The pursuit of automation in agriculture is a relentless drive towards efficiency, precision, and labor salvation. Among the various technological frontiers, robotic selective harvesting stands as a pinnacle challenge, requiring a harmonious integration of machine vision, sophisticated path planning, and, most critically, a dexterous and reliable end effector. The end effector is the final link in the robotic chain, the physical interface that directly interacts with the crop. Its performance dictates the success of the entire harvesting operation, influencing metrics such as pick success rate, fruit damage, and cycle time. For delicate, high-value crops where maturity is non-uniform and the plant structure is complex, the design of an effective end effector becomes paramount. This work delves into the intricate design, modeling, and experimental validation of a specialized cutting-detachment end effector developed for a challenging crop: the upward-growing fresh chili pepper.
Fresh chili peppers, particularly varieties that grow upright (often called “skyward-pointing” peppers), present a unique set of challenges for robotic harvesting. Unlike fruits like apples or oranges, these peppers are characterized by a slender, conical shape, a relatively thin and flexible stem (peduncle), and a growth habit that often places them singly amidst dense foliage. Manual harvesting relies on the human hand’s ability to precisely grasp the stem and snap it. Replicating this nuanced action with a machine requires a mechanism that can isolate the target fruit, securely engage its stem, and sever it without damaging the fruit or the plant. This paper chronicles the comprehensive development process of such an end effector, from initial biological investigation of the pepper’s physical and mechanical properties, through the conceptual and detailed mechanical design, to rigorous kinematic and dynamic analysis, optimization, and final field-style testing.
Biological Foundations: Understanding the Target
Any successful biomimetic or crop-specific robotic design must be rooted in a quantitative understanding of the target organism. Before a single component was modeled, a thorough biological characterization of the chosen pepper variety (‘High Spice 878’) was undertaken. The primary parameters of interest were the fruit’s dimensions and the stem’s morphological and biomechanical properties, as these directly inform the size, stroke, and force requirements of the end effector.
The physical parameters were measured from a sample of 200 fruits. The statistical distributions are summarized below:
| Parameter | Range | Percentage in Sample (%) |
|---|---|---|
| Fruit Length (L) | 55 – 80 mm | 85.5 |
| Fruit Max Diameter (D) | 9 – 13 mm | 91.5 |
| Stem Length (l) | 20 – 30 mm | 97.0 |
| Stem Diameter (d) | 2 – 3 mm | 89.5 |
This data was crucial for defining the envelope of the end effector’s capture and cutting zone. The stem length (20-30 mm) determined the necessary depth of the cutting mechanism, while the fruit dimensions defined the required internal volume of the capture funnel.
Beyond geometry, the stem’s resistance to cutting—its ultimate shear strength—is the fundamental load the end effector must overcome. This force is not constant; it varies with the stem’s diameter, moisture content, and critically, the angle of incision. The angle of incision (β) is defined as the angle between the stem’s central axis and the direction of the cutting blade’s motion. In practical harvesting, due to stem curvature and robotic positioning errors, this angle is not always a perfect 90°. Therefore, cutting tests were performed at incision angles of 70°, 80°, and 90°, and for different cutting types: transverse cut (stem axis orthogonal to cutting plane and direction), oblique cut (stem axis oblique to cutting plane, orthogonal to direction), and slicing cut (stem axis oblique to both plane and direction).
The maximum cutting force \( F_r \) was recorded using a universal testing machine. A representative force-displacement curve showed a characteristic double-peak profile, corresponding to the sequential failure of the outer epidermis and the inner vascular bundles. The key results for the highest-force scenario (transverse cut) are condensed below:
| Stem Diameter Range (mm) | Max Cutting Force at β=90° (N) | Max Cutting Force at β=80° (N) | Max Cutting Force at β=70° (N) |
|---|---|---|---|
| 2.0 – 2.3 | 8.5 | 6.8 | 6.8 |
| 2.6 – 2.9 | 11.9 | 10.7 | 10.5 |
| 2.9 – 3.2 | 13.1 | 11.9 | 12.4 |
The critical takeaway was that the maximum cutting force \( F_{r_{max}} \) observed was 13.1 N for a thick stem (≈3.2 mm) under a transverse cut. This value, with a safety margin factored in, became the key load parameter for the subsequent dynamic force analysis and actuation selection of the harvesting end effector. It was also noted that oblique and slicing cuts generally required lower forces than transverse cuts at the same stem diameter.
Conceptual and Detailed Design of the End Effector
Guided by the biological data, the core design philosophy for the end effector was established: Cutting-Detachment with Passive Tolerance. Instead of replicating the complex pinch-and-twist motion of human fingers, the design employs a simple, robust scissor-cut action. The “passive tolerance” is achieved through a conical capture sleeve, which accommodates slight positional inaccuracies from the robotic arm. The overall end effector consists of three main functional modules:
- Pose Adjustment Joints: Two serial revolute joints located at the base of the end effector. These allow the entire cutting assembly to be tilted, aligning its central axis with the upward-growing pepper fruit before the harvesting motion begins. This pre-alignment significantly increases the success rate of fruit capture.
- Tolerance Sleeve (Conical Funnel): This is the fruit capture interface. It is a conical tube with an internal diameter larger than the maximum fruit diameter. This design ensures that even if the robot’s approach is slightly off-center, the fruit will be guided into the sleeve. The sleeve’s length is designed to encompass the fruit’s stem fully.
- Cutting and Collection Mechanism: The core actuated module. It features a symmetrical four-bar linkage (crank-slider mechanism) that drives two opposing cutting blades. A flexible flap is attached to each blade. In the open position, the blades and flaps are retracted, allowing the fruit stem to enter. When activated, the blades close to cut the stem, and the flaps simultaneously close underneath the fruit, forming a temporary basket to prevent the harvested pepper from falling out during transport to the collection bin.
The working cycle is as follows: The robotic arm positions the end effector above the target fruit. The adjustment joints align the sleeve. The arm then moves down, enveloping the fruit within the conical sleeve. The cutting mechanism is activated, severing the stem and closing the retention flaps. The arm retracts and moves to a discharge point, where the cutting mechanism opens, releasing the fruit. This cycle combines precision (via pre-alignment) with robustness (via the tolerance sleeve and forceful cut).
Kinematic and Kineto-Static Analysis of the Cutting Mechanism
To size the components and select the appropriate actuator, a detailed mathematical model of the symmetrical cutting linkage was developed. The analysis focuses on one half of the mechanism due to its symmetry.
Kinematic Model: A vector loop analysis is performed on the crank-slider mechanism. Let the fixed pivot of the crank be point A, the crank-rocker joint be B, the slider (cutting blade assembly) be point C, and the tip of the cutting blade be point D. The crank length is \( l_1 \), the connecting rod length is \( l_2 \), and the blade extension from the slider joint is \( l_3 \). The crank angle is \( \theta_1 \), and the connecting rod angle is \( \theta_2 \). The offset of the crank pivot from the central axis is \( e \).
The position equations are derived as:
$$ x_B = e + l_1 \cos \theta_1, \quad y_B = l_1 \sin \theta_1 $$
$$ x_C = 0, \quad y_C = l_1 \sin \theta_1 + l_2 \sin \theta_2 $$
$$ x_D = l_3 \cos \theta_2, \quad y_D = l_1 \sin \theta_1 + (l_2 + l_3) \sin \theta_2 $$
The geometric constraint from the fixed vertical path of the slider gives the relationship between \( \theta_1 \) and \( \theta_2 \):
$$ l_2 \cos \theta_2 + l_1 \cos \theta_1 + e = 0 $$
By differentiating these position equations with respect to time, the velocities and accelerations of points C (slider) and D (blade tip) can be found. The angular velocity \( \omega_2 \) and angular acceleration \( \alpha_2 \) of the connecting rod are critical for the subsequent force analysis:
$$ \omega_2 = -\frac{l_1 \omega_1 \sin \theta_1}{l_2 \sin \theta_2} $$
$$ \alpha_2 = \frac{\omega_2^2 l_2 \cos \theta_2 – \omega_1^2 l_1 \cos \theta_1}{l_2 \sin \theta_2} $$
where \( \omega_1 \) and \( \alpha_1 \) are the crank’s angular velocity and acceleration, respectively.
Kineto-Static (Dynamic Force) Model: A dynamic force analysis is essential because the mechanism operates at speeds where inertial forces are significant relative to the cutting force. This inertia can be harnessed to reduce the required actuator torque. The analysis treats all links as rigid bodies and uses the D’Alembert principle to include inertial forces and moments. Each link’s free-body diagram is drawn, including:
- Weight forces (\( m_i g \)).
- Inertial forces (\( -m_i \mathbf{a}_{Si} \)) at the center of mass.
- Inertial moments (\( -J_{Si} \mathbf{\alpha}_i \)).
- The external cutting force \( F_r \) applied at the blade tip (point D). For analysis, a conservative value of 15 N was used.
- Joint reaction forces.
- The unknown driving torque \( M_b \) on the crank.
For the connecting rod (link BD), with mass \( m_3 \) and moment of inertia \( J_{S3} \), the inertial force and moment are:
$$ \mathbf{F}_{3}^{inertial} = -m_3 \mathbf{a}_{S3}, \quad M_{3}^{inertial} = -J_{S3} \alpha_2 $$
Similar equations apply to the crank and the slider/blade assembly. Writing the force and moment equilibrium equations for each link (\( \sum F_x=0, \sum F_y=0, \sum M_{Si}=0 \)) generates a system of linear equations. This system can be cast in matrix form \( \mathbf{C} \mathbf{F_R} = \mathbf{D} \), where \( \mathbf{F_R} \) is the vector of unknown forces and the driving torque \( M_b \), \( \mathbf{C} \) is the coefficient matrix from geometry, and \( \mathbf{D} \) is the vector containing known weights, inertial terms, and the external cutting force \( F_r \). Solving this system for a given crank position yields the instantaneous required driving torque \( M_b \).
Dimensional Optimization of the Linkage
With the kineto-static model established, the next step was to optimize the key link lengths (\( l_1, l_2, l_3 \)) to minimize the peak required driving torque \( M_b \) during the cutting stroke. This minimization allows for the selection of a smaller, lighter, and more energy-efficient actuator. A genetic algorithm (GA) was employed for this multi-variable, constrained optimization.
Objective Function: Minimize the absolute value of the required driving torque \( |M_b| \) throughout the cutting motion, particularly at the instant of peak cutting force.
Design Variables: \( l_1 \) (crank length), \( l_2 \) (connecting rod length), \( l_3 \) (blade extension).
Constraints:
- Transmission angle for the crank-rocker mechanism must be greater than 45° for good force transmission.
- Initial distance between blade tips (open position) > 40 mm to accommodate the fruit stem and sleeve.
- Final horizontal position of blade tip \( x_D \geq 0 \) at the end of the cut.
- Total vertical travel of blade tip \( \Delta y_D < 30 \) mm to match the biological stem length.
- Practical bounds on link lengths based on the end effector’s overall size envelope.
The optimization was run with the crank rotating uniformly (\( \omega_1 = 40^\circ/s \)) from an open position (\( \theta_1 = 90^\circ \)) to a closed position (\( \theta_1 = 150^\circ \)). The cutting force \( F_r = 15 N \) was applied when the blades engaged the stem. The GA successfully converged to an optimal set of dimensions.
| Optimized Parameter | Symbol | Optimized Value (mm) |
|---|---|---|
| Crank Length | \( l_1 \) | 33.0 |
| Connecting Rod Length | \( l_2 \) | 60.0 |
| Blade Extension Length | \( l_3 \) | 54.0 |
With these dimensions, the peak required driving torque was calculated to be 0.94 N·m. Based on this result, a high-torque servo motor (capable of providing a stall torque well above this value, e.g., a 10 kg-cm servo) was selected as the actuator for the cutting mechanism in the physical end effector prototype. The dynamic simulation using the optimized lengths showed the progression of angles, velocities, and the driving torque throughout the cutting cycle, confirming a smooth motion with a manageable torque peak corresponding to the cutting event.
Prototype Fabrication and Experimental Harvesting Trials
A functional prototype of the complete end effector was manufactured using a combination of 3D-printed polymers (for structural links, sleeve, and housing) and laser-cut steel (for the cutting blades). The end effector was mounted onto a Delta-type parallel robotic arm for testing. The trials aimed to validate the overall performance of the end effector in a controlled environment simulating real harvesting conditions.
Experimental Setup: Potted pepper plants with fruits at the target maturity stage were placed within the workspace of the robotic arm. A simple pick-and-place sequence was programmed: approach, align sleeve, descend, actuate cut, retract, and discharge. The primary performance metric was the stem cutting success rate, defined as the percentage of trials where the stem was completely severed by the cutting mechanism without the fruit being torn or pulled from the plant.
Results and Analysis: A total of 150 harvesting trials were conducted on fruits with varying stem diameters and natural growth angles (which affect the effective incision angle β). The results, stratified by stem diameter and effective incision angle range, are summarized below:
| Incision Angle Range (β) | Cutting Success Rate by Stem Diameter (mm) | ||||
|---|---|---|---|---|---|
| 2.2 – 2.4 | 2.4 – 2.6 | 2.6 – 2.8 | 2.8 – 3.0 | 3.0 – 3.2 | |
| 70° – 80° | 100% | 100% | 100% | 86.7% | 73.3% |
| 80° – 90° | 100% | 100% | 100% | 80.0% | 73.3% |
The overall cutting success rate across all 150 trials was 91.3%. This high rate validates the core functionality of the cutting-detachment end effector. The trials confirmed several design features: the conical tolerance sleeve effectively guided fruits into position despite minor alignment errors; the four-bar linkage provided the necessary mechanical advantage and motion trajectory for a clean cut; and the flexible retention flaps successfully contained the harvested fruit.
The failures (8.7%) were analyzed. They predominantly occurred with the thickest stem diameters ( > 2.8 mm) and were attributed to a combination of factors: 1) Minor manufacturing tolerances leading to a slight gap between the two cutting blades at full closure, 2) Occasional interference from small leaves or side shoots entering the sleeve alongside the stem, effectively increasing the cutting resistance, and 3) The inherent variability in stem tissue strength. In these failure cases, the stem was often partially cut, and the subsequent retraction of the robotic arm would pull and finally detach the fruit, causing undesirable plant disturbance.
Discussion and Impact
The development of this cutting-detachment end effector demonstrates a systematic, engineering-driven approach to solving a specific agricultural automation problem. The process underscores the critical importance of foundational biological data—without precise measurements of stem strength and fruit geometry, the design would be based on guesswork. The integration of kinematic and dynamic modeling with modern optimization techniques (genetic algorithm) allowed for a performance-driven design that minimized actuator requirements, leading to a more compact and efficient end effector.
The achieved success rate of 91.3% in controlled trials is highly promising for a prototype end effector targeting a challenging crop like upright chili peppers. It compares favorably with early-stage prototypes for other complex crops (e.g., sweet peppers, strawberries) reported in the literature. The key innovations contributing to this success are:
- The Passive Tolerance Conical Sleeve: This simple geometric feature drastically reduces the precision required from the vision system and robotic arm, making the system more robust to real-world noise and variability.
- The Symmetrical Four-Bar Cutting Mechanism: It provides a forceful, simultaneous cut from both sides, which is more reliable than a single moving blade. Its dynamic design leverages inertia to aid in the cutting process.
- The Integrated Retention Function: Combining the cutting action with an immediate collection function (the closing flaps) within the same mechanism streamlines the harvesting cycle, eliminating the need for a separate, complex gripper or suction device to hold the fruit after detachment.
Limitations and Future Work: While successful, the prototype reveals avenues for improvement. The failures point to the need for tighter manufacturing tolerances on the blades and potentially a blade sharpening study. The cutting force model could be refined to account for the dynamic change in stem resistance during the cut (rather than a constant force). The current end effector is tailored to a specific pepper variety; future work could involve creating adjustable parameters (e.g., sleeve diameter, stroke length) to adapt to a wider range of pepper types or similar crops. Finally, integration with a high-speed, robust machine vision system in a densely planted, unstructured greenhouse environment remains the next significant challenge for achieving a fully autonomous harvesting robot.
Conclusion
This work presents the complete journey of designing a specialized robotic end effector for harvesting fresh, upright-growing chili peppers. Beginning with a rigorous bio-mechanical characterization of the crop, a cutting-detachment end effector concept with passive positional tolerance was conceived. Detailed engineering followed, encompassing the design of a multi-functional mechanism integrating pose adjustment, fruit capture, stem cutting, and fruit retention. Kinematic and kineto-static mathematical models were derived and used to optimize the linkage dimensions, minimizing the required actuator torque. A physical prototype was built and tested, demonstrating a high stem-cutting success rate of 91.3% in controlled harvesting trials.
The significance of this end effector extends beyond the specific crop. It exemplifies a methodology that can be applied to the robotic harvesting of other delicate, stem-attached fruits and vegetables. The emphasis on tolerance, robust cutting action, and cycle integration provides a valuable template for end effector designers. As the global agricultural sector increasingly turns to robotics to address labor shortages and improve precision, the development of reliable, efficient, and crop-adapted end effectors like the one described here will be fundamental to unlocking the full potential of autonomous harvesting systems.