Research Status and Trends in Universal Robotic Picking End-Effectors for Various Fruits

Abstract

The land used for fruit cultivation now exceeds 120 million hectares globally, with an annual yield of nearly 940 million tons. Fruit picking, the most labor-intensive task in agricultural production, is gradually shifting toward automation using intelligent robotic systems. As the component in direct contact with crops, specialized picking end-effectors perform well for certain fruits but lack adaptability to diverse fruit types and canopy structures. This limitation has constrained technological progress and slowed industrial deployment. The diversity of fruit shapes and the wide variation in damage thresholds—2–4 N for strawberries, 15–40 N for apples, and about 180 N for kiwifruit—further highlight the challenge of universal end-effector design. This review examines two major technical pathways: separation mechanisms and grasping strategies. Research has focused on how fruits are detached and how they can be securely held. Recent advances and limitations in both approaches are systematically analyzed. Most prototypes have achieved picking success rates exceeding 80%, with average cycle times reduced to 4–5 s per fruit. However, most designs remain at Technology Readiness Levels (TRLs) 3–5, with only a few reaching TRLs 6–7 in greenhouse trials. A dedicated section also discusses advanced technologies, including tactile sensing, smart materials, and artificial intelligence, which are driving the next generation of picking end-effectors. Finally, challenges and future trends for highly universal agricultural end-effectors are summarized. Humanoid picking hands represent an important direction for the development of universal picking end-effectors. The insights from this review are expected to accelerate the industrialization and large-scale adoption of robotic picking systems.

1. Introduction

1.1. Background

As global agriculture accelerates toward intelligent and unmanned production, fruit picking remains a critical field task that is still heavily dependent on manual labor [1,2,3]. This reliance causes picking costs to account for a substantial proportion of the total production cost [4,5]. According to the Food and Agriculture Organization (FAO), in 2023, the global fruit cultivation area exceeded 120 million hectares [6]. Worldwide production reached approximately 1.12 billion tons of vegetables and 940 million tons of fruits, and the majority of this yield still depends on manual picking [7]. Consequently, under the combined pressures of labor scarcity and rising manual picking costs, fruit picking is undergoing a transition from traditional manual methods to intelligent and automated robotic systems [8,9].

In recent years, fruit-picking robots have advanced in applications for specific crops and controlled environments. In most European and American countries, where large-scale and standardized fruit production dominates, development has focused on task-specific robots [10,11]. These robots typically employ end-effectors designed around the growth patterns, physical properties, and picking patterns of target fruits [12,13,14]. However, their highly customized structures restrict adaptability to multiple fruit varieties and complex canopy environments. This limitation is particularly evident in regions such as China, Japan, and the Netherlands, where crop diversity and cultivation practices are more variable [15]. In these contexts, specialized end-effectors have become a major bottleneck to wider adoption. As a result, increasing research attention is being directed toward universal picking end-effectors capable of handling a broader range of fruits.

At present, the design of universal picking end-effectors mainly follows two approaches: fruit detachment and fruit grasping. The first focuses on how fruits are separated from plants, while the second emphasizes how they are stably held. End-effectors based on specific picking patterns are typically designed around a single detachment mode and are therefore applicable only to fruits with similar growth characteristics and separation mechanisms [16]. Although soft picking end-effectors offer clear advantages in adapting to variations in fruit shape and size [17], most are based on underactuated designs [18], so the fingers lack independent adjustability and differentiated control. In addition, these end-effectors generally suffer from limited load capacity and slow deformation response [19,20]. A systematic review and analysis of universal picking end-effectors is therefore essential not only to clarify research progress and limitations but also to provide theoretical guidance and a technical reference for advancing the industrial deployment of fruit-picking robots.

1.2. Literature Search Strategy

A systematic literature search was performed using the following electronic databases: IEEE Xplore, ScienceDirect, SpringerLink, MDPI, ASABE Technical Library, and Google Scholar. The search timeframe spanned from December 2008 to June 2025, covering the rapid development period of agricultural robotics. The keywords combined terms related to picking end-effectors (e.g., “agricultural picking robot,” “agricultural picking end-effectors,” “robotic hand,” “picking patterns”) with those referring to technical design aspects (e.g., “soft picking end-effectors,” “tendon-driven,” “pneumatic,” “rigid-flexible coupling,” “fin ray effect”).

Inclusion criteria: (1) peer-reviewed journal or conference papers; (2) studies proposing new design concepts, prototypes, or evaluation of picking end-effectors; (3) research directly addressing design aspects such as transmission mechanisms, structural characteristics, or sensor integration. Exclusion criteria: (1) studies focused solely on computer vision, path planning, or mobile platforms without details on the end-effectors; (2) studies not directly related to the design of picking end-effectors; (3) non-English and non-Chinese publications. The screening process is summarized in a PRISMA-like flow diagram (Figure 1), including the number of records identified, screened, excluded, and retained [21].

1.3. Paper Organization

The structure of this paper is as follows. Section 2 examines the challenges faced by specialized picking end-effectors, emphasizing the high complexity of the picking environment and the broad diversity of picking targets, which underscore the urgent need for universal picking end-effectors. Section 3 reviews end-effector designs based on picking patterns. Section 4 summarizes recent advances in soft picking end-effectors. Section 5 presents the application of advanced technologies in robotic picking end-effectors. Section 6 discusses the challenges and future trends in universal picking end-effectors. The overall framework of this study is illustrated in Figure 2.

2. Challenges of Specialized Picking End-Effectors

Specialized picking end-effectors are typically optimized for specific fruit shapes, sizes, and growth conditions. They achieve high efficiency and stability when applied to their target crops. However, the diversity of fruit species, the wide variation in geometry [22], and the complexity of canopy environments—such as leaf occlusion, spatial constraints, and unstable fruit positions—create major challenges [23,24,25]. Under these conditions, such devices often exhibit reduced grasping accuracy, insufficient stability, and an increased risk of fruit damage. Consequently, their universality remains limited, making it difficult to meet the requirements of cross-species picking and restricting their deployment in diverse agricultural scenarios.

2.1. A Wide Variety of Fruits

Fruit species in modern agricultural production are highly diverse [26,27]. They include large fruits such as apples and citrus, which have regular shapes and firm skins [28], as well as small fruits such as strawberries, grapes, and tomatoes, which are soft and fragile [29]. Significant variation exists in their morphology, size, surface texture, and growth patterns [4,5,6,7,8,9].

Table 1. Key biological characteristic parameters of typical fruits.

Fruit Name	Number of Main Varieties	Shape	Weight/(g)	Mechanical Damage Force/(N)	Photographs	References
Grape	Over 10,000	Nearly spherical	1–10	2–7		[30,31,32]
Apple	Over 7500	Spherical, ellipsoidal, hyperbolic paraboloid	90–315	15–40		[33,34,35]
Kiwifruit	Approximately 76	Ellipsoidal	50–115	180		[36,37,38]
Strawberry	Over 460	Spherical, conical	15–30	2–4		[39,40,41]
Mango	Over 1000	Ellipsoidal, rectangular	121–720	20		[42,43,44]
Cucumber	Over 3342	Oval, cylindrical	109–616	199–375		[45,46,47]
Cowpea	Approximately 768	Cylindrical	1–204	35–100		[48,49,50]

highlights typical biological differences among major fruits, including grapes, apples, and kiwifruits, focusing on key parameters such as shape, mass, and mechanical damage force.

2.2. Complex and Diverse Unstructured Canopy Environments of Fruits

In addition to fruit diversity, the complexity of unstructured canopies is a key factor limiting the development of specialized picking end-effectors [51]. As shown in

Table 2. Unstructured canopy characteristics of multiple types of fruits.

Fruit Crop Types	Main Cultivation Methods	Typical Plant Shapes	Photographs	References
Grape	Pergola and vertical trellis cultivation	V-shaped, T-shaped, U-shaped		[64,65]
Apple	Dense planting cultivation	Open layered shape, high spindle shape, open-center shape		[66,67]
Kiwifruit	Pergola and vertical trellis cultivation	Pergola support structure, T-bar support structure		[36,68]
Peach	Dense planting cultivation	Open vase, Quad-V, Hex-V, SSA, Y-shaped		[69,70]
Tomato	Vertical vine training	Erect type, dwarf vine type		[71,72]
Strawberry	Elevated cultivation, ridge cultivation	Creeping shape, upright open shape, upright spherical shape		[73,74,75]
Citrus	Trellis-training method with multiple leaders	Round-headed shape, open-center shape, modified central trunk shape		[76,77]
Mango	High-density espalier	Natural round-headed shape, central trunk shape, natural open-center shape		[78,79]
Cucumber	Vertical vine training	Erect type		[80,81]

, under natural cultivation conditions, most fruits grow within three-dimensional canopy structures characterized by interlaced branches, severe fruit occlusion, and irregular spatial distribution [52,53,54]. In such environments, fruit posture, position, and accessibility are highly uncertain [55,56,57]. Specialized end-effectors are typically designed around ideal picking postures and trajectories. However, they lack the sensing and adaptive capabilities needed to cope with unstructured scenarios such as occlusion, variable hanging angles, dense clustering, and branch entanglement [58,59,60]. Consequently, complex canopy structures place higher demands on the structural flexibility of specialized end-effectors [61,62,63] and further expose their limited adaptability in practical applications.

In summary, specialized picking end-effectors face significant limitations when confronted with diverse fruit types and complex unstructured canopy environments. These challenges restrict their applicability across crops and scenarios, underscoring the necessity of advancing research on universalized picking end-effectors.

3. Design of End-Effectors Based on Picking Patterns

In recent years, growing attention has been directed toward the design of picking end-effectors based on picking patterns, as conventional task-specific devices show limited adaptability to different fruit types and canopy conditions. This approach moves beyond optimizing structures for a single fruit species. Instead, it emphasizes analyzing the separation mechanisms of fruits, identifying the corresponding picking actions, and then designing end-effector structures that follow these action logics. Through this strategy, end-effectors can adapt to fruits that share the same picking pattern.

3.1. Design of End-Effectors Based on Single-Action Picking Patterns

3.1.1. Design of End-Effectors Based on Grasp-And-Pull Picking Patterns

End-effectors based on grasp-and-pull picking patterns are typically simple in both structure and motion, and they have been widely applied in fruit picking. Their grasping modules are usually designed with symmetrically arranged multi-finger structures. To prevent fruit damage caused by excessive grasping force [82,83], the fingers can be designed as compliant beams with optimized geometries (Figure 3a). This design allows a constant grasping force to be delivered within a specified displacement range [84]. However, the force plateau depends on a preset displacement threshold. When the fruit position deviates, the grasping performance may be compromised. To address this limitation, pressure sensors can be integrated into the finger surfaces to provide real-time feedback on contact forces (Figure 3b,c). Compared with three-finger adaptive end-effectors, two-finger grippers show limited ability to accommodate variations in fruit shape and exhibit lower grasping stability. Combined with control algorithms, these sensors help maintain the grasping force within a safe range [85,86].

Grasping modules alone, however, are often insufficient for fruit picking in complex canopy environments [87]. To overcome this limitation, vacuum suction modules are frequently incorporated. These modules assist in extracting fruits from dense foliage (Figure 3d), separating clustered fruits, and reducing the risk of grasping multiple fruits simultaneously. Their integration not only minimizes fruit damage but also enhances grasp stability and overall picking success rates [88].

Figure 3. End-effectors based on grasp-and-pull picking patterns. (a) An apple-picking actuator based on a compliant constant-force mechanism (reprinted/adapted with permission from Ref. [84]. 2020, Miao, Y.); (b) a two-finger picking end-effector with integrated force sensing (reprinted/adapted with permission from Ref. [85]. 2017, Taqi, F.); (c) a three-finger picking end-effector with integrated force sensing (reprinted/adapted with permission from Ref. [86]. 2023, Wang, T.); (d) a four-finger gripper with a built-in vacuum suction nozzle (reprinted/adapted with permission from Ref. [88]. 2020, Vu, Q.). The upper section (shaded in light blue) illustrates grasp-and-pull end-effectors, while the lower section (also shaded in light blue) illustrates vacuum-suction-and-pull end-effectors. Together, these designs illustrate two approaches to implementing the same separation principle.

Figure 3. End-effectors based on grasp-and-pull picking patterns. (a) An apple-picking actuator based on a compliant constant-force mechanism (reprinted/adapted with permission from Ref. [84]. 2020, Miao, Y.); (b) a two-finger picking end-effector with integrated force sensing (reprinted/adapted with permission from Ref. [85]. 2017, Taqi, F.); (c) a three-finger picking end-effector with integrated force sensing (reprinted/adapted with permission from Ref. [86]. 2023, Wang, T.); (d) a four-finger gripper with a built-in vacuum suction nozzle (reprinted/adapted with permission from Ref. [88]. 2020, Vu, Q.). The upper section (shaded in light blue) illustrates grasp-and-pull end-effectors, while the lower section (also shaded in light blue) illustrates vacuum-suction-and-pull end-effectors. Together, these designs illustrate two approaches to implementing the same separation principle.

3.1.2. Design of End-Effectors Based on Grasp-And-Twist Picking Patterns

Because grasp-and-pull methods apply large forces to fruits, they often cause damage and induce plant shaking [89,90]. As a result, research has gradually shifted toward grasp-and-twist picking end-effectors, where fruit detachment is achieved through rotation. These devices typically combine a flexible finger grasping module with a rotational module (Figure 4a,b) [91,92]. It has been paired with six-degree-of-freedom (6-DOF) and 3-DOF robotic arms to perform picking tasks. In some designs, springs are added at the finger joints to regulate the grasping force (Figure 4c) [93], thereby reducing compression damage to fruits. For delicate fruits such as cherry tomatoes, finger structures have been designed to mimic human fingers (Figure 4d). Their tapered tips can be inserted into narrow gaps between fruits, minimizing interference with surrounding targets [94]. However, these designs rely on mechanical stoppers to control the grasping range, making it difficult to adapt in real time to variations in fruit size. To address this limitation, pressure sensors can be integrated to provide real-time grasping force feedback and adjust stopper positions (Figure 4e), thereby reducing both slippage and damage caused by excessive force [95,96,97].

For highly fragile fruits such as oyster mushrooms, the grasping module can be replaced with a vacuum suction unit. Flexible suction cups with double-layer corrugated structures, supported by springs on both sides of a slider, allow adaptive deformation according to the height and shape of the target (Figure 4f). This configuration prevents rigid contact and effectively reduces fruit damage [98].

3.1.3. Design of End-Effectors Based on Grasp-And-Bend Picking Patterns

The grasp-and-twist picking pattern is inefficient when dealing with fruits that have thick and rigid stems, as the twisting step often requires more than 10 s, reducing operational efficiency. The grasp-and-bend picking pattern offers a faster alternative. For fruits with flexible stems that break along predictable directions, such as kiwifruit, a two-finger gripper combined with a reverse cam-slot mechanism (Figure 5a) has been developed. This design integrates grasping, detachment, and unloading into a continuous motion, reducing the average picking time to 4–5 s per fruit [99]. In contrast, oyster mushrooms lack a distinct fruit stem and have soft, fragile surfaces. They are more suited to an adsorption-and-bend structure. In this approach, bellows suction cups regulate the contact force, while bending motion is driven by an arc-shaped guide rail (Figure 5b). The system mimics the force application of human fingers and achieves a picking success rate of up to 90% [100].

3.1.4. Design of End-Effectors Based on Grasp-And-Cut Picking Patterns

For fruits with tough stems and uncertain breaking directions, grasp-and-cut end-effectors integrate cutting with grasping or suction modules to enhance fruit-picking efficiency and maintain fruit integrity [101]. For fruits with thick stems that are clearly separated from the canopy [102], a gripper combined with scissors is used to cut the stem (Figure 6a) [103,104]. For fruits with irregular orientations, guiding structures direct the stem into the cutting region (Figure 6b) [105]. When stems are difficult to identify or the fruits are heavy, direct grasping is employed to improve efficiency and stability. For small spherical fruits, hemispherical cup-shaped structures with integrated blades allow grasping and cutting to be performed simultaneously (Figure 6c) [106]. For larger and heavier fruits, enveloping grippers combined with upper and lower cutting mechanisms are used to complete the picking process (Figure 6d–f) [107,108,109].

For smooth, fragile, or soft fruits, suction-and-cut end-effectors are widely applied. They typically use vacuum suction cups to hold the fruit, while oscillating blades cut the stem (Figure 6g,h). Pressure and laser sensors are integrated to monitor the suction state in real time, ensuring reliable operation [110,111]. The sweet pepper end-effector incorporates a blade-based passive decoupling mechanism, which allows independent selection of grasping and cutting positions. In contrast, the tomato end-effector requires the fruit to be drawn into the cutting section before detachment can be performed. For unstable or bent steams, active traction-cutting units have gradually replaced conventional cutters (Figure 6i) [112]. Most end-effectors approach the fruit from a fixed direction, which limits adaptability to fruits growing in varied orientations [113]. To address this, guiding inlets and automatic posture-adjustment mechanisms have been designed (Figure 6j). These systems enable the picking of fruits tilted up to ±90°, achieving a success rate of 80.6%, significantly higher than the average reported in comparable studies (79.0%) [114]. To handle fruits with large surface curvature, such as cucumbers, the suction module has been designed with a paper-cut-inspired hyperbolic paraboloid geometry (Figure 6k). This design improves sealing performance during suction [115].

Figure 6. Grasp-and-cut picking end-effectors. (a) A grape-picking end-effector (reprinted/adapted with permission from Ref. [103]. 2021, Vrochidou, E.); (b) a sweet-pepper-picking end-effector (reprinted/adapted with permission from Ref. [105]. 2019, Lee, B.); (c) a cherry-tomato-picking end-effector (reprinted/adapted with permission from Ref. [106]. 2022, Yeshmukhametov, A.); (d) a specific end-effector for pineapple harvesting (reprinted/adapted with permission from Ref. [107]. 2020, Anh, N.P.T.); (e) an automatic pineapple-picking end-effector (reprinted/adapted with permission from Ref. [108]. 2021, Guo, A.F.); (f) a pumpkin-picking end-effector (reprinted/adapted with permission from Ref. [109]. 2020, Roshanianfard, A.); (g) a sweet-pepper-picking end-effector (reprinted/adapted with permission from Ref. [110]. 2017, Lehnert, C.); (h) a tomato fruit suction cutting device (reprinted/adapted with permission from Ref. [111]. 2021, Fujinaga, T.); (i) the tractional cutting unit for scissors (reprinted/adapted with permission from Ref. [112]. 2021, Jun, J.); (j) the end-effector structure integrating the cutting, suction, and transporting modules (reprinted/adapted with permission from Ref. [114]. 2023, Park, Y.); (k) a suction-cup-based robotic gripper for cucumber picking (reprinted/adapted with permission from Ref. [115]. 2024, Jo, Y.). The upper section (light blue) illustrates grasp-and-cut end-effectors. Within this category, two distinct approaches are used depending on fruit morphology and growth pattern: some designs grasp the stem for precise cutting, while others directly grasp the fruit body for stabilization, shown in purple. The lower section (light blue) presents vacuum-suction-and-cut designs.

Figure 6. Grasp-and-cut picking end-effectors. (a) A grape-picking end-effector (reprinted/adapted with permission from Ref. [103]. 2021, Vrochidou, E.); (b) a sweet-pepper-picking end-effector (reprinted/adapted with permission from Ref. [105]. 2019, Lee, B.); (c) a cherry-tomato-picking end-effector (reprinted/adapted with permission from Ref. [106]. 2022, Yeshmukhametov, A.); (d) a specific end-effector for pineapple harvesting (reprinted/adapted with permission from Ref. [107]. 2020, Anh, N.P.T.); (e) an automatic pineapple-picking end-effector (reprinted/adapted with permission from Ref. [108]. 2021, Guo, A.F.); (f) a pumpkin-picking end-effector (reprinted/adapted with permission from Ref. [109]. 2020, Roshanianfard, A.); (g) a sweet-pepper-picking end-effector (reprinted/adapted with permission from Ref. [110]. 2017, Lehnert, C.); (h) a tomato fruit suction cutting device (reprinted/adapted with permission from Ref. [111]. 2021, Fujinaga, T.); (i) the tractional cutting unit for scissors (reprinted/adapted with permission from Ref. [112]. 2021, Jun, J.); (j) the end-effector structure integrating the cutting, suction, and transporting modules (reprinted/adapted with permission from Ref. [114]. 2023, Park, Y.); (k) a suction-cup-based robotic gripper for cucumber picking (reprinted/adapted with permission from Ref. [115]. 2024, Jo, Y.). The upper section (light blue) illustrates grasp-and-cut end-effectors. Within this category, two distinct approaches are used depending on fruit morphology and growth pattern: some designs grasp the stem for precise cutting, while others directly grasp the fruit body for stabilization, shown in purple. The lower section (light blue) presents vacuum-suction-and-cut designs.

3.2. Design of End-Effectors Based on Combined-Action Picking Patterns

Compared with end-effectors based on single-action picking patterns, those based on combined-action strategies align more closely with the cooperative movements of the human hand and the mechanical characteristics of fruits.

The grasp–bend–pull end-effector has been applied to fruits such as kiwifruit and strawberries, which possess natural abscission layers. An asymmetric four-bar linkage enables fruit rotation (Figure 7a), thereby reducing both fruit loading and canopy oscillation [116]. For strawberries, whose delicate skin is highly susceptible to damage, distributed pressure sensors can be mounted on the surfaces of flexible fingers (Figure 7b) to monitor contact-force distribution and detect grasp deviations [117]. A V-shaped finger driven by linkages has been proposed for fruits with tougher stems (Figure 7c). This design combines the benefits of rigid separation (cutting) and flexible separation (bending) [118]. In cases where fruit posture increases the risk of picking failure, grasp–twist–pull end-effectors have been developed (Figure 7d). These systems use suction to achieve compliant alignment, improving adaptability to diverse fruit geometries [119]. Furthermore, integrated systems combining grasping, bending, twisting, and pulling have been introduced (Figure 7e). Such designs employ 3D-printed grippers, linear motion modules, and eccentric mechanisms to achieve coordinated multi-action picking [120].

Energy supply is a critical yet often overlooked factor in the design of robotic picking end-effectors [121]. Current solutions rely primarily on electric, pneumatic, and hydraulic actuation, each with distinct advantages and limitations. Electric drives provide precise control and fast response, but they consume considerable energy and are constrained by limited battery life, which reduces field portability. Pneumatic systems are lightweight and compliant, helping to minimize fruit damage. However, they depend on external compressors, resulting in low efficiency and poor convenience in orchard environments [122]. Hydraulic systems offer high power density and strong gripping capability, but their structural complexity and high maintenance costs hinder agricultural adoption [123]. The development of efficient energy management strategies and low-power actuation designs will be essential for transitioning from laboratory prototypes to commercially viable picking robots that are portable, durable, and capable of long-duration operation.

In conclusion, end-effectors based on picking patterns are typically designed from the perspective of fruit detachment, with structures tailored to a specific picking pattern, and thus are mainly limited to fruits with similar growth traits and detachment mechanisms. These designs often rely on accurate fruit pose recognition and precise end-effector positioning to ensure effective actions. However, in dense canopies with variable lighting and severe occlusion, such requirements are difficult to achieve, highlighting the constraints of these approaches in practical applications.

4. Soft Picking End-Effectors

End-effectors designed around picking patterns focus on fruit detachment, using specific motion strategies to adapt to different fruit types. In contrast, soft end-effectors emphasize the grasping process itself. By leveraging soft materials, compliant structures, and deformable actuation, these designs significantly improve adaptability and stability when handling fruits of varying shapes and sizes [124]. Currently, two main approaches predominate in the design of soft end-effectors for universal fruit picking. The first relies on pneumatic actuation, where chamber deformation of soft materials enables adaptive grasping [125]. The second exploits passive deformation through the fin ray effect (FRE), allowing natural shape adaptation [126]. The following sections review these two representative approaches, highlighting recent advances and key technical features.

4.1. Picking End-Effectors Based on Pneumatic Driving Mechanisms

Pneumatic soft end-effectors achieve flexible deformation by regulating the internal chamber pressure [127]. This approach offers unique advantages in lightweight design, fruit damage reduction, and adaptability to complex picking environments [128]. To enable soft contact and stable grasping of fruits, researchers have developed end-effectors based on inflatable chambers, as well as soft pneumatic actuators (SPAs).

4.1.1. Picking End-Effectors Based on Soft Air Chambers

In contrast to traditional rigid gripping structures, soft air chambers are constructed from elastic materials that adaptively conform to fruit surfaces during contact. This adaptation allows the contact area and grasping force to be adjusted, reducing the risk of bruising or compression damage [129]. Researchers designed a fruit-picking end-effector with a rigid circular shell lined with flexible air cushions (Figure 8a,b). Through rigid–flexible coupled interaction, this design enabled stable grasping, with grasping forces reaching 50 N [130,131]. In contrast to tomato-picking end-effectors, the soft circular-shell gripper can handle both spherical and irregularly shaped objects. It is suitable for diameters ranging from 42.5 to 76 mm. To further enhance actuation efficiency and structural flexibility, Wang et al. proposed an end-effector based on corrugated air chambers. The driving energy was concentrated by geometric structuring, while 3D-printed rigid components were integrated to form a rigid–flexible coupled module. This configuration allowed flexible adaptation to fruits of varying diameters (Figure 8c) [132].

4.1.2. Picking End-Effectors Based on SPAs

SPAs are generally classified into pneumatic network actuators and fiber-reinforced actuators. Pneumatic network actuators employ asymmetric multi-chamber structures [133]. When inflated, they generate continuous nonlinear deformations that conform naturally to fruit surfaces, making them widely used in fruit picking.

Current research is focused on arranging soft fingers symmetrically to form multi-finger end-effectors. These flexible fingers deform with fruit contours, achieving uniform wrapping and reducing pressure concentration (Figure 9a) [134,135,136]. To prevent slippage or damage, precise control of grasping force is required. Constitutive models defined by strain energy density functions, such as the Yeoh or Ogden model, are commonly applied. These models are used to construct energy-minimization-based bending models of pneumatic network actuators, enabling quantitative characterization of the nonlinear coupling among input pressure, bending angle, and output force. Such models provide theoretical support for performance analysis and control optimization [137]. However, they are highly sensitive to structural parameters of the material and lack real-time feedback adjustment. To address this limitation, thin-film pressure sensors are often integrated at the fingertips (Figure 9b), providing force feedback for accurate grasp control [138,139]. The flexible spherical-fruit end-effector is additionally equipped with pressure and torque sensors. Through multisensor fusion, adaptive force control is achieved.

Although pressure regulation and sensor feedback enable partial control of grasping force, the response speed, output range, and force direction of pneumatic soft end-effectors remain constrained by structural properties. As a result, research has shifted toward structural parameter optimization. Finite element analysis (FEA) and experiments have been employed to determine optimal values for chamber number, wall thickness, spacing, inclination angle, and constraint layer thickness. The appropriate number of chambers ensures uniform bending [140]. Wall thickness of approximately 2.5 mm or 3 mm prevents insufficient bending and leads to rupture. A spacing of 3 mm has been shown to yield the best performance. Larger inclination angles expand the workspace but reduce output force [141], and a constraint layer thickness of 2.5 mm or 3.5 mm provides a balance between stiffness and deformation [137,142]. Among these parameters, the constraint layer and chamber wall thickness have been identified as dominant factors. Together, they largely determine the bending angle and force output characteristics of soft fingers [142].

To address the low stiffness and weak grasping force of pneumatic soft end-effectors, several design strategies have been proposed. Tian et al. introduced constraint materials into a multi-chamber soft structure (Figure 9f), increasing the maximum grasped weight from 108 g to 414 g [143]. To reduce canopy interference caused by rigid components, a conical pneumatic finger with a gradually varying cross-section was designed, enabling graded bending curvature (Figure 9g) [144]. For slender and unstable cylindrical fruits, the chamber orientation was adjusted to generate a helical motion, allowing the finger to wrap around the fruit and achieve stable grasping (Figure 9h) [145].

Fiber-reinforced pneumatic actuators employ external winding materials to guide deformation paths and enhance mechanical performance, resulting in higher load capacity [146,147]. Liu designed a three-finger soft hand (Figure 9i), where dense winding allowed each finger to grasp objects weighing up to 568 g under 130 kPa [148]. To broaden the range of applicable fruits, Zhao et al. developed a five-finger anthropomorphic soft hand (Figure 9j). It was capable of three grasping modes and could stably grasp cylindrical or spherical fruits weighing up to 1 kg [149].

Figure 9. Picking end-effectors based on SPAs. (a) A flexible three-fingered end-effector (reprinted/adapted with permission from Ref. [134]. 2024, Ji, W.); (b) a force feedback soft gripper for tomato picking (reprinted/adapted with permission from Ref. [138]. 2021, Kultongkham, A.); (c–e) pneumatic soft end-effectors with optimized structural parameters (reprinted/adapted with permission from Ref. [140]. 2022, Zhou, K.; Ref. [142]. 2024, Li, M.; Ref. [137]. 2022, Zhu, Y.); (f) a soft picking end-effector with enhanced stiffness, 1. stiffness-enhancing structure, 2. strain-generating layer, 3. strain-limiting layer (reprinted/adapted with permission from Ref. [143]. 2021, Tian, H.); (g) a robust soft robotic gripper for apple picking (reprinted/adapted with permission from Ref. [144]. 2023, Wang, X.); (h) a soft pneumatic gripper for slender-fruit picking (reprinted/adapted with permission from Ref. [145]. 2021, Jia, J.); (i) a soft picking end-effector in a greenhouse (reprinted/adapted with permission from Ref. [148]. 2019, Liu, F.); (j) a pneumatic flexible fruit-picking end-effector (reprinted/adapted with permission from Ref. [149]. 2019, Zhao, Y.). According to different actuation principles, SPAs can be classified into pneumatic network actuators (shown in light blue in the upper section) and fiber-reinforced pneumatic actuators (shown in light blue in the lower section).

Figure 9. Picking end-effectors based on SPAs. (a) A flexible three-fingered end-effector (reprinted/adapted with permission from Ref. [134]. 2024, Ji, W.); (b) a force feedback soft gripper for tomato picking (reprinted/adapted with permission from Ref. [138]. 2021, Kultongkham, A.); (c–e) pneumatic soft end-effectors with optimized structural parameters (reprinted/adapted with permission from Ref. [140]. 2022, Zhou, K.; Ref. [142]. 2024, Li, M.; Ref. [137]. 2022, Zhu, Y.); (f) a soft picking end-effector with enhanced stiffness, 1. stiffness-enhancing structure, 2. strain-generating layer, 3. strain-limiting layer (reprinted/adapted with permission from Ref. [143]. 2021, Tian, H.); (g) a robust soft robotic gripper for apple picking (reprinted/adapted with permission from Ref. [144]. 2023, Wang, X.); (h) a soft pneumatic gripper for slender-fruit picking (reprinted/adapted with permission from Ref. [145]. 2021, Jia, J.); (i) a soft picking end-effector in a greenhouse (reprinted/adapted with permission from Ref. [148]. 2019, Liu, F.); (j) a pneumatic flexible fruit-picking end-effector (reprinted/adapted with permission from Ref. [149]. 2019, Zhao, Y.). According to different actuation principles, SPAs can be classified into pneumatic network actuators (shown in light blue in the upper section) and fiber-reinforced pneumatic actuators (shown in light blue in the lower section).

4.1.3. SPAs in Non-Agricultural Fields

Although soft pneumatic end-effectors have begun to demonstrate advantages in compliant grasping for agricultural picking, their development remains at an early exploratory stage. In contrast, significant progress has been achieved in non-agricultural fields, particularly in structural innovation, stiffness regulation, and multimodal actuation and control. These advances provide valuable insights for the design of universal end-effectors in agricultural applications.

Research on SPAs in non-agricultural fields has entered an advanced stage, characterized by structural innovation and the development of multimodal motion. Current designs focus on improving bending performance and output force. Yu et al. proposed a five-degree chamber structure (Figure 10a), which increased the actuator’s bending angle by 193.31% [150]. Keong et al. developed a multi-chamber corrugated structure that achieved a high output force of 25.21 N (Figure 10b) [151]. The inherent flexibility of SPAs has also enabled the design of integrated biomimetic double-joint actuators inspired by human fingers (Figure 10c). This design achieved a bending angle of 84.5° at 100 kPa [152]. However, its high activation pressure and limited rotation angle restricted the motion range and bending speed. To address these limitations, Liu et al. drew inspiration from spider legs and developed a folded membrane soft joint actuator capable of rotating 80° at only 5.25 kPa (Figure 10d) [153].

SPAs are commonly used for directional bending, but their structural flexibility makes them prone to twisting or lateral deformation under asymmetric loads. To address this issue, researchers have explored different structural modifications. As shown in Figure 10e, flexible connections between chambers were introduced to improve slip resistance [154]. Alternatively, herringbone-shaped diamond chambers were designed to enable both longitudinal and transverse conformal deformation, thereby enhancing grasping strength (Figure 10f) [155]. However, a single SPA can only generate a fixed bending curvature, limiting its ability to conform to diverse object profiles. To overcome this constraint, Tawk et al. combined an SPA with an FRE structure to achieve conformal grasping (Figure 10g), which improved adaptability to objects of varying shapes and stiffness [156].

SPAs are typically designed for single-mode motion, which limits their ability to meet the combined demands of multi-DOF output and environmental adaptability required in complex tasks. To address this limitation, researchers have begun exploring multimodal deformation mechanisms.

One strategy focuses on the regulation of strain-limiting layers. By altering their structure and distribution, bending, twisting, and elongation motions can be achieved [151]. DOFs can also be expanded by combining multiple actuator units (Figure 11a,b) [157,158]. Bidirectional bending is enabled by two actuation units, whereas three actuation units allow multidirectional bending, although this approach increases overall volume and restricts bending directions. To enable omnidirectional bending and extension, tendons have been arranged around square chambers and integrated with adjustable locking systems (Figure 11c) [159]. Another approach employs parameterized bending and twisting modules assembled in sequence and independently pressurized, enabling SPAs to perform a wide range of complex motions (Figure 11d) [160]. Advances in multi-material embedded 3D printing (ME3P) have further allowed the fabrication of intricate fiber-reinforced patterns that precisely control bending direction and curvature (Figure 11e) [161]. To improve load capacity and structural stability during diverse motion forms, bistable and multistable structures have been integrated with SPAs (Figure 11f,g). These designs allow extension and bending states to be locked without continuous pressure supply, while the parametric design enables a variety of motion behaviors [162,163].

In summary, pneumatic-driven picking end-effectors have demonstrated good compliance and effective fruit protection in agricultural applications. However, their development is still constrained by simple structural forms, limited load capacity, and low grasping stability. By contrast, non-agricultural fields have achieved substantial advances in multi-chamber structural design, functional coupling, and rigid–flexible topology. These achievements provide both theoretical support and transferable strategies for the design of universal picking end-effectors.

4.2. Picking End-Effectors Based on FRE Structures

4.2.1. Design of Picking End-Effectors Based on FRE Structures

The bioinspired FRE structure exhibits high compliance and shape adaptability and has been widely applied in soft picking end-effectors. Current designs often employ symmetrically arranged FRE fingers to form soft grippers (Figure 12a–d). These fingers achieve stable grasps of irregular fruits through passive geometric deformation [164,165,166,167]. The difference lies in the number of fingers and their relative positions. In addition, FRE-based flexible supports provide strong adaptability to heavy fruits such as bananas (Figure 12e) [168]. With advances in sensing technology, thin-film pressure sensors and ultrasonic sensors have been integrated into FRE fingers (Figure 12f). These additions enable force feedback and distance measurement, significantly improving grasp safety and success rates [169]. Furthermore, as shown in Figure 12g,h, FRE structures can be combined with shearing mechanisms or swallowing-type devices. This integration creates multifunctional picking end-effectors capable of grasping, cutting, or transporting in a single unit [170,171,172].

Researchers have continued to optimize the FRE structure to improve shape adaptability and achieve more uniform force distribution. In one design, compliant hinges in the FRE structure were replaced with pin joints and combined with a linkage mechanism (Figure 12i). This configuration enabled deformation under low force, allowing fruits to be uniformly wrapped and stably grasped [173]. Building on passive deformation of the FRE structure, Liu et al. introduced active actuation (Figure 12j), achieving a maximum payload of about 1.4 kg [174].

4.2.2. Optimization of Geometric Parameters in FRE Structures

Although end-effectors based on the FRE have shown notable compliance and adaptability in agricultural picking, their development has relied mainly on empirical design and simulation. Mechanisms for performance regulation remain underdeveloped. By contrast, research in non-agricultural fields has advanced further in structural optimization.

The geometric parameters of FRE fingers strongly influence their compliance and mechanical response, with interdependent relationships among different parameters. Figure 13 illustrates FRE finger structures with varying geometries. Studies have shown that the number of crossbeams and their filling density determine bending continuity and stiffness. Increasing these values enhances contact force, but excessive or insufficient numbers reduce adaptability [175]. Crossbeam thickness and inclination further regulate deformation trends and stiffness distribution. Thinner crossbeams provide greater softness and reduce local stress [176]. Inclined crossbeams generate larger deflections and improved conformity compared to parallel ones [177]. A gradient inclination can optimize fingertip displacement and contact stress, and under large deformations, it can trigger a layered jamming effect that increases stiffness and contact force [178,179]. In addition, when the angle between the front and rear beams and the base is set to 86°, maximum contact stress is achieved without compromising fingertip flexibility, balancing conformity with structural support [179].

Crossbeams are critical structural elements that determine the performance of FRE fingers. Compared with beamless designs, crossbeam-based structures provide larger contact areas and improved adaptability, but they may also be prone to deformation failure. A configuration in which the rear crossbeams are thicker than the front ones preserves contact area while maintaining structural integrity (Figure 14a) [180]. The geometry of the crossbeams also strongly influences wrapping and force transmission (Figure 14b). Among different shapes—rectangular, trapezoidal, straight-slot, and ±45°-inclined—the 2 mm thick straight-slot crossbeam achieved the best wrapping performance, producing a displacement of 31.97 mm under a 30 N external load [181]. Furthermore, branched and crossed crossbeam structures enhance adaptability and contact force, respectively (Figure 14c). The branched design provides a balance between the two, whereas the crossed design is more suitable for applications requiring higher grasping forces [182].

Picking end-effectors based on the FRE have shown strong adaptability in agricultural picking. However, their designs remain largely empirical and lack systematic performance control. In contrast, research in non-agricultural fields has advanced further in structural optimization. These efforts have established mature theoretical and experimental frameworks, which provide valuable guidance for the design of agricultural picking end-effectors.

Overall, soft picking end-effectors, driven by soft materials and compliant structures, provide notable advantages in adapting to fruits of varying shapes and sizes while ensuring grasping stability. However, their force application and magnitude are limited, and the slow deformation response reduces both success rate and picking efficiency.

In conclusion, we discuss and compare research on and technology for picking end-effectors for different fruits and present critical reflections and recommendations. More detailed information about references on fruit-picking end-effectors is shown in

Table 3. Comparison of picking end-effectors for different fruits.

End-Effector type	Picking Success Rate/(%)	Fruit Damage Rate/(%)	Average Time per Pick/(Seconds/Fruit)	Maximum Payload/(g)	Application Object	Year	References
Grasp-and-pull	100	/	30	/	Cherry tomato	2017	[85]
	95.3	0	/	400	Apple	2020	[84]
	100	0	/	230	Tomato	2023	[86]
Grasp-and-twist	60	0	23	/	Tomato	2016	[93]
	92.31	/	16	300	Apple	2019	[91]
	95	/	/	250	Apple	2022	[92]
	79.7	1.9	6.4	15.7	Cherry tomato	2022	[183]
	95.82	2.9	4.86	/	Cherry tomato	2022	[95]
	88.2	2.9	3.5	/	Button mushroom	2022	[98]
Grasp-and-bend	94.2	4.9	4–5	128.7	Kiwifruit	2019	[99]
	94.2	/	/	42.1	Button mushroom	2020	[100]
Grasp-and-cut	46	11.5–25	35–40	/	Sweet pepper	2016	[110]
	53.3	/	51.1	/	Sweet pepper	2019	[105]
	85	15.4	23	120	Tomato	2020	[111]
	41.67–100	0–58.33	5.874	/	Tomato	2020	[112]
	95.56	/	12	1500	Pineapple	2020	[107]
	/	/	2.2	1520	Pineapple	2021	[108]
	/	/	56	/	Cherry tomato	2022	[106]
	80.6	8	15.5	289.1	Tomato	2022	[114]
End-effectors based on combined-action picking patterns	65–100	/	/	/	Strawberry	2014	[117]
	51	24.6	5.5	/	Kiwifruit	2018	[116]
	100	/	/	/	Apple	2019	[120]
	64.06–82.47	0	8.8	/	Apple	2021	[119]
	80–100	0	9.6–12.5	/	Spherical fruits	2023	[118]
End-effectors based on pneumatic driving mechanisms	79.4–83.9	/	24	/	Tomato	2014	[130]
	/	0	/	1000	Spherical and cylindrical fruits	2019	[149]
	100	/	2.55–5.15	549	Food products	2020	[131]
	/	0	/	266	Tomato	2020	[138]
	/	0	3.6	340	Slender fruits	2021	[145]
	/	0	/	583	Spherical fruits	2022	[137]
	70.77	4.55	14.69	6710	Apple	2022	[144]
	/	3–11	1.9–3.9	/	Apple	2023	[134]
End-effectors based on FRE structures	70–85	0	/	/	Spherical fruits	2021	[165]
	80.17–82.93	0	12.53–17.17	/	Apple	2022	[164]
	80	0	/	309.8	Spherical fruits	2022	[169]
	/	0	/	/	Spherical fruits	2022	[172]
	67–84	0	6	840	Mango	2023	[166]
	/	0	/	/	Tomato	2024	[173]

.

5. Advanced Technology for Picking End-Effectors

5.1. Advanced Tactile Sensing Integration for Enhanced Perception and Feedback

Significant advances in tactile sensing have recently been achieved, providing picking end-effectors with critical force feedback and surface property recognition capabilities. To enable low-damage picking, end-effectors have incorporated tactile arrays—piezoresistive, capacitive, and fiber-optic—together with slip detection and opto-tactile techniques. These technologies allow grasping force monitoring and regulation, slip prediction, and hardness estimation even in dense canopies and clustered fruit environments [184,185]. For fragile fruits such as strawberries and tomatoes, closed-loop tactile feedback markedly reduces crushing and fruit drop compared with vision-based control alone. It also enhances the robustness and consistency of the approach, contact, and detachment phases [1,2,3,4,5,6]. For example, fingertip tactile sensing combined with time–frequency slip detection enables stable grasping under very small normal forces [184]. Similarly, opto-tactile systems can quantify surface deformation and hardness under occlusion, supporting real-time adjustment of grasp trajectories and force thresholds [186,187].

Two main integration strategies are currently employed. The first embeds sensors directly into the fingertip. Strain, pressure, fiber-optic, or gel-based vision-tactile devices are incorporated into the compliant finger pad, enabling the direct output of strain field, contact force, and slip features. This approach supports stable interaction under light contact and compliant adaptation. The second strategy applies electronic skin (e-skin) to the outer surface of soft grippers. Multimodal signals, including pressure, hardness, and temperature, are acquired from the external layer and fused with visual information. This combination enables real-time correction of picking posture and provides valuable cues for fruit quality assessment [188,189].

5.2. Advanced Material Technology for Field-Ready End-Effectors

In agricultural environments, ultraviolet radiation, humidity, and temperature cycling can cause mechanical drift and surface aging in elastomers such as silicone rubber. These effects reduce the repeatability of gripping actions and shorten the service life of end-effectors. Recent materials research has focused on the coupled mechanisms of UV–thermal–moisture aging, lifetime assessment, and interfacial failure. Systematic models have been proposed and validated through accelerated testing, providing critical data to guide material selection and protective design for picking end-effectors [190,191]. Postharvest handling and sorting impose strict requirements on food-contact safety [192,193]. To meet these standards, the industry commonly employs food-grade silicone and stainless-steel components.

To balance compliant contact with stable release, variable-stiffness and controllable-adhesion materials have been widely applied in the design of picking end-effectors. Typical strategies include layer- or particle-based jamming and magnetorheological elastomer (MRE) clamping, which enable rapid stiffening after contact to resist disturbances during peduncle detachment [194]. Gecko-inspired adhesion and electroadhesion techniques provide controllable attachment forces under very low normal loads, facilitating posture transitions and gentle handling [195]. Meanwhile, liquid crystal elastomers (LCEs), dielectric elastomers, and shape-memory alloys demonstrate clear advantages in energy density, compactness, and remote or low-voltage actuation [196]. Recent studies have further reported radio-frequency-selective actuation of LCEs and electronic skins with self-monitoring and self-healing capabilities, offering new pathways for integrating materials, sensing, and actuation [197,198,199]. Quantitative improvements in load-to-mass ratio, stability, and maintainability achieved by these advanced materials have enhanced the gripping capacity and damage control of picking end-effectors.

5.3. Learning-Enabled Control of Picking End-Effectors

Deep learning has become the dominant approach for fruit detection, segmentation, peduncle and fruit pose estimation, and obstacle avoidance planning [200,201]. Its application has recently extended to the end-effector level, enabling joint decision-making for approach posture and force-control parameters. In clustered fruits such as tomatoes, deep reinforcement learning (DRL) can autonomously learn collision-free “approach–grasp” strategies, which greatly reduce peduncle impacts and shorten picking cycles. With domain randomization and simulation-to-reality transfer, DRL further enables cross-domain adaptation with few or even zero real-world samples [202,203]. Systematic studies and recent reviews have also shown that co-modeling end-effector force control with visual or point-cloud estimation improves robustness in unstructured canopies. This synergy facilitates stable grasping and reliable fruit separation under complex agricultural conditions [204].

Incorporating slip, hardness, and contact-force features from tactile sensing or electronic skin into the state space or reward function can substantially accelerate policy convergence. This integration improves success rates while reducing the risk of fruit damage. In occluded scenes, a DRL planner designed for visibility maximization has already achieved zero-shot deployment on real plants [205]. End-to-end evaluations in orchard environments further show that integrating real-time perception, path planning, and end-effector control into a modular ROS2 framework enables reliable execution of the full picking chain—including detection, approach, detachment, and placement—under dwarfing cultivation systems [206]. Building on recent vision surveys, this framework also offers transferable perception–control configurations that can be adapted to multiple crops and diverse orchard conditions [207].

In summary, tactile sensing technologies provide real-time perception, advanced materials enhance adaptability and durability, and learning-based control methods improve intelligent decision-making. The integration of these approaches offers crucial support for performance breakthroughs and the universal development of picking end-effectors.

6. Challenges and Future Trends

6.1. Challenges

Current strategies for universal picking end-effectors are mainly guided by two design principles: the method of fruit detachment and the mode of grasp stabilization. These approaches focus on how the fruit is separated from the plant and how it is securely held. While such strategies have improved operational capacity across different fruit types, their effectiveness remains constrained by notable limitations.

End-effectors based on specific picking patterns can achieve a degree of universal applicability for fruits with similar detachment mechanisms, but their adaptability remains limited. These devices are typically designed around a single picking strategy—such as pulling, twisting, or bending—where combined loads are applied to the abscission layer through friction between the fingers and the fruit surface. This process often disturbs nearby fruits and branches and is restricted to fruits with comparable growth traits and separation characteristics [208]. In addition, their performance relies on precise fruit pose recognition and accurate end-effector positioning. Such accuracy is difficult to achieve in dense canopies with variable lighting and frequent occlusion, which increases the burden on visual perception and motion control and ultimately lowers the success rate of picking.

Soft end-effectors offer clear advantages in adapting to fruit shape and size, but their structural and mechanical limitations remain evident. Most designs employ symmetrically arranged multi-finger configurations with underactuated mechanisms. As a result, individual fingers lack independent adjustment and differentiated control. When handling fruits with irregular shapes, uneven surfaces, or atypical stem positions, some fingers often fail to make effective contact, while others may exert excessive pressure on localized regions. In addition, soft end-effectors typically exhibit limited load capacity and slow deformation response. For heavier fruits or tasks requiring higher detachment forces, excessive structural deformation or insufficient grasping strength may occur, reducing the picking success rate. The slow deformation response also introduces delays in actuation, further lowering picking efficiency.

In summary, both end-effectors based on picking patterns and soft end-effectors have achieved progress in enhancing adaptability across fruit varieties. Yet, a substantial gap remains when compared with the high universality and efficiency of human picking. Current approaches still fall short of meeting the practical requirements of agricultural production.

6.2. Future Trends

Inspired by the efficiency of the human hand in picking, a clear division of labor and functional complementarity exist among the fingers and palm. Through coordinated motion across multiple parts, the hand can flexibly perform a wide range of picking actions, including grasping, pinching, hooking, and pressing, while adapting to fruits of diverse shapes. As shown in Figure 15, analysis of picking behaviors reveals that the thumb and index finger typically act as the primary drivers of dexterous motion. They provide precise positioning and force control for fine operations such as pinching or severing stems. In contrast, the middle, ring, and little fingers, together with the palm, mainly serve as stabilizing supports, forming an auxiliary grasp. This pattern highlights a master–slave relationship, in which the thumb–index-finger pair performs dexterous primary movements, while the three ulnar-side fingers and palm provide auxiliary grasping. Furthermore, the rigid support of finger bones and joints ensures stability and efficient execution, whereas the compliance of soft tissues and tendons allows the fingertips to conform to fruit contours. This combination reduces excessive local pressure and minimizes fruit damage.

Humanoid dexterous hands represent an important direction for the development of universal picking end-effectors. Their design has advanced rapidly, with increasingly mature technologies. Most systems adopt multi-finger, multi-DOF configurations, where each DOF requires independent actuation and control. This results in complex structures and control schemes. Moreover, such hands rely heavily on high-precision perception and motion planning. In dense canopies with occluded fruits and variable lighting, accurate control is difficult to achieve, often leading to picking failure or fruit damage. An alternative strategy is to reduce the number of DOFs to simplify both structure and control. However, these designs typically depend on mechanical linkages with fixed joint motion sequences. As a result, hand flexibility is significantly reduced, making it difficult to perform coordinated multi-action picking, such as grasping, pinching, plucking, or hooking. Consequently, they fail to meet the diverse requirements of fruit detachment and grasping in complex agricultural environments.

Current humanoid dexterous hands fall far short of meeting the complex demands of agricultural picking. Dedicated research on structural optimization and control strategies is therefore required. The goal is to enable such hands to perform both detachment and grasping across multiple fruit types, thereby advancing toward truly universal picking. However, several key technical challenges remain, and corresponding development trends are emerging. The main issues are summarized as follows:

(1)

In the future, research should focus on master–slave design strategies for anthropomorphic picking hands. A central challenge lies in effectively integrating the “dexterous” master motion unit of the thumb and index finger with the “compliant” slave unit formed by the three ulnar fingers and the palm. A key breakthrough will be the dimensionality reduction mapping of human hand behaviors to the picking actions of anthropomorphic hands. This requires establishing clear relationships between human hand kinematic features and the core functional indicators of master–slave picking behavior. Furthermore, analyzing the degree of involvement of different DOFs and their interdependencies is expected to enable optimized combinations of motion, thereby providing theoretical guidance for simplified design. In addition, kinematic models of the master and slave units will likely play an important role in evaluating the motion performance of anthropomorphic picking hands. When combined with correlation models that characterize the interaction forces between human hand segments and fruit under different picking patterns, these models can serve as criteria for optimizing the structural design and control strategies of master–slave units, ultimately improving picking stability.

(2)

With advances in flexible materials and actuation technologies, the challenge of achieving compliant deformation of the palm is expected to be addressed. To enable palm deformation to support diverse picking behaviors, it is essential to investigate the coordination between the palm and fingers, for which the construction of a combined kinematic model is critical. Moreover, identifying how the interaction forces between fingers and fruit vary with palm motion will help optimize palm–finger coordination and refine palm control strategies. Such insights are vital for achieving distributed curved-surface contact and adaptive envelopment of fruits.

(3)

Although the introduction of flexible materials and structures has greatly improved compliance and adaptability, problems remain, including insufficient stiffness, limited load capacity, and slow actuation response. Rigid–flexible coupling offers new opportunities to address these limitations but also imposes higher demands on system design. Achieving a proper balance between the deformability of flexible structures and the load-bearing capacity of rigid elements remains a central challenge. One promising direction is to integrate anisotropic rigid materials or structures within flexible components, enabling deformation to vary significantly across different directions. Another direction is to optimize the layout of rigid–flexible coupled modules so that rigid elements are placed in non-critical deformation regions, while localized stress concentrations are introduced in flexible areas. This approach can accelerate structural deformation and reduce actuation delay.

While the development of anthropomorphic picking hands is a promising direction, it is not the only pathway for future research. The integration of advanced technologies into picking end-effectors offers equally important opportunities. For example, tactile sensing and multi-modal perception can provide fine-grained feedback for low-damage manipulation. Advanced material technologies, such as self-healing polymers, food-grade elastomers, and variable-stiffness composites, may significantly improve durability and adaptability under harsh agricultural conditions. Moreover, learning-based control methods, including reinforcement learning and zero-shot transfer, can enhance decision-making in unstructured environments. Together, these approaches complement the anthropomorphic design paradigm and broaden the roadmap for next-generation universal picking end-effectors.

Although technical performance is a prerequisite, the large-scale adoption of robotic picking end-effectors ultimately depends on economic viability and technology readiness. The primary economic value of universal end-effectors lies in their ability to operate across multiple crop types and seasons, thereby improving return on investment (ROI) by distributing costs over a broader range of tasks. To achieve payback periods attractive to farmers, however, hardware costs must be reduced, while picking speed (measured in fruits per second) and success rates must be significantly increased. At the same time, systems must demonstrate exceptionally high reliability with minimal human supervision. Looking forward, design standardization, mass production, and modular architectures are expected to reduce costs and accelerate the transition from experimental prototypes to commercially viable and field-ready systems.

Most designs remain at Technology Readiness Levels (TRL) 3–5, where prototypes are validated in laboratories or controlled agricultural environments. Only a limited number of commercial systems for specific crops, such as strawberries and tomatoes, have advanced to TRLs 6–7. Their deployment in open-field orchards, however, is still restricted by crop diversity, canopy complexity, and environmental variability. In comparison, more complex systems, such as soft pneumatic grippers, are mainly at TRLs 2–4, corresponding to the stages of technology development and component validation.

7. Conclusions

This review aims to summarize key advances and emerging directions in universal fruit-picking end-effectors under the challenges of fruit diversity and highly variable field environments. Two primary technical pathways are examined: end-effectors designed around specific picking patterns and soft end-effectors. Recent progress and limitations in both approaches are systematically analyzed. Finally, potential design concepts and technological pathways for highly universal picking end-effectors are proposed.

This paper first summarizes the key biomechanical parameters of multiple fragile fruits and their typical canopy structures under natural cultivation conditions. The analysis highlights that the great diversity in fruit types, morphologies, and growing environments results in poor adaptability and high damage rates when task transfer is required, which has significantly constrained the development and large-scale deployment of specialized picking end-effectors. Research progress on end-effectors designed around specific picking patterns and on soft picking end-effectors is then reviewed. Both approaches are found to exhibit clear gaps compared with manual picking, particularly in adaptability, stability, and the ability to perform diverse operations.

Finally, this review outlines the future directions for highly universal picking end-effectors. Their development is expected to converge toward humanoid dexterous hands for fruit picking, with three main aspects identified. First, greater attention should be given to master–slave design strategies that coordinate multiple fingers and the palm. Second, advances in flexible materials and actuation technologies may enable compliant deformation of the palm to be effectively addressed. Third, the integration of rigid–flexible coupled strategies holds promise for overcoming the limited load capacity of soft structures and the slow response of flexible actuators.

In summary, highly universal picking end-effectors hold broad prospects in agricultural robotics. Although significant challenges remain, advances in smart materials, multimodal perception, and adaptive control are expected to enable more efficient and intelligent cross-species fruit picking. These developments will also accelerate the industrialization and large-scale deployment of robotic picking systems.