The automation of fruit harvesting represents a critical frontier in modern precision agriculture, addressing challenges such as labor scarcity, rising operational costs, and the demand for consistent, high-quality produce. Among various fruits, the navel orange, with its significant economic value and specific growth characteristics—growing in clusters and requiring careful handling to prevent bruising—presents a unique set of challenges for robotic systems. The core component tasked with the physical interaction with the fruit is the end effector. Its performance directly dictates the success rate, speed, and, most importantly, the preservation of fruit quality during the harvesting cycle. This article details the comprehensive design, analytical modeling, control system integration, and experimental validation of a novel end effector developed specifically for the robotic harvesting of navel oranges.
The primary objective was to create an end effector capable of non-destructive picking within the typical equatorial diameter range of 50 mm to 100 mm. The design philosophy centers on a hybrid pneumatic-electric approach, leveraging the speed of pneumatic actuators and the precision of electric drives. The mechanism is fundamentally underactuated, meaning it possesses more degrees of freedom than control inputs. This simplifies the mechanical design and control complexity while maintaining adaptable and compliant grasping capabilities. The core sequence of operations for a single fruit involves three distinct phases executed by dedicated subsystems: separation from the cluster, stable grasping, and stem detachment.
The physical embodiment of the end effector consists of three integrated mechanisms mounted on a central frame that interfaces with a robotic manipulator. The Adsorption Mechanism is responsible for the initial fruit isolation. It comprises a pneumatic cylinder that extends a vacuum suction cup to make contact with the target orange. Once a seal is established and vacuum is applied, the cylinder retracts, pulling the fruit away from its neighboring oranges in the cluster. This pre-separation is crucial to prevent mechanical interference and potential damage during the subsequent grasping phase. The Grasping Mechanism is the heart of the end effector. It features two symmetrically arranged V-shaped fingers. Each finger has two phalanges with an internal angle of 140°, designed to create four potential contact points with the fruit’s equator. The fingers are driven by a single stepper motor via a ball screw assembly that converts rotary motion into linear translation of a nut. This translation is linked to the fingers through a slider-crank mechanism, causing them to close synchronously. The critical component for non-destructive operation is the array of thin-film resistive force sensors embedded on the inner surface of the finger phalanges, backed by a silicone rubber pad to increase friction and distribute contact pressure. The Rotary Cutting Mechanism severs the fruit’s stem. It consists of a high-speed DC motor driving a serrated circular blade, mounted on a second pneumatic cylinder. After the fruit is securely grasped, this cylinder advances the spinning blade to cleanly cut the stem, completing the detachment.
The design process began with establishing a precise geometrical and biomechanical model of the navel orange. Based on statistical measurements, the fruit was modeled as a standard prolate spheroid. The finger workspace was then analyzed parametrically. To ensure four-point contact across the entire target diameter range, the finger joint lengths and angles were optimized. The key parameters are summarized below:
| Parameter | Symbol | Value (mm or °) |
|---|---|---|
| Distal Phalanx Length | \(l_1\) | 35 |
| Proximal Phalanx Length | \(l_2\) | 35 |
| Internal Phalanx Angle | \(\theta\) | 140° |
| Finger Base Length (Crank) | \(R\) | 31 |
| Connecting Rod Length | \(L\) | 60 |
| Minimum Grasping Diameter | \(D_{min}\) | 50 |
| Maximum Grasping Diameter | \(D_{max}\) | 100 |
The force analysis for stable, non-destructive grasping involved two key aspects: the minimum force to prevent slippage and the maximum force to avoid bruising. For an orange of mass \(m\), the condition for static equilibrium against gravity is given by the sum of vertical frictional forces at the contact points equaling the weight:
$$ \sum_{i=1}^{4} f_i = mg $$
where \(f_i = \mu N_i\), \(\mu\) is the coefficient of static friction, and \(N_i\) is the normal force at contact point \(i\). Assuming symmetric contact, \(N_i = N\). Thus, the minimum required normal force is:
$$ N_{min} = \frac{m_{max}g}{4\mu} $$
where \(m_{max}\) is the mass of the largest target fruit (approx. 0.4 kg including a safety factor). The coefficient of friction \(\mu\) was experimentally determined by testing citrus rind against various compliant materials. Silicone rubber was selected for its superior performance.
| Compliant Material | Avg. Static Friction Coefficient (\(\mu\)) |
|---|---|
| Rubber | 0.806 |
| Polyethylene Foam | 0.500 |
| Silicone Rubber | 1.298 |
Using \(\mu = 1.298\) and \(g = 9.81 m/s^2\), \(N_{min} \approx 0.76 N\). Compression tests on fruit samples were conducted to establish the bioyield point and the elastic limit. To guarantee non-destructive operation, the maximum allowable deformation was set at 5 mm, corresponding to a maximum normal force \(N_{max} \approx 25.6 N\). This defines the safe operational range for the contact force: \(0.76 N \leq N \leq 25.6 N\).
A detailed static force analysis of the underactuated linkage was performed to size the driving motor. The system was modeled as a slider-crank mechanism. The force \(F\) required at the ball screw nut to produce a total normal force \(2N\) (from two fingers) at the fruit contact points is a function of the crank angle \(\alpha\):
$$ F = \frac{2N\left(l_2 \sin\frac{\theta}{2} + l_3\right)\left( \lambda^2 \sin^2\alpha – 1 \right)} {R\left(2\lambda \sin\alpha – \sin 2\alpha – \lambda^2 \sin^2\alpha \right)} $$
where \(l_3\) is the finger root length, \(R\) is the crank length, and \(\lambda = R/L\) is the linkage ratio. Calculating for the worst-case scenario (highest \(N\) and crank angle within the working range) yields the required axial force on the ball screw. The necessary motor torque \(T\) is then:
$$ T = \frac{F_a \cdot l}{2\pi\eta} $$
where \(F_a\) is the axial force on the screw, \(l\) is the screw lead (4 mm), and \(\eta\) is the screw efficiency (0.94). A stepper motor with a holding torque of 0.52 N·m was selected, providing a sufficient safety margin over the calculated requirement of approximately 0.31 N·m.
The control system was architected for high integration with a robotic manipulator controller. An embedded motion controller served as the main processor, handling both the trajectory planning for the manipulator and the sequential logic for the end effector. This integrated approach simplifies system architecture and improves synchronization. The I/O signals from the controller managed pneumatic solenoid valves for the cylinders and vacuum generator, a pulse generator for the stepper motor driver, and a relay for the DC cutting motor. Crucially, analog input channels read the voltage signals from the four force sensors. The voltage \(V_o\) from a sensor under force \(N\) follows a characteristic curve approximated by:
$$ V_o = V_{cc} – \frac{C}{N^{k}} $$
where \(V_{cc}\) is the supply voltage (5V), and \(C\) and \(k\) are constants determined through sensor calibration. This feedback is the cornerstone of the non-destructive grasp. The control algorithm closes the fingers until the average sensor reading exceeds a voltage threshold corresponding to a force safely above \(N_{min}\) but well below \(N_{max}\), then stops the motor and deactivates the vacuum. This constitutes a simple yet effective force-servo loop.
A full prototype end effector was constructed and mounted on a custom robotic arm for field testing. The primary variable investigated was the speed of the grasping motion, dictated by the stepper motor rotational speed \(n\). Performance was evaluated based on three metrics: average single-fruit picking cycle time \(T_{cycle}\), harvesting success rate \(S_r\), and fruit damage rate \(D_r\). A total of 105 picking trials were conducted on mature navel oranges across the target size spectrum.
| Stepper Motor Speed, \(n\) (rpm) | Avg. Cycle Time, \(T_{cycle}\) (s) | Success Rate, \(S_r\) (%) | Damage Rate, \(D_r\) (%) |
|---|---|---|---|
| 200 | 1.86 | 91.4 | 0 |
| 250 | 1.76 | 94.3 | 0 |
| 300 | 1.64 | 91.4 | 0 |
The results indicate that a speed of 250 rpm offered the optimal balance, maximizing the success rate while maintaining a fast cycle time. The slight drop in success rate at 300 rpm was attributed to increased inertial effects causing occasional misalignment during the initial contact phase. Importantly, the damage rate was zero across all tests, validating the effectiveness of the force-feedback grasping strategy. The main cause of failure was related to fruit stem geometry (exceptionally short stems), not a failure of the grasping or cutting mechanisms themselves.
In conclusion, this work presents the successful development of a specialized end effector for navel orange harvesting. The dual V-type underactuated finger design, complemented by a pre-separation adsorption system and a rotary cutter, demonstrates a viable and efficient approach to automated citrus picking. Key achievements include a rapid average picking cycle of under 1.8 seconds, a high success rate exceeding 94%, and a 0% damage rate in controlled tests, all accomplished through a mechanically simple and robust design. The use of an integrated embedded controller for both manipulator and end effector points towards a scalable and practical system architecture for agricultural robotics. Future work will focus on enhancing the end effector’s adaptability to a wider variety of stem attachment angles and lengths, improving the vision system’s robustness in cluttered environments, and conducting large-scale endurance trials in commercial orchard settings to fully validate economic viability.