beta =0.715$, $p < 0.001$) and NEE ($\beta = 0.155$, $p < 0.001$) each had a significant positive impact on farmers’ adoption intention for on-farm transport robots, validating H8 and H9 [38].

4.3.3. Analysis of Moderating Effects

This study examines whether the degree of robot anthropomorphism moderates the indirect paths from TTF to expectancy emotions and subsequently to farmers’ intention to adopt robot transport, as presented in

Table 6. Path Coefficients and Significant Levels of Regulatory Variables.

Pathway Relationship	Path Coefficient	t-Value	95% Confidence Intervals	p-Value	Hypothesis Validation
DATR × TTF → PEE → FBITR	−0.053	2.633	[−0.095, −0.015]	0.009	H6 established
DATR × TTF → NEE → FBITR	−0.027	3.566	[−0.045, −0.015]	0.000	H7 established

.

For the positive-emotion path, the path coefficient is −0.053 ($t = 2.633$, $p = 0.009$); the 95% bias-corrected bootstrap CI [−0.095, −0.015] does not include zero. These results indicate a significant negative moderating effect. Although TTF initially fosters PEE (e.g., pleasure, anticipation) and thereby increases acceptance intention, higher anthropomorphism—manifested in affective voice interaction, human-like facial expressions, or similar design cues—attenuates this mediation. Consequently, H6 is supported.

For the negative-emotion path, the path coefficient is −0.027 ($t = 3.566$, $p < 0.001$); the 95% CI [ −0.045, −0.015] again excludes zero. Thus, anthropomorphism also exerts a significant negative moderating effect. Whereas TTF normally suppresses NEE (e.g., anxiety, disappointment) and indirectly bolsters acceptance intention, increasing anthropomorphism weakens this suppressive mediation. H7 is therefore confirmed.

Taken together, robot anthropomorphism systematically dampens the mediating role of both positive and negative expectancy emotions in translating TTF into acceptance intention. Rather than providing a simple affective enhancement, heightened anthropomorphism interferes with the core mechanism through which technological appropriateness generates farmers’ willingness to adopt robot transport.

Although the moderating coefficients ($\beta = -0.053$ and $-0.027$) may appear small, they are statistically significant and carry meaningful practical implications. In real-world agricultural settings, where adoption decisions involve a complex weighing of cost, reliability, and perceived usefulness, the observed effects can translate to a measurable reduction in adoption likelihood. This occurs because anthropomorphism undermines the crucial link between task-fit and emotional confidence, even when the functional benefits of the technology remain unchanged. The consistently negative effects across both emotional channels suggest that anthropomorphism, despite being well-intentioned, ultimately dilutes the functional-emotional mechanism that supports adoption. Therefore, stakeholders should consider the following recommendations:

• Designers and manufacturers should prioritize functional clarity and reliability over highly anthropomorphic features. For example, simplifying emotional expressions, using task-specific auditory feedback instead of conversational language, and minimizing non-functional human-like design elements can help preserve the link between TTF and positive user emotions without compromising operational efficiency.
• Policymakers can promote the adoption of agricultural robots by establishing functional design standards or certification systems that discourage non-essential anthropomorphism. Subsidies, procurement incentives, and technical guidelines favoring robots with high task-appropriate functionality could accelerate the diffusion of robotic technologies in rural areas.
• Farmer training institutions and extension services should integrate these findings into instructional programs. Emphasizing functional benefits—such as adaptability, durability, and operational effectiveness—over anthropomorphic features can help farmers make more informed decisions aligned with practical needs. Training should highlight how task-robot fit enhances efficiency and reduces emotional uncertainty, supporting longer-term adoption.

In summary, these “small” coefficients reveal a critical insight: in agriculture, where utility drives adoption, even modest anthropomorphism can erode the mechanism through which technology fit translates to acceptance. By addressing this across design, policy, and training, stakeholders can strengthen the pathway from technological appropriateness to widespread adoption of farm transport robots.

5. Discussion

5.1. Explanation of the Mediating Effect of Expected Emotions

This study confirms that expectancy emotions significantly mediate the relationship between TTF and farmers’ acceptance intention. The standardized indirect effect via PEE ($\beta = 0.715$) is substantially stronger than that via NEE ($\beta = 0.155$), underscoring the dominant role of affective gain over loss avoidance in technology adoption decisions. When the functional attributes of a transport robot closely align with farmers’ transport task requirements, consumers anticipate pleasure and satisfaction, and this affective anticipation exerts a stronger influence on behavioral intention than concerns about potential negative experiences.

Theoretically, the finding extends the explanatory scope of TTF. Prior research has predominantly emphasized the objective alignment between technology and task characteristics [4]; our results reveal that this alignment must be emotionally translated to fully shape behavioral intentions. Specifically, the TTF → PEE path coefficient of 0.411 indicates that a one-unit increase in fit raises positive affective expectancies by 0.411 units, which in turn significantly elevates acceptance intention.

Practically, the outcome implies that firms should integrate “technical function–emotional experience” synergies into design strategies. For instance, visually presenting precise delivery-time estimates and real-time demand-matching data within the robot’s mobile application can amplify farmers’ anticipatory pleasure from efficient service, thereby promoting adoption more effectively than merely displaying technical specifications.

5.2. Mechanisms of the Moderating Effect of Anthropomorphization Level

The results reveal a significant, negative moderating role of robot anthropomorphism in the TTF → expectancy-emotion paths (H6 and H7 supported). Specifically, as anthropomorphism increases, TTF’s facilitative impact on PEE is attenuated ($\beta = -0.053$), and its inhibitory impact on NEE is likewise weakened ($\beta = -0.027$). This finding challenges the prevailing assumption that anthropomorphic design unequivocally enhances user experience and highlights a trade-off between technical fit and emotional design in human–robot interaction.

Mechanistically, the negative moderation is driven by an attention-diversion effect. When robots exhibit highly anthropomorphic cues, farmers allocate additional cognitive resources to these features—e.g., perceived cuteness or conversational style—thereby reducing attention to task-relevant technical attributes such as delivery speed or reliability. Consequently, the cognitive link between TTF and expectancy emotions is diluted.

This finding aligns with emerging robotics research prioritizing functional reliability and task-specific performance. For example, recent studies on logistics omnidirectional robots employ neurodynamics-based visual servo predictive control to optimize smooth movement and satisfy physical and visual constraints, emphasizing precision and reliability over anthropomorphic design [39]. Similarly, research on wheeled mobile robots introduces a worm-inspired creeping gait strategy for variable wheelbase control, enabling adaptive movement on deformable terrain by minimizing internal forces and stabilizing posture—addressing mobility through mechanical and control innovation rather than human-like features [40]. Further, work on collaborative robots focuses on robust approximate constraint-following control to enhance trajectory tracking under uncertainties and disturbances, again prioritizing functional accuracy over emotional engagement [41].

These examples illustrate a broader emphasis in field robotics on core technical capabilities—such as motion control, adaptability, and accuracy—which are critical for task performance in unstructured environments. In such contexts, highly anthropomorphic features may not contribute to functional goals and can even detract from user focus on technological appropriateness. These insights complement the task-anthropomorphism matching theory advanced by Ma et al. [20]: while high anthropomorphism benefits highly interactive tasks, moderate anthropomorphism appears optimal for efficiency-oriented tasks such as delivery, better balancing emotional appeal with perceived technical value.

5.3. Theoretical Contributions and Practical Implications

5.3.1. Theoretical Contributions

First, by integrating TTF theory with the goal-directed behavior model, this study is the first to demonstrate—both systematically and empirically—that expectancy emotions mediate the influence of TTF on consumers’ acceptance intention toward transport robots. This finding remedies the long-standing neglect of affective mechanisms in traditional TTF research and offers a new lens for understanding the technical-to-affective transformation pathway in technology-adoption decisions [42].

Second, we uncover a negative moderating role of robot anthropomorphism in the TTF → expectancy-emotion linkage, refuting the prevailing view that anthropomorphic design invariably enhances user experience. By delineating the boundary conditions of anthropomorphism in efficiency-oriented service contexts, the study extends the literature on human–robot interaction for intelligent services [43].

Third, we disentangle the differential mediating effects of positive versus negative expectancy emotions, confirming that PEE constitute the primary affective conduit through which TTF shapes acceptance intention [44]. This distinction deepens scholarly insight into the affective dynamics of technology adoption and provides empirical grounding for future theory development.

5.3.2. Practical Implications

For robot manufacturers, maximizing TTF should take precedence over anthropomorphic embellishment. Key actions include calibrating navigation algorithms to local terrain and accelerating response times for urgent deliveries [45], rather than indiscriminately increasing anthropomorphic features. When anthropomorphism is warranted, moderate cues—such as concise voice prompts—are recommended to prevent cognitive distraction from task-relevant attributes.

For delivery platforms, cultivating PEE should be a strategic priority. Interfaces can prominently display historical on-time delivery rates, task–matching scores, and other performance metrics to reinforce farmers’ anticipatory pleasure from efficient service and thereby heighten adoption willingness.

5.4. Limitations and Future Research Directions

This study offers valuable insights but has limitations. Data were collected from Chinese agricultural stakeholders, potentially limiting generalizability due to region-specific factors. Future research should include diverse contexts to validate the model’s applicability. The reliance on survey data may introduce biases, such as social desirability and selection biases, which could overstate certain relationships. Future studies could benefit from mixed-method approaches or objective measures. The cross-sectional design restricts causality establishment and dynamic change capture. Longitudinal or experimental designs are needed to confirm causal order and understand adoption processes. Addressing these limitations will enhance the findings’ validity and applicability, contributing to a more comprehensive understanding of agricultural technology adoption and informing policy-making and interventions.

6. Conclusions

This study empirically demonstrates that task–technology fit (TTF) significantly influences agricultural stakeholders’ (e.g., farm managers and workers) expectancy emotions and, consequently, their intention to adopt on-farm transport robots for reducing post-harvest losses in perishable crops. The key findings can be summarized as follows:

• Expectancy emotions, particularly positive ones, serve as a critical mediating mechanism between TTF and adoption intention.
• Robot anthropomorphism exerts a significant negative moderating effect on the relationship between TTF and expectancy emotions. While a certain level of anthropomorphism may be engaging, our results indicate that higher levels diminish the positive impact of TTF, likely by triggering emotional overload (e.g., perceived complexity, distraction) or diverting attention from functional attributes.
• A context-aware design approach—which prioritizes technical functionality while strategically deploying anthropomorphism—is essential for optimizing user acceptance and operational efficiency.

These results highlight the importance of an “emotion–technology dual-core” strategy in developing and deploying robotic transport systems. Rather than focusing solely on technical performance, designers and manufacturers should implement adaptive anthropomorphism mechanisms informed by real-time user feedback during operation. For instance, moderate anthropomorphism (e.g., intuitive status indicators or calming auditory feedback) can enhance trust during delicate handling of high-perishability crops, whereas minimal anthropomorphism is preferable in high-throughput operations to maximize efficiency and reduce spoilage.

Beyond robot manufacturers, the implications of this study extend to policymakers and agricultural training institutions. Policymakers are encouraged to integrate anthropomorphism guidelines into technology certification or subsidy frameworks to promote adoption of well-balanced robotic solutions. Farmer training institutions should develop educational modules that help users understand and interact with both functional and emotional aspects of robots, thereby improving operational fluency and trust.

From a strategic perspective, anthropomorphism should be treated as a core variable within farm technology ecosystems. Open-innovation platforms, such as co-design workshops and emotion-focused feedback sessions with farmers, can help align robot design with practical and affective user needs. A continuous improvement cycle—based on field data and user experience—will support broader adoption and maximize the contribution of robots to reducing post-harvest losses.

At a macro level, this research advocates for an industry-wide shift from purely technical-driven approaches to service ecosystems that integrate emotional intelligence with functional performance. By placing end-users’ needs and contextual task demands at the center of innovation, agricultural robots can evolve from functional tools into trusted partners, promoting sustainability across ecological, economic, and operational dimensions.