In recent years, the agricultural sector has faced increasing challenges due to labor shortages, particularly in the harvesting of premium teas such as Yinghong No. 9. Manual picking is not only time-consuming but also costly, which significantly impacts the economic viability of tea production. As a result, there is a pressing need for intelligent harvesting robots that can perform selective picking with high efficiency and accuracy. The end effector, as the critical component responsible for detaching the target buds from the stems, plays a pivotal role in determining the overall performance of such robots. In this work, we focus on the optimization and testing of an under-actuated end effector designed for tea harvesting, aiming to address issues of low success rates and bulky design observed in prior prototypes.
Our approach begins with a comprehensive analysis of the physical and mechanical properties of tea buds, which informs the design parameters. We then delve into the structural and operational principles of the end effector, followed by a detailed optimization process using constrained trust region algorithms. Finally, we conduct single-factor experiments to validate the improvements in performance. Throughout this article, we emphasize the importance of the end effector in robotic harvesting systems and seek to demonstrate how optimization can enhance its functionality.
The end effector under consideration is an under-actuated mechanism that integrates clamping, cutting, and collection functions into a compact design. Its working principle mimics manual picking actions, but with mechanical simplicity that allows for reliable operation. However, initial tests revealed that the end effector suffered from a relatively low success rate and excessive size, which could lead to increased inertia and potential damage during operation. Therefore, we undertook a systematic optimization of the link lengths to reduce the overall dimensions and improve the picking success rate.
To establish a foundation for the design, we first measured the physical and mechanical characteristics of Yinghong No. 9 tea buds. A sample of 100 buds (one bud with one leaf) was collected, and parameters such as length, width, height, and clampable length were recorded using vernier calipers. The results are summarized in Table 1.
| Parameter | Maximum | Minimum | Average | Standard Deviation |
|---|---|---|---|---|
| Height | 40.20 | 29.10 | 31.30 | 13.38 |
| Length | 35.50 | 10.20 | 22.50 | 22.70 |
| Width | 17.40 | 7.30 | 12.40 | 10.39 |
| Clampable Length | 11.90 | 8.10 | 10.50 | 4.19 |
In addition to physical dimensions, the mechanical strength of the stems was assessed using a tensile-compression testing machine. A custom-made blade with a 45-degree inclination angle was employed to shear the stems, and the required cutting force was recorded. The results indicated that the minimum cutting force needed to sever a stem was 2.8 N, the maximum was 4.7 N, and the average was 3.5 N. This data is crucial for ensuring that the end effector can generate sufficient cutting force during operation.
The end effector is essentially a planar five-bar linkage mechanism with two degrees of freedom, making it an under-actuated system. It consists of several key components: a collection drawer, limit blocks, a crank, a connecting rod, a stepped shaft, a flange plate, a servo arm, a servo motor, a stop lever, a lower jaw, a custom blade, a blade seat, a leak-proof fixed box, and a frame. The upper jaw is formed by the blade holder, custom blade, and blade seat. The working cycle of the end effector involves several states: initial opening, clamping and cutting, post-throwing, and resetting. During the clamping phase, the servo motor rotates counterclockwise, driving the crank and connecting rod to move the lower jaw towards the upper jaw, thereby shearing the tea stem. Subsequently, continued rotation causes the entire assembly to pivot, allowing the cut bud to be tossed into the collection drawer via a shaking motion. Finally, the servo motor rotates clockwise to reset the end effector to its initial position.
To optimize the end effector, we modeled it as a crank-rocker mechanism, focusing on the lengths of the crank (l1), connecting rod (l2), rocker (l3), and frame (l4). The additional link length (l5) is part of the lower jaw and affects the cutting geometry. The optimization goal was to minimize the total length of these links, thereby reducing the mass and size of the end effector while maintaining functionality. The objective function is expressed as:
$$ L = l_1 + l_2 + l_3 + l_4 $$
We established constraints based on the geometric relationships in three working states: initial, shearing, and post-throwing. In the initial state, the crank and connecting rod are collinear, and the angle between the rocker and the frame is given by α. In the shearing state, the crank and frame are collinear, and the angle between the rocker and the frame is β, with the transmission angle γ being critical for force transmission. The constraints are derived from cosine laws in the respective triangles.
For the initial state, we have:
$$ \cos(\alpha) = \frac{l_3^2 + l_4^2 – (l_1 + l_2)^2}{2 l_3 l_4} $$
where α = 90° + θ₁, and θ₁ is the angle between the lower jaw and the vertical line, set to 40° for symmetry.
For the shearing state:
$$ \cos(\beta) = \frac{(l_1 + l_4)^2 + l_3^2 – l_2^2}{2 l_3 (l_1 + l_4)} $$
where β = 90° – θ₂, and θ₂ is also set to 40°.
The transmission angle γ must satisfy 40° ≤ γ ≤ 140° to ensure efficient force transmission, leading to the inequality constraint:
$$ \cos(40^\circ) \leq \frac{l_3^2 + l_2^2 – (l_1 + l_4)^2}{2 l_2 l_3} \leq \cos(140^\circ) $$
Additionally, we imposed constraints based on the assemblability of the crank-rocker mechanism and practical design limits:
$$ l_1 \geq 18 \text{ mm} $$
$$ l_1 \leq l_4 – 10 $$
$$ l_1 \leq l_2 $$
$$ l_1 \leq l_3 $$
$$ l_1 + \max(l_1, l_2, l_3, l_4) \leq \frac{l_1 + l_2 + l_3 + l_4}{2} $$
To solve this constrained optimization problem, we employed a trust region algorithm implemented in Python using the SciPy library. The algorithm iteratively adjusts the link lengths to minimize the objective function while satisfying all constraints. The convergence was rapid, as shown in the iterative curve, and the optimized lengths were obtained as summarized in Table 2.
| Link | Before Optimization | After Optimization |
|---|---|---|
| Crank (l₁) | 18.0 | 18.0 |
| Connecting Rod (l₂) | 50.0 | 35.3 |
| Rocker (l₃) | 24.0 | 30.7 |
| Frame (l₄) | 58.0 | 28.0 |
| Total Length | 150.0 | 112.0 |
The optimization resulted in a 25.33% reduction in the total link length, significantly compacting the end effector. To determine the length of the lower jaw link (l₅), we considered the geometry of the opening formed by the upper and lower jaws. The width (w) and height (h) of the opening must accommodate the maximum dimensions of the tea buds. Based on the measured data, w = 35.5 mm and h = 40.2 mm. Using trigonometric relationships, we derived the minimum required length for l₅:
$$ \frac{l_5}{\sin(90^\circ)} \geq \frac{35.5}{2 \sin(40^\circ)} $$
$$ \frac{l_5}{\sin(90^\circ)} \geq \frac{40.2}{\sin(50^\circ)} $$
Solving these inequalities yields l₅ ≥ 52.4 mm. To avoid interference with the tea canopy, we set l₅ = 65 mm.
With the optimized dimensions, we proceeded to analyze the cutting force generated by the end effector. The servo motor provides a torque of M = 400 N·mm. During the shearing state, the force transmission through the linkage can be analyzed using static equilibrium. The force F₁ exerted by the servo on the crank at point B is given by:
$$ F_1 = \frac{M}{l_1} $$
This force is transmitted through the connecting rod to the rocker, resulting in a cutting force F₄ at the blade. The relationship can be derived as:
$$ F_4 = \frac{M l_3 \cos(\phi_1) \cos(\phi_2)}{l_1 l_5} $$
where φ₁ and φ₂ are angles related to the geometry of the linkage. Using the optimized lengths and the calculated angles φ₁ = 48.08° and φ₂ = 1.92°, we compute:
$$ F_4 = \frac{400 \times 30.7 \times \cos(48.08^\circ) \times \cos(1.92^\circ)}{18 \times 65} \approx 7.0 \text{ N} $$
This cutting force exceeds the maximum required force of 4.7 N measured for tea stems, confirming that the optimized end effector is capable of performing the shearing action effectively.
To evaluate the performance of the optimized end effector, we conducted single-factor experiments focusing on three key variables: clamping position, growth angle, and the number of leaves per bud. The clamping position refers to the distance H between the opening of the end effector and the bud-leaf junction. The growth angle ψ is the angle between the bud and the horizontal plane. We compared the performance of the end effector before and after optimization in terms of success rate and average picking time. Each test was repeated 20 times for statistical reliability.
The results for clamping position are presented in Table 3. The optimized end effector showed significant improvements in success rate across all clamping positions, with the best performance observed at 4–8 mm. The average picking time also decreased, indicating enhanced efficiency.
| Clamping Position (mm) | Optimized Success Rate (%) | Optimized Time (s) | Original Success Rate (%) | Original Time (s) |
|---|---|---|---|---|
| 0–4 | 90.00 | 0.65 | 64.17 | 0.71 |
| 4–8 | 92.50 | 0.63 | 78.33 | 0.73 |
| 8–12 | 90.80 | 0.66 | 73.33 | 0.70 |
For growth angle, the results are summarized in Table 4. The optimized end effector consistently outperformed the original design, particularly at growth angles between 70° and 90°, where the success rate reached 91.67%. The lower success rates at smaller angles are attributed to the difficulty in accessing the clamping zone, but such angles are less common in practice.
| Growth Angle (°) | Optimized Success Rate (%) | Optimized Time (s) | Original Success Rate (%) | Original Time (s) |
|---|---|---|---|---|
| 30–50 | 70.00 | 0.66 | 58.33 | 0.72 |
| 50–70 | 87.50 | 0.63 | 74.17 | 0.71 |
| 70–90 | 91.67 | 0.65 | 79.17 | 0.70 |
Finally, we tested the end effector on different types of tea buds: single bud, one bud with one leaf, and one bud with two leaves. The results are shown in Table 5. The optimized end effector achieved high success rates for single buds and one-bud-one-leaf samples, but performance dropped significantly for one-bud-two-leaf samples due to clogging issues. This indicates that the end effector is best suited for selective harvesting of premium tea types.
| Bud Type | Optimized Success Rate (%) | Optimized Time (s) | Original Success Rate (%) | Original Time (s) |
|---|---|---|---|---|
| Single Bud | 95.00 | 0.63 | 80.00 | 0.72 |
| One Bud One Leaf | 93.30 | 0.63 | 76.67 | 0.71 |
| One Bud Two Leaves | 39.17 | 0.65 | 39.17 | 0.75 |
The optimization of the end effector has led to substantial improvements in both design and performance. By reducing the total link length by 25.33%, we have created a more compact and lightweight end effector that minimizes inertial forces and potential impacts during operation. The trust region algorithm proved effective in solving the constrained optimization problem, yielding dimensions that satisfy all geometric and functional requirements. Furthermore, the static analysis confirmed that the cutting force generated by the optimized end effector is sufficient to shear tea stems reliably.
The experimental results demonstrate that the optimized end effector achieves success rates of up to 95% for single buds and 93.3% for one-bud-one-leaf samples under optimal conditions (clamping position of 4–8 mm and growth angle of 70–90°). The average picking time of approximately 0.64 s is also commendable, indicating that the end effector can operate efficiently in field conditions. However, limitations remain when dealing with larger buds or multiple leaves, suggesting areas for future refinement.
In conclusion, this work presents a comprehensive approach to optimizing an under-actuated end effector for tea harvesting. The integration of physical property analysis, kinematic modeling, constrained optimization, and experimental validation has resulted in a robust design that enhances picking success rates while reducing size. The end effector is a critical component in the development of autonomous tea harvesting robots, and its improved performance contributes to the broader goal of agricultural automation. Future research could explore adaptive control strategies or further design modifications to handle a wider variety of tea bud configurations, ultimately advancing the efficiency and intelligence of robotic harvesting systems.
Throughout this study, the importance of the end effector in robotic harvesting has been emphasized repeatedly. The end effector’s design directly influences the success rate, speed, and reliability of the picking process. By optimizing the end effector, we have taken a significant step toward practical and efficient mechanized harvesting of premium teas. The methodologies and results presented here may also inspire similar optimizations for other crop harvesting end effectors, promoting innovation in agricultural robotics.