Electrical impedance imaging haptic sensing learnable reconstruction method and system

Abstract

The application discloses a kind of electric impedance imaging tactile sensing learnable reconstruction methods, it includes the following steps: S1: the conductivity distribution model of electric impedance imaging flexible tactile sensor under the external contact state is established, to obtain the first boundary electric response data in the state without external contact and the second boundary electric response data in the external contact state;S2: first, the second boundary electric response data is differentially processed to obtain boundary difference measurement data;S3: the forward model of electric impedance imaging flexible tactile sensor is constructed, and sensitivity matrix is generated;S4: physical priori constraint information is extracted by sensitivity matrix;S5: based on boundary difference measurement data, sensitivity matrix and physical priori constraint information, electric impedance imaging tactile sensing learnable reconstruction network is constructed, and perception solution is carried out, to output tactile representation information.Correspondingly, the application also discloses a kind of electric impedance imaging tactile sensing learnable reconstruction system, to be used for executing the above-mentioned method of the application.

Description

Technical Field

[0001] This invention relates to the field of tactile perception, and more particularly to a learnable reconstruction method and system for electrical impedance imaging tactile sensing. Background Technology

[0002] As is well known, tactile sensing is one of the important technical approaches to acquiring contact information, and it is widely used in flexible electronics, intelligent sensing, human-computer interaction, and surface condition detection. Among these applications, Electrical Impedance Tomography (EIT) tactile sensing technology is commonly used for detection. This technology employs continuously conductive or piezoresistive flexible materials as the sensing medium. By arranging multiple electrodes at the boundary of the field and applying an excitation current, the boundary voltage response is collected, and the changes in internal conductivity or impedance distribution are inverted. Compared with traditional array-structured tactile sensors, EIT tactile sensors have advantages such as no need for dense internal wiring, simple structure, and ease of achieving large-area continuous sensing.

[0003] The core of EIT tactile perception is the reconstruction algorithm. Deep learning-based EIT reconstruction methods improve reconstruction speed and quality by learning the mapping relationship between boundary response and internal distribution. However, existing pure data-driven methods treat reconstruction as a "black box mapping," lacking explicit utilization of EIT physical mechanisms, boundary measurement constraints, positive model consistency, and spatial sensitivity distribution characteristics. This results in problems such as strong dependence on training data, insufficient generalization ability, poor stability across working conditions, and weak physical interpretability of reconstruction results.

[0004] EIT reconstruction is essentially a nonlinear, underdetermined, ill-conditioned inverse problem. There is a complex "soft field" coupling relationship between the boundary measurement data and the changes in conductivity within the field, which leads to many problems: the boundary measurement dimension is limited, making it difficult to accurately recover internal local changes; the reconstruction results are sensitive to noise, contact position, and electrode distribution, resulting in insufficient stability; the sensitivity distribution near and far from the electrodes is usually uneven, leading to a significant positional dependence of the tactile detection results; traditional iterative reconstruction methods rely on prior assumptions and parameter adjustment, resulting in limited computational efficiency and difficulty in balancing resolution, real-time performance, and robustness.

[0005] Therefore, existing deep learning-based EIT tactile perception reconstruction methods ignore the physical model information of EIT tactile sensing, such as sensitivity distribution characteristics, pressure area block correlation characteristics, and pressure area conductivity variation characteristics, resulting in low perception accuracy and poor generalization in practical application scenarios. Summary of the Invention

[0006] The technical problem to be solved by the present invention is that the present invention discloses a learnable reconstruction method and system for electrical impedance imaging tactile sensing, in order to solve the problems of low perception accuracy, poor generalization and difficulty in balancing computational efficiency and reconstruction performance of the existing deep learning-based EIT tactile sensing reconstruction method.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a learnable reconstruction method for resistive impedance imaging tactile sensing, comprising the following steps: S1: Establish a conductivity distribution model of the electrical impedance imaging flexible tactile sensor under external contact conditions to obtain the first boundary electrical response data under no external contact conditions and the second boundary electrical response data under external contact conditions. S2: Perform differential processing on the first boundary electrical response data and the second boundary electrical response data to obtain boundary differential measurement data; S3: Construct a forward model of the resistive impedance imaging flexible tactile sensor and generate a sensitivity matrix; S4: Extract physical prior constraint information using the sensitivity matrix; S5: Construct a learnable reconstructable network for electrical impedance imaging tactile sensing based on the boundary difference measurement data, the sensitivity matrix, and the physical prior constraint information, and perform perception solution to output tactile representation information.

[0008] To address the shortcomings of existing deep learning-based EIT tactile sensing reconstruction methods, such as low sensing accuracy, poor generalization, and difficulty in balancing computational efficiency and reconstruction performance, this invention proposes a learnable reconstruction method for electrical impedance tomography (EIT) tactile sensing based on physical constraints. Specifically, it integrates EIT physical model information, such as sensitivity distribution characteristics, block correlation characteristics of the pressed area, and conductivity variation characteristics of the pressed area, into the learnable reconstruction method. This improves EIT tactile sensing accuracy and generalization while maintaining computational efficiency, enhancing sensing performance in real-world scenarios, and improving the physical consistency, spatial resolution, and cross-regional sensing consistency of the reconstruction results. Ultimately, this enhances the application performance of EIT tactile sensors in practical tactile sensing tasks, demonstrating promising prospects and application value.

[0009] Furthermore, the learnable reconstruction method for electrical impedance imaging tactile sensing described in this invention further includes step S6, which involves training the learnable reconstruction network for electrical impedance imaging tactile sensing using a multi-constraint joint loss function; wherein the multi-constraint joint loss function is expressed as:

[0010] in, To monitor the losses from the reconstruction; This results in a loss of data consistency. This is a physical feasibility constraint. For structural regularization terms; These are the weighting coefficients corresponding to each type of loss.

[0011] Furthermore, in the learnable reconstructable method for resistive impedance imaging tactile sensing described in this invention, in step S1, the conductivity distribution model of the resistive impedance imaging flexible tactile sensor under external contact state is as follows:

[0012] in, This describes the conductivity distribution of the sensing area of ​​a resistive impedance imaging flexible tactile sensor under external contact conditions. This represents the initial conductivity distribution of the sensing area of ​​a resistive impedance imaging flexible tactile sensor in the absence of external contact. The change in local conductivity caused by contact. , Indicates pressure, Indicates material state parameters.

[0013] Furthermore, in the resistive impedance imaging tactile sensing learnable reconstruction method of the present invention, step S1 specifically includes: An initial model is performed on the sensing area of ​​the electrical impedance imaging flexible tactile sensor in the absence of external contact, and its conductivity distribution is represented as the initial conductivity distribution. The electrodes of the control impedance imaging flexible tactile sensor apply current excitation to the sensing area according to a preset excitation sequence, and acquire boundary voltage response data in the absence of external contact, as the first boundary electrical response data. When the sensing area comes into contact with the outside, the electrode continues to apply current excitation to the sensing area according to the preset excitation sequence in order to collect the second boundary electrical response data under the external contact state.

[0014] Furthermore, in the resistive impedance imaging tactile sensing learnable reconstruction method of the present invention, step S3 specifically includes: A forward model is established based on the geometry of the sensing area, electrode distribution, and flexible conductive material parameters of the impedance imaging flexible tactile sensor. The forward model is discretized and linearized around the initial state to obtain the linearized relationship between the boundary difference measurement data and the internal conductivity change, so as to generate the sensitivity matrix; The linearized relationship between boundary difference measurement data and changes in internal conductivity is as follows:

[0015] in, For boundary difference measurement data, For the sensitivity matrix, This is the discretized conductivity change vector. This refers to the measurement noise and modeling error terms.

[0016] Furthermore, in the electrical impedance imaging tactile sensing learnable reconstruction method of the present invention, in step S4, physical prior constraint information is extracted based on the sensitivity matrix and tactile contact mechanism, and the physical prior constraint information includes at least: spatial sensitivity distribution information, contact area block correlation information, conductivity change direction information of the pressure area, and local sparse distribution information.

[0017] Furthermore, in the electrical impedance imaging tactile sensing learnable reconstructive method of the present invention, in step S5, the boundary difference measurement data, the sensitivity matrix, and the physical prior constraint information are combined, and the sensing solution process is expanded into a multi-layer cascaded electrical impedance imaging tactile sensing learnable reconstructive network based on the iterative shrinking threshold algorithm:

[0018] in, The output result of the reconstruction at level k is... This is the reconstruction output for the (k-1)th layer. Indicates the direction of the gradient. For boundary difference measurement data, For the sensitivity matrix, Step size, This is a soft threshold operator.

[0019] Furthermore, in the learnable reconstruction method for resistive impedance imaging tactile sensing described in this invention, in step S5, for the k-th layer reconstruction process, specifically: First, data consistency correction is performed using the boundary difference measurement data and the sensitivity matrix. Then, prior information fusion is performed. Finally, a threshold constraint is applied in conjunction with the physical prior constraint information to preserve significant contact features and suppress noise and artifacts, generating the k-th layer reconstruction output. .

[0020] Furthermore, in the learnable reconstruction method for electrical impedance imaging tactile sensing described in this invention, in step S5, the final layer reconstruction output result is... The tactile characterization information is extracted based on the conductivity change distribution at the target time, which is the result of the conductivity change distribution at the target time. The tactile characterization information includes at least the contact position, contact range, contact intensity, and contact trajectory.

[0021] Accordingly, another objective of the present invention is to disclose a learnable reconstruction system for electrical impedance imaging tactile sensing, which also addresses the problems of low perception accuracy, poor generalization, and difficulty in balancing computational efficiency and reconstruction performance in existing deep learning-based EIT tactile sensing reconstruction methods.

[0022] In this invention, the electrical impedance imaging tactile sensing learnable reconstructable system includes: An electrical impedance imaging flexible tactile sensor includes multiple electrodes and a sensing region made of conductive sensing material, wherein the multiple electrodes are distributed at the region boundary of the sensing region. An excitation measurement circuit is electrically connected to the electrode; A control processing unit is electrically connected to the excitation measurement circuit and the electrode, respectively, and the control processing unit is used to execute the resistive impedance imaging tactile sensing learnable reconstruction method as described above in this invention.

[0023] The beneficial effects of the present invention are as follows: The electrical impedance imaging tactile sensing learnable reconstruction method and system designed in this invention are suitable for reconstructing the boundary electrical response generated by flexible tactile sensors under pressure, sliding or multi-point contact conditions, so as to obtain tactile characterization information more accurately, such as contact position, contact distribution, contact intensity and contact trajectory.

[0024] Specifically, this invention integrates the physical model information of the electrical impedance imaging flexible tactile sensor, such as sensitivity distribution characteristics, block correlation characteristics of the pressing area, and conductivity variation characteristics of the pressing area, into a learnable reconstructable network. This improves the accuracy and generalization of EIT tactile perception while ensuring computational efficiency, enhances perception performance in real-world scenarios, and strengthens the physical consistency, spatial resolution, and cross-regional perception consistency of the reconstruction results. As a result, it improves the application performance of the electrical impedance imaging tactile sensor in practical tactile perception tasks, and has good prospects for promotion and application value. Attached Figure Description

[0025] Figure 1 This is a flowchart of the steps in one embodiment of the electrical impedance imaging tactile sensing learnable reconstruction method of the present invention; Figure 2 The schematic diagram shows a structural block diagram of the electrical impedance imaging tactile sensing learnable reconfigurable system according to one embodiment of the present invention; Figure 3 This is a schematic diagram of the spatial sensitivity distribution of the electrical impedance imaging tactile sensing learnable reconstruction method described in this invention under one implementation. Figure 4This is a block diagram of a learnable reconstructable network guided by physical constraints in one embodiment of the electrical impedance imaging tactile sensing learnable reconstructable method of the present invention. Figure 5 This diagram illustrates the application of the electrical impedance imaging tactile sensing learnable reconfigurable system described in this invention on a carrier such as a robot arm or robotic arm in one embodiment. Detailed Implementation

[0026] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0027] To address the problems of low sensing accuracy, poor generalization, and difficulty in balancing computational efficiency and reconstruction performance in existing EIT tactile sensing reconstruction methods, this invention designs a novel learnable reconstruction method for electrical impedance imaging tactile sensing, which incorporates physical model information from EIT tactile sensing, thereby improving EIT tactile sensing accuracy while ensuring computational efficiency.

[0028] like Figure 1 As shown, this invention designs a learnable reconstruction method for electrical impedance imaging tactile sensing, which specifically includes the following steps: S1: Establish a conductivity distribution model of the electrical impedance imaging flexible tactile sensor under external contact conditions to obtain the first boundary electrical response data under no external contact conditions and the second boundary electrical response data under external contact conditions. S2: Perform differential processing on the first boundary electrical response data and the second boundary electrical response data to obtain boundary differential measurement data; S3: Construct a forward model of the resistive impedance imaging flexible tactile sensor and generate a sensitivity matrix; S4: Extract physical prior constraint information using the sensitivity matrix; S5: Construct a learnable reconstructable network for electrical impedance imaging tactile sensing based on the boundary difference measurement data, the sensitivity matrix, and the physical prior constraint information, and perform perception solution to output tactile representation information.

[0029] Therefore, the core of this invention, a learnable reconstructable method for impedance imaging tactile sensing, lies in constructing a learnable reconstructable network for impedance imaging tactile sensing guided by physical constraints. This method first establishes a conductivity distribution model of the impedance imaging flexible tactile sensor under external contact conditions. Then, it acquires first and second boundary electrical response data under both non-external and external contact conditions to obtain boundary difference measurement data. Subsequently, it establishes a corresponding forward model based on the geometry, electrode distribution, and material conductivity parameters of the impedance imaging flexible tactile sensor, generating a sensitivity matrix. Simultaneously, it extracts and obtains prior physical constraint information, which is then used as the guiding input for the subsequent learnable reconstructable network. This constructs a new learnable reconstructable network for impedance imaging tactile sensing, and based on this network, it performs sensory solutions to output tactile representation information.

[0030] Based on this, it is easy to see that the core improvement of this invention lies in integrating the physical prior constraint information of the electrical impedance imaging flexible tactile sensor into the learnable reconstruction method, which can improve its perception accuracy and generalization while ensuring computational efficiency, thereby enhancing its perception performance in real-world scenarios.

[0031] Accordingly, see Figure 2 As shown, in this invention, a learnable reconstruction system for electrical impedance imaging tactile sensing is also designed to execute the learnable reconstruction method for electrical impedance imaging tactile sensing described above, thereby realizing the output of tactile representation information by utilizing the operation of the system.

[0032] like Figure 2 As shown, in practical applications, this electrical impedance imaging tactile sensing learnable reconfigurable system can specifically include: an electrical impedance imaging flexible tactile sensor, an excitation measurement circuit, and a control processing unit. The aforementioned electrical impedance imaging flexible tactile sensor includes multiple electrodes and a sensing region composed of conductive sensing material, with these electrodes distributed at the boundary of the sensing region as boundary electrodes.

[0033] In specific implementation, the excitation measurement circuit is electrically connected to each of the above electrodes, and the control processing unit is electrically connected to the excitation measurement circuit and the electrodes respectively, so as to control the start and stop of the excitation measurement circuit and detect the data of each electrode, thereby enabling the system to effectively execute steps S1-S5 in the resistive impedance imaging tactile sensing learnable reconstruction method of the present invention.

[0034] It should be noted that, in this embodiment, when the system specifically executes step S1 of the learnable reconstruction method for resistive impedance imaging tactile sensing described in this invention, it can specifically be as follows: The system's control and processing unit first initializes the sensing area of ​​the aforementioned electrical impedance imaging flexible tactile sensor in a non-contact state, and represents its conductivity distribution in a non-external contact state as the initial conductivity distribution. When external contact acts on the sensing area of ​​the electrical impedance imaging flexible tactile sensor, a spatially non-uniform conductivity change occurs in a local area of ​​the sensing area under the combined effects of pressure and material state parameters. Therefore, the conductivity distribution model of the electrical impedance imaging flexible tactile sensor under external contact conditions can be specifically represented as follows:

[0035] in, This describes the conductivity distribution of the sensing area of ​​a resistive impedance imaging flexible tactile sensor under external contact conditions. This represents the initial conductivity distribution of the sensing area of ​​a resistive impedance imaging flexible tactile sensor in the absence of external contact. This refers to the change in local conductivity caused by contact.

[0036] Accordingly, during computation, the change in local conductivity caused by the aforementioned external contact acting on the sensing area It can be expressed as a function of contact information and material state parameters, that is:

[0037] in, Indicates pressure, Indicates material state parameters.

[0038] Therefore, after completing the above-mentioned modeling of the conductivity distribution of the resistive impedance imaging flexible tactile sensor under external contact conditions, the system controls the boundary electrodes to apply current excitation to the sensing area according to a preset excitation sequence, and acquires the boundary voltage response data under non-contact conditions as the first boundary electrical response data; at the same time, when the sensing area of ​​the resistive impedance imaging flexible tactile sensor receives external contact, that is, when the sensing area is pressed, slid, or subjected to multi-point contact, the system continues to collect the second boundary electrical response data under contact conditions according to the preset excitation sequence, and performs differential processing between the current second boundary electrical response data and the first boundary electrical response data to obtain boundary differential measurement data.

[0039] In other words, in step S2, the boundary difference measurement data can be expressed as:

[0040] in, This refers to the boundary voltage response under contact conditions. The reference boundary voltage response under the initial contactless state. Boundary difference measurement data characterizing the current contact changes.

[0041] Accordingly, see Figure 3 As shown, when the control processing unit of the system executes step S3 of the learnable reconstruction method of electrical impedance imaging tactile sensing, it can specifically establish a forward electrical impedance imaging model based on the geometry of the sensing area, the electrode distribution, and the parameters of the flexible conductive material.

[0042] It should be noted that when establishing the forward model, according to the quasi-static conduction field equations, the potential distribution within the field should satisfy:

[0043] in, For the field conductivity distribution, This represents the electric potential distribution within the field.

[0044] Accordingly, such as Figure 3 As shown, after constructing the forward model, the system's control processing unit discretizes the forward model and linearizes it around the initial state, obtaining the linearized relationship between the boundary difference measurement data and the internal conductivity change, in order to generate the sensitivity matrix.

[0045] At this point, the linearized relationship between the boundary difference measurement data and the change in internal conductivity can be expressed as:

[0046] in, For boundary difference measurement data, For the sensitivity matrix, This is the discretized conductivity change vector. This refers to the measurement noise and modeling error terms.

[0047] It should be noted that in the learnable reconstruction method for resistive impedance imaging tactile sensing designed in this invention, during step S3, in order to solve the subsequent inverse problem, the system's control processing unit further expresses the conductivity reconstruction problem as an optimization problem with prior constraints:

[0048] in, This represents the conductivity distribution obtained from the reconstruction. Indicates a positive mapping. Represents prior constraints. This represents the regularization parameter.

[0049] Furthermore, based on the finite element model of the electrical impedance imaging flexible tactile sensor, the system's control processing unit can further convert the above inverse problem into a linearized solution form, which is expressed as:

[0050] Accordingly, such as Figure 3 As shown, after the sensitivity matrix is ​​generated in step S3 of the impedance imaging tactile sensing learnable reconstructable system designed in this invention, the system's control processing unit can further execute step S4 of the method, that is, extract physical prior constraint information based on the sensitivity matrix and the tactile contact mechanism; wherein, the physical prior constraint information includes at least: spatial sensitivity distribution information, contact area block correlation information, conductivity change direction information of the pressure area, and local sparse distribution information.

[0051] It should be noted that, in the aforementioned physical prior constraint information, the aforementioned spatial sensitivity distribution information is specifically derived from the sensitivity matrix. The calculations are used to characterize the response differences of different locations within the sensing area to boundary measurements; the aforementioned contact area block correlation information is used to characterize the local continuous distribution characteristics of the actual contact area in space; the aforementioned information on the direction of conductivity change in the pressure area is used to constrain the conductivity change to meet the unidirectional change law after the material is actually under pressure; the aforementioned local sparse distribution information is used to represent the sparsity of the area where conductivity changes, because in the actual tactile perception process, the area where contact actually occurs usually only accounts for a small part of the entire sensing area.

[0052] Therefore, see Figure 4 As shown, after obtaining the physical prior constraint information, the control processing unit in the system can further use the physical prior information as the guiding input for the subsequent learnable reconstruction network in step S5 of the electrical impedance imaging tactile sensing learnable reconstruction method, so as to construct a learnable reconstruction network based on physical constraints.

[0053] Therefore, in step S5, the construction of a learnable reconstructable network guided by physical constraints can be specifically as follows: combining the above-mentioned boundary difference measurement data, the above-mentioned sensitivity matrix and the above-mentioned physical prior constraint information, and expanding the perception solution process into a multi-layer cascaded electrical impedance imaging tactile sensing learnable reconstructable network based on the Iterative Shrinkage Thresholding Algorithm (ISTA).

[0054] In practical applications, it can specifically combine the sparsity of conductivity changes during the tactile perception process of impedance imaging, representing the perception solution process as a LASSO-type (least squares optimization form with L1 norm sparse constraints) optimization problem, and expanding the solution process into a multi-layered cascaded learnable reconstructible network based on the iterative shrinking threshold algorithm:

[0055] in, The output result of the reconstruction at level k is... This is the reconstruction output for the (k-1)th layer. Indicates the direction of the gradient. For boundary difference measurement data, This is the sensitivity matrix. Step size, This is a soft threshold operator.

[0056] Therefore, the idea behind this invention's learnable reconstruction method for resistive impedance imaging tactile sensing is that, when expanding the traditional iterative solution process into a learnable network structure, each layer first uses a sensitivity matrix to correct the data consistency of the current result, and then completes structure restoration and noise suppression through learnable linear transformation and threshold constraints, thereby ensuring model constraints and learning capabilities. Therefore, see [link to relevant documentation]. Figure 4 As shown, for the first k The layer reconstruction process can first use the obtained boundary difference measurement data and sensitivity matrix to perform data consistency correction, then perform prior information fusion, and finally apply threshold constraints in combination with physical prior constraint information to retain significant contact features and suppress noise and artifacts, and generate the k-th layer reconstruction output. .

[0057] In other words, for the first k During the layer reconstruction process, the system's control processing unit can execute the following steps: S51: Data consistency correction; like Figure 4 As shown, the system performs gradient updates based on the reconstruction results from the previous layer, boundary difference measurement data, and sensitivity matrix to obtain intermediate results:

[0058] in, This is the output result for the (k-1)th layer. For the preprocessed boundary difference measurement data, This is the intermediate result after data consistency correction for layer k. Let be the step size parameter for the k-th layer.

[0059] S52: Prior information fusion; The system's control processing unit will process the above intermediate results. Mapped to the feature space, and will be determined by the sensitivity matrix The pre-calculated spatial sensitivity map is incorporated into the feature representation process to obtain features that incorporate physical priors. Represented as:

[0060] in, Denotes the linear transformation operator at the k-th level. Represents the spatial sensitivity map. Represents the sensitivity mapping function. This represents the Hadama product operator.

[0061] S53: Threshold Constraints and Structural Recovery; like Figure 4 As shown, the system's control processing unit integrates the features after physical priors. This indicates that a threshold constraint is applied to preserve significant contact features and suppress noise and artifacts.

[0062] in, This indicates the results of further structural constraints and artifact removal on the features; This represents the threshold operator at the k-th layer. This represents the learnable threshold parameter for the k-th layer.

[0063] Subsequently, the system's control processing unit will constrain the features Map back to the reconstructed space via inverse transformation:

[0064] in, This indicates the corrected conductivity distribution; This represents the inverse transform operator at the k-th level.

[0065] Finally, the system's control processing unit generates the k-th layer output using a residual update method:

[0066] in, The output result of the reconstruction of the k-th layer is used as the input for the next layer until the iterative reconstruction of all expanded layers is completed.

[0067] It should be noted that, as Figure 4 As shown, after the control processing unit of the system completes the iterative reconstruction of all expanded layers, it can output the reconstruction result of the last layer. The tactile characterization information is extracted based on the conductivity change distribution at the target time, which is the result of the conductivity change distribution at the target time. The tactile characterization information includes at least the contact position, contact range, contact intensity, and contact trajectory.

[0068] Accordingly, in the practical application of the impedance imaging tactile sensing learnable reconstruction method designed in this invention, step S6 can be further configured as follows: During the model training phase, the system's control processing unit constructs a multi-constraint joint loss function to train the impedance imaging tactile sensing learnable reconstruction network; wherein, the aforementioned multi-constraint joint loss function can be specifically expressed as:

[0069] in, To monitor the reconstruction loss, gas is used to constrain the consistency between the reconstruction results and the actual contact distribution; This is the data consistency loss, which is used to constrain the reconstruction results to match the boundary difference measurement data after being mapped by the sensitivity matrix. This is a physical feasibility constraint term, which is used to constrain the output results to satisfy the sparsity, continuity and unidirectional variation characteristics. This is a structure regularization term used to enhance the local smoothness and spatial continuity of the output. These are the weighting coefficients corresponding to each loss term.

[0070] Finally, the system's control processing unit extracts tactile characterization information such as contact position, contact range, contact intensity, and contact trajectory based on the conductivity change distribution at the target time. The control processing unit can also communicate with external devices to output the tactile characterization information to a display terminal or host computer for use in tactile display, contact recognition, collision detection, human-computer interaction control, or robot surface safety perception, thus facilitating user operation.

[0071] In this invention, for ease of understanding of the learnable reconstruction method and system for resistive impedance imaging tactile sensing designed in this invention, please refer to... Figure 5 The inventors also used the following specific embodiment for explanation and illustration: In this embodiment, the electrical impedance imaging tactile sensing learnable reconstruction method and system of the present invention are specifically applied to the "robot forearm surface". In this case, when the electrical impedance imaging tactile sensing learnable reconstruction system is arranged, a non-array flexible EIT tactile sensor can be covered on the outer surface of the robot forearm to perform large-area flexible tactile perception. Of course, the present invention is not limited to this scenario.

[0072] In practical applications, the flexible EIT tactile sensor also consists of a sensing area made of conductive sensing material and multiple electrodes distributed at the boundary of the sensing area. Specifically, the flexible conductive sensing material can be a porous foam composite conductive material, a hydrogel material, or other flexible material whose conductivity parameter changes measurably under pressure.

[0073] In this embodiment, multiple boundary electrodes of the system's flexible EIT tactile sensor are evenly distributed along the sensor's edge and connected to the excitation measurement circuit and control processing unit. After the system is powered on, the control processing unit first executes an initialization program: the control excitation circuit applies weak AC excitation to each group of boundary electrodes in a preset sequence, and collects the first boundary electrical response data when the robot is in a state without external contact, as the reference boundary response data; subsequently, the system establishes a corresponding forward model based on the sensor geometry, electrode distribution, and material conductivity parameters, and generates a sensitivity matrix and spatial sensitivity distribution map.

[0074] When the robot comes into contact with external objects during operation, such as the robot's forearm contacting the edge of a table, the surface of a workpiece, or a human arm, the contact area causes a change in the local conductivity of the flexible conductive material of the flexible EIT tactile sensor. At this time, the system automatically enters an online sensing state, continuously collecting second boundary electrical response data under external contact conditions, and performing differential processing with the first boundary electrical response data, which serves as the reference boundary response data, to obtain the boundary differential data corresponding to the current contact event.

[0075] Next, the control processing unit inputs the boundary difference data into the learnable reconfigurable network model based on physical constraints constructed in this invention. This learnable reconfigurable network model first uses the sensitivity matrix to correct data consistency, and then combines spatial sensitivity distribution information, contact area block correlation characteristics, and constraints on the direction of piezoelectric conductivity change to perform layer-by-layer recovery and artifact suppression of the contact area, finally outputting a two-dimensional or three-dimensional tactile distribution map of the robot forearm surface.

[0076] It should be noted that, in practical applications, in single-point contact scenarios, the impedance imaging tactile sensing learnable reconstruction system designed in this invention can output the contact point position and contact intensity; in multi-point contact scenarios, the impedance imaging tactile sensing learnable reconstruction system can distinguish multiple contact areas and provide their respective spatial distributions; in sliding contact scenarios, the system can output the contact trajectory change process based on the reconstruction results at consecutive time points; and in different contact force scenarios, the system can characterize the difference in contact strength based on the amplitude change of the reconstruction results. Subsequently, the system can send the results to the robot control module for collision warning, compliant interactive control, touch feedback recognition, or external surface safety perception.

[0077] For example, in human-robot collaboration scenarios, when a human arm touches the surface of a robot's forearm, the electrical impedance imaging tactile sensing learnable reconstruction method and system designed in this invention can quickly reconstruct the contact position and contact distribution range, and send it to the robot control system to trigger deceleration, stopping, or avoidance strategies. In grasping operation scenarios, when the grasped object slides along the robot's surface, this invention can reflect the sliding trajectory and contact change process in real time, providing a basis for grasping stability analysis and control. In large-area coverage scenarios, because the electrical impedance imaging tactile sensing learnable reconstruction method of this invention introduces spatial sensitivity equalization and physical prior constraints, it can further reduce the difference in recognition performance between the central and edge regions compared with traditional methods, and improve the perceptual consistency of the entire tactile skin.

[0078] In addition, it should be noted that, apart from the surface area of ​​the robot forearm, the electrical impedance imaging tactile sensing learnable reconstruction method and system designed in this invention can also be applied to robot palms, robotic arm shells, service robot torsos, smart interactive surfaces, flexible wearable device surfaces, and other flexible structural surfaces that require large-area continuous contact sensing.

[0079] In summary, the learnable reconstruction method for electrical impedance imaging tactile sensing designed in this invention addresses the problems of low sensing accuracy, poor generalization, and difficulty in balancing computational efficiency and reconstruction performance in existing EIT tactile sensing reconstruction methods. This invention designs a novel learnable reconstruction method for electrical impedance imaging tactile sensing that incorporates physical model information from EIT tactile sensing, thereby improving EIT tactile sensing accuracy while ensuring computational efficiency. It is suitable for reconstructing the boundary electrical response generated by flexible tactile sensors under pressure, sliding, or multi-point contact conditions to obtain spatial characterization results of contact position, contact distribution, contact intensity, and their change process. It has good prospects for promotion and application value.

[0080] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A learnable reconstruction method for electrical impedance imaging tactile sensing, characterized in that, Including the following steps: S1: Establish a conductivity distribution model of the electrical impedance imaging flexible tactile sensor under external contact conditions to obtain the first boundary electrical response data under no external contact conditions and the second boundary electrical response data under external contact conditions. S2: Perform differential processing on the first boundary electrical response data and the second boundary electrical response data to obtain boundary differential measurement data; S3: Construct a forward model of the resistive impedance imaging flexible tactile sensor and generate a sensitivity matrix; S4: Extract physical prior constraint information using the sensitivity matrix; S5: Construct a learnable reconstructable network for electrical impedance imaging tactile sensing based on the boundary difference measurement data, the sensitivity matrix, and the physical prior constraint information, and perform perception solution to output tactile representation information; In the process of solving the perception problem, the reconstruction of the k-th layer includes the following steps: S51: Perform data consistency correction using the boundary difference measurement data and the sensitivity matrix: Perform gradient update based on the (k-1)th layer reconstruction result, boundary difference measurement data, and sensitivity matrix to obtain intermediate results: in, This is the output result for the (k-1)th layer. For the preprocessed boundary difference measurement data, This is the intermediate result after data consistency correction for layer k. Let be the step size parameter for the k-th layer. This is the sensitivity matrix. for transpose; S52: Perform prior information fusion: The control processing unit integrates the above intermediate results. Mapped to the feature space, and will be determined by the sensitivity matrix The pre-calculated spatial sensitivity map is incorporated into the feature representation process to obtain features that incorporate physical priors. Represented as: in, Denotes the linear transformation operator at the k-th level. Represents the spatial sensitivity map. Represents the sensitivity mapping function. This represents the Hadama product operator; S53: The feature fused with physical priors is processed through the control processing unit. This indicates that a threshold constraint is applied to preserve significant contact features and suppress noise and artifacts. in, Represents the characteristics Further structural constraints and artifact removal results, This represents the threshold operator at the k-th layer. This represents the learnable threshold parameter for the k-th layer; The constrained features are processed by the control processing unit. Map back to the reconstructed space via inverse transformation: in, This represents the corrected conductivity distribution. This represents the inverse transform operator at the k-th level; The output of the k-th layer is generated by the control processing unit using a residual update method: in, The output result of the reconstruction of the k-th layer is used as the input of the next layer until the iterative reconstruction of all expanded layers is completed; After the control processing unit completes the iterative reconstruction of all expanded layers, it outputs the reconstruction result of the last layer. The tactile representation information is extracted based on the conductivity change distribution at the target time, and the tactile representation information includes at least the contact position, contact range, contact intensity, and contact trajectory. S6: The electrical impedance imaging tactile sensing learnable reconstructable network is trained using a multi-constraint joint loss function; wherein, the multi-constraint joint loss function is expressed as: in, To monitor the losses from the reconstruction; This results in a loss of data consistency. This is a physical feasibility constraint. For structural regularization terms; These are the weighting coefficients corresponding to each type of loss.

2. The learnable reconstruction method for electrical impedance imaging tactile sensing according to claim 1, characterized in that, In step S1, the conductivity distribution model of the resistive impedance imaging flexible tactile sensor under external contact is as follows: in, This describes the conductivity distribution of the sensing area of ​​a resistive impedance imaging flexible tactile sensor under external contact conditions. This represents the initial conductivity distribution of the sensing area of ​​a resistive impedance imaging flexible tactile sensor in the absence of external contact. The change in local conductivity caused by contact. , Indicates pressure, Indicates material state parameters.

3. The learnable reconstruction method for electrical impedance imaging tactile sensing according to claim 2, characterized in that, In step S1, specifically: An initial model is performed on the sensing area of ​​the electrical impedance imaging flexible tactile sensor in the absence of external contact, and its conductivity distribution is represented as the initial conductivity distribution. The electrodes of the control impedance imaging flexible tactile sensor apply current excitation to the sensing area according to a preset excitation sequence, and acquire boundary voltage response data in the absence of external contact, as the first boundary electrical response data. When the sensing area comes into contact with the outside world, the electrode continues to apply current excitation to the sensing area according to the preset excitation sequence in order to collect the second boundary electrical response data under the contact state with the outside world.

4. The learnable reconstruction method for electrical impedance imaging tactile sensing according to claim 1, characterized in that, In step S3, specifically: A forward model is established based on the geometry of the sensing area, electrode distribution, and flexible conductive material parameters of the impedance imaging flexible tactile sensor. The forward model is discretized and linearized around the initial state to obtain the linearized relationship between the boundary difference measurement data and the internal conductivity change, so as to generate the sensitivity matrix; The linearized relationship between boundary difference measurement data and changes in internal conductivity is as follows: in, For boundary difference measurement data, For the sensitivity matrix, This is the discretized conductivity change vector. This refers to the measurement noise and modeling error terms.

5. The learnable reconstruction method for electrical impedance imaging tactile sensing according to claim 1, characterized in that, In step S4, physical prior constraint information is extracted based on the sensitivity matrix and tactile contact mechanism, and the physical prior constraint information includes at least: spatial sensitivity distribution information, contact area block correlation information, conductivity change direction information of the pressure area, and local sparse distribution information.

6. The learnable reconstruction method for electrical impedance imaging tactile sensing according to claim 1, characterized in that, In step S5, the boundary difference measurement data, the sensitivity matrix, and the physical prior constraint information are combined, and the sensing solution process is expanded into a multi-layered cascaded learnable reconstructable network for electrical impedance imaging tactile sensing based on an iterative shrinking threshold algorithm. in, The output result for the k-th layer reconstruction. This is the reconstruction output for the (k-1)th layer. Indicates the direction of the gradient. For boundary difference measurement data, This is the sensitivity matrix. Step size, This is a soft threshold operator.

7. A learnable and reconfigurable electrical impedance imaging tactile sensing system, characterized in that, include: An electrical impedance imaging flexible tactile sensor includes multiple electrodes and a sensing region made of conductive sensing material, wherein the multiple electrodes are distributed at the region boundary of the sensing region. An excitation measurement circuit is electrically connected to the electrode; A control processing unit is electrically connected to the excitation measurement circuit and the electrode, respectively, and the control processing unit is used to execute the electrical impedance imaging tactile sensing learnable reconstruction method as described in any one of claims 1-6.

Images