In modern agricultural production, achieving efficient and non-destructive mechanized harvesting remains a significant challenge, particularly for fragile fruits like citrus. The primary goal of our research was to design a novel harvesting solution that overcomes the inherent limitations of traditional rigid robotic grippers. Conventional end effectors, typically constructed from rigid materials, often suffer from poor compliance, limited adaptability to complex orchard environments, and a high risk of bruising or damaging the delicate fruit skin during the grasping process. To address these critical issues, we turned to the field of soft robotics. Soft robotic manipulators, inspired by biological systems like octopus tentacles, offer inherent compliance, adaptability, and safe interaction with objects. Our work focused on the development and comprehensive evaluation of a pneumatically actuated soft end effector specifically designed for the non-destructive harvesting of citrus fruit. This end effector integrates a soft gripper for gentle holding and a separate cutting mechanism for severing the peduncle.
The overall architecture of our harvesting end effector is a synergistic combination of a soft manipulator and a cutting assembly. The soft manipulator forms the core grasping unit and consists of three soft fingers, a customizable palm, and connection components. A key design feature for enhancing versatility is the detachable interface between the soft manipulator and the cutting mechanism. The palm is designed to be adjustable, allowing for modifications in the relative positioning of the fingers to accommodate fruits of varying sizes. The cutting assembly, driven by a stepper motor via a lead screw, is responsible for cleanly severing the fruit stem. The operational workflow of the end effector is sequential: first, the system positions the end effector adjacent to the target fruit. Then, the pneumatic system is activated, inflating the soft fingers to gently envelop and clamp the citrus. Once a pre-set grasping force is achieved (monitored by sensors), the cutting mechanism is engaged to detach the fruit. This integrated approach ensures a secure hold without damage before the cutting action.
The design of the soft finger is paramount to the performance of the end effector. Among various soft actuator designs, we selected a multi-chamber pneumatic configuration due to its ability to achieve larger bending angles compared to fiber-reinforced designs, which is crucial for enveloping fruits of different shapes and sizes. The internal chamber structure is a combination of a rectangular and a semi-circular segment. To enforce directional bending upon pressurization and prevent undesirable extension at the base, a layer of paper is embedded along the neutral axis of the finger’s bottom layer, acting as a strain-limiting layer. The geometric parameters of the finger were initially defined based on the physical characteristics of the target fruit, Gannan navel oranges. After measuring a sample population, key dimensions like total finger length were set. Several structural parameters were identified as variables for subsequent optimization to maximize performance.
| Parameter Meaning | Symbol | Value / Range (mm) |
|---|---|---|
| Total Finger Width | w | 24 |
| Total Finger Length | L | 95 |
| Air Chamber Total Length | l | 83 |
| Single Chamber Width | L1 | 20 |
| Chamber Wall Thickness | L5 | [1.75, 2.25] |
| Strain-Limiting Layer Thickness | s | [2.0, 4.0] |
| Airbag Height (Rectangular Part) | h2 | [4.0, 8.0] |
| Air Chamber Thickness | L3 | [2.5, 3.5] |
To predict and understand the bending behavior of the soft finger, we established a mechanical model based on hyperelastic material theory. The silicone material (Dragon Skin 30) used for fabricating the fingers is characterized as incompressible and hyperelastic. The Yeoh constitutive model, which describes the strain-energy density function (W), was employed due to its effectiveness in modeling the stress-strain behavior of such elastomers over moderate strains. The model is expressed as:
$$W = C_{10}(I_1 – 3) + C_{20}(I_1 – 3)^2$$
where $C_{10}$ and $C_{20}$ are material constants (0.11 MPa and 0.02 MPa, respectively), and $I_1$ is the first invariant of the Cauchy-Green deformation tensor. Under the assumption of constant curvature bending and plane strain conditions, the relationship between the internal air pressure ($P$) and the resulting bending angle ($\phi$) was derived through moment balance analysis. The theoretical model incorporates a correction factor ($K$) to account for simplifications in the stress distribution assumption:
$$P = K A D [(1+\mu\phi) – (1+\mu\phi)^{-3}] + K B D [(1+\mu\phi)^3 – (1+\mu\phi)^{-5}]$$
where $A=2C_{10}-8C_{20}$, $B=4C_{20}$, $\mu = s/l$, and $D$ is a geometric parameter grouping. This model provides a foundational understanding of the soft finger’s actuation mechanics for the end effector.
To determine the optimal structural parameters for the soft finger within the defined ranges, we employed a simulation-driven optimization approach using Finite Element Analysis (FEA) in ANSYS. A central composite design (CCD) within the Response Surface Methodology (RSM) framework was conducted. The airbag height ($h2$), limiting layer thickness ($s$), and air chamber thickness ($L3$) were chosen as the independent variables, with the maximum bending angle under a reference pressure as the response. The FEA simulations provided the data for constructing a predictive regression model. Analysis of variance (ANOVA) confirmed the high significance of the model.
| Source | Sum of Squares | F-value | p-value | Significance |
|---|---|---|---|---|
| Model | 0.6025 | 23.86 | 0.0002 | ** |
| A-Airbag Height | 0.2211 | 78.79 | <0.0001 | ** |
| B-Limiting Layer Thickness | 0.2145 | 76.44 | <0.0001 | ** |
| C-Air Chamber Thickness | 0.0420 | 14.98 | 0.0061 | ** |
| BC | 0.0506 | 18.04 | 0.0038 | ** |
The optimization routine yielded an optimal parameter set: $h2 = 6.93$ mm, $s = 3.65$ mm, $L3 = 2.88$ mm. For practical fabrication, these were rounded to $h2 = 7$ mm, $s = 3.5$ mm, $L3 = 3$ mm. Using this optimized geometry, a series of FEA simulations were run at different pressures. The pressure-angle data from these high-fidelity simulations was then used to calibrate the theoretical mechanical model. By fitting the simulation data with MATLAB and comparing it to the theoretical equation, the correction factor $K$ was determined to be 0.7. The final, calibrated relationship for the soft finger in the end effector is:
$$P = 0.0193D[(1+0.042\phi) – (1+0.042\phi)^{-3}] + 0.0258D[(1+0.042\phi)^3 – (1+0.042\phi)^{-5}]$$
Following the design and simulation phase, physical soft fingers were fabricated using a molding process with Dragon Skin 30 silicone rubber. The paper limiting layer was embedded during the casting process. To validate the FEA results and the calibrated mechanical model, bending performance tests were conducted on the fabricated finger. A bending sensor attached to the finger’s surface measured the bending angle at various input pressures supplied by a regulated air pump. The experimental results showed a strong correlation with the FEA simulation data and followed the trend predicted by the theoretical model, especially in the lower pressure range. Some deviation at higher pressures was attributed to the model’s simplifying assumptions, such as not accounting for the slight expansion of the limiting layer. This validation confirmed the reliability of both the simulation and the model for predicting the behavior of the end effector‘s key component.
A critical requirement for the harvesting end effector is to apply sufficient force to securely hold the fruit without causing compression damage. To evaluate this, we conducted a two-part FEA study. First, a stiffness simulation was performed where a load equivalent to the weight of a large citrus fruit (3 N) was applied to the soft finger while it was pressurized. The simulation aimed to find the minimum pressure at which the finger’s deformation under load became acceptable (converged solution). Results indicated that a pressure of 0.035 MPa provided adequate structural stiffness for the task. Second, a contact simulation modeled a single soft finger pressing against a detailed model of a citrus fruit, which included distinct material properties for the peel and the pulp. The simulated contact pressures and induced stresses within the fruit tissue at operational pressures (0.035 MPa and 0.07 MPa) were far below the experimentally measured failure thresholds of citrus peel and pulp. This confirmed that the designed soft end effector could operate within a pressure range of 0.035 MPa to 0.07 MPa, ensuring both secure grip and non-damaging interaction with the fruit.
The final stage involved integrating the soft fingers into a three-fingered gripper assembly, combining it with the cutting mechanism, and building a complete control system for the harvesting end effector. The control system featured two main loops: a pneumatic control loop for finger actuation (using a pump, solenoid valve, and pressure sensor) and a motor control loop for scissor operation (using a stepper motor driver and microcontroller). An initial bench test verified the non-damaging grip by placing a force sensor on the palm; the maximum squeezing force at 0.07 MPa was 18.13 N, significantly lower than the citrus’s damaging force. The end effector also demonstrated excellent adaptability by successfully grasping various spherical fruits (apples, pears, tomatoes) of different sizes and firmness using either enveloping or fingertip grasps. For the final harvesting test, the end effector was mounted on a collaborative robot arm (Dobot CR5). In a series of 30 trials with citrus fruit attached to branches, the system achieved a 96.67% success rate in securely and non-destructively clamping the fruit. The primary cause of failure was obstruction from nearby branches during the approach phase. For all successful grasps, the cutting mechanism reliably severed the peduncle. The average total time to complete a single fruit harvesting cycle was approximately 3.54 seconds.
In conclusion, this research successfully designed, modeled, optimized, fabricated, and tested a novel soft robotic end effector for citrus harvesting. The development process involved establishing a theoretical Yeoh model-based mechanical analysis, performing finite element simulation and response surface optimization to identify optimal geometric parameters (airbag height: 7 mm, limiting layer: 3.5 mm, air chamber: 3 mm), and validating the models through physical experiments. Comprehensive FEA contact simulations proved that the end effector operates safely within a pressure range of 0.035-0.07 MPa, providing necessary stiffness without causing fruit damage. The physical prototype demonstrated high effectiveness, with a 96.67% successful non-destructive clamping rate and an efficient cycle time. This work confirms the significant potential of soft robotics in agricultural applications, offering a compliant, adaptable, and damage-free alternative to traditional rigid end effectors for delicate fruit harvesting.