In recent years, the integration of robotics into agricultural practices has gained significant traction, particularly in fruit harvesting. Traditional manual picking is labor-intensive and time-consuming, prompting the development of automated solutions. Among these, the end effector—a critical component of harvesting robots—plays a pivotal role in determining picking efficiency, success rates, and fruit integrity. I have focused on designing an end effector specifically for clustered crabapple picking, addressing challenges such as low success rates, fruit damage, and unsuitable picking methods. This article presents my comprehensive research on a rigid-flexible coupled end effector, detailing its design, experimental optimization, and validation.
The crabapple, a fruit rich in nutrients and widely cultivated in northern regions, typically grows in clusters of one to five fruits. Its small size and clustered arrangement pose unique challenges for mechanical harvesting. Conventional rigid end effectors often cause bruising or incomplete picking, while existing soft grippers may lack the necessary shear force for stem cutting. Therefore, I aimed to develop an end effector that combines flexible grasping with precise cutting, leveraging pneumatic actuation for both functions. The core innovation lies in the synergy between a silicone-based soft finger mechanism for enveloping fruits and a rigid shear mechanism for stem severance, ensuring minimal damage and high success rates.
My approach involved systematic design, simulation, and experimentation. I began by measuring key parameters of crabapples, such as fruit diameter and stem length, to inform the end effector’s dimensions. The design phase included kinematic simulations to validate the shear mechanism’s effectiveness. Subsequently, I conducted single-factor experiments to identify the influence of critical parameters—shear air pressure, blade thickness, and driving air pressure—on picking success. Building on this, I employed response surface methodology (RSM) to optimize these parameters and explore their interactions. Finally, I validated the optimized end effector in a laboratory setting using a robotic arm. Throughout this study, the term “end effector” is emphasized to underscore its centrality in automated harvesting systems.
Background and Motivation
Agricultural robots, especially for fruit picking, have evolved from industrial applications, yet their adoption in orchards remains limited due to design constraints. End effectors for harvesting must be adaptable, gentle, and efficient. For clustered fruits like crabapples, the end effector must accommodate variable cluster sizes while applying sufficient force to cut stems without harming adjacent fruits. Previous studies have explored rigid grippers, suction devices, and hybrid systems, but few address the specific needs of small, clustered fruits. My work builds on existing research by introducing a rigid-flexible coupled end effector that integrates grasping and cutting into a single, compact unit. This design aims to enhance compatibility and reduce mechanical complexity, addressing gaps in current technology.
The motivation stems from the economic and practical demands of crabapple harvesting. Manual picking is costly and prone to inconsistencies, whereas automated systems can improve yield and quality. By focusing on the end effector, I contribute to the broader goal of sustainable agriculture through robotics. The following sections detail the design process, experimental methodology, and results, with an emphasis on data-driven optimization. Tables and formulas are used extensively to summarize findings and illustrate relationships between variables.
Design of the End Effector
The end effector was designed with two main subsystems: a flexible mechanism for fruit envelopment and a rigid mechanism for stem cutting. The overall structure ensures that the end effector is lightweight, durable, and capable of handling cluster variations. I selected materials such as silicone for flexibility and aluminum alloy for rigidity, balancing strength and weight. The working principle involves positioning the end effector around the fruit cluster, inflating the soft fingers to grasp the fruits, and then activating the shear mechanism to cut the stem. This sequence minimizes fruit movement and damage, enhancing picking reliability.
The flexible mechanism consists of soft fingers made from HY-E620 silicone, chosen for its elasticity and durability. The fingers are fabricated using a mold-based process, with a strain layer that bends upon pneumatic inflation. The design parameters, derived from crabapple measurements, ensure that the fingers can envelop clusters up to 121.6 mm in diameter. The strain layer’s geometry, as summarized in Table 1, was optimized through iterative prototyping to achieve sufficient curvature without overstress. The fingers are mounted on support frames within the end effector, allowing simultaneous actuation via a single air source.
| Parameter | Value (mm) |
|---|---|
| Length (a1) | 3 |
| Width (b1) | 3 |
| Overall Length (l1) | 113 |
| Width (m1) | 17 |
| Radius 1 (r11) | 3 |
| Radius 2 (r12) | 6 |
| Radius 3 (r13) | 9 |
The rigid mechanism comprises a shear assembly with tungsten steel blades, a pneumatic cylinder, and sliding rails. The cylinder (model CU25-5) provides the force for blade movement, with its output calculated using the formula:
$$ F = P \times S $$
where \( F \) is the force (N), \( P \) is the air pressure (MPa), and \( S \) is the effective area (mm²). For the cylinder, the internal area is 412 mm², so the force ranges from 72.1 N to 123.6 N for pressures of 0.175 MPa to 0.300 MPa. The blades are mounted on tool plates connected via linkages to a sliding plate, ensuring synchronized opposing motion for clean stem cutting. The end effector’s housing is made from aluminum alloy 6063 for lightness, and all fasteners are standard steel components to maintain structural integrity.
Key dimensions of the end effector were based on crabapple measurements. I collected data from 20 crabapples, as shown in Table 2, to determine average fruit diameter (43.14 mm) and stem length (37.20 mm). For clusters, measurements indicated a maximum diameter of 121.6 mm for five-fruit clusters, guiding the envelopment cavity width of 125 mm. The shear mechanism height was set to 30 mm to accommodate stems within the cutting zone. These dimensions ensure that the end effector can handle typical cluster sizes without redesign.
| Sample Number | Fruit Diameter (mm) | Stem Length (mm) |
|---|---|---|
| 1 | 43.20 | 37.60 |
| 2 | 43.44 | 41.36 |
| 3 | 46.98 | 37.20 |
| 4 | 46.70 | 27.66 |
| 5 | 42.70 | 39.00 |
| 6 | 41.98 | 44.00 |
| 7 | 41.38 | 39.30 |
| 8 | 38.38 | 28.82 |
| 9 | 43.90 | 36.96 |
| 10 | 39.90 | 31.44 |
| 11 | 41.22 | 36.98 |
| 12 | 46.34 | 29.90 |
| 13 | 43.72 | 48.34 |
| 14 | 47.10 | 42.34 |
| 15 | 42.90 | 38.52 |
| 16 | 41.00 | 30.94 |
| 17 | 41.90 | 43.50 |
| 18 | 44.02 | 33.52 |
| 19 | 42.86 | 34.72 |
| 20 | 42.66 | 41.88 |
To validate the shear mechanism, I performed a kinematic simulation using SolidWorks Motion. The simulation assumed ideal conditions, with the stem modeled as lightweight wood and the blades as tungsten steel. A force of 82.4 N was applied to the cylinder rod, corresponding to 0.2 MPa pressure, to assess blade displacement and stem contact force. The results, illustrated in Figure 4 (simulation output), showed a maximum contact force of 29 N at 0.1 seconds, with blade displacement of 14 mm. This exceeds the measured stem shear force range of 12.567 N to 26.167 N (Table 3), confirming that the end effector can effectively cut stems.
| Sample Number | Shear Force (N) |
|---|---|
| 1 | 22.298 |
| 2 | 23.969 |
| 3 | 24.519 |
| 4 | 16.628 |
| 5 | 15.443 |
| 6 | 16.826 |
| 7 | 12.567 |
| 8 | 24.859 |
| 9 | 13.574 |
| 10 | 20.133 |
| 11 | 17.179 |
| 12 | 25.999 |
| 13 | 16.705 |
| 14 | 21.142 |
| 15 | 17.592 |
| 16 | 23.733 |
| 17 | 26.167 |
| 18 | 20.764 |
| 19 | 18.009 |
| 20 | 21.381 |
The end effector’s performance hinges on the coordination between the flexible and rigid parts. During operation, the soft fingers inflate to a bending angle that depends on driving air pressure. I characterized this relationship through preliminary tests, yielding the curve in Figure 5, which shows that angles above 60 degrees are achieved at pressures over 0.07 MPa. This ensures adequate envelopment for cluster grasping. The shear mechanism, meanwhile, relies on precise blade alignment and force control to prevent stem slippage or fruit crushing. Overall, the end effector’s design emphasizes robustness and adaptability, key for field deployment.
Experimental Methodology
To optimize the end effector, I conducted a series of experiments, starting with single-factor tests to determine parameter ranges, followed by response surface methodology for multi-factor optimization. All experiments were performed in a controlled laboratory environment using artificial crabapple clusters attached to a simulated tree. The end effector was mounted on an AUBO-E5 robotic arm for positioning, and picking success was defined as the fruit being cleanly cut and falling into a collection tube. Each test involved 20 picking attempts per cluster type (three-, four-, and five-fruit clusters), with success rates averaged over three repetitions.
The single-factor experiments focused on three parameters: shear air pressure (A), blade thickness (B), and driving air pressure (C). Based on preliminary trials, I established ranges: A from 0.175 MPa to 0.300 MPa, B from 0.2 mm to 0.5 mm, and C from 0.07 MPa to 0.10 MPa. For each parameter, I held the others constant at intermediate values and measured picking success rates. The results, summarized in Table 4, guided the selection of levels for RSM. For instance, shear air pressure showed an optimal range around 0.275 MPa, while blade thickness of 0.4 mm yielded high success. Driving air pressure above 0.09 MPa provided consistent grasping but with diminishing returns.
| Parameter | Value | Three-Fruit Success (%) | Four-Fruit Success (%) | Five-Fruit Success (%) |
|---|---|---|---|---|
| Shear Air Pressure (MPa) | 0.175 | 100 | 85.0 | 75.0 |
| 0.200 | 100 | 88.3 | 78.3 | |
| 0.225 | 100 | 91.7 | 85.0 | |
| 0.250 | 100 | 95.0 | 88.3 | |
| 0.275 | 100 | 98.3 | 90.0 | |
| 0.300 | 100 | 91.7 | 85.0 | |
| Blade Thickness (mm) | 0.2 | 100 | 90.0 | 81.7 |
| 0.3 | 100 | 100 | 88.3 | |
| 0.4 | 100 | 100 | 90.0 | |
| 0.5 | 100 | 95.0 | 83.3 | |
| Driving Air Pressure (MPa) | 0.07 | 95.0 | 88.3 | 80.0 |
| 0.08 | 98.3 | 91.7 | 85.0 | |
| 0.09 | 100 | 98.3 | 88.3 | |
| 0.10 | 100 | 100 | 90.0 |
For RSM, I used a Box-Behnken design (BBD) with three factors at three levels, as shown in Table 5. This design required 15 experimental runs, each repeated three times for reliability. The response variables were picking success rates for four- and five-fruit clusters (three-fruit success was consistently 100%, so it was excluded from analysis). I employed Design-Expert V8 software to fit quadratic models and analyze interactions. The general form of the quadratic model is:
$$ Y = \beta_0 + \sum \beta_i X_i + \sum \beta_{ii} X_i^2 + \sum \sum \beta_{ij} X_i X_j + \epsilon $$
where \( Y \) is the response, \( \beta \) are coefficients, \( X \) are factors, and \( \epsilon \) is error. The models were evaluated using analysis of variance (ANOVA) to assess significance and fit.
| Code | Shear Air Pressure (MPa) | Blade Thickness (mm) | Driving Air Pressure (MPa) |
|---|---|---|---|
| -1 | 0.250 | 0.3 | 0.08 |
| 0 | 0.275 | 0.4 | 0.09 |
| 1 | 0.300 | 0.5 | 0.10 |
The experimental matrix and results are presented in Table 6. Each run involved testing the end effector under specified conditions, with success rates recorded. For example, Run 1 with shear air pressure of 0.250 MPa, blade thickness of 0.3 mm, and driving air pressure of 0.09 MPa yielded 98.3% success for four-fruit clusters and 88.3% for five-fruit clusters. These data points were used to derive regression equations.
| Run | Shear Air Pressure (MPa) | Blade Thickness (mm) | Driving Air Pressure (MPa) | Four-Fruit Success (%) | Five-Fruit Success (%) |
|---|---|---|---|---|---|
| 1 | 0.250 | 0.3 | 0.09 | 98.3 | 88.3 |
| 2 | 0.300 | 0.3 | 0.09 | 91.7 | 85.0 |
| 3 | 0.250 | 0.5 | 0.09 | 95.0 | 81.7 |
| 4 | 0.300 | 0.5 | 0.09 | 90.0 | 75.0 |
| 5 | 0.250 | 0.4 | 0.08 | 91.7 | 90.0 |
| 6 | 0.300 | 0.4 | 0.08 | 83.3 | 85.0 |
| 7 | 0.250 | 0.4 | 0.10 | 98.3 | 88.3 |
| 8 | 0.300 | 0.4 | 0.10 | 91.7 | 83.3 |
| 9 | 0.275 | 0.3 | 0.08 | 95.0 | 88.3 |
| 10 | 0.275 | 0.5 | 0.08 | 90.0 | 80.0 |
| 11 | 0.275 | 0.3 | 0.10 | 98.3 | 85.0 |
| 12 | 0.275 | 0.5 | 0.10 | 93.3 | 78.3 |
| 13 | 0.275 | 0.4 | 0.09 | 100 | 90.0 |
| 14 | 0.275 | 0.4 | 0.09 | 98.3 | 90.0 |
| 15 | 0.275 | 0.4 | 0.09 | 100 | 91.7 |
Results and Discussion
The RSM analysis yielded quadratic regression models for four-fruit and five-fruit picking success rates. Using stepwise regression, non-significant terms were removed, resulting in the following equations:
$$ Y_1 = 99.43 – 5A – 4.16B + 1.66C – 0.025AB – 1.68AC – 2.22A^2 – 5.54B^2 – 8.04C^2 $$
$$ Y_2 = 90.57 – 1.66A – 3.75B – 0.41C – 0.85AB + 0.025AC – 3.20A^2 – 4.87B^2 – 1.55C^2 $$
where \( Y_1 \) is four-fruit success (%), \( Y_2 \) is five-fruit success (%), \( A \) is shear air pressure (coded), \( B \) is blade thickness (coded), and \( C \) is driving air pressure (coded). The coefficients indicate the magnitude and direction of each factor’s effect. For instance, in \( Y_1 \), shear air pressure has a negative linear effect, suggesting that increasing pressure beyond an optimum reduces success, possibly due to mechanical vibrations.
ANOVA results for the models are shown in Tables 7 and 8. Both models are highly significant (p < 0.01), with low lack-of-fit values (p > 0.05), indicating good fit. The R² values of 0.9505 for \( Y_1 \) and 0.9923 for \( Y_2 \) imply that the models explain most of the variance. The coefficient of variation (CV) is below 2%, demonstrating experimental precision. These statistics validate the models for optimization.
| Source | Sum of Squares | Degrees of Freedom | Mean Square | F-value | p-value | Significance |
|---|---|---|---|---|---|---|
| Model | 293.89 | 9 | 32.65 | 10.66 | 0.0090 | ** |
| A | 88.45 | 1 | 88.45 | 28.87 | 0.0030 | ** |
| B | 28.13 | 1 | 28.13 | 9.18 | 0.0291 | * |
| C | 58.32 | 1 | 58.32 | 19.04 | 0.0073 | ** |
| AB | 0.64 | 1 | 0.64 | 0.21 | 0.6668 | ns |
| AC | 0.81 | 1 | 0.81 | 0.26 | 0.6290 | ns |
| BC | 0.000 | 1 | 0.000 | 0.000 | 1.0000 | ns |
| A² | 68.01 | 1 | 68.01 | 22.20 | 0.0053 | ** |
| B² | 7.15 | 1 | 7.15 | 2.33 | 0.1871 | ns |
| C² | 55.92 | 1 | 55.92 | 18.25 | 0.0079 | ** |
| Residual | 15.32 | 5 | 3.06 | |||
| Lack of Fit | 13.39 | 3 | 4.46 | 4.63 | 0.1826 | ns |
| Pure Error | 1.93 | 2 | 0.96 | |||
| Total | 309.21 | 14 |
| Source | Sum of Squares | Degrees of Freedom | Mean Square | F-value | p-value | Significance |
|---|---|---|---|---|---|---|
| Model | 322.06 | 9 | 36.90 | 71.88 | <0.0001 | *** |
| A | 50.00 | 1 | 50.00 | 97.40 | 0.0002 | *** |
| B | 124.82 | 1 | 124.82 | 243.16 | <0.0001 | *** |
| C | 8.82 | 1 | 8.82 | 17.18 | 0.0090 | ** |
| AB | 2.89 | 1 | 2.89 | 5.63 | 0.0637 | ns |
| AC | 0.000 | 1 | 0.000 | 0.000 | 1.0000 | ns |
| BC | 0.64 | 1 | 0.64 | 1.25 | 0.3149 | ns |
| A² | 17.20 | 1 | 17.20 | 33.51 | 0.0022 | ** |
| B² | 128.89 | 1 | 128.89 | 251.09 | <0.0001 | *** |
| C² | 11.42 | 1 | 11.42 | 22.24 | 0.0053 | ** |
| Residual | 2.57 | 5 | 0.51 | |||
| Lack of Fit | 0.64 | 3 | 0.21 | 0.22 | 0.8755 | ns |
| Pure Error | 1.93 | 2 | 0.96 | |||
| Total | 334.63 | 14 |
Response surface plots were generated to visualize factor interactions. For instance, Figure 9a shows the effect of shear air pressure and blade thickness on success rates at fixed driving air pressure. The plots reveal that success rates peak at intermediate values, forming convex surfaces. The contours indicate that shear air pressure has the steepest gradients, followed by blade thickness and driving air pressure. This aligns with the coefficient magnitudes, confirming that shear air pressure is the most influential factor. Interaction terms like AB and AC are generally non-significant, implying that factors act largely independently within the tested ranges.
To optimize the end effector, I used the models to maximize picking success. The optimal conditions, as predicted by Design-Expert, are shear air pressure of 0.260 MPa, blade thickness of 0.4 mm, and driving air pressure of 0.09 MPa. At these settings, the predicted success rates are 100% for three- and four-fruit clusters, and 91.29% for five-fruit clusters. The desirability function approach confirmed this as a global optimum, balancing all responses. The ranking of factor effects, from largest to smallest, is shear air pressure, blade thickness, and driving air pressure, consistent across both models.
I validated the optimization through laboratory experiments. The end effector was configured with the optimal parameters and tested on 100 clusters per type (20 clusters per group, five groups). The results, summarized in Table 9, show success rates of 100% for three-fruit clusters, 100% for four-fruit clusters, and 91% for five-fruit clusters. The slight deviation from the predicted 91.29% for five-fruit clusters is within experimental error, confirming the model’s accuracy. Compared to pre-optimization trials, success rates improved by up to 18.3% for four-fruit clusters and 11% for five-fruit clusters, demonstrating the value of RSM.
| Cluster Type | Group 1 Success (%) | Group 2 Success (%) | Group 3 Success (%) | Group 4 Success (%) | Group 5 Success (%) | Average Success (%) |
|---|---|---|---|---|---|---|
| Three-Fruit | 100 | 100 | 100 | 100 | 100 | 100 |
| Four-Fruit | 100 | 100 | 100 | 100 | 100 | 100 |
| Five-Fruit | 90 | 90 | 90 | 90 | 95 | 91 |
The end effector’s performance can be attributed to its rigid-flexible coupling. The soft fingers provide compliant grasping, adapting to cluster shapes without excessive force, while the rigid shear mechanism delivers precise cutting. The pneumatic actuation allows for fine control over both functions. However, limitations include susceptibility to vibration at high shear pressures and wear on silicone fingers over time. Future iterations could incorporate feedback sensors or alternative materials to enhance durability.
Conclusion and Future Work
In this study, I designed and optimized a rigid-flexible coupled end effector for clustered crabapple picking. The end effector integrates silicone soft fingers for fruit envelopment and a pneumatic shear mechanism for stem cutting, addressing key challenges in automated harvesting. Through single-factor experiments and response surface methodology, I identified optimal parameters: shear air pressure of 0.260 MPa, blade thickness of 0.4 mm, and driving air pressure of 0.09 MPa. These settings yield picking success rates of 100% for three- and four-fruit clusters, and 91% for five-fruit clusters, meeting practical requirements.
The research highlights the importance of parameter optimization in end effector design. Shear air pressure emerged as the most critical factor, followed by blade thickness and driving air pressure. The quadratic models developed through RSM provide a reliable framework for predicting performance and guiding adjustments. The end effector’s success demonstrates the potential of rigid-flexible coupling for delicate fruit harvesting, offering a solution that reduces labor and minimizes damage.
For future work, I plan to explore several avenues. First, reducing the number of soft fingers could streamline the end effector and save space, as noted in the conclusion. Second, integrating machine vision for automatic cluster detection and positioning would enhance autonomy. Third, field testing in real orchards is necessary to assess durability under environmental conditions. Finally, the design principles could be adapted for other clustered fruits, such as cherries or grapes, by scaling dimensions and adjusting material properties.
In summary, this end effector represents a step forward in agricultural robotics. By combining flexibility with precision, it offers a versatile tool for fruit picking. The experimental methodology and optimization approach can inform future developments, contributing to more efficient and sustainable farming practices. As robotics continue to evolve, end effectors like this will play a crucial role in bridging the gap between manual labor and full automation.