Tactile Recognition of Shape and Texture on the Same Substrate

Abstract

Due to the difficulties in designing sensor arrays with a wide detection range, high sensitivity, and the sensing ability to convert tangential force into normal force on the same substrate, the integration of shape and texture recognitions in one electronic skin (e-skin) has not been realized so far. Herein, an e-skin tactile-sensing system is presented, based on resistive pressure-sensing units (serving as Meissner corpuscles) unevenly distributed on a bionic hand-shaped polyimide substrate, which can realize shape and texture recognitions concurrently. A multilayer microporous structure with different pores is designed and introduced into the sensor, which enables each sensing unit ultrahigh sensitivity and a wide detection range. Meanwhile, a customized micro-pyramid array is developed and assembled to the sensor array, which realizes the transformation from tangential force to normal force. With the help of artificial intelligence technology, the recognition accuracy reaches 100% for 8 different shapes, and 99.7% for 10 different textures, respectively. The proposed design strategy enables compatible fabrication, simple signal processing, and convenient extension of bionic free-shaped e-skin, which paves a promising way for the popularization of e-skin in large-scale intelligent wearable fields.

Introduction

Tactile-sensing systems are composed of an electronic skin (e-skin), signal acquisition circuits and artificial intelligence technology, and are becoming an exciting frontier for the next generation of wearable electronics owing to their potential applications in human–machine interface,^[1,2] health care,^[3,4] sports detection,^[5–7] and soft robotics.^[8,9] However, although remarkable breakthroughs have been realized for e-skin tactile-sensing systems to imitate the perception of human skin and the recognition of the human brain, most of them are limited to a single function such as temperature perception,^[10–13] press perception,^[14–16] shape recognition,^[17,18] and texture recognition.^[19–21] To date, two important tactile recognition functions (shape and texture) for collecting external stimulus information have not been integrated into a single-tactile e-skin on the same substrate. This is because the perceptual ability of skin differs in different parts of the human body.^[22] The fingertip skin is supersensitive in distinguishing the texture of objects, which depends synergistically on the densely distributed Meissner corpuscles and fingerprints. Fingerprints convert the tangential force generated during friction into tiny normal pressures, whereas Meissner corpuscles sense the distribution of pressure with high resolution. Palm skin is good at perceiving more object shape features when grasping (high pressure) because of the relatively scattered but widely distributed tactile corpuscles over the palm. The design and fabrication of sensing units (serving as tactile corpuscles) to perceive the aforementioned various parameters and realize the two identification functions on the same substrate, such as a human being, are considerably difficult.

Specifically, the sensing units in the bionic palm e-skin part must possess a wide detection range to detect normal pressures.^[23] Simultaneously, the sensing units in the bionic fingertip e-skin must have the ability to convert tangential friction into normal pressure, and a tiny normal pressure should be distinguished with a high sensitivity in a narrow pressure range to identify the texture and roughness of objects.^[24] The tradeoff between the sensitivity and the detection range, and the pressure sensing of different directions are typical technical difficulties in the design of the sensor.

In addition to the difficulties in designing, manufacturing is a critical barrier in the development of e-skin, and the preparation process needs to satisfy the following characteristics. First, it should be suitable for processing with a customized flexible substrate to satisfy the shape and curvature requirements of different human perception areas. Second, it is better to use a standard process to ensure the stability and consistency of the sensing units used in e-skin. Third, it should be a compatible process, that manufactures sensing arrays with different unit density/sizes on a flexible substrate and in the same batch to achieve multi-recognition demands. Traditional methods, such as vacuum filtration, casting, and dip coating, have excellent advantages in fabricating devices with agile microstructure features to obtain a high performance. However, the poor consistency and complex process of individually assembling arrays makes them unsuitable for integration and large-scale manufacturing.^[25] High-precision processing methods used in the microelectronics industry, such as photolithography and etching, are capable of controlled manufacturing. However, they have limitations in the construction of microstructures and the selection of materials, which hinders the improvement of sensor performance. Thus, the aforementioned methods are not the best choices for use in the fabrication of large-sized and free-shaped bionic e-skin.^[26] Designing a sensor unit to obtain pressure sensing in different directions with a high sensitivity, a wide detection range, and continuous fabrication into arrays with high consistency on one substrate to realize shape and texture recognition functions is the most significant challenge in the research and application of e-skin.

In this study, an e-skin was constructed by unevenly distributing resistive pressure-sensing units on a hand-shaped polyimide (PI) substrate to realize shape and texture recognition functions. Each sensing unit consists of a reduced graphene oxide/MXene/reduced graphene oxide (RGMRG)-sensitive layer with a microporous structures and flexible electrodes on the same PI substrate. A multilayer microporous structure with different pore sizes was designed and fabricated using inkjet-printing and high-temperature reduction. The different pores were realized by the film-forming characteristics of the materials, which resulted in different degrees of foaming during the thermal annealing process. Benefiting from the synergistic effect of microporous structures, both the single-unit and array formats exhibited a high sensitivity (0.934 kPa⁻¹ in 0–0.6 kPa) and wide detection range (0–50 kPa). Intelligent shape recognition was obtained by the palm-shaped e-skin part, which consists of 5 × 4 irregularly distributed circular sensing units (1 cm in diameter) that cover the grasping position on demand. A 4 × 4 sensor array with dense small units (2 mm × 2 mm) served as the fingertip e-skin part to achieve advanced intelligent texture recognition after being equipped with the designed micro-pyramid module, which imitated the function of fingerprints. The recognition accuracies were 100% for 8 different shapes and 99.7% for 10 different textures.

Results and Discussion Fabrication Process and Characterization

Figure 1a,b displays a schematic of the tactile-sensing system based on the RGMRG e-skin and the specific structure of the RGMRG e-skin. The proposed sensor unit was composed of five layers: an upper electrode, an upper reduced graphene oxide (rGO) layer, an MXene layer, an lower rGO layer, and an lower electrode. The inkjet-printing technology was used to print sensitive layers including two rGO layers and an MXene layer on a flexible printed circuit board (FPCB) electrode array. High-temperature thermal reduction was used to introduce microporous structures. Each sensing unit was approximately a Meissner corpuscle that can help this skin perceive extremely slight contact. The smaller sensor array at the fingertip was similar to the dense distribution of Meissner corpuscles in the fingertip, which can sense more touch information. The cross-sectional scanning electron microscopy (SEM) image in Figure 1c shows that the rGO layer exhibited a multilayered structure whereas MXene was more disordered, which was due to the better film-forming properties of graphene oxide (GO) compared with those of MXene. The RGMRG in Figure 1c is thicker than that of the sample without thermal reduction (as shown in Figure S1, Supporting Information) because of the loose porous structure of the film induced by high temperature (the thicknesses are 59.3 and 80.2 μm before and after thermal reduction, respectively). At high reduction temperature, the elimination of oxygen groups on GO and MXene nanosheets led to the production of CO, CO₂, and H₂O. The gaseous species were released rapidly and established a high pressure between the nanosheets, which overcame the van der Waals forces that held the nanosheets together. A microporous structure was eventually formed in the RGMRG film. The incorporation of porous structures in sensitive films can decrease the Young's modulus, thereby endowing the films with greater deformation under the same pressure stimulation. The closure of pores generates massive conductive paths, leading to a significant reduction in resistance. Hence, the introduction of porous structures can greatly improve the sensitivity of sensors. The pore size could be controlled by adjusting the thermal reduction time. The longer the thermal reduction time, the larger the pore size, as previously demonstrated in our research.^[27] Moreover, we captured the surface morphologies of the different layers of the e-skin. As shown in Figure 1d, the more prominent wrinkles on the surface of the top rGO layer compared with the bottom rGO layer may be attributed to the non-film-forming properties of the MXene layer. The bottom rGO layer appeared as a flat film, whereas the MXene layer appeared as disordered stacked nanosheets, which resulted in more wrinkles in the top rGO layer.

Figure 1. The design of the tactile-sensing system and the characterization of the reduced graphene oxide/MXene/reduced graphene oxide (RGMRG) electronic skin (e-skin). a) The schematic of the human tactile system and the bionic tactile-sensing system based on the RGMRG e-skin. b) The structure of the RGMRG e-skin. c) The cross-sectional scanning electron microscopy (SEM) images of the e-skin sensor structure. d) The surface SEM images of the e-skin sensor structure. e) Apparent viscosity of the graphene oxide (GO) ink (5, 10, 20 mg mL−1) as a function of shear rate. f) Apparent viscosity of the MXene ink (5, 10, 20 mg mL−1) as a function of shear rate. g) The contact angle of GO and MXene inks.

The development of printable inks with suitable viscoelasticity and surface tension to adapt to the surface energy of the substrate is a prerequisites for the inkjet-printing technology. In general, the ideal ink for inkjet printing must have a viscosity between 1 and 20 mPa s.^[28] Therefore, GO and MXene inks of different concentrations were prepared and analyzed. Figure 1e,f shows the rheological properties of GO and MXene inks with the loadings of 5, 10, and 20 mg mL⁻¹. The apparent viscosity of all the inks decreased approximately linearly with increasing shear rate (from 0.1 to 10 s⁻¹), demonstrating that the ink exhibited an apparent shear-thinning behavior. The shear-thinning behavior showed that the ink could be smoothly ejected from the nozzle under an appropriate pressure. To match the requirement of inkjet printing, 10 mg mL⁻¹ GO ink (19.8 mPa s at 1 s⁻¹) and 20 mg mL⁻¹ MXene ink (19.3 mPa s at 1 s⁻¹) were chosen. In addition, to adapt to the surface energy of the substrate and the surface tension of the ink, the contact angle between the GO ink and the copper foil used as the electrode plate on the FPCB was tested, and was determined to be 75°. The contact angle between the MXene ink and the GO film was 82°. The optical images of the contact angles of the GO and MXene inks are shown in Figure 1g. The enhanced hydrophobicity between MXene ink and GO film can be attributed to the rougher surface of the printed GO layer in comparison to the smooth surface of the copper foil.^[29] The rougher surface of GO film comes from the microscopic wrinkles formed during the stacking and assembly process. In fact, the contact angle has no effect on the performance of the sensor but will affect the good matching of the ink and the substrate, which is crucial for controlling the line width accuracy and ensuring the ink's adhesion during printing. Too small contact angle will lead to ink halo, while too large contact angle will result in poor adhesion.

Sensing Properties of the RGMRG e-Skin

The porous structure significantly improved the sensitivity of the sensor. As shown in Figure 2a, the sensitive layer has a lower elastic modulus than the solid structure owing to the existence of pores (the stress–strain curve is shown in Figure S2, Supporting Information). Under pressure stimulation, the closure of pores generated numerous conductive paths, and the resistance decreased sharply, resulting in higher sensitivity. Figure 2b displays the current–voltage (I–V) curves of a printed single sensing unit from −1 to 1 V recorded at different static pressures. The linear relationship and stable response showed that the printed e-skin obeyed Ohm's law and that the resistance decreased with increasing pressure. Benefiting from the low elastic modulus of the porous structure, the e-skin exhibited an ultrahigh sensitivity (0.934 kPa⁻¹) in the pressure range of 0–0.6 kPa. Figure 2c illustrates the sensitivity curve for a single sensing unit. The sensitivity is defined as S = −(ΔR/R₀)/ΔP, where ΔR/R₀ is the resistance change rate under pressure, and ΔP is the applied pressure. R₀ is the initial resistance value of the sensor (4.39 kΩ). In addition, the sensitivity values of the device in the ranges of 0.6–7 and 7–50 kPa were 0.027 and 0.001 kPa⁻¹, respectively, which are consistent with its mechanism. When a low pressure was applied to the sensor, the pores in the rGO and MXene layers began to close, generating a large number of conductive paths. When the pressure increased to 0.6–7 kPa, the disorderly stacked MXene nanosheets were compressed together, and the pores were almost completely closed, whereas the rGO layer had a relatively large Young's modulus and continued to generate new conductive paths under pressure. When the pressure was greater than 7 kPa, almost all the pores in the sensitive layer were closed, and the resistance change was mainly due to the tunneling effect between the remaining untouched nanosheets. For e-skin, it is critical to ensure that each sensing unit in the array exhibits approximately the same sensing performance. Each unit in the 4 × 4 array was tested and the sensitivity curves are shown in Figure 2d. The e-skin array prepared via inkjet printing demonstrated outstanding consistency, which ensured the simplification of its calibration process in touching recognition. Figure S3, Supporting Information, displays the interference of temperature and humidity on the performance of the sensor. Thanks to the good encapsulation of the PI layer, the proposed device is not sensitive to humidity (30%–90%) and temperature (30–60 °C). At higher temperatures (60–90 °C), due to the expansion of air in the porous structure, the resistance slightly increased.

Figure 2. Electrical properties and sensing performance of the printed RGMRG e-skin. a) The schematic diagram of the sensing mechanism. b) Current–voltage (I–V) curves of a single sensing unit under various pressures. c) The sensitive curve of a single sensing unit. d) The sensitive curves of each unit of the 4 × 4 sensing array. e) Real-time relative resistance variation ratios (ΔR/R0) of the single sensing unit under different pressure (from 100 Pa to 5 kPa). f) The response and recovery times of the single sensing unit under different pressure (50, 100, and 500 Pa). g) Durability test of the single sensing unit under 10 kPa pressure for 5000 cycles with a pressure loading rate of 0.5 mm min−1.

To investigate the real-time reliability of a single sensing unit, the relative resistance changes ratios (ΔR/R₀) of four loading–unloading cycles under the pressures of 100 Pa, 500 Pa, 1 kPa, and 5 kPa are shown in Figure 2e (each loading and unloading process lasted for approximately 10 s). The value of ΔR/R₀ remained stable at each applied pressure. As the pressure increased, ΔR/R₀ also increased gradually. Furthermore, owing to the high toughness of the porous structure, the single sensing unit exhibited good reproducibility and a stable dynamic performance under different loading–unloading frequencies. As shown in Figure S4, Supporting Information, under pressure of10 kPa, the ΔR/R₀ of the RGMRG sensor was stable, and the frequencies were 0.17, 0.83, 1.6, and 3.3 Hz, respectively. Figure 2f shows that the response and recovery times were approximately 143 and 87 ms. Loading–unloading pressures of 50, 100, and 500 Pa were exerted on the sensing unit to verify the fast response property of the e-skin under different pressures. Stability and durability are critical factors for wearable sensor applications. Figure 2g shows the ΔR/R₀ for a single sensing unit at a pressure of10 kPa for 5000 cycles. The ΔR/R₀ was stable and repeatable throughout the durability test. The inset shows the detailed signal output during the test. The comparison of performance parameters with other sensors in Table 1 displays that the proposed sensor had ultrahigh sensitivity in a small range and presents a medium width detection range. In durability testing, the parameters such as testing pressure and number of cycles employed were similar to other studies, which proved that the proposed sensor could meet the durability requirements for flexible sensors in daily applications.

Table 1 Comparison of performance parameters with other sensors

Materials	Sensitivity	Detection range	Durability	References
Polydimethylsiloxane (PDMS)/Ag	0.636 kPa−1 (0–1 kPa) 0.106 kPa−1 (1–60 kPa) 0.0135 kPa−1 (60–500 kPa)	0–500 kPa	6000 (10 kPa)	[37]
PDMS/Ag	1.005 kPa−1 (0–1 kPa) 0.625 kPa−1 (1–100 kPa) 0.082 kPa−1 (100–200 kPa)	0–200 kPa	6000 (10 kPa)	[38]
multiwalled carbon nanotubes (MWCNTs)/graphene (GR)/Fe3O4/silicone rubber (SR)	8.43 (0%–120% strain) 100.56 (120%–160% strain)	0.16–160% strain	9000 (10% strain)	[39]
PDMS/Ag	1.08 N−1	0–0.7 N	100 (0.7 N)	[40]
rGO/polyacrylic acid (PAA)	0.18 kPa−1 (0–1.5 kPa); 0.023 kPa−1 (3.5–6.5 kPa)	0–6.5 kPa	2000 (6.5 kPa)	[41]
Carbon black (CB)/polyurethane (PU)	0.068 kPa−1 (0–2.3 kPa); 0.023 kPa−1 (2.3–10 kPa); 0.036 kPa−1 (10–16 kPa);	0–16 kPa	50 000 (40% strain)	[42]
CuNWs/rGO/PDMS	0.144 kPa−1	0.1–15 kPa	1000 (15 kPa)	[43]
Polypyrrole (ppy)/rGO/fabric–sponge–fabric (FSF)	0.51 kPa−1 (0–1.5 kPa) 0.026 kPa−1 (1.5–8 kPa) 0.0015 kPa−1 (8–30 kPa)	0–30 kPa	7500 (1.5 kPa)	[44]
MXene/rGO	0.934 kPa−1 (0–0.6 kPa) 0.027 kPa−1 (0.6–7 kPa) 0.001 kPa−1 (7–50 kPa)	0–50 kPa	5000 (10 kPa)	This work

Applications of the Tactile-Sensing System in Multipoint Pressing Perception and Object Shape Touching Recognition

The position mapping detection of mechanical stimuli is considered an important function of the human skin. As a simple proof of concept for this large-scale perception, Figure 3a illustrates the structural frame diagram of the signal acquisition and processing circuit (a detailed circuit diagram and resistance measurement circuit are shown in Figure S5, Supporting Information). The system consists of an amplifying/filtering circuit, a multichannel switch, an analog-to-digital converter (ADC) module, an microcontroller unit (MCU) processor, a Bluetooth module, and a host computer. After the host computer receives the signal through Bluetooth, the received signal can be displayed in real-time through the user interface designed using LabVIEW software. Figure 3b shows the excellent pressure spatial distribution perception ability of the proposed e-skin in accurately identifying the value and position of the pressure. Objects with different weights (wooden beads 0.4 g, chips 1.5 g, small magnets 9 g, and weights 50 g) were placed on different units of the e-skin, and the units illustrated different feedback. In addition, the letters “X”, “U”, and “O” were placed on the surface of each unit with small magnets (300 Pa) to test the detection ability of the pressure spatial distribution. Electrical signal changes were detected at each pixel, and the color contrast corresponding to the voltage is shown in Figure 3c. The local pressure distribution could be mapped by the color contrast, which was consistent with the shapes of the letter samples. These results indicated that this inkjet-printed array has the potential to be used to construct a large-area flexible e-skin. A more detailed demonstration can be found in Video S1–S3, Supporting Information.

Figure 3. Application of tactile-sensing system in multipoint pressing perception and object shape recognition. a) Structural frame diagram of signal acquisition and a processing circuit. b) Detection of objects with different weights and positions. c) Detection of the letters X, U, and O as well as their results. d) Photo of the signal acquisition and processing circuit. e) Schematic illustration of the convolutional neural network for object shape recognition. f) Photos of the smart glove recognition system. g) Photos of the smart glove grasping different objects. h) Classification confusion matrices of the smart glove in object shape recognition.

The e-skin can be mounted directly on latex gloves with a medical tape to map tactile information when grasping objects, such as the human palm skin (Figure 3f). A tactile-sensing system can detect synchronized signals from different sensing units, thereby mapping gestures and force distributions when grasping an object. Using the convolutional neural network (CNN) algorithm, eight objects, as shown in Figure 3g, were recognized, and the framework of the algorithm is shown in Figure 3e. The CNN contained an input layer, two adjacent CNN layers, a flattened layer, a fully connected layer, and a sigmoid layer.^[30,31] The input layer input the preprocessed data into the network after standardization and Butterworth filtering. Each CNN layer contained a convolution layer, an ReLU layer, and a batch standardization layer. The convolution kernel size of the convolution layer of the first CNN layer was 5 × 5, and that of the second CNN layer was 3 × 3 (detailed convolutional layer parameters are listed in Table S1, Supporting Information). The data input to the CNN layer was extracted through the convolution layer, further filtered through the ReLU layer, and finally output after the normalization of the batch standardization layer. The flattened layer input the data conversion size output from the CNN layer to the full-link layer, which integrated the extracted features into the sigmoid layer. The sigmoid layer classified the data and output the results. The classification confusion matrices are shown in Figure 3h. The recognition accuracy was 100% (and the training and testing processes are shown in Figure S6, Supporting Information). Video S4, Supporting Information, illustrates the pressure distribution of the smart gloves when grasping different objects.

Application of Tactile-Sensing System in Texture Touching Recognition

The perception ability of the skin differs for different parts of the human body. The skin of the fingertips, for example, can identify the surface of objects with different textures, which places higher requirements on the sensitivity and resolution of the e-skin in a very narrow pressure range. When the distance between the protrusions on the surface of an object is much smaller than the size of sensing unit, it is difficult for the original array to obtain effective information, which reduces the recognition accuracy. Inspired by the fingerprint of the finger pulp, we expanded the initial e-skin by assembling a layer of a pyramidal microstructure module made of polydimethylsiloxane as a bionic fingerprint to identify different textures. Intelligent texture recognition, such as for fingertips, can be realized using a small sensor array as the compact Meissner corpuscle in the fingertip and an expandable micro-pyramid module as a fingerprint. As shown in Figure 4a, when the expandable e-skin is attached to the surface of the object and rubbed, the compressible pyramid tip is squeezed upward along with the undulations of the surface topography, which converts the force generated by tangential friction with the object surface into normal pressure and transmits it to the array units. Although the presence of the micro-pyramid module weakened the sensitivity of individual units (Figure S7, Supporting Information), it enabled the array to obtain more detailed information. Figure 4b shows the signals collected by a single unit when the two arrays before and after expandability were rubbed on the waffle (initial pressure: 1 Pa, sliding speed: 0.05 mm s⁻¹, the sliding speed needs to match the response time of the device to obtain more friction information). An array with a micro-pyramid module can be used to distinguish more undulations on the surface of an object.

Figure 4. Applications of the tactile-sensing system in texture touching recognition. a) Schematic diagram of the structure and identification mechanism of the expandable sensing array. b) Collected signals as the e-skin rubs the Waffle before and after improvement. c) Schematic illustration of the tactile-sensing system for texture recognition. d) Schematic illustration of the long short-term memory deep-learning algorithm. e) Texture photos used in classification tasks (samples from S1 to S10 are Canvas, Waffle, Leather, Denim, Sandpaper P240, Sandpaper P600, Sandpaper P1000, Silk, Copper, and Suit, respectively). f) Classification confusion matrices of the RGMRG e-skin in texture recognition.

The schematic in Figure 4c shows the detailed process of signal acquisition and processing of the tactile perception system. The original electronic signal was low-pass filtered through a first-order Butterworth filter with a cutoff frequency of 3 Hz owing to the low sliding speed, and then processed using the Z-score normalization algorithm. The processed data were divided into training and test sets at a ratio of 8:2, which were used as the input for the long short-term memory (LSTM) model to recognize the fabric roughness. Figure 4d shows a schematic of the LSTM deep-learning algorithm.^[32–36] The input size was set to 16, which corresponds to the 16-unit data of the RGMRG array. The LSTM cell represented the LSTM network at a certain time and contained 64 neurons. The output of the last time-step LSTM cell was used as the input for the full connection layer. The units of the full connection layer were set to 64 (the effect of different training parameters on the recognition accuracy can be obtained, as shown in Figure S8, Supporting Information). The output layer completed the recognition of the 10 classes of fabric roughness (namely, Canvas, Waffle, Calf Leather, Denim, Sandpaper P240, Sandpaper P600, Sandpaper P1000, Silk, Copper, and Suit, which can be observed in Figure 4e). Through the gating mechanism, the LSTM model can selectively forget and retain information to solve the gradient disappearance or explosion problem and is suitable for the classification of sequential signals. The confusion matrix for texture recognition is shown in Figure 4f. Almost all the samples were correctly classified. Only two S7 test samples were incorrectly classified as S6, because S7 and S6 were sandpapers with similar roughness values. The overall classification accuracy was 99.7% (training and testing processes are shown in Figure S9, Supporting Information). The classification results demonstrated that the proposed method is effective for texture recognition. Informed consent was obtained from the participants who volunteered to participate in all the experiments and studies (i.e., wearable testing and image publication).

The proposed e-skin tactile-sensing system can be widely used in different application scenarios such as pressure distribution perception, shape recognition, and texture touching recognition. A functional comparison with recent studies is presented in Table 2. Most e-skin sensing systems can only realize the function of pressure distribution perception, and almost none is suitable for both shape and texture touch recognitions of the object.

Table 2 Comparison with recent e-skin tactile-sensing system

Sensor structure	Fabrication process	Pressure perceptiona)	Shape recognition	Texture recognition	Complete system	References
Array	Easy	√	×	×	Yes	[45]
Array	Complex	√	×	×	No	[46]
Array	Complex	√	×	×	–	[47]
Array	Easy	√	×	×	No	[48]
Array	Easy	√	×	×	No	[49]
Array	Middle	√	×	×	No	[50]
Single	Easy	√	×	×	–	[51]
Array	Complex	√	×	×	No	[52]
Array	Complex	√	√	×	Yes	[53]
Array	Complex	√	×	×	No	[54]
Array	Complex	√	×	√	–	[55]
Array	Middle	√	×	×	–	[56]
Array	Easy	√	√	√	Yes	This work

^a)“√” and “×” indicate whether e-skin tactile-sensing system can provide the kind of recognition or not.

Conclusion

We developed an e-skin by unevenly distributing resistive pressure-sensing units on a hand-shaped substrate to realize shape and texture recognition functions. Benefiting from the synergistic effect of the designed multilayer microporous structure by inkjet-printing and high-temperature reduction, both the single-unit and array formats exhibited a high sensitivity (0.934 kPa⁻¹ in 0–0.6 kPa) and wide detection range (0–50 kPa), which satisfied the performance requirements of shape and texture recognition functions simultaneously. The inkjet-printing method enables the design of sensor units to be interconnected into arrays of different sizes, scales, and resolutions on a continuous substrate through an all-in-one process. This simplifies the compatible fabrication and acquisition circuit and extends the potential toward free-shaped e-skin. With the help of the expandable micro-pyramid module assembled on the e-skin, the recognition accuracies were 100% for 8 different shapes and 99.7% for 10 different textures. Although advanced intelligent systems integrate only two types of tactile recognition functions, they have shown extraordinary potential for developing more functions based on bionic design strategies. In future research work, we will focus on the target to integrate more recognition functions into the e-skin tactile system and explore more specific applications such as intelligent robots, intelligent prosthetics, and virtual reality.

Experimental Section Materials

GO nanosheets with an average diameter of 0.5–5 μm and an average thickness of 4 nm and MXene nanosheets with an average diameter of 0.05 μm and an average thickness of 4 nm were purchased from XFNANO Inc. (Nanjing, China). All the materials were utilized as received without any further purification.

Preparation of the Printing Ink

The preparation processes of the printing ink are displayed in Figure S10a, Supporting Information. Under the proposed experimental conditions, GO and MXene nanosheets had excellent dispersibility due to their hydrophilicity and could be used as inks for inkjet printing without adding any additives. In addition, to match the surface energy of the substrate and the surface tension of the ink, 10 mg mL⁻¹ GO dispersion and 20 mg mL⁻¹ MXene dispersion were prepared. The specific process was as follows: 100 mg GO nanosheets were mixed in 10 mL deionized water. The mixed solution was sonicated and stirred for 1 h to obtain 10 mg mL⁻¹ GO dispersion at room temperature (20 °C). The MXene dispersion (20 mg mL⁻¹) could be achieved by the same procedure.

The Fabrication Process of the RGMRG e-Skin by Inkjet Printing

The printing processes of the RGMRG e-skin are schematically displayed in Figure S10b, Supporting Information. We demonstrated the patterning approach by printing patterns onto a 4 × 4 flexible electrode array (the size of each array was 3 × 3 mm, and the spacing between the arrays was 2 mm) which was fabricated by FPCB technology. First, the GO ink was deposited through a 180 μm nozzle. The whole printing process was realized by a microelectronic printer (Scientific 3, Shanghai Mi Fang Electronics, China), the pressure used in printing was 32 kPa, and the printing speed was 1.5 mm s⁻¹. Second, the MXene layers were printed in the same way, but with different printing parameters. Specifically, a nozzle diameter of 150 μm was used and the pressure during printing was 28 kPa. These printing parameters depended on the viscosity of the ink and were tested several times to ensure a continuous pattern could be printed. Subsequently, another GO layer was printed on the MXene layer with the same parameters. These three printed layers formed a clear and precise array of sensitive thin films on the surface of the FPCB substrate through the van der Waals forces.

After the printing process, the high-temperature reduction method was used to prepare the RGMRG thin-film array. Through heating and reducing at 300 °C for 1 h in a vacuum-drying oven, the RGMRG thin-film array was successfully prepared. Figure S10c, Supporting Information, displays the optical photos of the RGMRG e-skin. The schematic illustration of the sensors array in Figure S9c, Supporting Information, shows that the whole device includes the upper and lower PI encapsulation layers, the upper and lower electrodes prepared with FPCB technology, and the RGMRG thin film as the pressure-sensitive layer.

Characterization

The morphology of the RGMRE composite films was observed via SEM (Helios 5 CX) at an acceleration voltage of 15 kV. The mechanical measurements were accomplished by a Mark-10 mechanical testing machine (ESM303 force test stands and SERICE 5 digital force gauges). Electrical measurements were tested by a semiconductor parameter analyzer (Keithley 4200-SCS). The OMEGA 205 C340 temperature–humidity chamber was used to test the interference of temperature and humidity.

Informed consent was obtained from the participants who volunteered to perform all experiments and studies (i.e., wearable testing and image publication). All testing reported conformed to the ethical requirements of Southeast University. No animal or medical experiments were performed in this study.

Acknowledgements

The authors acknowledge the funding provided by the National Key Research and Development Program of China (grant no. 2020YFB2008502), National Natural Science Foundation of China (grant nos. 62274031, 12174050, and 12234005), Jiangsu Provincial Natural Science Foundation of China (grant no. BK20201268), the Key Research and Development Program of Jiangsu Province (grant no. BE2021007-2), and the Fundamental Research Funds for the Central Universities.

Conflict of Interest

The authors declare no conflict of interest.

Data Availability Statement

The data that support the findings of this study are available in the supplementary material of this article.