The automated harvesting of greenhouse tomatoes presents a significant challenge for agricultural robotics, primarily due to the delicate nature of the fruit and the cluttered, occluded growth environment within facilities. A critical component determining the success of such robotic systems is the end effector—the device responsible for the final interaction with the crop. Common issues with traditional grippers include low success rates, fruit damage during grasping or pulling, and interference from surrounding foliage and adjacent fruits. To address these limitations, this work details the design, analysis, and experimental validation of a novel three-fingered robotic end effector integrated with a telescopic suction mechanism. This design aims to achieve reliable, high-success-rate, and low-damage harvesting of tomatoes in greenhouse settings.
The target cultivar for this study is ‘Fentailang,’ a common greenhouse variety known for its firm texture and suitability for mechanical handling. The physical and mechanical properties of this tomato were first characterized to inform the end effector design. Geometrical parameters, including the horizontal diameter (the maximum width, $h_t$) and vertical diameter (height from calyx to blossom end, $h_v$), were measured for a sample of 100 fruits. Their mass was also recorded. The statistical ranges are summarized in Table 1.
| Parameter | Minimum | Maximum | Mean (approx.) |
|---|---|---|---|
| Mass ($m$) | 182.0 g | 336.5 g | ~250 g |
| Horizontal Diameter ($h_t$) | 62.0 mm | 98.3 mm | ~80 mm |
| Vertical Diameter ($h_v$) | 52.4 mm | 72.3 mm | ~62 mm |
These dimensions directly dictate the required workspace and finger length of the harvesting end effector. The mechanical property, specifically the force required to damage the fruit, was determined using a texture analyzer. A compression test on the fruit’s horizontal axis revealed a critical threshold. The force-deformation curve showed an initial elastic region until a deformation of approximately 13 mm. The first significant peak, indicating epidermal rupture, occurred at a force of $F_d = 9.8 \, \text{N}$. This value establishes the upper limit for the permissible grasping force exerted by the end effector to avoid bruising or damaging the tomato.
The core innovation of the proposed end effector lies in its hybrid approach, combining a rigid grasping mechanism with a flexible pre-positioning suction system. The overall architecture comprises three main subsystems: the grasping mechanism, the drive mechanism, and the adsorption (suction) mechanism. The grasping mechanism consists of three rigid fingers arranged symmetrically at 120-degree intervals, mimicking a human hand’s grip. Each finger is coated with a silicone pad to increase friction and cushion the contact. The fingers are actuated by a single stepper motor, which drives a ball screw. The linear motion of the screw nut is converted into the rotational opening and closing of the fingers via a linkage-rocker system.
The telescopic suction unit is mounted centrally. Its purpose is to actively locate and gently extract the target fruit from its often occluded position within the cluster before the fingers close. This proactive step significantly reduces the chance of gripping leaves or neighboring fruits. The general workflow is: 1) The vision system identifies and locates a ripe tomato. 2) The robot arm positions the end effector near the fruit. 3) The suction cup extends, attaches to the fruit’s surface, and retracts slightly to pull the fruit into a more accessible position. 4) The three fingers close to securely grip the fruit. 5) A combined twisting and pulling motion detaches the fruit from the stem. 6) The fruit is transported and released into a collection bin.
The design of the finger mechanism required a detailed kinematic analysis. A schematic of a single finger is established in a Cartesian coordinate system. The primary design goal was to achieve a workspace accommodating tomatoes with horizontal diameters between 50 mm and 110 mm. The key parameters are the initial finger opening $d_i$, the slider travel distance $S$, and the rocker rotation angle $\delta$. Based on the measured vertical diameter, the finger length $L_5$ was set to 75 mm. The initial opening was set to the average target diameter:
$$ d_i = \frac{D_{\text{max}} + D_{\text{min}}}{2} = 80 \, \text{mm} $$
The slider’s travel is derived from the rocker’s rotation ($\delta = 15^\circ$) and length $L_2$:
$$ S_{\text{max}} = L_2 \sin \delta $$
Through kinematic modeling and considering spatial constraints from the motor and cylinder, the final link lengths were determined, as shown in Table 2.
| Joint / Link | Symbol | Value |
|---|---|---|
| Slider to First Joint | $L$ | 30 mm |
| First Link (Rocker) | $L_1$ | 60 mm |
| Second Link (Crank) | $L_2$ | 45 mm |
| Connecting Link | $L_3$ | 65 mm |
| Finger Base Link | $L_4$ | 30 mm |
| Finger Length | $L_5$ | 75 mm |
| Max Slider Travel | $S_{\text{max}}$ | ~11.6 mm |
| Rocker Rotation | $\delta$ | 0° to 15° |
A crucial aspect of the end effector design is determining the safe and effective grasping force range. A force analysis on the tomato during gripping is essential. When the suction cup detaches and the fruit is held solely by the fingers, the static friction force must balance the fruit’s weight to prevent slipping. With three symmetric contact points, the normal force $N_i$ at each point is assumed equal. The minimum required normal force $N_{\text{min}}$ is:
$$ \sum_{i=1}^{3} f_i = m g $$
where $f_i = \mu N_i$ is the friction force and $\mu$ is the coefficient of friction.
$$ N_{\text{min}} \geq \frac{m_{\text{max}} \, g}{3 \mu} $$
The coefficient of friction $\mu$ between the tomato skin and potential cushioning materials was tested. Silicone was chosen over rubber due to its higher friction, with $\mu \approx 0.8$. Using the maximum fruit mass $m_{\text{max}} = 0.3365 \, \text{kg}$, the lower force bound is:
$$ N_{\text{min}} \geq \frac{0.3365 \times 9.8}{3 \times 0.8} \approx 1.37 \, \text{N} $$
Combined with the damage threshold force $F_d = 9.8 \, \text{N}$, the permissible grasping force range for the end effector finger is therefore:
$$ 1.37 \, \text{N} \leq F_N < 9.8 \, \text{N} $$
This range ensures a stable grip without damage.
The interaction between the end effector and the tomato during the harvesting cycle was analyzed in three distinct phases: Pulling, Separation, and Pre-placement. In the Pulling phase, the fruit is subjected to the stem attachment force $T_1 \approx 9.73 \, \text{N}$, finger forces $F_N$, gravity, and friction. Successful detachment requires:
$$ 3 (F_N \cos \theta + f \sin \theta) + mg \geq T_1 $$
For a large tomato ($\theta \approx 78^\circ$, $m=0.3 \, \text{kg}$), solving this inequality gives $F_N \geq 2.28 \, \text{N}$, which lies within the safe range. During Separation, with the stem detached, the fruit is in equilibrium between finger forces, suction force $F_p$, and gravity:
$$ 3F_N \cos \theta + mg = F_p $$
Substituting $F_N = 2.28 \, \text{N}$ yields $F_p \approx 4.36 \, \text{N}$, well below the damage threshold. In the Pre-placement phase (fruit oriented downward), the condition to prevent dropping is:
$$ 3 (F_N \cos \theta + f \sin \theta) > mg $$
This holds true for the entire range of fruit sizes and the corresponding contact angles $\theta$ (68° to 78°), confirming the end effector‘s ability to securely hold the fruit after harvest.
To validate the kinematic and dynamic performance prior to physical prototyping, a simulation of the grasping process was conducted using ADAMS software. A 3D model of the end effector and a tomato (modeled with a mass of 300 g, elastic modulus of 0.762 MPa, and Poisson’s ratio of 0.45) was imported. The simulation tracked the motion of the fingers and slider, and more importantly, the contact forces on the tomato. The displacement curves for the three fingers showed smooth and symmetric motion, with a total opening/closing travel consistent with the theoretical design (~23 mm). The slider travel matched the predicted $S_{\text{max}}$ of approximately 11.6 mm. The contact force analysis was critical. The simulation showed an initial spike in force upon impact (up to ~16 N), which rapidly stabilized. During the steady gripping and pulling phase, the contact force oscillated around 6 N. This value is within the safe range of less than 9.8 N, confirming that the end effector design should not cause damage during operation. Simulations with different fruit sizes (70, 80, 90 mm diameter) yielded similar stable force profiles without fruit drop, verifying the robustness of the grasp.
Based on the design and simulation results, a functional prototype of the tomato harvesting end effector was manufactured. The fingers and linkages were machined from aluminum alloy 6061 for strength, while housings were 3D-printed with ABS plastic. The prototype was integrated with a six-degree-of-freedom robotic arm and a vision system for field testing in a greenhouse environment. The primary performance metrics were the single-fruit harvesting success rate and the average cycle time. A series of 10 trials were conducted on different plants, involving a total of 113 tomato fruits. The results are summarized in Table 3.
| Trial Set | Total Attempts | Successful Suction | Successful Pick |
|---|---|---|---|
| 1 | 14 | 12 | 12 |
| 2 | 8 | 8 | 8 |
| 3 | 13 | 12 | 12 |
| 4 | 12 | 10 | 10 |
| 5 | 7 | 6 | 6 |
| 6 | 18 | 15 | 15 |
| 7 | 10 | 10 | 10 |
| 8 | 13 | 9 | 9 |
| 9 | 7 | 7 | 7 |
| 10 | 11 | 10 | 10 |
| Total / Rate | 113 | 99 (88%) | 99 (88%) |
The data shows a clear correlation: whenever the telescopic suction mechanism successfully adhered to and extracted the fruit, the grasping mechanism subsequently achieved a successful harvest. Therefore, the overall 88% success rate is primarily determined by the suction subsystem’s performance. Further analysis revealed that success rate varied with fruit size, as shown in Table 4. The end effector performed best on medium-sized fruits (70-90 mm diameter), with a 91.8% success rate. For smaller fruits, the high surface curvature likely reduced the effective suction seal. For larger fruits, their increased mass may have exceeded the suction force or required grasping forces closer to the damage threshold.
| Fruit Size Category | Attempts | Successes | Success Rate |
|---|---|---|---|
| Large (>90 mm) | 15 | 11 | 73.3% |
| Medium (70-90 mm) | 86 | 79 | 91.8% |
| Small (<70 mm) | 12 | 9 | 75.0% |
The average cycle time for a single fruit harvest was broken down as follows: suction extension and retraction (1.8 s), finger closing (0.6 s), and twisting/pulling (1.5 s), with short intervals between actions. The total average harvest time was approximately 5.4 seconds per fruit, which is promising for automated harvesting applications.
In conclusion, this work presented the comprehensive design and testing of a specialized end effector for automated greenhouse tomato harvesting. The three-fingered, telescopic-suction end effector was designed based on a thorough analysis of the tomato’s physical and mechanical properties. Kinematic and force analyses established design parameters and confirmed that the grasping forces would remain within a safe, non-damaging range while being sufficient for stable manipulation. Dynamic simulations validated these models, showing stable grasping with forces below the damage threshold. Finally, field trials with a functional prototype demonstrated the practical efficacy of the end effector, achieving an overall harvest success rate of 88% and an average cycle time of 5.4 seconds per fruit. The design effectively addresses key challenges of occlusion and fruit damage, contributing to the advancement of robust robotic solutions for facility agriculture. The integration of proactive fruit positioning via suction prior to grasping is a key feature that enhances the reliability of this robotic end effector in complex growing environments.