Human skin-like super-resolution tactile sensor and tactile reconstruction method

Abstract

Disclosed in the present invention are a human skin-like super-resolution tactile sensor and a tactile reconstruction method. According to a biomimetic mechanism of embedding different types of receptors in soft tissues of human skin, the sensor realizes functional biomimicry of various human receptors (such as Pacinian corpuscles and Meissner corpuscles) by embedding different types of sensing units in different arrangement modes in a soft silicone material, thereby realizing large-area and super-resolution sensing of forward pressure, shear force, and torsional force by means of a tactile reconstruction method. The characteristic of the provided sensor is to focus not only on the biomimicry of the functions of human receptors but also on the biomimicry of the functions of human skin tissues; that is, by selecting a soft silicone material having a maximum force-traction coupling effect, the same external contact can cause as many sensing units as possible to respond, thereby realizing human skin-like large-area and super-resolution tactile sensing.

Description

A human skin-like super-resolution tactile sensor and tactile reconstruction method Technical Field

[0001] This invention relates to the field of robot tactile perception, and more particularly to a human skin-like super-resolution tactile sensor and a tactile reconstruction method. Background Technology

[0002] As a sensory modality parallel to vision and hearing, tactile perception provides robots with irreplaceable and rich information about the contacted environment, such as the shape, surface texture, roughness, temperature, and humidity of the objects being touched; the location and magnitude of contact force between the robot and the external object; and whether the robot's fingers slip during grasping. Tactile information plays a crucial role in tasks requiring direct physical contact with the external environment, such as robot grasping and human-robot interaction.

[0003] Currently, to acquire tactile information, various tactile sensors based on different principles have been developed in the robotics field, mainly including piezoresistive, capacitive, photoelectric, and visual-tactile sensors. The fundamental principle of piezoresistive tactile sensors is based on the change in resistivity of single-crystal silicon material when subjected to external pressure. This change can be converted into a pressure change through calibration techniques. However, it cannot achieve large-area flexible adhesion and cannot directly and accurately measure shear force. Capacitive pressure sensors typically consist of two parallel electrodes. When the sensor receives external pressure, deformation causes a change in capacitance, and the pressure magnitude is sensed by detecting this change in capacitance. However, its anti-interference capability is poor, and it is difficult to accurately measure shear and torsional forces. Piezoelectric sensors are based on the piezoelectric effect, where positive and negative charges are generated on opposite sides of a piezoelectric material when subjected to pressure, and the amount of charge generated is positively correlated with the applied pressure. By detecting the voltage values ​​on both sides of the material, the relationship between voltage and external pressure can be established. Although piezoelectric sensors have high sensitivity, their manufacturing cost and complexity are also higher. They cannot detect static pressure, and vibration and high-frequency noise can significantly affect them. The aforementioned tactile sensors typically simulate the function of a single receptor in the skin. Therefore, to improve spatial resolution performance, traditional methods often involve reducing the size of the sensing unit or increasing the density of the sensor array. However, traditional methods can never surpass the physical resolution of the array and cannot achieve the infinite resolution of human skin.

[0004] Currently, visual-tactile sensors, which are attracting more attention from scholars and researchers, infer contact force information by designing markers on the surface of a soft, elastic layer and capturing the movement of these markers with a camera. The development of visual-tactile sensors requires utilizing the visual imaging technology and data processing methods of miniature cameras to increase the measurement dimensions of tactile information. They offer superior performance, possessing the ability to perceive complex tactile forces such as shear and slip forces, as well as texture recognition capabilities. However, their drawbacks are also significant, such as difficulty in adapting to different surface shapes and challenges in miniaturization, making them unsuitable for large-area deployment. Summary of the Invention

[0005] The purpose of this invention is to address the problems of insufficient spatial resolution of current tactile sensors, the reliance on increasing sensor array density to improve resolution, and the difficulty in miniaturization, by proposing a human skin-like super-resolution tactile sensor and a tactile reconstruction method.

[0006] The objective of this invention is achieved through the following technical solution: a human skin-like super-resolution tactile sensor, which is assembled from a top-to-bottom soft silicone layer, a bottom bonding layer, and a rigid outer shell layer;

[0007] The soft silicone layer contains a spatially arranged sensor array, and the bottom bonding layer includes an array of sensor units, which are horizontally attached to the inner bottom surface of the rigid outer shell layer.

[0008] A soft silicone layer with a spatially arranged sensor array embedded in it is placed in a rigid shell with a horizontal array of sensor units attached.

[0009] Furthermore, within the soft silicone layer, the spatial arrangement of the sensor array includes a three-dimensional geometric arrangement pattern in the form of a biomimetic four-leaf clover, petal shape, or ring shape.

[0010] The biomimetic four-leaf clover arrangement includes: four flexible thin-film pressure sensors arranged at an angle around the four sides of a sensing point, with each flexible thin-film pressure sensor slightly higher near the center of the sensing point and slightly lower further away.

[0011] The biomimetic petal-shaped arrangement includes: arranging eight flexible thin-film pressure sensors around a sensing point, with four arranged in a biomimetic four-leaf clover pattern, and the other four embedded at the four apex corners of a rectangle formed by the aforementioned four-leaf clover pattern. The embedding method is as follows: first, the direction of the flexible film is aligned with the direction surrounding the sensing point; then, it is tilted, with the tilt direction of a group of flexible films symmetrical about the center of the sensing point remaining consistent.

[0012] The circular arrangement pattern includes: eight flexible thin-film pressure sensors are arranged at an angle around a sensing point, with equal spacing between each flexible thin-film pressure sensor, forming a circular arrangement in a clockwise or counterclockwise direction.

[0013] Furthermore, the soft silicone layer has the thickness and Shore hardness for optimal force traction coupling effect. By applying a fixed normal force to the soft silicone layer, the maximum distance that the pressure unit can generate is selected as an indicator.

[0014] On the other hand, this specification also provides a tactile reconstruction method based on the aforementioned sensor, the method comprising: normal force reconstruction achieved by a horizontal array of sensors in the bottom bonding layer, shear force reconstruction achieved by a sensor array arranged in a spatial structure, and torsional force reconstruction.

[0015] Furthermore, the normal force reconstruction includes: the normal force forming a binary Gaussian pressure distribution along the surface of the soft silicone layer.

[0016] Where x and y represent the positions below the surface coordinate plane of the tactile sensor, μ x and μ y σ represents the mean in the x and y directions. x and σ y ρ represents the standard deviation in the x and y directions, and ρ represents the correlation between x and y.

[0017] Let σ x =σ y If σ = 0 and ρ = 0, then the pressure distribution simplifies to the following form:

[0018] From the pressure distribution model above, we can obtain that the coordinates of the contact center of the normal force are (μ x ,μ y The normal pressure magnitude is When a normal force is applied to the surface of the soft silicone layer, the optimal force-coupled effect will cause at least two sensing units in the bottom pressure sensing layer to generate coupling response signals. The microcontroller reads the coupling response signals and sends them to the computer via serial port. The pressure change curve is displayed in real time on the computer. Based on the received coupling response signals, the least squares method is used to fit the optimal Gaussian distribution parameters, transforming the problem of estimating the magnitude and position of the normal force into an optimization problem with minimum error.

[0019] Based on the optimal parameters obtained through optimization, the contact position coordinates (μ) are displayed in real time through visualization processing. x ,μ y Simultaneously, the contact simulation effect is drawn.

[0020] Furthermore, the shear force reconstruction process includes shear force reconstruction of a biomimetic four-leaf clover array, a biomimetic petal array, and a ring-shaped array. The shear force reconstruction methods for different sensor arrays are as follows:

[0021] A shear force along a certain direction is applied to the surface of a soft silicone layer, and the embedded sensor array generates a response signal. Based on the fixed position and fixed angle of the pressure sensing unit that generates the response signal, a supervised learning method with manual labeling is used, combined with a multilayer perceptron to learn the nonlinear relationship between the electrical signal of the sensor array and the actual force. Through large-scale data acquisition, the regularity and similarity features are extracted, and a shear force reconstruction model is trained. The magnitude and direction of the shear force are obtained using the trained model.

[0022] Furthermore, the torsional force reconstruction includes torsional force reconstruction of a biomimetic four-leaf clover array, a biomimetic petal array, and a ring-shaped array. The shear force reconstruction methods for different sensor arrays are all:

[0023] When a torsional force is applied to a point on the surface of the soft silicone layer, the torsional force radiates outwards through the soft silicone layer, which has the best force traction coupling effect. The three-dimensional pressure sensor array embedded in the internal layer generates regular response signals. Using a supervised learning method with human labels, combined with a multilayer perceptron to learn the nonlinear relationship between the electrical signals of the sensor array and the actual force, through large-scale data acquisition, the regularity and similarity features are extracted to train a torsional force reconstruction model. The magnitude and direction of the torsional force are obtained using the trained model.

[0024] Furthermore, the supervised learning method with manual labeling uses mean squared error as a loss function to measure the difference between the predicted value and the actual value.

[0025] The beneficial effects of this invention are:

[0026] 1. This invention features a simple structure, small size, and light weight, enabling its application in tactile sensing in grippers, robotic arms, and other applications, thereby improving operational safety and reliability. Through a specific spatial arrangement, this invention can effectively respond to shear forces parallel to the contact surface, simulating the function of human skin soft tissue. This surpasses the physical resolution capabilities of sensor arrays, achieving human-skin-like super-resolution tactile sensing. This invention achieves the perception of multi-dimensional force information through a simple and effective combination of different sensor arrangements. The surface structure of the soft silicone layer can be adjusted according to the actual contact surface requirements, adapting to different needs. This invention can reproduce the contact force state information in real time within a visualization window, improving the user's intuitive perception of the contact state. Attached Figure Description

[0027] Figure 1 is a schematic diagram of the structure of a human skin-like large-area, super-resolution tactile sensor proposed in this invention;

[0028] Figure 2 shows the arrangement patterns of the horizontal array, the biomimetic four-leaf clover shape, the biomimetic petal shape, and the circular shape of the present invention;

[0029] Figure 3 is a data acquisition curve diagram in a specific example of the present invention;

[0030] Figure 4 is a visualization interface of the contact position center estimated by the tactile reconstruction algorithm of the present invention;

[0031] Figure 5 shows the effect of the contact force estimated by the tactile reconstruction algorithm of the present invention;

[0032] Figure 6 shows the representative response curves when shear force is applied to the biomimetic four-leaf clover arrangement pattern;

[0033] Figure 7 shows representative curves when shear force is applied to the biomimetic petal arrangement pattern;

[0034] Figure 8 shows a representative curve when a torsional force is applied to a circular arrangement pattern. Detailed Implementation

[0035] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0036] This invention draws inspiration from the biomimetic mechanism of human skin embedding different types of receptors in soft tissue. By embedding different types and arrangement patterns of sensing units in soft silicone material, it achieves functional biomimicry of various human receptors (such as Pacinian bodies, Messner bodies, etc.). Furthermore, through a tactile reconstruction algorithm based on physical models and machine learning, it achieves large-area and super-resolution perception of positive pressure, shear force, and torsional force.

[0037] As shown in Figure 1, the present invention provides a human skin-like super-resolution tactile sensor, which is assembled by a soft silicone layer 1, a pressure sensing layer 2 and a rigid outer shell layer 3.

[0038] The pressure sensing layer 2 includes a bottom bonding layer 4 and an internal embedding layer 5 within a soft silicone layer.

[0039] The internal embedded layer 5 is a sensor array with a spatial structure, and the bottom bonding layer 4 includes an array of sensor units, which are horizontally attached to the inner bottom surface of the rigid outer shell layer 3.

[0040] A soft silicone layer with a spatially arranged sensor array embedded in it is placed in a rigid shell with a horizontal array of sensor units attached.

[0041] Inside the soft silicone layer, the spatial structure arrangement of the sensor array includes: a three-dimensional geometric arrangement pattern with biomimetic four-leaf clover, petal shape, or ring shape.

[0042] The biomimetic four-leaf clover arrangement includes: four flexible thin-film pressure sensors arranged at an angle around the four sides of a sensing point, with each flexible thin-film pressure sensor slightly higher near the center of the sensing point and slightly lower further away.

[0043] The biomimetic petal-shaped arrangement includes: arranging eight flexible thin-film pressure sensors around a sensing point, with four arranged in a biomimetic four-leaf clover pattern, and the other four embedded at the four apex corners of a rectangle formed by the aforementioned four-leaf clover pattern. The embedding method is as follows: first, the direction of the flexible film is aligned with the direction surrounding the sensing point; then, it is tilted, with the tilt direction of a group of flexible films symmetrical about the center of the sensing point remaining consistent.

[0044] The circular arrangement pattern includes: eight flexible thin-film pressure sensors are arranged at an angle around a sensing point, with equal spacing between each flexible thin-film pressure sensor, forming a circular arrangement in a clockwise or counterclockwise direction.

[0045] The soft silicone layer 1 is obtained by mixing commercially available food-grade silicone A and B in a 1:1 mass or volume ratio, stirring until homogeneous, and then directly pouring the mixture into a mold for curing. The pressure sensing unit in the internal embedded layer 5 is embedded in the liquid silicone layer according to a spatial arrangement before the liquid silicone cures. The shape of the soft silicone layer 1 can be designed and cast according to actual needs; for example, its shape can be designed to the dimensions of a given robotic arm.

[0046] To utilize the soft silicone layer 1 with optimal force-traction coupling effect, the relevant parameters of the prepared soft silicone layer 1, including but not limited to thickness and Shore hardness, need to be determined before designing the casting model. For soft silicone layers 1 with different thicknesses and Shore hardnesses, a pressure sensing unit is placed at the bottom, and a fixed normal force is applied to the soft silicone layer 1. The maximum distance that causes the pressure sensing unit to respond is used as the evaluation index to select the thickness and Shore hardness of the soft silicone layer 1 with optimal force-traction coupling effect.

[0047] The bottom bonding layer 4 is composed of a 3×3 array of sensing units and is directly and horizontally attached to the inner bottom surface of the rigid outer shell layer 3. The rigid outer shell is 3D printed. Subsequently, a soft silicone layer 1, which has been cured and embedded with a spatially arranged sensing array, is placed in the rigid outer shell with the horizontal array of sensing units attached. Each sensing unit is then connected to the pressure conversion module via a wired connection, and then connected to a microcontroller for data reading and processing.

[0048] Once the hardware fabrication process is complete, super-resolution, multi-dimensional force perception can be achieved through tactile reconstruction methods.

[0049] Based on the aforementioned embodiment of a human skin-like super-resolution tactile sensor, the present invention also provides an embodiment of a tactile reconstruction method. This method reconstructs normal force through a horizontal array of sensors in the bottom bonding layer 4, and reconstructs shear force and torsional force through a sensor array arranged in a spatial structure.

[0050] Part 1: Taking a 3×3 horizontal array as an example, the magnitude and position of the normal force are estimated using a binary Gaussian physical model.

[0051] The implementation process is as follows:

[0052] The normal force forms a binary Gaussian pressure distribution along the surface of the soft silicone layer 1:

[0053] Where x and y represent the positions below the surface coordinate plane of the tactile sensor, μ x and μ y σ represents the mean in the x and y directions. x and σ y Let σ represent the standard deviation in the x and y directions, and ρ represent the correlation between x and y. Based on the isotropic nature of silicone material, and also to simplify the model, let σ... x =σ y If σ = 0 and ρ = 0, then the pressure distribution simplifies to the following form:

[0054] From the pressure distribution model above, we can obtain that the coordinates of the contact center of the normal force are (μ x ,μ y The normal pressure magnitude is When a normal force is applied to the surface of the soft silicone layer 1, the bottom bonding layer 4 will generate a coupling response signal due to the optimal force-traction coupling effect. That is, for the same normal contact force, at least two sensing units will generate response signals. The coupling signal is read by a microcontroller and sent to a computer via a serial port, where the pressure change curve is displayed in real time, as shown in Figure 3. The sensor design uses a total of nine channels, and the data from these nine channels is transmitted in real time. Based on the received data from the nine channels, the optimal Gaussian distribution parameters are fitted using the least squares method, transforming the problem of estimating the magnitude and position of the normal force into an optimization problem with minimal error.

[0055] in, The parameter represents the actual value of the normal force measured in the i-th measurement, P(x i ,y i) is the estimated value of the i-th normal force, and Err(,) represents the sum of squares of the errors between the actual and estimated values ​​of the normal force. Based on the optimal parameters obtained through optimization, the contact position coordinates (μ) are displayed in real time through visualization processing. x ,μ y Simultaneously, the contact simulation effect is drawn, as shown in Figures 4 and 5.

[0056] Part Two: Estimation of the magnitude and direction of shear and torsional forces through a spatially arranged sensor array.

[0057] The implementation process is as follows:

[0058] First, when a shear force along a certain direction is applied to the surface of the soft silicone layer 1, this shear force will cause the soft silicone layer 1 to deform, and under the action of traction coupling effect, the shear force will be transmitted to the pressure sensing unit of the internal embedded layer 5, thereby generating a response signal. Therefore, shear forces of different directions and magnitudes will cause pressure sensing units with different positions and angles to generate different but regular response signals. Based on the fixed position and fixed angle of the pressure sensing unit that generates the response signal, combined with the multilayer perceptron (MLP) method, the magnitude and direction of the shear force can be obtained through large-scale data acquisition. For example, for a sensor array arranged in a biomimetic four-leaf clover pattern, when a shear force is applied along the surface of the soft silicone layer 1 at a 45-degree angle to the upper right, a response curve as shown in Figure 6 will be generated; for a sensor array arranged in a biomimetic petal pattern, when a shear force is applied along the surface of the soft silicone layer 1 at a 45-degree angle to the upper right, a response curve as shown in Figure 7 will be generated. From the response curves of different arrangement patterns, it can be found that for the same shear force, the generated response curve signals have obvious regularity, and the curves of the coupled response sensing units are similar. By using supervised learning methods with human labels, and through large-scale data collection, regularity and similarity features can be extracted, thereby enabling shear force detection along the entire surface of the tactile sensor.

[0059] Secondly, when a torsional force is applied to a point on the surface of the soft silicone layer 1, the torsional force radiates outwards through the soft silicone layer 1, which has the best force-traction coupling effect. This causes the three-dimensional pressure sensing array of the internal embedded layer 5 to generate a regular response signal. A circularly arranged sensor array is more suitable for torsional force reconstruction. For example, as shown in Figure 8, when a torsional force is applied to the lower left of the center of the array, the sensing units near the contact center will respond, showing that multiple response curves have basically the same trend, with differences in amplitude, indicating similarity. Similarly, using labeled data (response curves, sensor positions, sensor arrangement angles, etc.) as input, the output model of the tactile sensor is obtained through a multilayer perceptron (MLP) method. This allows for the estimation of the magnitude and location of the torsional force.

[0060] Constructing a Gaussian-like physical model for a spatially arranged sensor array is extremely difficult. Therefore, based on the general approximation theorem, a multilayer perceptron (MLP) is chosen to obtain the mapping relationship between sensor electrical signals and actual force information. The MLP is a typical feedforward artificial neural network, consisting of an input layer, several hidden layers, and an output layer. Each layer comprises several neurons connected by weighted sums. MLPs possess powerful nonlinear modeling capabilities, effectively handling complex input-output mapping problems. The number of nodes in the input layer corresponds to the sampling data dimension of the sensor array, i.e., time t. i The response value F of each sensor i Location (x) i ,y i ) and angle θ i The system employs two hidden layers: the first layer contains 64 neurons, and the second layer contains 16 neurons. The activation function is the sigmoid function. The output layer contains two dimensions of force information: magnitude and direction. Mean squared error (MSE) is used as the loss function to measure the difference between the predicted and actual values. Through this process, the MLP network can learn the complex nonlinear relationship between the electrical signals of the sensor array and the actual force, thereby providing accurate force estimation for tactile sensors and significantly improving the accuracy and reliability of the sensor system, especially in complex multidimensional force coupling environments.

[0061] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0062] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this application. This application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A human skin-like super-resolution tactile sensor, characterized in that, The sensor is assembled from a top-to-bottom soft silicone layer, a bottom bonding layer, and a rigid outer shell layer; The soft silicone layer contains a spatially arranged sensor array, and the bottom bonding layer includes an array of sensor units, which are horizontally attached to the inner bottom surface of the rigid outer shell layer. A soft silicone layer with a spatially arranged sensor array embedded in it is placed in a rigid shell with a horizontal array of sensor units attached.

2. The human skin-like super-resolution tactile sensor according to claim 1, characterized in that, Inside the soft silicone layer, the spatial structure arrangement of the sensor array includes: a three-dimensional geometric arrangement pattern with biomimetic four-leaf clover, petal shape, or ring shape. The biomimetic four-leaf clover arrangement includes: four flexible thin-film pressure sensors arranged at an angle around the four sides of a sensing point, with each flexible thin-film pressure sensor slightly higher near the center of the sensing point and slightly lower further away. The biomimetic petal-shaped arrangement includes: arranging eight flexible thin-film pressure sensors around a sensing point, with four arranged in a biomimetic four-leaf clover pattern, and the other four embedded at the four apex corners of a rectangle formed by the aforementioned four-leaf clover pattern. The embedding method is as follows: first, the direction of the flexible film is aligned with the direction surrounding the sensing point; then, it is tilted, with the tilt direction of a group of flexible films symmetrical about the center of the sensing point remaining consistent. The circular arrangement pattern includes: eight flexible thin-film pressure sensors are arranged at an angle around a sensing point, with equal spacing between each flexible thin-film pressure sensor, forming a circular arrangement in a clockwise or counterclockwise direction.

3. The human skin-like super-resolution tactile sensor according to claim 1, characterized in that, The soft silicone layer has the optimal thickness and Shore hardness for the force traction coupling effect. By applying a fixed normal force to the soft silicone layer, the maximum distance that the pressure unit can generate is selected as an indicator.

4. A tactile reconstruction method based on the sensor described in any one of claims 1-3, characterized in that, The method includes: normal force reconstruction achieved by a horizontal array of sensors in the bottom bonding layer, shear force reconstruction and torsional force reconstruction achieved by a sensor array arranged in a spatial structure.

5. The tactile reconstruction method according to claim 4, characterized in that, The normal force reconstruction includes: the normal force forming a binary Gaussian pressure distribution along the surface of the soft silicone layer. Where x and y represent the positions below the surface coordinate plane of the tactile sensor, μ x and μ y σ represents the mean in the x and y directions. x and σ y represents the standard deviation in the x and y directions, and p represents the correlation between x and y; Let σ x =σ y If σ = 0 and ρ = 0, then the pressure distribution simplifies to the following form: From the pressure distribution model above, we can obtain that the coordinates of the contact center of the normal force are (μ x ,μ y The normal pressure magnitude is When a normal force is applied to the surface of the soft silicone layer, the optimal force-traction coupling effect causes at least two sensing units in the bottom bonding layer to generate coupling response signals. The coupling response signals are read by a microcontroller and sent to a computer via a serial port. The pressure change curve is displayed in real time on the computer. Based on the received coupling response signals, the optimal Gaussian distribution parameters are fitted using the least squares method, transforming the problem of estimating the magnitude and position of the normal force into an optimization problem with minimum error. in, The parameter represents the actual value of the normal force measured in the i-th measurement, P(x i ,y i ) is the estimated value of the i-th normal force, and Err(,) represents the sum of squares of the errors between the actual and estimated values ​​of the normal force. Based on the optimal parameters obtained through optimization, the contact position coordinates (μ) are displayed in real time through visualization processing. x ,μ y Simultaneously, the contact simulation effect is drawn.

6. The tactile reconstruction method according to claim 4, characterized in that, The shear force reconstruction process includes shear force reconstruction of a biomimetic four-leaf clover array, a biomimetic petal array, and a circular array. The shear force reconstruction method for different sensor arrays is as follows: A shear force along a certain direction is applied to the surface of a soft silicone layer, and the embedded sensor array generates a response signal. Based on the fixed position and fixed angle of the pressure sensing unit that generates the response signal, a supervised learning method with manual labeling is used, combined with a multilayer perceptron to learn the nonlinear relationship between the electrical signal of the sensor array and the actual force. Through large-scale data acquisition, the regularity and similarity features are extracted, and a shear force reconstruction model is trained. The magnitude and direction of the shear force are obtained using the trained model.

7. The tactile reconstruction method according to claim 6, characterized in that, The torsional force reconstruction includes torsional force reconstruction of a biomimetic four-leaf clover array, a biomimetic petal array, and a circular array. The shear force reconstruction method for different sensor arrays is as follows: When a torsional force is applied to a point on the surface of the soft silicone layer, the torsional force radiates outwards through the soft silicone layer, which has the best force traction coupling effect. The three-dimensional pressure sensor array embedded in the internal layer generates regular response signals. Using a supervised learning method with human labels, combined with a multilayer perceptron to learn the nonlinear relationship between the electrical signals of the sensor array and the actual force, through large-scale data acquisition, the regularity and similarity features are extracted to train a torsional force reconstruction model. The magnitude and direction of the torsional force are obtained using the trained model.

8. The tactile reconstruction method according to claim 6, characterized in that, The supervised learning method that uses manual labeling employs mean squared error as a loss function to measure the difference between predicted and actual values.

Images