Previous Article in Journal
AI-Driven Innovations in 3D Printing: Optimization, Automation, and Intelligent Control
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Novel Pneumatic Soft Gripper Integrated with Mechanical Metamaterials for Enhanced Shape Matching Performance

1
College of Mechanical and Electrical Engineering, Hohai University, Changzhou 213200, China
2
College of Mechanics and Engineering Science, Hohai University, Nanjing 213022, China
*
Author to whom correspondence should be addressed.
J. Manuf. Mater. Process. 2025, 9(10), 330; https://doi.org/10.3390/jmmp9100330
Submission received: 12 September 2025 / Revised: 30 September 2025 / Accepted: 3 October 2025 / Published: 8 October 2025

Abstract

Traditional pneumatic soft grippers often suffer from a limited contact area and poor shape-matching performance, restricting their effectiveness in handling objects with complex or delicate surfaces. To address this problem, this study proposed an integrated soft gripper that combines pneumatic actuators with specially designed mechanical metamaterials, aiming to optimize deformation characteristics and enhance gripping surface conformity to target objects. The key contributions are as follows: (1) A novel integrated structure is designed, incorporating pneumatic actuators and mechanical metamaterials. (2) A highly efficient design framework based on deep learning is developed, incorporating forward and inverse neural networks to enable efficient performance prediction and inverse design. (3) The novel gripper is fabricated using stereolithography (SLA) and silicone casting, with experimental validation conducted via machine vision and multi-shape object tests. FEA simulations and experiments demonstrate significant improvements in shape matching: average deviations of gripping surfaces from targets are greatly reduced after optimization. This work validates that integrating mechanical metamaterials with data-driven design enhances the gripper’s adaptability, providing a feasible solution for high-performance soft gripping systems.

1. Introduction

Pneumatic soft grippers are generating significant research interest due to a number of benefits over traditional rigid grippers, including greater flexibility, adaptability, and safer human–machine interaction [1]. Various types of pneumatic soft grippers have been developed and applied in important engineering fields, including biomedical instrument [2,3], agricultural harvesting [4] and precision manufacturing [5]. These fields involve fragile or complex-shaped objects, which traditional rigid grippers struggle with, creating an ideal application for pneumatic soft grippers [6].
Currently, most grippers rely on positive pressure and the resulting frictional forces to hold objects. However, their fixed structures, once manufactured, often fail to fully conform to object contours, leading to limited contact area and poor shape-matching performance [7]. A common limitation of many pneumatic soft grippers is their insufficient contact with object surfaces [8]. To compensate, operators often increase positive pressure, which raises the risk of damaging delicate objects [9]. Thus, improving gripping performance without increasing pressure requires an alternative approach: expanding the effective contact area by enhancing shape adaptability has become a critical direction for advancing pneumatic soft gripper design.
Al-Rubaiai et al. [10] developed a gecko-inspired soft gripper that uses a basal adhesive layer for object manipulation, eliminating the need for continuous pressure. Ilievski et al. [11] created a starfish-like gripper with six pneumatically actuated fingers. Each finger bends upon pressurization due to a differential expansion mechanism, allowing adaptable grasping of various objects. Hoang et al. [12] put forward a fabric-based soft gripper that integrates gecko-inspired adhesion and adjustable stiffness, utilizing hydraulic systems for actuation and thermally responsive materials in its design. These studies show that although various soft gripper designs have significantly improved shape adaptability and gripping performance, they also increase complexity in materials, structural design, and control systems [13]. Thus, achieving higher shape-matching capabilities with simpler structures remains a key challenge.
Metamaterials have attracted widespread attention from researchers due to their unique shape-regulation capabilities [14,15]. Metamaterials are artificial materials with extraordinary physical properties not found in natural or conventional materials [16]. By designing and optimizing the structural units, researchers can precisely tailor their various properties. Among these, Poisson’s ratio-tunable metamaterials [17,18] enable controlled deformation under tension or compression. Jenett et al. [19] designed planar mechanical metamaterial unit cells and assembled them into 3D structures; after optimization, these structures showed high flexibility, spring-like large deformation under compression, and excellent shape recovery. Liu et al. [20] proposed a curved-wall lattice honeycomb metamaterial, whose effective elastic modulus was reduced by multiple orders of magnitude via geometric design to meet diverse deformation matching needs. Dikici et al. [21] integrated positive and negative Poisson’s ratio unit cells into a tubular lattice honeycomb, enabling it to show different transverse deformations under axial loads. Dudek et al. [22] further examined the current state of the field of shape-morphing metamaterials and proposed a unified classification system for the involved mechanisms, as well as the underlying design principles; more detailed information can be found in their review.
Meanwhile, the application of mechanical metamaterial structures in soft actuators and soft grippers has also been explored. Pan et al. [23] proposed an innovative approach by integrating mechanical metamaterial unit cells with both positive and negative Poisson’s ratio characteristics on two sides of a rubber tube. Under uniform pneumatic pressure, the tube exhibited corresponding buckling deformation. In another study, Charbel et al. [24] directly incorporated a mechanical metamaterial structure as an additional component at the tip of a zipper-type pneumatic soft actuator to fabricate a soft gripper, increased the gripping surface area and enhanced interaction capabilities.
Inspired by progress in shape-matching mechanical metamaterial sensors, the incorporation of mechanical metamaterial structures into pneumatic soft grippers presents a promising avenue for enhancing mechanical intelligence. This approach has the potential to mitigate existing limitations such as insufficient contact area and limited shape adaptability. Furthermore, advancements in fabrication technologies, particularly additive manufacturing, offer feasible ways to produce complex metamaterial architectures, thereby supporting the development of more intelligent gripper systems.
Therefore, this study proposes an integrated soft gripper that incorporates specially designed mechanical metamaterials at the distal ends of the soft actuators. This integration optimizes the deformation characteristics of the soft gripper and enhances the matching between the gripping surface and the target object profile. The main contributions include the following: (1) We designed an integrated soft gripper combining pneumatic actuators and mechanical metamaterials based on Gibson hexagonal units, with structural feasibility validated through finite element simulations in ABAQUS. (2) We developed a deep learning framework for performance prediction and inverse design, enabling automatic data generation and high-quality initial solutions for optimized gripper design. (3) We fabricated the gripper via 3D printing and silicone casting and experimentally validated its deformation and gripping performance using a machine vision system and multi-shape object tests.
The remainder of this paper is structured as follows: Section 2 presents the configuration of the chosen pneumatic soft actuator, the design approach for the mechanical metamaterial structure, and the finite element analysis are prese. Section 4 establishes a deep learning network framework to design mechanical metamaterial structures for desired gripping surfaces. Section 5 describes the fabrication process of actuators and metamaterial structures and presents loading experiments to verify validity by extracting coordinate values of sampling points on the gripping surface. Finally, Section 6 provides conclusions.

2. Gripper Enhanced by Mechanical Metamaterial

2.1. Pneumatic Soft Actuator

In our previous work, various pneumatic soft actuators with different structural forms have been proposed, the details of which can be found in Refs. [25,26]. As shown in Figure 1, the selected finger-based actuator is designed based on the pneumatic network principle, incorporating a specialized soft pneumatic structure with both slow-response and fast-response elements. Its internal structure is simple and efficient, featuring air chambers with a semi-circular cross-section and a built-in rectangular air channel connecting all chambers. The differential compliance between the stretchable top layer and non-stretchable bottom layer enables effective actuation. In summary, this design allows the actuator to withstand high pressure, respond quickly to pressure changes, achieve significant deformation, simplify manufacturing, ensure smooth airflow, and promote uniform inflation, enhancing responsiveness and overall performance.
For pneumatic soft actuators, geometric parameters affect performance. In this design, actuator parameters are determined based on empirical data and experiments, as shown in Table 1.

2.2. Attached Mechanical Metamaterial for Higher Shape Match Performance

The primary goal of this design is to control the deformed shape of the gripping surface. Since the actuator’s structure is fixed and difficult to modify, a more feasible and flexible approach is to add an external controllable, shape-matching attachment. Thus, we propose integrating mechanical metamaterial structures as attachments (Figure 2). Metamaterial units are beam-like, widely used in artificial material design. The attached mechanical metamaterial structures include two types of units: positive Poisson’s ratio (PPR) and negative Poisson’s ratio (NPR). Under air pressure, units in the metamaterial structure stretch or compress, causing deformation. The structural arrangement converts bending strain into tensile or compressive strain, enabling precise control of the gripping surface. Through careful design, metamaterial deformation can be precisely regulated, allowing the assembled gripper to adapt its gripping surface to objects of various geometries and material properties.
Both PPR and NPR units can be characterized by the same set of parameters. In detail, the crucial geometric parameters of both PPR cells and NPR cells involve the beam width (b) and the cell shape control parameters (g1, g2). Results indicate that when the parameters g1 and g2 vanish, a single metamaterial cell exhibits the most optimal deformation effect, either a positive Poisson ratio or negative Poisson ratio, during deformation. However, for practical reasons related to fabrication and deformation needs, suitable values of g1 and g2 are restricted based on experience and experiments (1.5 ≤ g1 ≤ 3.5, 1.5 ≤ g2 ≤ 3.5). At the same time, cells in the same row must have the same g2 parameter value.
In the mechanical metamaterial structure, there are m cells arranged along the length (m = 6), and n cells along the height (n = 1 or 2). Each cell occupies a plane space of s × s (s = 10 mm), matching the full width (W) of the actuator. Various positive Poisson ratio or negative Poisson ratio cells are programmatically arranged and naturally interconnected to form different mechanical metamaterial structures, as shown in Figure 3. As a result, these different architectures may generate different gripping surfaces under the bending actuation of the actuator. This study used a two-layer graded design—by adjusting the two unit cell layers’ parameters differently, it enhances the soft gripper’s deformation performance and suits more grasping targets, unlike single-layer structures. Theoretically, more layers allow richer deformation control, but two layers were chosen due to manufacturing/actuation constraints (more layers with current unit cell sizes would make the gripper too large and air pressure unable to drive deformation) and research focus (this work verifies the graded design’s effectiveness, which two layers sufficiently demonstrate).

3. Finite Element Analysis (FEA) of Traditional and Novel Assembled Gripper

3.1. FEA on Traditional Gripper

Finite element analysis was performed to understand deformation mechanisms in the novel assembled gripper and establish a simulation model for optimizing metamaterial structures (enabling design of structures that form specific gripping surfaces under the same pressure). Simulations were conducted using ABAQUS/Standard. Both the actuator and mechanical metamaterial structures were simulated using the super-elastic constitutive model to determine the material properties of Dragon Skin30 silicone gel and DPI8400 polyurethane TPU. The Yeoh model was applied for the silica gel material, with material constants set as C10 = 0.11, C20 = 0.02, and other parameters as 0. Similarly, the Yeoh model was used for the polyurethane TPU materials, with material constants set as C10 = 0.5, C20 = 0.1, and other parameters as 0. As shown in Figure 4a, The gripper was meshed with high-order tetrahedral elements, and convergence tests ensured simulation accuracy. The actuator’s fixed end was fully constrained, and normal pressure was applied to the inner wall. Frictionless hard contact was defined between adjacent surfaces of the actuator and metamaterial structure due to their interaction.
As shown in Figure 4b, Simulation results demonstrate the soft actuator’s bending deformation and driving performance under varying pressures. Due to differential axial elongation between the upper strain layer and lower strain-limiting layer, the actuator bends gradually toward the strain-limiting layer as pressure increases. Characterizing the bending angle (between the initial/final tip positions and the coordinate origin) intuitively reflects performance, validating structural parameters.

3.2. FEA on Novel Assembled Gripper

Simulation analyses of various mechanical metamaterial structures were performed under the same air pressure load (P = 100 kPa) to demonstrate their ability to regulate the gripping surfaces of pneumatic soft grippers. By extracting coordinate values from specific sampling points on the gripping surfaces, the significant impact of incorporating mechanical metamaterial structures was effectively confirmed. Preliminary finite element simulation analyses were conducted on three mechanical metamaterial structures composed of identical cell arrays: structure A (g1 = g2 = 1.5), structure B (g1 = g2 = 2.5), and structure C (g1 = g2 = 3.5). Static simulations in ABAQUS resulted in three distinct outcomes, as shown in Figure 5.
Obviously, these three distinct mechanical metamaterial structures incorporate unit cells with varying Poisson’s ratios. Even under the same air pressure load applied to the soft actuator, the stretching and compression of the internal beam structures within the unit cells result in the formation of gripping surfaces with different shapes: flat, concave, and convex. The results of this finite element simulation analysis inspire the present study to explore whether the mechanical metamaterial structure—specifically, the shape-control parameters of its individual unit cells—can be intelligently designed based on the desired gripping surface shape. This approach aims to better address the issue of low shape-matching degree between the gripping surface formed after the actuator’s deformation and the contour of the object to be grasped.

4. Data-Driven Inverse Design of Novel Assembled Gripper

The simulated results presented in Section 2 demonstrate that the deformation response of the gripping surfaces under specific air pressure load are primarily determined by the architectures of the mechanical metamaterial structure. The core goal is to enhance shape-matching capability by modifying the attached metamaterial, improving gripping performance. This requires rapid design of metamaterials tailored to specific gripping surfaces. However, the relationship between gripping surface deformation and metamaterial structure is complex, with no explicit function. Thus, we developed a systematic optimization framework, including a parametric model, forward design neural network, and inverse design neural network.

4.1. Parametric Finite Element Analysis Model

As shown in Figure 6, shape control parameters were defined to determine the architecture of the mechanical metamaterial. The intersection points between the centerline of each cell column and the gripping surface were selected as sampling points, whose vertical coordinates accurately describe the target deformed shape of the assembled gripper. Based on this parametric model, the gripper’s deformation response was computed through static simulations using Abaqus. With a fixed unit array (2 rows and 6 columns) and unit beam width (b = 1 mm), the architecture is influenced by 14 shape control parameters g i (1 ≤ i ≤ 14). Figure 7 presents samples of mechanical metamaterial structures generated with different shape control parameters.

4.2. Generation of Dataset for Deep Learning

The deep learning neutral network requires large datasets to support the training of the models in order to achieve the desired results. Using the Latin Hypercube Sampling method, 10,000 sets of different shape control parameter combinations were randomly generated. Meanwhile, joint simulations using MATLAB R2022b and ABAQUS 2022 were performed to extract the corresponding vertical coordinate values of different sampling points on the gripping surface for these 10,000 sets. After excluding data sets that failed to regenerate the model and those that were interrupted during static simulation, we obtained a total of 5800 valid datasets to serve as samples for the deep learning neural network. The data collection was arbitrarily split into three subsets following an 8:1:1 proportion. Specifically, 80% of the entire dataset was allocated as the Training Set, serving as the foundational data for training the deep learning neural network. Another 10% was designated as the validation set, which plays a crucial role in adjusting and optimizing the parameters of the deep learning neural network during the training process. The remaining 10% was set aside as the test set, whose primary function is to evaluate the accuracy of the already trained deep learning neural network.

4.3. Forward Design Neutral Network

The shape control parameters of the mechanical metamaterial structures exhibit a complex nonlinear relationship with the vertical coordinate values of the sampling points on the gripping surfaces. By leveraging multiple hidden layers in a deep learning neural network, the relationship between input and output features can be created.
Figure 8 shows the forward design neural network model used in this design. Shape control parameters g i are put into the input layer. The data is then transmitted through 3 hidden layers, each containing 200 nodes. In the end, it reaches the output layer, which consists of the vertical coordinate values R i of sampling points on the gripping surfaces. During the propagation, the output of each layer serves as the input for the next layer. Weights are added between nodes of adjacent layers, and the input for each node in the i th layer is obtained by summing the output values from the nodes in the ( i 1 ) th layer, multiplied by their corresponding weights, and then adding a bias value. The output value of the current node is calculated using an activation function as follows:
u k i = j = 1 n i 1 w j k i 1 × z j i 1 + b k i
z k i = f ( u k i )
where u k i and z k i represent the input and output values of the k th node in the i th layer, respectively; w j k i 1 denotes the connected weight between the j th node of the ( i 1 ) th layer and the k th node of the i th layer; b k i signifies the bias value for the k th node in the i th layer; n i 1 is the total number of nodes in the ( i 1 ) th layer; and f is the activation function.
It is worth noting that, in order to achieve better convergence, all data values should be normalized before training, so that these values are distributed between 0 and 1. Both input and output features are scaled proportionally as follows:
p ¯ i = p i p m i n p m a x p m i n
where p m i n and p m a x are the minimum and maximum values of each set of data features, while p ¯ i is the scaled value for the data p i .
The Mean Squared Error (MSE) formula is used to assess the magnitude of the error between the predicted output values and the corresponding reference values. It is described as follows:
M S E = 1 n i = 1 n ( y i y ¯ i ) 2
where i presents the i th observation; y i denotes the vertical coordinate value of a sampling point on the gripping surface of the novel assembled gripper in the dataset; y ¯ i represents the corresponding vertical coordinate value predicted by the forward design neutral network; and n is the size of the training dataset multiplied by the number of output nodes.
Rectified Linear Unit (Re-LU) activation function and Adaptive Moment Estimation (Adam) backpropagation algorithm are selected for parameter optimization in deep learning neural networks, aiming to find the optimal weights and biases that minimize the loss function. To be specific, the learning rate for the deep learning neutral network is set to 0.0001, with weight decay at 0.002. At the same time, the dropout techniques and the early stopping algorithm with a “patience” parameter of 10 are applied to prevent overfitting and enhance convergence. The training process was carried out on a computer equipped with an Intel Core i9-13900HX central processing unit and an NVIDIA GeForce RTX 4060 graphics processing unit. After 500 iterations, training/validation losses reached 0.0025/0.0032, with converged curves indicating effective training. Total training time was 150 s (0.3 s/iteration). Test set results confirmed accuracy and efficiency.
As illustrated in Figure 9, following 500 iterations, the loss function values for the training set and validation set stabilized at 0.0025 and 0.0032, respectively. This outcome serves as evidence of the deep neural network model’s precision in forecasting the deformation of the gripping surface of an integrated soft gripper—equipped with diverse mechanical metamaterial structures—when subjected to specific air pressure loads. The convergence of loss curves indicates the model exhibited strong stability and effectively optimized parameters within the set iterations. Total training time was 150 s (0.3 s/iteration), reflecting high computational efficiency. Test set evaluation showed predictions closely matched actual data with acceptable errors, confirming strong generalization and reliability for practical use.

4.4. Inverse Design Neutral Network

The aforementioned forward design neural network, trained on a large dataset, maps mechanical metamaterial shape-control parameters to vertical coordinates of gripping surface sampling points. This trained network is integrated into a new inverse design neural network, enabling efficient global search: inputting desired sampling point coordinates for a target gripping surface directly yields the corresponding metamaterial’s shape-control parameters, achieving target-to-architecture active control. However, inverse design often faces a one-to-many mapping issue—different parameter sets can produce identical gripping surface deformation under the same load—risking learning failures. To resolve this, the forward network is used to assist training the inverse network, ensuring the combined network meets inverse design requirements.
As shown in Figure 10, the connected neutral network consists of two parts: the front part is the inverse design neutral network, and the back part is the forward design neutral network that has been trained and frozen in the previous section. In the inverse design neutral network, the vertical coordinate values R i of the sampling points corresponding to the desired gripping surface are used as the input layer. The data is processed through two hidden layers with 1024 and 200 nodes, respectively. Then it passes to the output layer composed of shape control parameters g i . In the forward design neutral network, the shape control parameters g i obtained before are used as inputs to derive the predicted vertical coordinate values R ¯ i of the sampling points on the gripping surface. To achieve better convergence all the data values required for the connected neutral network have been normalized. The Rectified Linear Unit (Re-LU) is selected as the activation function to enhance learning performance. The connected neural network uses the Adam algorithm (learning rate 0.0001) for backpropagation to optimize weights and biases in the inverse network, minimizing the loss function. Dropout and early stopping (patience = 50) are applied to prevent overfitting and enhance convergence. Unlike in the previous section, the loss function here expresses the magnitude of the error between the vertical coordinate values R i of the sampling points corresponding to the desired gripping surface and the vertical coordinate values R ¯ i obtained through the forward design neural network. Therefore, we aim for these two values to be as close as possible.
As shown in Figure 11, after 500 iterations, training and validation losses reached 0.0014 and 0.0042, with converging curves indicating effective training. The loss function quantified differences between desired and predicted vertical coordinates of gripping surface sampling points. Total training time was 180 s (0.36 s/iteration). With satisfactory accuracy and efficiency, the inverse network within the connected network was separated as an independent model, capable of determining optimal metamaterial shape-control parameters for desired gripping surfaces.

4.5. Design Results

Two common types of gripping surfaces—planar and wedge-type—were selected. These two types, which typically exhibit suboptimal gripping performance in traditional soft grippers, served to validate the rationality and feasibility of the proposed integrated soft gripper in terms of gripping surface shape-matching functionality. Two common types of gripping surfaces—planar and wedge-type—were selected. Based on the two sets of shape control parameters, mechanical metamaterial configurations corresponding to the two gripping surfaces were generated. Finite element simulations of the integrated soft grippers (with both initial and optimized configurations) were conducted using ABAQUS, and the vertical coordinates of sampling points on the gripping surfaces under a certain air pressure load were collected as shown in Figure 12 and Figure 13.
Finite element simulations in ABAQUS under 100 kPa (standard operating pressure) were performed for both initial and optimized configurations. Results showed substantial improvements post-optimization. For planar surfaces shown in Figure 12a,b, initial sampling points deviated from targets by −1.80 mm to +0.24 mm, while optimized points narrowed to −0.06 mm to +0.13 mm—a 90% reduction in average error. For wedge-type surfaces shown in Figure 13a,b, initial deviations ranged from −0.43 mm to +1.20 mm, and optimized deviations shrank to −0.05 mm to +0.21 mm—an 80% reduction. Deformation profiles visually confirmed that optimized metamaterial-integrated grippers formed surfaces far closer to targets than initial neural network-generated configurations. These results—with maximum deviations reduced from 1.80 mm to 0.13 mm (planar) and 1.20 mm to 0.21 mm (wedge-type)—validate ISIGHT/2022 software’s effectiveness in enhancing deformation-matching performance.

5. Experiments and Validation

5.1. Preparation of the Novel Assembled Gripper

Pneumatic soft actuators feature complex geometries that are difficult to fabricate using traditional methods, but advances in additive manufacturing (AM) technology have made this possible [27]. As shown in Figure 14, the process involves three molds: the stretchable top layer mold, the air chamber mold, and the non-stretchable bottom layer mold. All of these molds are fabricated via stereolithography (SLA) 3D printing, with the equipment model being Anycubic Photon Mono M7 Max. The key process parameters are as shown in Table 2. The specific preparation process is as follows: First, Dragon Skin 30 silicone gel is prepared by mixing components A and B in a 1:1 ratio and poured into the mold. The mixture is defoamed in a vacuum kettle, then the mold is covered and placed in an incubator for curing. Once the curing process is complete, the mold is released, resulting in the preparation of the actuator’s stretchable top layer and non-stretchable bottom layer. After both parts are prepared, they will be bonded together using a silicone gel binder to complete the actuator.
The attached mechanical metamaterial structure is prepared using a vacuum remolding process as shown in Figure 15. First, a high-precision mother mold with smooth, natural surfaces is created using stereolithography (SLA) 3D printing technology. The mother mold is then fixed in the mold molding box, and liquid silicone is poured in to encapsulate it. After curing, the mold is removed, leaving behind a cavity that exactly matches the architecture of the mother mold. Next, liquid polyurethane (DPI8400) is poured into the cavity, and the mold is placed in a vacuum environment to eliminate air bubbles. After curing, a complete mechanical metamaterial structure is obtained.
When the pneumatic soft actuator and the specific mechanical metamaterial structure are prepared, the attached metamaterial structure is bonded to the front end of the actuator using silicone adhesive. When the contact surfaces of the two parts are securely bonded, the assembly of the novel assembled gripper enhanced by mechanical metamaterial is complete.

5.2. Performance Testing and Results Discussion

In the performance testing of the pneumatic soft actuator, a small air compressor (JKY003-400W) from Fujiwara, Japan—with a maximum supply pressure of 800 kPa—was connected to a French Elve Flow pressure sensor to achieve accurate control of the actuator’s internal pressure. The actuator’s end was fixed, and its stable-state deformation was measured multiple times under air pressure loads ranging from 0 to 120 kPa, with results recorded as shown in Figure 16a and specific deformation magnitudes in Figure 16b. The data points now represent the mean values of three independent measurements, and the error bars represent the relative standard deviation (RSD) of these triplicate measurements.
Observations showed the actuator’s bending angle and deformation magnitude increased nonlinearly with rising air pressure, reaching maximum values at 120 kPa. This nonlinearity stems from the nonlinear constitutive properties of the superelastic Dragon Skin 30 silicone used and the influence of the actuator’s own weight during deformation. Despite this, the actuator’s performance met the study requirements: its sufficient bending strain can convert into tensile or compressive strain in the attached mechanical metamaterial structure, enabling precise control of the gripping surface shape for the integrated soft gripper.

5.3. Deformation of Soft Grippers with Integrated Mechanical Metamaterials

As shown in Figure 17, performance tests of integrated soft grippers (with different mechanical metamaterials) under loading were conducted via the soft actuator test platform. Halcon 20.11 collected coordinates of sampling points on the clamping surface under specific air pressure, intuitively quantifying deformation; the simple coordinate acquisition platform and image processing results are in Figure 18 and Figure 19. During testing, deformation of the two grippers under 0–100 kPa air pressure (after stabilization) was recorded (Figure 18d and Figure 19d). With increasing pressure, the soft actuator induced gradual deformation of metamaterials; at 100 kPa (working load), their macroscopic clamping surface deformation met the study’s targets.
After image acquisition of deformation under working load, sampling point coordinates on the clamping surfaces were collected. Image threshold segmentation enhanced grayscale differences between sampling points and surroundings; noise filtering improved contour clarity, boosting acquisition accuracy and algorithm efficiency. Real ordinate values of sampling points were obtained and compared with target values and finite element simulation results (Figure 18c,e and Figure 19c,e). Analysis showed that specific clamping surfaces, suboptimal in traditional grippers, performed effectively in integrated ones, with results consistent with simulations within error ranges. In summary, integrating mechanical metamaterials enables matching the soft gripper’s clamping surface to the target shape, verifying the rationality and feasibility of adjusting metamaterials to form ideal clamping surfaces per the contours of clamped objects.
The experimental results clearly demonstrate that the two specific gripping surface requirements are largely met and fall within the error bounds of the simulation results. Overall, this confirms the validity and feasibility of the design.

5.4. Experimental Validation of Shape-Matching Control in Soft Grippers

To evaluate the performance of the soft gripper with integrated mechanical metamaterials in practical applications, a specialized fixture for bilateral fixation of soft actuators was designed, as shown in Figure 20a. Depending on the required gripping surface shapes, the soft grippers integrated with corresponding mechanical metamaterial structures were symmetrically mounted into the fixture slots and secured with bolts, completing the functional testing platform (Figure 20b).
The fixture was fabricated using light-curing 3D printing (Figure 20a). After assembly, pneumatic tubes were connected to supply air pressure to the soft actuators, with silicone adhesive applied at the interfaces to ensure airtightness. The final setup comprised a pair of integrated soft grippers with specific mechanical metamaterial structures (Figure 20b). To address the poor performance of traditional soft grippers in handling planar and wedge-shaped objects, optimized mechanical metamaterials were developed to enhance deformation adaptability. For functional validation, planar and wedge-shaped test specimens were fabricated using fused deposition modeling (Figure 21) and used to assess the gripper’s adaptability and gripping effectiveness.
Both specimen types were tested on the platform. Under a 100 kPa air pressure, the soft grippers with different metamaterial structures achieved stable gripping states, conforming closely to the specimen contours (Figure 22). Compared to conventional soft grippers, the designed system demonstrated significantly improved stability and adaptability in industrial settings.

5.5. Discussion and the Limitations

Based on the experimental results above, we can observe that traditional soft grippers can only achieve specific arch deformations under certain air pressures, exhibiting limited shape adaptability and difficulty in altering their shape characteristics, as illustrated in Figure 23a. However, by incorporating mechanical metamaterials, diverse deformation modes—such as custom horizontal (Figure 23b) and inclined (Figure 23c) shapes—can be attained. This approach allows for the customization of different metamaterial attachments based on the actual shape of the target object, significantly enhancing deformation matching performance.
While the above simulations and experiments confirm the proposed metamaterial-enhanced soft gripper has better shape-matching performance, it still has key limitations for practical industrial use:
(1)
Structural Design: Metamaterials are silicone-bonded “patches” on the fingertip (not integral). Long-term cyclic bending may cause adhesive failure (shear fatigue, peeling, creep); adhesive thickness/coating errors affect stiffness/Poisson’s ratio, with no quantified error sensitivity or tolerance control yet.
(2)
Overly Simple Optimization Objective: The deep learning loss function only uses sampling point z-coordinate deviation (focusing on geometric fit), ignoring gripping force, peak contact stress, fatigue life, and airtight safety margin. This risks crushing fragile objects (despite good profile matching) or early cracking.
(3)
Inadequate Experimental Coverage: Experiments only used two samples (planar, wedge-shaped; monotonic curvature, no depressions), failing to verify “high adaptability to complex/fragile targets.” Tests used only single-pressure loading (no cyclic tests) and lacked pressure retention/profile drift records, providing no basis for industrial service life.
We aim to investigate these issues more deeply in future work.

6. Conclusions

This study presents an integrated pneumatic soft gripper enhanced by mechanical metamaterials, addressing the critical limitation of poor shape-matching in traditional soft grippers. By combining pneumatic actuators with tunable metamaterial structures (featuring positive and negative Poisson’s ratios), the gripper achieves optimized deformation to conform to target object contours.
(1)
This study presents the structural design of an integrated gripper, validated through FEA to ensure feasible deformation under pneumatic loads. The incorporation of Gibson hexagonal unit-based metamaterials enables tailored surface deformation, significantly improving contact conformity with complex object geometries.
(2)
This study develops a robust deep learning framework, comprising forward and inverse neural networks, which efficiently predicts gripping surface deformation and enables inverse design of metamaterial parameters for desired shapes, significantly reducing design complexity. This data-driven approach allows rapid generation of high-quality design solutions, overcoming the limitations of traditional trial-and-error methods.
(3)
This study demonstrates successful fabrication via 3D printing and silicone casting, with experimental results confirming that the optimized gripper achieves remarkable shape-matching performance and stable gripping of test objects. Machine vision-based validation demonstrated average shape deviation reductions of over 80% compared to initial designs, highlighting the practical effectiveness of the proposed method.
This research demonstrates that integrating mechanical metamaterials with data-driven design methodologies effectively enhances the shape adaptability of soft grippers. Future work will optimize material combinations to improve durability, expand the range of target object shapes and develop multi-objective optimization models, supplement tests on complex/fragile samples and cyclic loading experiments, and explore multi-finger gripper configurations for more complex grasping tasks.

Author Contributions

Conceptualization, Z.H. and X.Z.; Methodology, Z.H. and X.Z.; Software, Z.H., B.Z. and X.Z.; Validation, X.Z.; Formal analysis, X.C.; Investigation, B.Z.; Resources, B.Z. and W.S.; Data curation, B.Z. and W.S.; Writing—original draft, S.W.; Writing—review & editing, Z.H., W.S., Z.X. and S.W.; Visualization, Z.X.; Supervision, Z.X. and X.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the China Postdoctoral Science Foundation (Grant No. 42408902), Natural Science Research of JiangSu Higher Education Institutions of China (Grant No. 524047011), Changzhou Sci & Tech Program (Grant No. CJ20240092), the Fundamental Research Funds for the Central Universities (Grant No. B240201183), and National Natural Science Foundation of China (Grant No. 12502159).

Data Availability Statement

The data that support the findings of this study are available upon reasonable request.

Acknowledgments

We sincerely thank the anonymous reviewers for their insightful comments that greatly improved this manuscript and the editors for their professional guidance and efficient handling of the submission and revision process.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Kim, D.; Baek, S.R.; Kim, M.S.; Park, C.Y.; Lee, I.H. Design and manufacturing process of pneumatic soft gripper for additive manufacturing. Sens. Actuators A Phys. 2024, 370, 115218. [Google Scholar] [CrossRef]
  2. Sinuo, Z.; Cong, N.C.; Thien, H.T.; Nho, D.T.; HoangPhuong, P. Transparent Pneumatic Tactile Sensors for Soft Biomedical Robotics. Sensors 2023, 23, 5671. [Google Scholar] [CrossRef]
  3. Li, X.; Fan, D.; Sun, Y.; Xu, L.; Li, D.; Sun, B.; Nong, S.; Li, W.; Zhang, S.; Hu, B. Porous magnetic soft grippers for fast and gentle grasping of delicate living objects. Adv. Mater. 2024, 36, 2409173. [Google Scholar] [CrossRef]
  4. Zhang, D.; Zhang, W.; Yang, H.; Yang, H. Application of Soft Grippers in the Field of Agricultural Harvesting: A Review. Machines 2025, 13, 55. [Google Scholar] [CrossRef]
  5. Schreiber, F.; Frohn-Sörensen, P.; Engel, B.; Manns, M. Applicability of models to predict the bending behavior of soft pneumatic grippers. Int. J. Adv. Manuf. Technol. 2025, 138, 273–286. [Google Scholar] [CrossRef]
  6. Qu, J.; Yu, Z.; Tang, W.; Xu, Y.; Mao, B.; Zhou, K. Advanced technologies and applications of robotic soft grippers. Adv. Mater. Technol. 2024, 9, 2301004. [Google Scholar] [CrossRef]
  7. Knospler, J. Design and Control of a Versatile Modular Soft Robotic System with Integrated Resource Sharing and Self-Reconfiguration Capabilities. Master’s Thesis, Rowan University, Glassboro, NJ, USA, 2025. [Google Scholar]
  8. Dzedzickis, A.; Petronienė, J.J.; Petkevičius, S.; Bučinskas, V. Soft grippers in robotics: Progress of last 10 years. Machines 2024, 12, 887. [Google Scholar] [CrossRef]
  9. Kozhemyatov, K.; Bulauka, Y. The improving of the safety level of the equipment working under excessive pressure. In Topical Issues of Rational Use of Natural Resource; CRC Press: Boca Raton, FL, USA, 2019; Volume 2, pp. 822–831. [Google Scholar]
  10. Al-Rubaiai, M.; Pinto, T.; Qian, C.; Tan, X. Soft Actuators with Stiffness and Shape Modulation Using 3D-Printed Conductive Polylactic Acid Material. Soft Robot. 2019, 6, 318–332. [Google Scholar] [CrossRef] [PubMed]
  11. Ilievski, F.; Mazzeo, A.D.; Shepherd, R.F.; Chen, X.; Whitesides, G.M. Soft robotics for chemists. Angew. Chem. (Int. Ed. Engl.) 2011, 50, 1890–1895. [Google Scholar] [CrossRef]
  12. Thien, H.T.; Sheng, Q.J.J.; Thanh, T.M.; Thien, P.P.; Hamilton, L.N.; Nho, D.T. Soft robotic fabric gripper with gecko adhesion and variable stiffness. Sens. Actuators A Phys. 2021, 323, 112673. [Google Scholar]
  13. Kumar, S.; Wang, X.; Strachan, J.P.; Yang, Y.; Lu, W.D. Dynamical memristors for higher-complexity neuromorphic computing. Nat. Rev. Mater. 2022, 7, 575–591. [Google Scholar] [CrossRef]
  14. Duan, F.; Wei, D.; Chen, A.; Zheng, X.; Wang, H.; Qin, G. Efficient modulation of thermal transport in two-dimensional materials for thermal management in device applications. Nanoscale 2023, 15, 1459–1483. [Google Scholar] [CrossRef]
  15. Yao, H.; Guo, B.; Zhang, T.; Tao, W. Photocurrent enhancement of topological insulator by femtosecond laser controlled surface structure. Eur. Phys. J. Plus 2023, 138, 562. [Google Scholar] [CrossRef]
  16. Valipour, A.; Kargozarfard, M.H.; Rakhshi, M.; Yaghootian, A.; Sedighi, H.M. Metamaterials and their applications: An overview. Proc. Inst. Mech. Eng. Part L J. Mater. Des. Appl. 2022, 236, 2171–2210. [Google Scholar] [CrossRef]
  17. Liu, G.-Q.; Liu, H.-T. Synergistic design of curved beam metastructure with tunable nonlinearity deformation and Poisson’s ratio. Eng. Fract. Mech. 2025, 316, 110897. [Google Scholar] [CrossRef]
  18. Shunshun, R.; Zhao, G. Mechanical characterization of a metamaterial with negative Poisson’s ratio under compressive loading: Experimental along with FEM. Mech. Adv. Mater. Struct. 2025, 1–10. [Google Scholar] [CrossRef]
  19. Jenett, B.; Cameron, C.; Tourlomousis, F.; Rubio, A.P.; Ochalek, M.; Gershenfeld, N. Discretely assembled mechanical metamaterials. Sci. Adv. 2020, 6, eabc9943. [Google Scholar] [CrossRef]
  20. Liu, W.; Li, H.; Zhang, J.; Bai, Y. In-plane mechanics of a novel cellular structure for multiple morphing applications. Compos. Struct. 2019, 207, 598–611. [Google Scholar] [CrossRef]
  21. Dikici, Y.; Jiang, H.; Li, B.; Daltorio, K.A.; Akkus, O. Piece-by-piece shape-morphing: Engineering compatible auxetic and non-auxetic lattices to improve soft robot performance in confined spaces. Adv. Eng. Mater. 2022, 24, 2101620. [Google Scholar] [CrossRef]
  22. Dudek, K.K.; Kadic, M.; Coulais, C.; Bertoldi, K. Shape-morphing metamaterials. Nat. Rev. Mater. 2025, 10, 783–798. [Google Scholar] [CrossRef]
  23. Qi, P.; ShiTong, C.; FeiFei, C.; XiangYang, Z. Programmable soft bending actuators with auxetic metamaterials. Sci. China (Technol. Sci.) 2020, 63, 2518–2526. [Google Scholar]
  24. Charbel, T.; Rahim, M.; Gursel, A. A 3D Printed Modular Soft Gripper Integrated with Metamaterials for Conformal Grasping. Front. Robot. AI 2022, 8, 799230. [Google Scholar] [CrossRef] [PubMed]
  25. Xinjie, Z.; Elisha, O.A.; Ke, M.; Shouyi, Y. Entirely soft valve leveraging snap-through instability for passive flow control. Sens. Actuators B Chem. 2022, 367, 132035. [Google Scholar]
  26. Zhang, X.; Yu, S.; Dai, J.; Oseyemi, A.E.; Liu, L.; Du, N.; Lv, F. A Modular Soft Gripper with Combined Pneu-Net Actuators. Actuators 2023, 12, 172. [Google Scholar] [CrossRef]
  27. Vesco, S.; Salvi, D. Fuzzy skin in fused filament fabrication: Enhancing morphology, wettability, and friction through a full-factorial experimental plan. Prog. Addit. Manuf. 2025, 1–25. [Google Scholar] [CrossRef]
Figure 1. Structure and key parameters of the soft actuator.
Figure 1. Structure and key parameters of the soft actuator.
Jmmp 09 00330 g001
Figure 2. Geometry and parametric variables of the mechanical metamaterial and assembled novel soft gripper: (a) Geometry of metamaterials; (b) Attachment process.
Figure 2. Geometry and parametric variables of the mechanical metamaterial and assembled novel soft gripper: (a) Geometry of metamaterials; (b) Attachment process.
Jmmp 09 00330 g002
Figure 3. Multi-layer model of the mechanical metamaterial structures.
Figure 3. Multi-layer model of the mechanical metamaterial structures.
Jmmp 09 00330 g003
Figure 4. (a) Mesh generation of the soft actuator; (b) Simulation results of the soft actuator under different pneumatic loads.
Figure 4. (a) Mesh generation of the soft actuator; (b) Simulation results of the soft actuator under different pneumatic loads.
Jmmp 09 00330 g004
Figure 5. Results of different assembled soft grippers under the same pneumatic load: (a) mesh generation of the integrated soft gripper; (b) concave deformation; (c) straight-line deformation; (d) convex deformation.
Figure 5. Results of different assembled soft grippers under the same pneumatic load: (a) mesh generation of the integrated soft gripper; (b) concave deformation; (c) straight-line deformation; (d) convex deformation.
Jmmp 09 00330 g005
Figure 6. Shape control parameters of the arrayed metamaterials.
Figure 6. Shape control parameters of the arrayed metamaterials.
Jmmp 09 00330 g006
Figure 7. (a) Representative samples of mechanical metamaterial structures. (b) Schematic diagram of sampling point locations on the gripping surface.
Figure 7. (a) Representative samples of mechanical metamaterial structures. (b) Schematic diagram of sampling point locations on the gripping surface.
Jmmp 09 00330 g007
Figure 8. Deep neural network model for performance prediction.
Figure 8. Deep neural network model for performance prediction.
Jmmp 09 00330 g008
Figure 9. Loss function of the deep neural network model for performance prediction.
Figure 9. Loss function of the deep neural network model for performance prediction.
Jmmp 09 00330 g009
Figure 10. Serial deep neural network model.
Figure 10. Serial deep neural network model.
Jmmp 09 00330 g010
Figure 11. Loss function of the serial deep neural network model.
Figure 11. Loss function of the serial deep neural network model.
Jmmp 09 00330 g011
Figure 12. Configuration and simulation results of the planner griper with (a) Initial geometry parameters: (b) Optimized geometry parameters.
Figure 12. Configuration and simulation results of the planner griper with (a) Initial geometry parameters: (b) Optimized geometry parameters.
Jmmp 09 00330 g012
Figure 13. Configuration and simulation results of the wedge-type griper with (a) Initial geometry parameters: (b) Optimized geometry parameters.
Figure 13. Configuration and simulation results of the wedge-type griper with (a) Initial geometry parameters: (b) Optimized geometry parameters.
Jmmp 09 00330 g013
Figure 14. Three-dimensional model and actual molds of the soft actuator molds: (a) Mold A; (b) Mold B; (c) Mold C.
Figure 14. Three-dimensional model and actual molds of the soft actuator molds: (a) Mold A; (b) Mold B; (c) Mold C.
Jmmp 09 00330 g014
Figure 15. Key steps in mechanical metamaterial structure fabrication and prepared product diagram: (a) Fixation; (b) Mold Box Preparation; (c) Demolding; (d) Replication; (e) Prepared Mechanical Metamaterial Product.
Figure 15. Key steps in mechanical metamaterial structure fabrication and prepared product diagram: (a) Fixation; (b) Mold Box Preparation; (c) Demolding; (d) Replication; (e) Prepared Mechanical Metamaterial Product.
Jmmp 09 00330 g015
Figure 16. Displacement test results of the soft actuator under different pneumatic loads: (a) Deformation process; (b) Bending angle.
Figure 16. Displacement test results of the soft actuator under different pneumatic loads: (a) Deformation process; (b) Bending angle.
Jmmp 09 00330 g016
Figure 17. (a) Coordinate extraction platform based on Halcon software. (b) Image processing results.
Figure 17. (a) Coordinate extraction platform based on Halcon software. (b) Image processing results.
Jmmp 09 00330 g017
Figure 18. Coordinates extracted results of the planar integrated soft gripper: (a) Optimized design result; (b) Fabricated specimen; (c) Simulation result; (d) Actual deformed configurations under different air pressures; (e) Comparison of target deformation, simulated deformation, and actual deformation, showing close agreement and validating the design objectives.
Figure 18. Coordinates extracted results of the planar integrated soft gripper: (a) Optimized design result; (b) Fabricated specimen; (c) Simulation result; (d) Actual deformed configurations under different air pressures; (e) Comparison of target deformation, simulated deformation, and actual deformation, showing close agreement and validating the design objectives.
Jmmp 09 00330 g018
Figure 19. Coordinates extracted results of the wedge-type integrated soft gripper: (a) Optimized design result; (b) Fabricated specimen; (c) Simulation result; (d) Actual deformed configurations under different air pressures; (e) Comparison of target deformation, simulated deformation, and actual deformation, showing close agreement and validating the design objectives.
Figure 19. Coordinates extracted results of the wedge-type integrated soft gripper: (a) Optimized design result; (b) Fabricated specimen; (c) Simulation result; (d) Actual deformed configurations under different air pressures; (e) Comparison of target deformation, simulated deformation, and actual deformation, showing close agreement and validating the design objectives.
Jmmp 09 00330 g019
Figure 20. Working platform of the integrated soft grippers: (a) Prepared product diagram; (b) Assembly diagram.
Figure 20. Working platform of the integrated soft grippers: (a) Prepared product diagram; (b) Assembly diagram.
Jmmp 09 00330 g020
Figure 21. Prepared specimens for gripping tests: (a) Planar type; (b) Wedge type.
Figure 21. Prepared specimens for gripping tests: (a) Planar type; (b) Wedge type.
Jmmp 09 00330 g021
Figure 22. Gripping performance of the integrated soft gripper: (a) Planar type; (b) Wedge-type.
Figure 22. Gripping performance of the integrated soft gripper: (a) Planar type; (b) Wedge-type.
Jmmp 09 00330 g022
Figure 23. Comparison of (a) traditional and (b,c) novel grippers in shape matching performance. Note that the dotted lines represent the target deformed shape.
Figure 23. Comparison of (a) traditional and (b,c) novel grippers in shape matching performance. Note that the dotted lines represent the target deformed shape.
Jmmp 09 00330 g023
Table 1. Key Structural parameters values of the soft actuator.
Table 1. Key Structural parameters values of the soft actuator.
LWHRtitodrw
110 mm26 mm19 mm16 mm2 mm3 mm3 mm13 mm2 mm
Table 2. Key parameters and their corresponding values of the SLA manufacturing process.
Table 2. Key parameters and their corresponding values of the SLA manufacturing process.
Exposure TimeLight-Off TimeBottom Exposure TimeNumber of Bottom LayersZ-Axis Lift HeightZ-Axis Lift SpeedZ-Axis Retract Speed
2.2 s2 s40 s55 mm4 mm/s4 mm/s
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Han, Z.; Zhang, B.; Sun, W.; Xu, Z.; Chen, X.; Weng, S.; Zhang, X. Novel Pneumatic Soft Gripper Integrated with Mechanical Metamaterials for Enhanced Shape Matching Performance. J. Manuf. Mater. Process. 2025, 9, 330. https://doi.org/10.3390/jmmp9100330

AMA Style

Han Z, Zhang B, Sun W, Xu Z, Chen X, Weng S, Zhang X. Novel Pneumatic Soft Gripper Integrated with Mechanical Metamaterials for Enhanced Shape Matching Performance. Journal of Manufacturing and Materials Processing. 2025; 9(10):330. https://doi.org/10.3390/jmmp9100330

Chicago/Turabian Style

Han, Zhengtong, Boqing Zhang, Wentao Sun, Ze Xu, Xiang Chen, Shayuan Weng, and Xinjie Zhang. 2025. "Novel Pneumatic Soft Gripper Integrated with Mechanical Metamaterials for Enhanced Shape Matching Performance" Journal of Manufacturing and Materials Processing 9, no. 10: 330. https://doi.org/10.3390/jmmp9100330

APA Style

Han, Z., Zhang, B., Sun, W., Xu, Z., Chen, X., Weng, S., & Zhang, X. (2025). Novel Pneumatic Soft Gripper Integrated with Mechanical Metamaterials for Enhanced Shape Matching Performance. Journal of Manufacturing and Materials Processing, 9(10), 330. https://doi.org/10.3390/jmmp9100330

Article Metrics

Article metric data becomes available approximately 24 hours after publication online.
Back to TopTop