In the global tea industry, the harvesting of premium teas, such as famous teas requiring selective plucking of tender shoots (e.g., single buds or one bud with one leaf), remains heavily reliant on manual labor. This dependence poses significant challenges due to seasonal labor shortages, high costs, and inefficiencies, ultimately limiting production scalability. To address these issues, robotic harvesting systems have emerged as a promising solution. However, many existing end effectors for tea picking exhibit drawbacks, including complex structures, low efficiency in single-workstation operations, or poor harvesting quality (e.g., oxidized cut surfaces that turn red). Therefore, there is a pressing need for an end effector that is structurally simple, efficient, and capable of high-quality harvesting and collection. In this work, we present the design and experimental optimization of a novel rotating two-workstation integrated end effector. This end effector combines picking and collection functions into a single, compact unit, aiming to enhance operational efficiency and preserve tea quality by mimicking manual plucking actions.
The core innovation of our end effector lies in its rotating two-workstation mechanism, which eliminates the need for separate, complex collection devices. The end effector consists of several key components: a motor, a temporary storage box, a connecting piece, two soft picking fingers, an installation plate, a servo motor, a servo arm, and transmission links. The motor is fixed to the connecting piece, with its output shaft attached to the installation plate. The temporary storage box is mounted on the opposite end of the connecting piece. Two soft picking fingers are symmetrically installed on both ends of the installation plate, aligned with its rotational center. The servo motor is fixed on the installation plate, with its output shaft connected to a servo arm that hinges two transmission links. These links are, in turn, hinged to bolts on the respective picking fingers. The two workstations refer to the picking station and the collection station. Through a forward-and-reverse rotation cycle, the two picking fingers alternately occupy these stations, enabling continuous picking and collection.
During operation, a vision system identifies target tea shoots. The robotic arm positions the end effector above the target. Initially, one picking finger is at the lower picking station. The servo arm rotates counterclockwise by a set angle, causing the transmission link to pull the lower picking finger closed via a slider-crank mechanism, thereby gripping the tender shoot’s stem. Meanwhile, the upper picking finger remains open. The installation plate then rotates 180 degrees, driven by the motor. This rotation moves the lower finger (now holding the plucked shoot) to the upper collection station and the upper finger to the lower picking station. At the collection station, the servo arm adjusts to open the finger, releasing the shoot into the temporary storage box via gravity. Simultaneously, at the picking station, the other finger closes to grip a new shoot. This cycle repeats, allowing one end effector to pick while the other collects, significantly improving efficiency compared to single-workstation designs.
The picking finger drive mechanism is critical for precise gripping. It comprises a sliding mechanism and a gripping mechanism. The sliding mechanism includes a support block, slider, and fixed rail. The support block, attached to the slider, moves linearly along the fixed rail under force from the transmission link. The gripping mechanism consists of a silicone installation plate, a finger mounting frame, links, and silicone pads. When the support block moves upward, the links cause the finger mounting frame to pivot, closing the silicone pads to clamp the stem. After rotation, the tensile force breaks the shoot. The clamped shoot is held until rotation brings it over the storage box, where the finger opens for collection. The silicone pads are designed with micro-protrusions to increase hysteresis friction, enhancing grip and reducing slippage during the plucking action.
To inform the end effector design, we measured physical parameters of tender tea shoots (e.g., one bud with two leaves) from a tea plantation. Key parameters include bud-leaf horizontal distance, stem diameter, bud length, leaf length, and first-segment stem length. Statistical data are summarized in Table 1. These measurements guided dimensions such as the storage box opening and silicone pad size. The storage box features a 45° chamfered edge to facilitate shoot entry, with an opening diameter of 150 mm and a depth of 250 mm, capable of holding at least 50 shoots. The silicone pads are sized at 8 mm × 8 mm in length and width, with thickness to be optimized, to grip the first stem segment without damaging leaves or buds.
| Parameter | Maximum | Minimum | Average | Standard Deviation |
|---|---|---|---|---|
| Bud-Leaf Horizontal Distance (mm) | 32.0 | 8.5 | 15.9 | 7.62 |
| Stem Diameter (mm) | 2.1 | 1.1 | 1.7 | 0.33 |
| Bud Length (mm) | 14.7 | 7.5 | 10.6 | 2.28 |
| Leaf Length (mm) | 26.2 | 16.5 | 20.8 | 3.06 |
| First-Segment Stem Length (mm) | 15.8 | 10.2 | 11.6 | 1.1 |
| Stem Elastic Modulus (MPa) | 3.2 | 1.9 | 2.6 | 0.5 |
| Stem Density (kg/m³) | 813.0 | 785.3 | 805.0 | 11.6 |
Optimizing the finger drive mechanism parameters was essential for ensuring proper opening and closing at both workstations. We modeled the mechanism as a symmetric slider-crank system. Let point O be the origin at the servo motor shaft, with coordinates as shown. The servo arm OA has length \(l_1\) and rotation angle \(\Delta\theta_1\). Other link lengths include \(l_2\) (link AB), \(l_3\) (distance from slider center to hinge B), \(l_4\) (distance from slider center to hinge D), \(l_5\) (link DE), \(l_6\) (segment FE of finger mount), and \(l_7\) (segment EK). Angles \(\theta_2\), \(\theta_3\), and \(\theta_4\) define orientations. Coordinates of points A, B, D, E, F are given by:
$$ x_A = 0, \quad y_A = l_1 \cos(\Delta\theta) $$
$$ x_B = 0, \quad y_B = y_A + l_2 \cos(\theta_2) $$
$$ x_D = l_4, \quad y_D = y_B + l_3 $$
$$ x_E = x_D + l_5 \cos(\theta_3), \quad y_E = y_D + l_5 \sin(\theta_3) $$
$$ x_F = x_E + l_6 \cos(\theta_4), \quad y_F = y_E + l_6 \sin(\theta_4) $$
The constraint equations from link lengths are:
$$ (x_E – x_D)^2 + (y_E – y_D)^2 = l_5^2 $$
$$ (x_F – x_E)^2 + (y_E – y_F)^2 = l_6^2 $$
We aimed to maximize the transmission angle between link DE and segment FE when the finger is closed, subject to constraints on motion range and link lengths. Using a genetic algorithm in MATLAB, we optimized the vector \(\mathbf{l} = (l_1, l_2, l_3, l_4, l_5, l_6, l_7)\). The objective function was \(\max(\theta_4 – \theta_3)\) under constraints: \(90^\circ \geq \theta_4 – \theta_3 \geq 60^\circ\) (transmission angle limits), \(0.03 \geq x_F – (l_6 + l_7) \cos \theta_4 \geq 0.003\) (x-displacement of point K for adequate opening), \(0.01 – (l_6 + l_7)(\sin \theta_4 – \sin \theta_4′) \geq 0\) (y-displacement limit to reduce interference), and others for linkage validity. Optimization yielded rounded lengths: \(l_1 = 25\) mm, \(l_2 = 55\) mm, \(l_3 = 10\) mm, \(l_4 = 33\) mm, \(l_5 = 31\) mm, \(l_6 = 50\) mm, \(l_7 = 100\) mm. This design ensured a transmission angle of \(60^\circ\) and point K trajectory with 28 mm horizontal and 5 mm vertical movement, suitable for gripping.
Selection of silicone material for the picking fingers was crucial for quality. Unlike blade cutting, which can cause oxidation and reddening at cut surfaces, silicone gripping mimics manual “pull-breaking,” preserving shoot quality. We evaluated silicone hardness options (30 HS, 40 HS, 50 HS, 60 HS) via finite element analysis (FEA) in Abaqus to assess stress on tea stems during clamping. The stem was modeled with density \(\rho = 805 \, \text{kg/m}^3\) and elastic modulus \(E = 2.6 \, \text{MPa}\). Silicone material constants were input based on literature. Results showed that 30 HS silicone induced minimal stress and no visible deformation on stems, whereas harder silicones caused higher stress and potential damage. Experimental validation involved picking 30 shoots with each hardness; only 30 HS silicone left no oxidative marks after 20 minutes. Thus, 30 HS silicone was chosen for the end effector fingers. Additionally, micro-protrusions on the silicone surface enhance hysteresis friction \(F_h\), as per Moore’s friction model for rubber-like materials:
$$ F = F_a + F_h $$
where \(F\) is total friction, \(F_a\) is adhesive friction, and \(F_h\) is hysteresis friction due to surface deformation. This design improves grip success during pull-breaking.
To optimize the end effector’s harvesting performance, we identified key factors influencing the harvesting success rate \(\gamma\), defined as the percentage of successfully picked and collected shoots relative to total attempts. From preliminary tests, the main factors are: installation plate angular velocity \(\omega\) (affecting tangential pull force), servo arm rotation angle \(\Delta\theta\) (affecting clamping force), and silicone thickness \(d\) (affecting grip and stress distribution). The tangential force during rotation is given by:
$$ F_t = \frac{m \omega^2}{129600} r $$
where \(m\) is the rotating mass (kg), \(\omega\) is in degrees per second, and \(r\) is the rotation radius (m). Higher \(\omega\) increases pull force but may cause instability; we set range \(180 \leq \omega \leq 300\) (°)/s. \(\Delta\theta\) affects clamping force; range \(30^\circ \leq \Delta\theta \leq 50^\circ\). Silicone thickness range is \(3 \leq d \leq 7\) mm to balance deformation and grip.
We conducted a Box-Behnken experimental design with three factors at three levels (-1, 0, 1) to study their effects on \(\gamma\). The factors and levels are coded in Table 2. Experiments were performed in a tea field using the end effector on premium tea shoots (e.g., one bud with one leaf). For each trial, 80 pick attempts were made, and success rate was calculated as \(\gamma = (p_s / p_t) \times 100\%\), where \(p_s\) is number of shoots successfully collected in the storage box, and \(p_t\) is total attempts. Results are shown in Table 3.
| Code | Angular Velocity \(\omega\) (°/s) | Servo Arm Angle \(\Delta\theta\) (°) | Silicone Thickness \(d\) (mm) |
|---|---|---|---|
| -1 | 180 | 30 | 3 |
| 0 | 240 | 40 | 5 |
| 1 | 300 | 50 | 7 |
| Run | \(\omega\) (°/s) | \(\Delta\theta\) (°) | \(d\) (mm) | \(p_s\) | \(p_t\) | \(\gamma\) (%) |
|---|---|---|---|---|---|---|
| 1 | 180 | 30 | 5 | 45 | 80 | 56.25 |
| 2 | 240 | 40 | 5 | 64 | 80 | 80.00 |
| 3 | 240 | 50 | 7 | 36 | 80 | 45.00 |
| 4 | 240 | 40 | 5 | 65 | 80 | 81.25 |
| 5 | 180 | 40 | 3 | 34 | 80 | 42.50 |
| 6 | 180 | 40 | 7 | 33 | 80 | 41.25 |
| 7 | 240 | 30 | 7 | 57 | 80 | 71.25 |
| 8 | 240 | 50 | 3 | 63 | 80 | 78.75 |
| 9 | 240 | 30 | 3 | 26 | 80 | 32.50 |
| 10 | 300 | 50 | 5 | 70 | 80 | 87.50 |
| 11 | 180 | 50 | 5 | 51 | 80 | 63.75 |
| 12 | 300 | 40 | 3 | 48 | 80 | 60.00 |
| 13 | 300 | 30 | 5 | 54 | 80 | 67.50 |
| 14 | 240 | 40 | 5 | 64 | 80 | 80.00 |
| 15 | 300 | 40 | 7 | 45 | 80 | 56.25 |
| 16 | 240 | 40 | 5 | 68 | 80 | 85.00 |
| 17 | 240 | 40 | 5 | 65 | 80 | 81.25 |
Using Design-Expert software, we performed analysis of variance (ANOVA) on the data to develop a quadratic regression model for \(\gamma\). The model was highly significant (p < 0.0001) with a coefficient of determination \(R^2 = 0.9927\), indicating excellent fit. The regression equation in terms of coded factors (A for \(\omega\), B for \(\Delta\theta\), C for \(d\)) is:
$$ \gamma = 81.40 + 8.50A + 5.75B – 0.25C + 2.75AB – 0.75AC – 17.25BC – 10.33A^2 – 2.32B^2 – 21.33C^2 $$
ANOVA revealed that factors A (\(\omega\)), B (\(\Delta\theta\)), interaction BC, and quadratic terms A² and C² had extremely significant effects (p < 0.01), while AB and B² were significant (p < 0.05). The order of factor significance was: installation plate angular velocity (\(\omega\)), servo arm angle (\(\Delta\theta\)), then silicone thickness (\(d\)). This highlights that the rotational speed of the end effector is most critical for harvesting success, followed by clamping force and material thickness.
Interaction effects were analyzed through response surface plots. For instance, at fixed silicone thickness (\(d = 5\) mm), increasing both \(\omega\) and \(\Delta\theta\) enhanced \(\gamma\), as higher angular velocity provides greater pull force and larger servo arm angle improves clamping grip. Conversely, at high silicone thickness (e.g., \(d = 7\) mm), increasing \(\Delta\theta\) could reduce \(\gamma\) due to excessive silicone deformation, reducing effective grip. This underscores the importance of optimizing these parameters jointly for the end effector.
To maximize harvesting success rate, we used the optimization function in Design-Expert with the goal of maximizing \(\gamma\). The optimal parameters were: angular velocity \(\omega = 265.329\) (°)/s, servo arm angle \(\Delta\theta = 40^\circ\), and silicone thickness \(d = 4.986\) mm, with a predicted success rate of \(83.194\%\). Considering manufacturing precision, we rounded these to: \(\omega = 265\) (°)/s, \(\Delta\theta = 40^\circ\), \(d = 5\) mm. Verification experiments were conducted with three replicates, yielding an average success rate of \(85\%\), which is within 5% relative error of the prediction, confirming model reliability. The end effector achieved a picking time of approximately 1.2 seconds per shoot, demonstrating high efficiency.
We compared our rotating two-workstation end effector with a previously developed single-workstation cutting end effector. Both were tested on similar tea shoots. Results, summarized in Table 4, show that our end effector improved efficiency by 85% (0.833 shoots/second vs. 0.451 shoots/second) while producing no reddening at break surfaces, unlike the cutter which caused oxidation. Moreover, the integrated collection eliminated need for separate suction devices, simplifying the system. This confirms that our end effector design effectively balances speed, quality, and simplicity.
| End Effector Type | Efficiency (shoots/s) | Break Surface Quality | Collection Method |
|---|---|---|---|
| Single-Workstation Cutting End Effector | 0.451 | Reddened (oxidized) | Separate suction device |
| Rotating Two-Workstation Integrated End Effector | 0.833 | No reddening (like manual) | Integrated gravity-based |
In conclusion, this work presents a novel rotating two-workstation integrated end effector for premium tea harvesting. The end effector design combines picking and collection into a single mechanism, reducing complexity and improving operational efficiency. Through kinematic optimization, material selection, and parameter tuning via Box-Behnken experiments, we achieved a robust design with high harvesting success. Key advantages include: (1) Enhanced efficiency via simultaneous picking and collection, reducing cycle time; (2) Preservation of tea quality through silicone-based pull-breaking, avoiding cut surface oxidation; (3) Simplified structure without external collection systems. The optimized parameters (\(\omega = 265\) (°)/s, \(\Delta\theta = 40^\circ\), \(d = 5\) mm) yielded an 85% success rate in validation tests. This end effector demonstrates potential for practical deployment in robotic tea harvesting, addressing labor shortages and quality concerns. Future work may focus on scalability, adaptability to varied tea cultivars, and integration with advanced vision systems for autonomous operation in field conditions.