Estimation device, estimation method, estimation program, and learning model generation device

Abstract

An estimation device detects an electrical characteristic between a plurality of detection points of a soft material having electrical conductivity by a detection section. An estimation section estimates a shape of an estimation target object using the soft material, a learning model, and based on the electrical characteristic of the soft material. The learning model uses, as learning data, an electrical characteristic that changes in a time series corresponding to a deformation of the soft material and shape information of a shape at the soft material that represents a pressure stimulus that applies the deformation to the soft material, to learn in a manner that the electrical characteristic is input and the shape information is output.

Description

Technical Field

[0001] This disclosure relates to an estimation device, estimation method, estimation procedure, and learning model generation device. Background Technology

[0002] Deformation detection has traditionally focused on soft materials such as polyurethane components. However, it is difficult to detect deformation in soft materials like polyurethane without hindering it. Furthermore, strain sensors used in detecting deformation of rigid bodies such as metals are difficult to apply to soft materials. Therefore, specialized detection devices are needed to detect deformation in soft materials. For example, International Patent Publication No. 2017029905 discloses a technique for extracting deformation by measuring the displacement and vibration of an object using a camera and acquiring deformation images. Additionally, Japanese Patent Application Publication No. 2013-101096 discloses a technique related to a soft tactile sensor that estimates deformation based on light transmittance. Summary of the Invention

[0003] The problem the invention aims to solve

[0004] However, when detecting deformation of soft materials using methods such as detecting displacement of an object through cameras and image analysis, the system, including cameras and image analysis, becomes undesirable due to the increased size of the device required. Furthermore, optical methods using cameras cannot measure hidden parts that cannot be captured by the camera. Therefore, there is room for improvement in detecting deformation of soft materials.

[0005] This disclosure eliminates the need for special detection devices and enables the estimation of shape information from pressure stimuli by utilizing the electrical properties of conductive soft materials during deformation.

[0006] Solution for solving the problem

[0007] The technology disclosed herein includes: a detection unit that detects electrical properties between multiple detection points of a conductive soft material; and an estimation unit that estimates the shape information of an estimated object by taking the electrical properties and shape information as learning data as learning data and learning the electrical properties as input and the shape information as output, and taking the electrical properties as input. The electrical properties change in a time series in accordance with the deformation of the soft material, and the shape information represents the shape at the soft material when a pressure stimulus of deformation is applied to the soft material. Attached Figure Description

[0008] Figure 1 This is a diagram illustrating an example of the structure of a shape estimation device for soft materials according to an embodiment.

[0009] Figure 2 This is a diagram illustrating the concept of the electrical path of the flexible material involved in the embodiment.

[0010] Figure 3 This is a diagram illustrating the learning process involved in the implementation method.

[0011] Figure 4 This is a diagram illustrating an example of the measuring device involved in the embodiment.

[0012] Figure 5 This is a diagram illustrating an example of the pressing member involved in the embodiment.

[0013] Figure 6 This is a diagram illustrating an example of the detection points of the soft material involved in the embodiment.

[0014] Figure 7 This is a flowchart illustrating an example of the learning data collection and processing involved in the implementation method.

[0015] Figure 8 This diagram illustrates the learning process in the learning processing unit according to the embodiment.

[0016] Figure 9 This is a flowchart illustrating an example of the learning process involved in the implementation method.

[0017] Figure 10 This diagram illustrates the learning process in the learning processing unit according to the embodiment.

[0018] Figure 11 This is a diagram illustrating an example of the structure of a shape estimation device for soft materials according to an embodiment.

[0019] Figure 12 This is a flowchart illustrating an example of the estimation process involved in the implementation.

[0020] Figure 13A It is a graph showing the time characteristics of pressure values ​​in relation to the pressure stimulation and electrical properties of the soft material involved in the embodiment.

[0021] Figure 13B This is a graph showing the time characteristics of the depth of the soft material in relation to the pressure stimulation and electrical properties of the soft material involved in the embodiment.

[0022] Figure 13C It is a graph showing the time characteristics of the resistance values ​​in relation to the pressure stimulation and electrical properties of the soft material involved in the embodiment.

[0023] Figure 13DIt is a graph showing the time characteristics of the change in resistance value in relation to the pressure stimulation and electrical properties of the soft material involved in the embodiment.

[0024] Figure 14 This is a graph showing the verification results related to the shape estimation involved in the implementation.

[0025] Figure 15 This is a graph showing the verification results related to the shape estimation involved in the implementation.

[0026] Figure 16 This is a graph showing the verification results related to the shape estimation involved in the implementation.

[0027] Figure 17A It is a graph showing the time characteristics of the resistance values ​​related to the electrical properties of the flexible material involved in the embodiment.

[0028] Figure 17B It is a graph showing the time characteristics of the change in resistance value in relation to the electrical properties of the flexible material involved in the embodiment.

[0029] Figure 18A This is a graph showing the analysis results of the electrical characteristics involved in the implementation method, illustrating the results of multiple pressing components.

[0030] Figure 18B This is a graph showing the analysis results of the electrical characteristics involved in the embodiment, showing the results for a portion of the pressing member.

[0031] Figure 19 This is a graph showing the verification results related to the shape estimation involved in the implementation.

[0032] Figure 20 This is a graph showing the verification results related to the shape estimation involved in the implementation.

[0033] Figure 21A This is a graph showing the analysis results of the electrical characteristics involved in the implementation method, showing the results of the resistance values.

[0034] Figure 21B This is a graph showing the analysis results of the electrical characteristics involved in the implementation method, showing the results obtained by referring to the analysis results from different directions. Detailed Implementation

[0035] The embodiments of the technology of this disclosure will now be described in detail with reference to the accompanying drawings.

[0036] Furthermore, throughout all the accompanying drawings, the same reference numerals are used to denote components and processes that have the same effect or function, and repeated descriptions are sometimes appropriately omitted. Additionally, this disclosure is not limited to any of the embodiments described below, and modifications and implementations can be appropriately made within the scope of the purpose of this disclosure. Furthermore, while this disclosure primarily describes the estimation of physical quantities for non-linearly deforming components, it can of course be applied to the estimation of physical quantities for linearly deforming components.

[0037] In this disclosure, "soft material" is a concept encompassing materials that can deform, at least partially, by being subjected to an external force, such as flexing. This includes soft elastomers such as rubber, structures with a fibrous framework, and structures containing numerous microbubbles. An example of an external force is pressure. Examples of structures with a fibrous framework and structures containing numerous microbubbles include polymers such as polyurethane. "Soft material with conductivity" is a concept encompassing materials with conductivity, including materials obtained by adding conductive materials to a soft material to impart conductivity, and soft materials that are conductive. Furthermore, a soft material with conductivity has the function of changing its electrical properties in accordance with deformation. Additionally, an example of a physical quantity that generates the function of changing electrical properties in accordance with deformation is a pressure value based on pressure stimulation. When a soft material is deformed, it can deform according to the shape of the pressure stimulation. Furthermore, an example of a physical quantity representing the electrical properties that change in accordance with deformation is a resistance value. This resistance value can be considered as the volume resistivity of the soft material.

[0038] Flexible materials exhibit electrical properties corresponding to deformation due to the imparting of conductivity. That is, such as... Figure 2 As shown, in a flexible material endowed with conductivity, electrical paths are complexly coupled, and these paths expand or contract in response to deformation. Furthermore, sometimes electrical paths are temporarily severed, and connections different from those previously observed are formed. Therefore, the flexible material exhibits different electrical properties at locations spaced at predetermined distances (e.g., detection points) in response to applied forces (e.g., pressure stimulation). Thus, from the viewpoint of shape change in the flexible material, it can be considered that the electrical properties change in accordance with the force applied to the flexible material (e.g., pressure stimulation) and the shape of the force applied to the flexible material.

[0039] The estimation device disclosed herein uses a learning model that takes electrical properties corresponding to the deformation of a soft material and shape information of the pressure stimulus applied to the soft material as learning data, and learns in a manner that takes the electrical properties as input and outputs shape information. The estimation device takes the electrical properties of the soft material as the object of estimation as input to the learning model, and estimates its output as the shape information of the object of estimation.

[0040] Furthermore, in this disclosure, in order to understand the physical quantities of the elastomer, a case is described where a component obtained by permeating a conductive material into a polyurethane component (hereinafter referred to as a conductive flexible component) is used as an example of a flexible material. Additionally, a case is described where the pressure stimulus applied to a component with a predetermined shape is used as a physical quantity to deform the flexible material, and the resistance value is used as a physical quantity that changes accordingly with the deformation of the flexible material.

[0041] exist Figure 1 The diagram shows an example of the structure of a shape estimation device 1 for a soft material, which is the estimation device of this disclosure.

[0042] In the estimation process of the soft material shape estimation device 1, a learned model that has been trained and trained with assigned data (at least shape values) labeled with pressure stimulation applied to the conductive soft member 2 and resistance data (i.e., resistance values) of the conductive soft member 2 as input is used to estimate and output shape data of unknown pressure stimulation applied to the conductive soft member 2.

[0043] That is, the shape estimation device 1 for soft materials estimates the shape of the pressure stimulus applied to the soft material based on the electrical properties of the soft material, which changes shape due to the pressure stimulus applied to it. Thus, the shape of the pressure stimulus applied to the soft material can be determined without the need for special or large-scale devices to directly measure the deformation of the soft component.

[0044] In this embodiment, the conductive flexible member 2 is used as a detection unit. That is, as shown in the figure. Figure 1 As shown, the shape estimation device 1 for soft materials includes an estimation unit 5. Input data 4, representing the magnitude (resistance value) of the resistance corresponding to the pressure stimulus 3 applied to the conductive soft member 2, is input to the estimation unit 5. Output data 6, representing the physical quantity (shape value) of the pressure stimulus 3 applied to the conductive soft member 2 as an estimation result, is output from the estimation unit 5. The estimation unit 5 includes a learned model 51.

[0045] Learning model 51 is a model that has completed learning to derive the shape (output data 6) of the pressure stimulus applied to the conductive soft member 2 based on the resistance of the conductive soft member 2 to which the pressure stimulus 3 is applied (input data 4). Learning model 51 is, for example, a model for defining a learned neural network, and is represented as a set of information on the weights (strengths) of the connections between the nodes (neurons) constituting the neural network.

[0046] Learning model 51 is achieved through learning processing unit 52 ( Figure 3 The learning processing unit 52 generates the learning data by using the time-series resistance (input data 4) of the conductive soft member 2 to which pressure stimulation 3 is applied. That is, a large amount of data obtained by measuring the resistance between detection points of the conductive soft member 2 at predetermined distances, labeled with the shape of pressure stimulation 3, in a time sequence, is used as learning data. Specifically, the learning data includes a large number of combinations of input data and information (output data 6), where the input data includes resistance values ​​(input data 4) and the information indicates the shape of the pressure stimulation 3 corresponding to the input data. Here, for example, by adding information indicating the measurement time to the resistance values ​​(input data 4) of the conductive soft member 2, a correspondence is established with the time-series information. In this case, a correspondence can also be established with the time-series information by adding information indicating the measurement time to the combination of successive resistance values ​​when pressure stimulation is applied to the conductive soft member 2.

[0047] Next, the learning processing performed by the Learning Processing Department 52 will be explained.

[0048] First, the learning data used in the learning process will be explained.

[0049] exist Figure 4 An example of a measuring device 7 for measuring the physical quantities of a conductive flexible member 2 is shown.

[0050] The measuring device 7 has a pressure-applying part 73 mounted on a fixing part 72 fixed to a base 71 for applying pressure stimulation (a physical quantity that deforms the conductive flexible member 2) to the conductive flexible member 2. The pressure-applying part 73 includes a pressure-applying body 73A, an arm 73B that can extend and retract relative to the pressure-applying body 73A, and a front end 73C mounted at the front end of the arm 73B. In the pressure-applying part 73, the pressure-applying body 73A is fixed to the fixing part 72, and the arm 73B extends and retracts according to an input signal, thereby causing the front end 73C to move in a predetermined direction (arrow F direction).

[0051] A conductive flexible member 2 is disposed on the base 71, and a pressing member 74 of a predetermined shape is disposed between the front end portion 73C of the pressure-applying portion 73 and the conductive flexible member 2. Furthermore, in this embodiment, as an example of the pressing member 74 of a predetermined shape, a... Figure 5 The pressing members 74A to 74E are shown. Pressing member 74A applies circular pressure stimulation to the conductive soft member 2, pressing member 74B applies quadrilateral pressure stimulation, and pressing member 74C applies triangular pressure stimulation. Furthermore, in this embodiment, considering the orientation of the pressing members 74, pressing member 74D is designated as applying pressure stimulation in an orientation different from pressing member 74B, and pressing member 74E is designated as applying pressure stimulation in an orientation different from pressing member 74C.

[0052] The pressure-applying part 73 operates by extending the arm 73B, causing the front end 73C to push the pressing member 74 against the conductive flexible member 2. The surface of the conductive flexible member 2 on the side near the base 71 has detection points 75 for detecting electrical characteristics (i.e., a physical quantity representing the electrical characteristics of the conductive flexible member 2, in this case, resistance value). The detection points 75 are arranged at multiple different locations spaced at predetermined distances to detect the resistance value of the conductive flexible member 2.

[0053] In this embodiment, as an example of the detection point 75 of the conductive flexible member 2, an application is used. Figure 6 The multiple (shown) arranged in a dot matrix pattern Figure 6 There are 8 detection points 75 in total. Any two of these detection points 75 can be selected to detect the resistance value (e.g., volume resistivity) of the conductive flexible member 2. Figure 6 In the diagram, each of the eight detection points 75 is marked with a symbol representing the first to eighth point within the circle representing detection point 75. Additionally, in... Figure 6 In the example shown, a first detection group #1 is illustrated, where resistance values ​​are detected via a first detection point 75 and a second detection point 75. The second and third detection points 75 represent the second detection group #2; the third and fourth detection points 75 represent the third detection group #3; and the fourth and first detection points 75 represent the fourth detection group #4. The fifth and seventh detection points 75 represent the fifth detection group #5; and the sixth and eighth detection points 75 represent the sixth detection group #6. The second and fourth detection points 75 represent the seventh detection group #7; and the first and third detection points 75 represent the eighth detection group #8.

[0054] Furthermore, in this embodiment, the case of using any of the above-described detection groups to detect the electrical characteristics (resistance value) of the conductive flexible member 2 will be described.

[0055] The measuring device 7 includes an electrical characteristic detection unit 76 connected to the detection point 75 to detect electrical characteristics (i.e., resistance value). The measuring device 7 also includes a controller 70 connected to the pressure application unit 73 and the electrical characteristic detection unit 76. The controller 70 controls the pressure application unit 73 to apply pressure stimulation to the conductive flexible member 2, acquires and stores the resistance value under pressure stimulation of the conductive flexible member 2. Furthermore, the stored resistance value is correlated with information indicating the shape of the pressure stimulation on the conductive flexible member 2, i.e., the shape of the pressing member 74.

[0056] Under the pressing control of the pressing member 74, the measuring device 7 is able to acquire multiple data combinations of the resistance values ​​of the conductive flexible member 2 in a time sequence corresponding to the shape of the pressing member 74.

[0057] The controller 70 can be configured as a computer (not shown) including a CPU to perform learning data collection and processing. Figure 7 The diagram illustrates an example of learning data collection and processing. In step S100, the controller instructs the pressure member 74 to apply pressure stimulation to the conductive soft member 2. In step S102, the resistance values ​​of the conductive soft member 2 are acquired in a time sequence. In the following step S104, the shape of the pressure member 74 is attached as a label to the acquired time-series resistance values ​​and stored. The controller 70 repeats the above process until the combination of the shape of the pressure member 74 with the resistance values ​​of the conductive soft member 2 reaches a predetermined number or a predetermined time (from a negative judgment in step S106 until a positive judgment is made).

[0058] Therefore, by performing pressure control through which the pressing member 74 applies pressure stimulation to the conductive soft member 2, the controller 70 can acquire and store the resistance value of the conductive soft member 2 in a time sequence according to the shape of each pressing member 74. The combination of the resistance values ​​of the conductive soft member 2 stored in the controller 70 according to the shape of each pressing member 74 and in a time sequence constitutes learning data.

[0059] Next, refer to Figure 8 Let's explain the Learning Processing Department 52.

[0060] The learning processing unit 52 includes a generator 54 and an arithmetic unit 56. The generator 54 has the function of generating an output by taking into account the sequential relationship of the resistance values ​​obtained as input in a time series.

[0061] In addition, the learning processing unit 52 holds a large number of combinations of input data 4 (resistance value) measured by the measuring device 7 and output data 6 (shape) related to the pressing member 74 applied as pressure stimulation to the conductive soft member 2 as learning data.

[0062] exist Figure 8 In the example shown, generator 54 includes an input layer 540, an intermediate layer 542, and an output layer 544, constituting a well-known neural network (NN). Neural networks are a well-known technology, so detailed descriptions are omitted, but the intermediate layer 542 includes a large number of node groups (neuron groups) with inter-node connections and feedback connections. Data from the input layer 540 is input to the intermediate layer 542, and the data from the intermediate layer 542, as the result of the computation, is output to the output layer 544.

[0063] Generator 54 is a neural network that generates output data 6A representing the shape of the pressing member 74 based on the input data 4 (resistance). Output data 6A is data that estimates the shape of the pressing member 74 after pressure stimulation is applied to the conductive soft member 2 based on the input data 4 (resistance). Generator 54 generates output data representing a shape close to the shape of the pressing member 74 after pressure stimulation is applied to the conductive soft member 2, based on the input data 4 (resistance) input in a time sequence. By learning through a large amount of input data 4 (resistance), generator 54 is able to generate output data 6A that more closely approximates the shape of the pressing member 74 after pressure stimulation is applied to the conductive soft member.

[0064] The arithmetic unit 56 compares the generated output data 6A with the output data 6 of the learning data and calculates the error as the comparison result. The learning processing unit 52 inputs the generated output data 6A and the output data 6 of the learning data to the arithmetic unit 56. Accordingly, the arithmetic unit 56 calculates the error between the generated output data 6A and the output data 6 of the learning data and outputs a signal representing the calculation result.

[0065] The learning processing unit 52 enables the generator 54 to learn and adjust the weight parameters of the connections between nodes based on the error calculated by the arithmetic unit 56. Specifically, using methods such as gradient descent or backpropagation, the generator 54 feeds back the weight parameters of the connections between nodes in the input layer 540 and the intermediate layer 542, the weight parameters of the connections between nodes in the intermediate layer 542, and the weight parameters of the connections between nodes in the intermediate layer 542 and the output layer 544. That is, with the output data 6 of the learning data as the target, the generator optimizes all the connections between nodes to minimize the error between the generated output data 6A and the output data 6 of the learning data.

[0066] The learning model 51 is generated through the learning processing of the learning processing unit 52. The learning model 51 is a set of information on the weight parameters (weights or strengths) of the connections between nodes, which are the learning results of the learning processing unit 52.

[0067] The learning processing unit 52 is configured as a computer including a CPU (not shown) and is capable of performing learning processing. For example, such as Figure 9 As shown in one example of the learning process, in step S110, the learning processing unit 52 acquires learning data as a result of time-series measurement, i.e., input data 4 (resistance) labeled with information representing the shape of the pressing member 74. In step S112, the learning processing unit 52 uses the learning data as a result of time-series measurement to generate a learning model 51. That is, it obtains a set of information on the weight parameters (weights or strengths) of the connections between nodes as a learning result obtained by learning with a large amount of learning data as described above. Then, in step S114, the data representing the set of information on the weight parameters (weights or strengths) of the connections between nodes as a learning result is stored as the learning model 51.

[0068] Furthermore, generator 54 can use either a recurrent neural network that has the ability to generate outputs by taking into account the sequential relationship of the inputs over time, or other methods.

[0069] Furthermore, in the aforementioned soft material shape estimation device 1, a learned generator 54 (i.e., data representing a set of information on the weight parameters of the connections between nodes as a learning result) generated by the method illustrated above is used as the learning model 51. If a sufficiently learned learning model 51 is used, it is not impossible to determine the shape of the pressing member 74 pressing the conductive soft member 2 based on the time-series resistance values ​​in the pressure stimulation of the pressing member 74 on the conductive soft member 2.

[0070] Furthermore, the processing of the learning processing unit 52 is an example of the processing of the learning model generation apparatus of this disclosure. Additionally, the shape estimation device 1 for soft materials is an example of the estimation unit and estimation apparatus of this disclosure.

[0071] Furthermore, as mentioned above, in the conductive flexible member 2, the electrical paths are complexly combined (e.g., refer to...). Figure 2The conductive flexible member 2 exhibits behaviors such as stretching, contraction, temporary disconnection, and the formation of new connections in accordance with the deformation of the electrical path. As a result, it exhibits different electrical characteristics in response to the applied force (e.g., pressure stimulation). This means that the conductive flexible member 2 can be used as a reservoir for storing data related to the deformation of the conductive flexible member 2. That is, the shape estimation device 1 of the flexible material can apply the conductive flexible member 2 to a network model called Physical Reservoir Computing (PRC) (hereinafter referred to as PRCN). PRC and PRCN are well-known technologies, so detailed descriptions are omitted. In other words, PRC and PRCN can be better applied to the estimation of information related to the deformation of the conductive flexible member 2.

[0072] exist Figure 10 The diagram shows an example of a learning processing unit 52 that uses the conductive flexible member 2 as a reservoir for storing data related to the deformation of the conductive flexible member 2 for learning. The conductive flexible member 2 transforms into electrical characteristics (resistance values) corresponding to each type of pressure stimulus, functioning as an input layer for inputting resistance values, and also as a reservoir layer for storing data related to the deformation of the conductive flexible member 2. The conductive flexible member 2 outputs different electrical characteristics (input data 4) corresponding to the applied pressure stimulus 3 (the shape of the pressing member), thus enabling the estimation layer to estimate the applied pressure stimulus 3 (the shape of the pressing member) based on the resistance value of the conductive flexible member 2. Therefore, in the learning process, learning can be performed on the estimation layer.

[0073] The shape estimation device 1 for the aforementioned soft material can be implemented, for example, by having a computer execute a program representing the aforementioned functions.

[0074] exist Figure 11 The diagram shows an example of a case in which an execution device, which performs various functions of a shape estimation device 1 for realizing a soft material, is configured to include a computer.

[0075] As Figure 8 The computer that functions as the shape estimation device 1 for the soft material shown has Figure 11The computer main body 100 is shown. The computer main body 100 includes a CPU 102, RAM 104 and ROM 106 (including volatile memory), an auxiliary storage device 108 such as a hard disk drive (HDD), and input / output interfaces (I / O) 110. These components—CPU 102, RAM 104, ROM 106, auxiliary storage device 108, and I / O 110—are connected via a bus 112 in a manner capable of transmitting data and instructions to each other. Furthermore, a communication unit 114 for communicating with external devices and an operation display unit 116 such as a display and keyboard are connected to the I / O 110. The communication unit 114 functions to obtain input data 4 (resistance) from the conductive flexible member 2. That is, the communication unit 114 can obtain input data 4 (resistance) from an electrical characteristic detection unit 76, which is a detection unit that includes the conductive flexible member 2 and is connected to a detection point 75 of the conductive flexible member 2.

[0076] The auxiliary storage device 108 stores a control program 108P for enabling the computer main body 100 to function as an example of the estimation device of this disclosure, namely, the shape estimation device 1 for soft materials. The CPU 102 reads the control program 108P from the auxiliary storage device 108, expands it in the RAM 104, and executes the processing. Thus, the computer main body 100, having executed the control program 108P, operates as an example of the estimation device of this disclosure, namely, the shape estimation device 1 for soft materials.

[0077] In addition, the auxiliary storage device 108 stores a learning model 108M including a learning model 51 and data 108D including various types of data. The control program 108P can also be provided via a recording medium such as a CD-ROM.

[0078] Next, the estimation process in the shape estimation device 1 for soft materials implemented by a computer will be explained.

[0079] exist Figure 12 The diagram shows an example of the estimation process based on the control program 108P executed in the computer body 100.

[0080] Figure 12 The estimation process shown is the process executed by the CPU 102 when the computer body 100 is powered on. That is, the CPU 102 reads the control program 108P from the auxiliary storage device 108 and expands it in the RAM 104 to execute the process.

[0081] First, in step S200, the CPU 102 reads the learning model 51 from the learning model 108M in the auxiliary storage device 108 and expands it in the RAM 104, thereby obtaining the learning model 51. Specifically, a network model based on the inter-node connections represented as the weight parameters of the learning model 51 is expanded in the RAM 104. Thus, a learning model 51 based on the weight parameters and inter-node connections is constructed.

[0082] Next, in step S202, the CPU 102 acquires, via the communication unit 114, unknown input data 4 (resistance) of the object that estimates the shape of the pressing member based on the pressure stimulus applied to the conductive soft member 2 in a time sequence.

[0083] Next, in step S204, the CPU 102 uses the learning model 51 obtained in step S200 to estimate the output data 6 (the shape of the unknown estimated object) corresponding to the input data 4 (resistance) obtained in step S202.

[0084] Then, in the next step S206, output data 6 (estimated object shape) as the estimation result is output via communication unit 114, and this processing routine ends.

[0085] also, Figure 12 The estimation process shown is an example of a process performed using the estimation method of this disclosure.

[0086] As explained above, according to this disclosure, the shape of an object to which pressure stimulation has been applied to the conductive flexible member 2 can be estimated based on input data 4 (resistance) that varies in response to the pressure stimulation 3 applied to the conductive flexible member 2. That is, without the need for special or large-scale equipment, it is possible to directly measure the deformation of the flexible member and thus estimate the pressure stimulation applied to the flexible member, for example, the shape of the object.

[0087] Next, the verification results of the shape estimation of the object when the conductive soft member 2 is pressed are explained in the shape estimation device 1 of the soft material described above.

[0088] exist Figures 13A to 13D The diagram shows the pressure stimulation and electrical characteristics (resistance value) of the conductive flexible member 2 when pressure stimulation is applied to it by the pressing member 74. Figure 13A The time characteristics of the pressure value as a pressure stimulus applied to the conductive soft member 2 are shown in the figure. Figure 13B The diagram shows the application of a conductive flexible member 2. Figure 13A The time-dependent characteristics of the depth in the deformation of the conductive flexible member 2 under the pressure value shown. Figure 13CThe image shows the application of a conductive flexible member 2. Figure 13A The time characteristics of the resistance value of the conductive flexible component 2 in group #1 under the pressure value shown. Figure 13D The diagram shows the time characteristics related to the change in resistance value from the start of pressure stimulation on the conductive flexible member 2, i.e., the start of pressing. An example of the time characteristics related to this change in resistance value is given, showing the time characteristics with discrepancies.

[0089] like Figures 13A to 13D As shown, it can be confirmed that when pressure is applied to the conductive soft member 2, the resistance value (volume resistivity) of the conductive soft member 2 changes in response to the pressure stimulation. Furthermore, based on the time characteristics (e.g., difference time characteristics representing the difference) related to the amount of change in resistance value from the start of pressure, it can also be confirmed that the resistance value changes sequentially over time in accordance with the deformation of the soft material.

[0090] Next, the estimation results obtained based on different shapes of the pressing member 74, which applies pressure stimulation to the conductive soft member 2, were verified.

[0091] exist Figure 14 The validation results are shown for the estimation results obtained based on different shapes of the pressing member 74, which is subjected to pressure stimulation applied to the conductive soft member 2. In this validation, for each of the pressing members 74 with different shapes 74A to 74E, 40 pressing experiments were conducted, and the resistance values ​​of each changing over time were collected as learning data. Then, a learning model was constructed using the collected learning data, and 30 shape estimation experiments were carried out using this learning model. Furthermore, in Figure 14 In the diagram, pressing member 74A is denoted as C0, pressing member 74B as S0, pressing member 74C as T0, and pressing member 74D as T. 90 The pressing component 74E is denoted as S. 45 The estimation result represents the number of estimations performed on the shape of the estimated component when the pressing component 74 is pressed.

[0092] like Figure 14 As shown, according to the soft material shape estimation device 1 described above, it can be confirmed that even when pressure is applied to the conductive soft member 2 by pressing member 74 of different shapes, the shape can be estimated.

[0093] exist Figure 15The diagram shows the validation results of the shape estimation of the pressing member 74 in different detection groups. In these validations, the same pressing member 74 was subjected to 40 pressing experiments for different detection groups #1 to #8, and the resistance values ​​that varied over time were collected as learning data. Then, a learning model was constructed using the collected learning data, and 30 shape estimation experiments were carried out using this learning model.

[0094] like Figure 15 As shown, according to the soft material shape estimation device 1 described above, it can be confirmed that by setting a detection group, i.e. a detection position (e.g., detection group #1), even a single detection group can be used to estimate the shape well.

[0095] exist Figure 16 The diagram shows the verification results of estimating the shape of the pressing member 74 using a combination of different detection groups. In this verification, the aforementioned 40 pressing experiments were also performed, and 30 shape estimation experiments were conducted using a learning model learned from the collected learning data.

[0096] like Figure 16 As shown, according to the soft material shape estimation device 1 described above, it can be confirmed that by combining multiple detection groups, the shape can be estimated well regardless of which detection group is used.

[0097] exist Figure 17A and Figure 17B The diagram shows the time characteristics related to the electrical properties (resistance value) of the conductive soft member 2 when pressure is applied to it by multiple pressing members 74. Figure 17A The diagram shows the time-dependent resistance characteristics of the conductive soft member 2 when pressure is applied to it by pressing members 74A to 74E. Figure 17B The changes in resistance value of the pressing components 74A to 74E are shown respectively from the start of pressure stimulation, i.e., the start of pressing.

[0098] like Figure 17A and Figure 17B As shown, for each pressing member 74A to 74E with different shapes, it can be confirmed that the characteristics of the change in resistance value (volume resistivity value) of the conductive flexible member 2 are different.

[0099] exist Figure 18A and Figure 18B The text shows about Figure 17A and Figure 17B The electrical characteristics shown are used to display the results of principal component analysis in three dimensions. Figure 18AThe results of principal component analysis related to each pressing component 74A-74E are shown in the figure. Figure 18B The results of principal component analysis related to pressing members 74D and 74E are shown in the figure.

[0100] like Figure 18A and Figure 18B As shown, it can be confirmed that the principal component analysis results for each pressing component 74A to 74E of different shapes are separate structures, and the principal component analysis results for each pressing component 74A to 74E of different shapes are different.

[0101] Next, the estimation results were validated when pressure stimulation was applied to the conductive soft component 2 under different conditions than during learning, using the learned model.

[0102] exist Figure 19 The diagram shows the verification results of the estimation results obtained based on different sizes and pressing positions of the pressing member 74. Figure 19 Regarding the circular pressing member 74A, the pressing member 74A at a position separated from the learned position by a predetermined distance (e.g., 5 mm) is denoted as C1, and the pressing member 74A at a position further separated by an interval (e.g., 10 mm) is denoted as C2. Furthermore, the pressing member 74A at a size reduced from the learned size (e.g., diameter 50 mm) to a size of 40 mm is denoted as C3, and the pressing member 74A at a size further reduced to a size of 30 mm is denoted as C4. For the quadrilateral pressing member 74B, the pressing member 74B at a position separated from the learned position by a predetermined distance (e.g., 5 mm) is denoted as S1, and the pressing member 74B at a position further separated by an interval (e.g., 10 mm) is denoted as S2. Similarly, for the triangular pressing member 74C, the pressing member 74C under the condition of being separated from the learned position by a specified distance (e.g., 5 mm) is denoted as T1, and the pressing member 74C under the condition of being further separated by an interval (e.g., 10 mm) is denoted as T2.

[0103] like Figure 19 As shown, it can be confirmed that conditions such as the shape of the pressing member 74 affect the estimation results. Therefore, it is confirmed that the estimation process may sometimes differ under conditions different from those during learning.

[0104] Therefore, the shape estimation results obtained by the learning model, which was trained using different conditions such as the shape of the mesh pressing member 74, were verified.

[0105] exist Figure 20The diagram shows the validation results of the estimation results obtained using a learning model trained under various conditions based on the size and pressing position of the aforementioned pressing member 74. For example... Figure 20 As shown, it can be confirmed that the shape can be estimated well for almost all conditions.

[0106] exist Figure 21A and Figure 21B The text shows about Figure 20 The results shown are a three-dimensional display of the principal component analysis results, representing the verification results. Figure 21A The results of resistance value detection using detection group #1 are shown in the figure. Figure 21B The image shows references from different directions. Figure 21A The result.

[0107] like Figure 21A and Figure 21B As shown, even under multiple conditions, by using learning models that have been learned separately, the results related to the shape of each pressing member 74A to 74E are separated into structures, and it is possible to distinguish each pressing member 74A to 74E with different shapes.

[0108] To achieve the above objective, the first approach is an estimation device, comprising:

[0109] The testing department detects the electrical properties between multiple test points of a conductive, flexible material; and

[0110] The estimation unit, which uses electrical characteristics and shape information as learning data and learns by taking the electrical characteristics as input and outputting the shape information, takes the electrical characteristics of the object to be estimated as input and the detection unit as input to estimate the shape information of the object to be estimated, wherein the electrical characteristics change in a time series in accordance with the deformation of the soft material, and the shape information represents the shape of the soft material at the location of the soft material under the pressure stimulus of applying deformation.

[0111] Regarding the second method, in the estimation device of the first method,

[0112] The flexible material is one whose electrical properties change accordingly with the deformation.

[0113] The learning model learns by outputting shape information corresponding to the detected electrical characteristics.

[0114] Regarding the third method, in the estimation device of the first or second method,

[0115] The electrical property of the soft material is volume resistivity.

[0116] Regarding the fourth method, in the estimation device of any of the first to third methods,

[0117] The soft material is a material obtained by imparting conductivity to a polyurethane material with a fibrous skeleton structure or a structure in which multiple microbubbles are dispersed inside.

[0118] Regarding the fifth method, in the estimation device of any of the first to fourth methods,

[0119] The learning model is generated by using the soft material as a reservoir and learning from a network based on reservoir computation that utilizes the reservoir.

[0120] The sixth approach is an estimation method in which the computer performs the following processing:

[0121] The electrical characteristics of the detection unit are obtained from multiple detection points that detect the electrical properties of a conductive soft material; and

[0122] A learning model is used to learn electrical characteristics with time-series information that change in accordance with the deformation of the soft material and shape information of the pressure stimulus applied to the soft material to cause deformation, and the electrical characteristics are used as input and the shape information is output. The acquired electrical characteristics of the estimated object are input to estimate the shape information of the estimated object.

[0123] The seventh method is an estimation procedure used to cause a computer to perform the following processes:

[0124] The electrical characteristics of the detection unit are obtained from multiple detection points that detect the electrical properties of a conductive soft material; and

[0125] A learning model is used to learn electrical characteristics with time-series information that change in accordance with the deformation of the soft material and shape information of the pressure stimulus applied to the soft material to cause deformation, and the electrical characteristics are used as input and the shape information is output. The acquired electrical characteristics of the estimated object are input to estimate the shape information of the estimated object.

[0126] The eighth approach is a learning model generation device, comprising:

[0127] An acquisition unit acquires electrical characteristics and shape information, wherein the electrical characteristics are obtained from a detection unit that detects the electrical characteristics between multiple detection points of a conductive soft material, and the shape information is shape information obtained from pressure stimulation applied to the soft material to induce deformation; and

[0128] The learning model generation unit generates a learning model based on the acquisition results of the acquisition unit. The learning model is a learning model that takes as input the electrical characteristics with time-series information that change in accordance with the deformation of the soft material and outputs the shape information of the object.

[0129] According to this disclosure, the following effect is achieved: without the need for special detection devices, the shape information of pressure stimulation can be estimated by utilizing the electrical properties of a conductive soft material during deformation.

[0130] As described above, this disclosure describes the application of a conductive flexible member as an example of a flexible member, but the flexible member is of course not limited to a conductive flexible member.

[0131] Furthermore, the scope of this disclosure is not limited to the scope described in the above embodiments. Various changes or modifications can be made to the above embodiments without departing from the spirit of the matter, and the manner obtained by making such changes or modifications is also included within the scope of this disclosure.

[0132] Furthermore, in the above embodiments, the case where the inspection process is implemented by a software structure based on the process using a flowchart is described, but it is not limited to this. For example, it is also possible to implement each process by a hardware structure.

[0133] Alternatively, a portion of the estimation device, such as a neural network or a learning model, can be incorporated into the hardware circuitry.

[0134] Furthermore, all documents, patent applications, and technical standards described in this specification are incorporated herein by reference to the same extent as those specifically and individually described therein.

[0135] In addition, the entire disclosure of Japanese Patent Application No. 2020-210737, filed on December 18, 2020, is incorporated herein by reference.

Claims

1. An estimation device, comprising: The testing department detects the electrical properties between multiple test points of a conductive, flexible material. as well as The estimation unit, which uses electrical characteristics and shape information as learning data and learns by taking the electrical characteristics as input and outputting the shape information, takes the electrical characteristics detected by the detection unit when the estimated object presses the soft material as input, and estimates the shape information of the pressing member corresponding to the pressure stimulus applied by the estimated object to the soft material as the shape information of the estimated object. The electrical characteristics change in a time series with respect to the deformation of the soft material, and the shape information represents the shape of the pressing member that imparts the pressure stimulus that applies deformation to the soft material.

2. The estimation device according to claim 1, wherein, The flexible material is one whose electrical properties change accordingly with the deformation. The learning model learns by outputting shape information corresponding to the detected electrical characteristics.

3. The estimation device according to claim 1, wherein, The electrical property of the soft material is volume resistivity.

4. The estimation device according to claim 2, wherein, The electrical property of the soft material is volume resistivity.

5. The estimation apparatus according to any one of claims 1 to 4, wherein, The soft material is a material obtained by imparting conductivity to a polyurethane material with a fibrous skeleton structure or a structure in which multiple microbubbles are dispersed inside.

6. The estimation apparatus according to any one of claims 1 to 4, wherein, The learning model is generated by using the soft material as a reservoir and learning from a network based on reservoir computation that utilizes the reservoir.

7. The estimation apparatus according to claim 5, wherein, The learning model is generated by using the soft material as a reservoir and learning from a network based on reservoir computation that utilizes the reservoir.

8. An estimation method in which a computer performs the following processing: The electrical characteristics of the detection unit are obtained from multiple detection points that detect the electrical properties of a conductive soft material; and A learning model is used to learn electrical characteristics that change in accordance with the deformation of the soft material and shape information representing the shape of a pressing member that applies pressure stimulation to the soft material, which is given time-series information, as learning data, and in a manner that takes the electrical characteristics as input and outputs the shape information. The electrical characteristics obtained when the estimated object presses the soft material are input to estimate the shape information of the pressing member corresponding to the pressure stimulation applied by the estimated object to the soft material.

9. An estimation program for causing a computer to perform the following processes: The electrical characteristics of the detection unit are obtained from multiple detection points that detect the electrical properties of a conductive soft material; and A learning model is used to learn electrical characteristics that change in accordance with the deformation of the soft material and shape information representing the shape of a pressing member that applies pressure stimulation to the soft material, which is given time-series information, as learning data, and in a manner that takes the electrical characteristics as input and outputs the shape information. The electrical characteristics obtained when the estimated object presses the soft material are input to estimate the shape information of the pressing member corresponding to the pressure stimulation applied by the estimated object to the soft material.

10. A learning model generation apparatus, comprising: The acquisition unit acquires electrical characteristics and shape information, wherein the electrical characteristics are obtained from a detection unit that detects the electrical characteristics between multiple detection points of a conductive soft material, and the shape information is shape information representing the shape of a pressing member that imparts a pressure stimulus that applies deformation to the soft material; as well as The learning model generation unit generates a learning model based on the acquisition results of the acquisition unit. The learning model is a learning model that takes as input the electrical characteristics with time-series information that change in response to the deformation of the soft material when the object is pressed, and outputs the shape information of the pressing member corresponding to the pressure stimulus applied by the object to the soft material as the shape information of the object.

Images