Pressure calculation method based on integration of spherical / arc surface structure and flexible pressure array

Abstract

Disclosed in the present invention is a pressure calculation method based on integration of a spherical / arc surface structure and a flexible pressure array. The method comprises: collecting pressure array information of interaction between a pressing end and a target, wherein the pressure array information reflects sensed pressures that are in different directions of a three-dimensional space and generated when the pressing end comes into contact with the target; inputting the pressure array information into a trained neural network model, so as to obtain a normal force and a tangential force of a corresponding contact surface; and controlling a subsequent operation of the pressing end on the basis of the normal force and the tangential force. The present invention can establish an intrinsic relationship between a three-dimensional force and a pressure sensing array, thereby realizing accurate calculation of a three-dimensional pressing force that is based on flexible pressure array information.

Description

Pressure calculation method based on the fusion of spherical arc surface structure and flexible pressure array Technical Field

[0001] The present invention relates to the field of pressure analysis technology, and more specifically, to a pressure calculation method based on the fusion of a spherical arc surface structure and a flexible pressure array. Background Art

[0002] With the development of automation and artificial intelligence technologies, robots for various medical, service, industrial process and other application fields have been widely studied and applied. Force sensors, as important "senses" for robots to obtain tactile information, play an important role in the process of robot intelligence, especially in some application fields that require high accuracy of robot force control and force perception (such as rehabilitation and physiotherapy). It is necessary to accurately obtain the normal force and tangential force of the control end to ensure the safety and effectiveness of the operation. Pressing robots have many application scenarios, such as pressing to stop bleeding, pressing acupoints in traditional Chinese medicine, and muscle massage and relaxation. These scenarios require accurate perception of the normal vector and tangential vector of the force, as well as the corresponding pressing force, and feedback control.

[0003] Currently, methods for sensing three-dimensional forces on pressing robots primarily include force / torque sensors, visual force sensors, and pressure sensors. Force / torque sensors sense external forces acting on the robot by measuring the forces and moments in the robot's joints. Visual force sensors use cameras to detect the displacement and deformation of objects to infer the magnitude and direction of external forces. Pressure sensors sense the magnitude and direction of external forces by measuring the pressure distribution when the robot contacts the environment.

[0004] However, existing force sensing methods cannot meet the application requirements of high precision and precise control of pressure force in complex environments. Force / torque sensors are large in size, which makes them inconvenient to integrate into terminal devices with limited space. They are also expensive and require calibration. Whether integrated internally or externally, visual and force sensors will be affected by the end-point blocking the contact object or ambient light. Pressure sensors are an effective way to achieve unobstructed direct perception of contact force in a limited space at the end. The use of high-density flexible pressure array sensors can fit the end well and ensure accurate measurement. However, due to the simple sensing principle, flexible pressure sensors cannot sense normal and tangential forces.

[0005] In the prior art, patent application CN202210690911.6 provides a decoupling analysis method for an array tactile sensor. This method determines the coupling relationship of each tactile unit through the array position relationship and data signal, and then determines the coupling relationship between each tactile unit based on the applied force and output force. This method then determines the input-output relationship of the array sensor under coupling, constructs a compensating decoupling function based on the detection error, decouples the coupling relationship, and determines the actual output force of the sensor. However, as the number of sensing units in the array sensor increases, the complexity of this solution increases dramatically, and the accumulated error affects the measurement accuracy.

[0006] Patent application CN202011490592.1 provides a neural network-based force-position tracking algorithm for a 5-DOF robotic arm. This algorithm establishes a link coordinate system by defining various parameters in the robotic arm, obtains the end contact force through a force sensor, and establishes a mathematical model between the end contact force and the robotic arm's reference coordinate system, describing the matrix of the end contact force in the reference coordinate system. However, this solution can only sense one-dimensional force conditions, and the indirect method of obtaining the force through coordinate transformation is relatively cumbersome.

[0007] In summary, existing three-dimensional pressure sensors are bulky and inconvenient to integrate into space-constrained press terminals. High-density flexible pressure (tactile) array sensors, on the other hand, have a simple sensing principle and can only sense a single dimension of force, failing to meet the application requirements of press robots that require simultaneous sensing of both normal and tangential forces.

[0008] Summary of the Invention

[0009] The purpose of the present invention is to overcome the above-mentioned defects of the prior art and provide a pressure calculation method based on the fusion of a spherical arc surface structure and a flexible pressure array, which is particularly suitable for the calculation of three-dimensional pressing force.

[0010] The technical solution of the present invention is:

[0011] Compared with existing technologies, the present invention has the advantage of proposing a three-dimensional force calculation method based on a spherical arc surface structure of a pressing end and a high-density flexible pressure array sensor. First, by fusing the pressing end's spherical / arc surface structure design with a flexible high-density force tactile sensor array, one-dimensional tactile force information is converted from a planar space to a three-dimensional space. Furthermore, the one-dimensional pressure distribution information sensed by the spherical arc surface pressing end of the integrated flexible tactile pressure array sensor is fused with the three-dimensional spatial structure information of the pressing end, and a convolutional neural network model is constructed to calculate the normal and tangential forces of the pressing end. This invention improves the accuracy and efficiency of calculating pressing force.

[0012] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0014] FIG1 is a flow chart of a pressure calculation method based on the fusion of a spherical arc surface structure and a flexible pressure array according to one embodiment of the present invention;

[0015] FIG2 is a schematic diagram of a normal force and a tangential force calculation process according to an embodiment of the present invention;

[0016] FIG3 is a schematic diagram of the design of a curved and spherical pressing end according to an embodiment of the present invention;

[0017] FIG4 is a schematic diagram of an end spherical structure and a high-density flexible pressure array sensor according to an embodiment of the present invention;

[0018] FIG5 is a schematic diagram of a data set collection method according to an embodiment of the present invention;

[0019] FIG6 is a schematic diagram of a neural network model structure according to an embodiment of the present invention;

[0020] FIG. 7 is a heat map of a high-density flexible pressure array sensor according to one embodiment of the present invention. DETAILED DESCRIPTION

[0021] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention.

[0022] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0023] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0024] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0025] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0026] This invention designs a spherical arc-shaped structure for pressing endpieces, a flexible pressure sensor array, and an array-force-to-three-dimensional force mapping scheme for three-dimensional pressing force sensing and calculation. In short, the proposed three-dimensional force calculation method based on the spherical structure of the pressing endpiece and a high-density flexible pressure sensor array uses the spherical structure of the endpiece covered with a high-density pressure sensor array to collect complete interactive force distribution information, namely a force matrix image. This is then used to establish a neural network model to convert the high-dimensional spherical sensing force into normal and tangential forces on the contact surface.

[0027] Specifically, in conjunction with FIG1 and FIG2 , the pressure calculation method based on the fusion of the spherical arc surface structure and the flexible pressure array includes the following steps:

[0028] Step S110 , designing a spherical arc surface pressing structure to sense the distribution of external forces in different directions.

[0029] In order to facilitate the combination of the pressure array film and the pressing end structure, a spherical / arc surface end pressing structure is designed, which can ensure the close combination of the pressure array film and the pressing end structure without causing folding and deformation of the pressure array film. The design of the spherical and arc surface pressing ends is shown in Figure 3.

[0030] First, the arc / spherical structure can evenly distribute external force without concentrating it on a certain point, and the surface of the arc / spherical structure can sense forces from different directions, so it can better sense the magnitude and direction of external forces. In addition, flexible array sensing has high precision and high sensitivity, can be flexibly applied to a variety of scenarios, and obtain high-dimensional tactile information. The combination of arc / spherical structure and flexible array sensor can have better fit and surface contact, thereby improving measurement accuracy, and can also enhance the reliability of the structure, disperse stress and load, thereby increasing the life of the pressing structure and sensor.

[0031] Figure 4 shows a physical image of the spherical structure of the pressure tip and the high-density flexible pressure sensor array. This spherical design more easily reflects the external force generated when the pressure tip contacts an object, enabling the perception of pressure in different directions in three-dimensional space. The integration of high-density flexible pressure sensors into the spherical structure of the tip enables the direct acquisition of high-resolution pressure distribution images.

[0032] Step S120 , using the designed pressing structure to collect a sample data set and train a neural network model, wherein the sample data set reflects the correspondence between the three-dimensional pressure array information and the normal force and the tangential force.

[0033] First, a sample dataset for training and testing is constructed. The dataset acquisition equipment is shown in Figure 5. The data acquisition steps are as follows: two identical robotic arms are used, one with a high-density flexible array sensor covered in a spherical structure at the end, and the other with an existing three-dimensional force sensor at the end as a gold standard; the two robotic arms are controlled to perform pressing actions in the same way at the same time, where the pressure array information obtained by the pressure array sensor is used as the input quantity X_i of the i-th sample in the dataset, and the normal force F obtained by the three-dimensional force sensor is used as the input quantity X_i of the i-th sample in the dataset. n _i and tangential force F x _i、F y _i is the label Yi of the i-th sample in the dataset.

[0034] In addition, data augmentation can be used to expand the number of samples in the dataset. For example, based on a small dataset, mirroring or rotation operations can be used to expand the number of samples, thereby improving the generalization and robustness of model training.

[0035] The neural network model can be of various types, such as a convolutional neural network or a recurrent neural network. For example, a neural network regression model is used to explore the intrinsic relationship between pressure array information and three-dimensional force.

[0036] In one embodiment, the designed convolutional neural network (CNN) is used for binary regression task, with the input being the pressure array information of size 18x18x1 and the output being a vector of size 3x1x1, namely the normal force F n and tangential force F x 、F y As shown in Figure 6, the convolutional neural network generally includes a convolutional layer, a pooling layer, a flatten layer, a fully connected layer, and an output layer. This model can achieve end-to-end feature extraction and learn more complex nonlinear mapping relationships.

[0037] Specifically, Convolutional Layer 1 uses 16 3x3 convolution kernels with a stride of 1 to extract features from the input pressure array, producing an output feature map of size 16x16x16. Pooling Layer 1 downsamples the output of Convolutional Layer 1, for example, using max pooling with a size of 2x2 and a stride of 2, resulting in an output feature map of size 8x8x16. Convolutional Layer 2 uses 32 3x3 convolution kernels with a stride of 1 to extract features from the output feature map of Pooling Layer 1, producing an output feature map of size 6x6x32. Pooling Layer 2 downsamples the output of Convolutional Layer 2, for example, using max pooling with a size of 2x2 and a stride of 2, producing an output feature map of size 3x3x32. The Flatten Layer expands the output of Pooling Layer 2 into a one-dimensional vector of size 288x1x1. Fully Connected Layer 1 linearly combines this expanded vector with some weights to produce a smaller vector of size 64x1x1. The fully connected layer 2 again linearly combines the output vector of the previous layer with some weights to produce a final output vector of size 3x1x1. The convolutional layer, the fully connected layer, and the output layer can all use the rectified linear unit (ReLU) as the activation function. The purpose of this CNN is to map the 18x18x1 pressure array information to an output vector of size 3x1x1. By adjusting the weight parameters, the CNN can better regress and predict the input pressure array information. By integrating the convolutional neural network with the regression task into a deep learning network model, the calculation of the normal force and tangential force can be directly obtained from the pressure array information.

[0038] The pressure array information obtained by the pressure array sensor is used as the input quantity Xi, and the normal force F obtained by the three-dimensional force sensor is used as the gold standard. n _i and tangential force F x _i、F y _i is used as the label Yi, and the output of the neural network regression model is used as the predicted value Explore the intrinsic connection between pressure array information and three-dimensional forces.

[0039] There are also many types of loss functions for training neural network models, such as mean square error function, cross entropy loss function, weighted cross entropy loss function, etc.

[0040] In one embodiment, a mean squared error (MSE) function is used as a loss function to measure the gap between the model prediction value and the true value. Y_i={F n _i F x _i F y _i} (2)

[0041] The training data is fed into the model for prediction. The mean squared error between the predicted result and the true value is calculated. The model parameters are then updated using the backpropagation algorithm to gradually reduce the mean squared error. In this way, the model is continuously optimized and its regression prediction ability is improved.

[0042] To further validate the neural network model's ability to mine the intrinsic connection between pressure array information and three-dimensional forces, a three-dimensional force regression task test was conducted. As shown in Figure 7, the pressure array information contains complex and high-dimensional feature information. Comparing Figure 7(a) and Figure 7(b), it can be seen that the increase in contact area in Figure 7(b) compared to Figure 7(a) is due to the increase in force in the normal direction, but the precise normal force value cannot be determined. Comparing Figure 7(a) with Figure 7(c) and Figure 7(d), it can be seen that Figure 7(c) has a certain amount of offset in the Y-axis direction compared to Figure 7(a), and Figure 7(d) has a certain amount of offset in the X-axis direction compared to Figure 7(c). This distribution of pressure array information indicates that a tangential force is generated at the end, but the precise tangential force value cannot be determined. It has been verified that by utilizing the neural network model, normal and tangential force information can be effectively mined based on pressure array information.

[0043] It should be understood that the number of layers and convolution kernel size of the neural network model can be designed based on the model accuracy and computational efficiency requirements. Other types of loss functions can also be used.

[0044] In step S130 , pressure array information of the interaction between the pressing structure and the target is collected and input into a trained neural network model to obtain the normal force and tangential force of the corresponding contact surface.

[0045] After the neural network model is trained, optimized model parameters such as weights and biases are obtained. The trained neural network model can be used for real-time normal and tangential force calculations. This involves collecting pressure array information from the interaction between the pressing tip and the target; inputting this pressure array information into the trained neural network model to obtain the normal and tangential forces of the corresponding contact surface; and then using the calculated normal and tangential forces as feedback to control the subsequent operation of the pressing tip. The pressing tip can be mounted on a robotic arm and applied in a variety of scenarios, including hemostasis, acupressure, and muscle massage.

[0046] In summary, the present invention proposes a three-dimensional force calculation method based on the spherical / arc surface structure of the pressing robot end and the high-density flexible pressure array sensor. Through the special spherical / arc spatial structure design of the pressing end and the high-density flexible pressure array sensor, the multi-dimensional force information perceived by the planar pressure sensor is three-dimensionally mapped, and the mapping relationship between the perception quantity of the flexible pressure sensor array and the three-dimensional force under different pressing end structures is established through deep learning methods.

[0047] It should be noted that the training process of the neural network model involved in the present invention can be performed offline on a server or in the cloud. By embedding the trained model into an electronic device, the normal force and tangential force corresponding to the pressure array information can be obtained in real time. The electronic device can be a terminal device or a server, and the terminal device includes any terminal device such as a robot, a mobile phone, a tablet computer, a personal digital assistant (PDA), a point of sale (POS), an in-vehicle computer, a smart wearable device, etc. The server includes but is not limited to an application server or a web server, and can be a standalone server, a cluster server, or a cloud server.

[0048] In summary, compared with the prior art, the present invention has the following advantages:

[0049] 1) The spherical / arc-shaped pressing end structure is designed to facilitate integration with high-density pressure array sensors. The direction of the pressing force can be determined by the pressure at different locations on the spherical / arc-shaped surface. Using a spherical structure as the end effector ensures uniform distribution of external forces, effectively measuring three-dimensional forces while also extending the life of the end effector.

[0050] 2) The proposed method for acquiring three-dimensional force based on high-dimensional pressure array information effectively integrates the design of the terminal's spherical structure with a high-density flexible pressure array sensor. Using the three-dimensional structural information of the pressing terminal as a known condition, artificial intelligence methods are used to map the information sensed by the flexible pressure array sensor into three-dimensional space, thereby acquiring the tangential and normal pressure forces exerted by the pressing terminal on the pressed object. By using a high-density flexible pressure array sensor, the method perfectly matches the terminal's spherical structure and fully perceives the force distribution when external forces act on the array sensor.

[0051] 3) An efficient neural network regression model was designed, using a convolutional neural network to extract features from the pressure array information. This model further established the intrinsic connection between 3D force and the pressure sensing array, enabling accurate calculation of 3D pressing force based on the flexible pressure array information. By employing a convolutional neural network regression model, end-to-end feature learning was achieved, fully exploiting the intrinsic connection between the pressure array information and the normal and tangential forces, improving the accuracy of 3D force estimation.

[0052] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0053] Computer-readable storage medium can be a tangible device that can keep and store the instructions used by the instruction execution device.Computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device or any suitable combination thereof.More specific examples (non-exhaustive list) of computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove having instructions stored thereon, and any suitable combination thereof.Computer-readable storage medium used herein is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagated by waveguides or other transmission media (for example, light pulses by fiber optic cables), or electrical signals transmitted by wires.

[0054] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0055] The computer program instructions for performing the operation of the present invention can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, Python, and conventional procedural programming languages ​​such as "C" language or similar programming languages. The computer readable program instructions can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), is personalized by utilizing the state information of the computer readable program instructions, and the electronic circuit can execute the computer readable program instructions, thereby realizing various aspects of the present invention.

[0056] Various aspects of the present invention are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0057] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0058] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0059] The flowcharts and block diagrams in the accompanying drawings show the possible implementation architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of an instruction, and the module, program segment or part of the instruction contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

[0060] While various embodiments of the present invention have been described above, the foregoing description is intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A pressure calculation method based on the fusion of a spherical arc surface structure and a flexible pressure array, comprising the following steps: Collecting pressure array information of the interaction between the pressing end and the target, where the pressure array information reflects the sensed pressure in different directions of the three-dimensional space generated when the pressing end contacts the target; Inputting the pressure array information into a trained neural network model to obtain the normal force and tangential force of the corresponding contact surface; Subsequent operation of the pressing tip is controlled based on the normal force and the tangential force.

2. The method according to claim 1, characterized in that The neural network model is trained according to the following steps: Acquire a sample data set, wherein the sample data set reflects the corresponding relationship between the pressure array information and the normal force and the tangential force collected using the three-dimensional force sensor as the gold standard; Expanding and enhancing the sample data set to obtain an enhanced sample data set, wherein the enhancement method includes a mirroring operation and a rotation operation; The neural network model is trained using the enhanced sample data set with minimization of the set loss function as the optimization goal.

3. The method according to claim 1, characterized in that The loss function for training the neural network model is set to: Y_i={F n _i F x _i F y _i} Among them, Yi represents the true value, represents the predicted value, n represents the number of samples, i is the sample index, and F n _i represents the normal force corresponding to sample i, F x _i and F y _i represents the tangential force on the x-axis and y-axis corresponding to sample i.

4. The method according to claim 1, characterized in that The neural network model includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a Flatten layer, a first fully connected layer, a second fully connected layer and an output layer, wherein the first convolutional layer is used to extract features from the input pressure array information and output a first feature map; the first pooling layer is used to downsample the first feature map and output a second feature map; the second convolutional layer is used to extract features from the second feature map and output a third feature map; the second pooling layer is used to downsample the third feature map and output a fourth feature map; the Flatten layer is used to expand the fourth feature map into a one-dimensional vector; the first fully connected layer is used to linearly combine the expanded vector with the weight; the second fully connected layer is used to linearly combine the output vector of the first fully connected layer with the weight to generate an output vector of size 3x1x1; the output layer is used to output a 3x1x1 vector, which respectively indicates the calculated normal force, x-axis tangential force and y-axis tangential force.

5. The method according to claim 1, characterized in that The pressing end is a spherical arc surface structure and fits the flexible pressure array sensor.

6. The method according to claim 1, characterized in that The pressing end is arranged on a robot arm as an end effector.

7. The method according to claim 5, characterized in that The first convolutional layer, the second convolutional layer, the first fully connected layer, the second fully connected layer and the output layer of the neural network model all use rectified linear units as activation functions.

8. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

Images