DETAILED ACTION
This action is responsive to the claims filed on 06/15/2026. Claims 1, 4-8, and 15 are pending for examination.
This action is Final.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 03/12/206 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Arguments
Applicant’s remarks concerning the rejections under 35 U.S.C. §112 have been considered. Claims 17 and 18, which were rejected under 35 U.S.C. §112(b), and claims 12 and 16, which were rejected under 35 U.S.C. §112(d), have been canceled. Accordingly, the rejections of claims 12 and 16-18 under 35 U.S.C. §112 are withdrawn as moot.
Applicant’s arguments concerning the rejection under 35 U.S.C. §101 have been considered and are persuasive in view of the amendments. As amended, the pending independent claims recite a particular physical tactile-sensing configuration that detects, in a time series, volume resistance values between selected detection points in a conductive flexible material while the material is deformed by a pressure stimulus. The claims further apply a learning model trained using corresponding physical measurement data and pressure-stimulus shape information to estimate shape information of a target object. Accordingly, the claims are no longer merely directed to mental evaluation or generic data gathering and outputting, and the rejection of pending claims 1, 4-8, and 15 under 35 U.S.C. §101 is withdrawn.
Applicant’s arguments concerning the newly added limitations directed to time-series volume-resistance values, selected pairs of detection points, and shape-labeled learning data are moot because the previous ground of rejection has been replaced by a new ground based on Park in view of Drimus, Chen, and Liu. To the extent Applicant also argues that Park does not teach the previously recited estimation arrangement, the argument is not persuasive. Park expressly teaches that its trained DNN maps “boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures,” and further teaches that “[t]he training maps input of 208 voltage measurements to output of n × n conductivity distribution,” including preset spatial distributions such as “a shape of square that contributes on reconstruction of areal distribution.” Drimus further teaches the chronological electrical response relied upon in the present rejection, explaining that resistance “changes as a result of producing strain in the material with external force,” that squeezing the object “results in a sequence of measurements,” and that its classifier is used “to classify the time series resulting from the palpation procedure.” Thus, Park continues to teach applying detected electrical measurements to a trained model to estimate pressure-related spatial information, while Drimus supplies the chronological, deformation-responsive electrical-measurement and time-series processing teachings. Chen and Liu address the remaining volume-resistance and predetermined-shape limitations, respectively. Accordingly, Applicant’s arguments do not overcome the rejection as presently stated.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or non-obviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 4, and 6-8 are rejected under 35 U.S.C. 103 as being unpatentable over Park et al., (Park, H., Lee, H., Park, K., Mo, S., & Kim, J. (2019, November). Deep neural network approach in electrical impedance tomography-based real-time soft tactile sensor. In 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (pp. 7447-7452). IEEE.),), hereafter referred to as Park in view ofDrimus et al. (Drimus, A., Kootstra, G., Bilberg, A., & Kragic, D. (2014). “Design of a Flexible Tactile Sensor for Classification of Rigid and Deformable Objects.” Robotics and Autonomous Systems, 62, 3-15), hereinafter referred to as Drimus, and in further view of Chen et al., (“A Highly Sensitive Piezoresistive Pressure Sensor Based on Graphene Oxide/Polypyrrole@Polyurethane Sponge,” Sensors 2020, 20, 1219), hereinafter referred to as Chen, and Liu et al., (Liu, H., Greco, J., Song, X., Bimbo, J., Seneviratne, L., & Althoefer, K. (2012, September). Tactile image based contact shape recognition using neural network. In 2012 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI) (pp. 138-143). IEEE.), hereinafter referred to as Liu.
Claim 1: Park teaches:
An estimation device, (Park page 7448, col. 1, paragraph 3, “This paper presents a novel nonlinear EIT algorithm using Deep Neural Network (DNN) approach to improve sensing performance of tactile sensing applications. The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures.”, Park teaches a conductive-fabric tactile-sensing device having electronic measurement equipment and a trained deep neural network used to estimate an applied pressure distribution.
Park, page 7451, col. 1, section D, paragraph 1, “To show the applicability of the proposed model, the piezoresistive tactile sensor was demonstrated to estimate external normal pressure. Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure.” Park further demonstrates the trained DNN in an actual tactile-sensing device, thus, Park teaches an estimation device containing a tactile detection arrangement and a trained model that estimates pressure-related spatial information.)
comprising: a detection unit that detects an electrical property between a plurality of detection points in a conductive flexible material; and (Park, page 7448, section B, “To implement EIT-based tactile sensors, conductive medium and boundary electrodes are required. In this study, a stretchable piezoresistive fabric (COM-14112, Eeonyx, USA) is used to ensure large-scale manufacturability for whole-body tactile sensing in the future. The resistance of this fabric sensitively changes from normal pressure and lateral stretch… After that, conductive threads are stitched to the piezoresistive fabric as boundary electrodes…. In total, 16 electrodes were attached on the boundary of the piezoresistive fabric.”; Park, page 7449, section C, “Implementing EIT method requires electronic equipment to inject currents and measure voltages from the boundary electrodes.”, Park teaches a stretchable piezoresistive fabric having a plurality of boundary electrodes and electronic equipment that obtains electrical measurements between the electrodes. Accordingly, Park’s electronic equipment corresponds to the claimed detection unit, the sixteen boundary electrodes correspond to the plurality of detection points, and the piezoresistive fabric corresponds to the conductive flexible material.)
and an estimation unit that inputs, to a learning model that has been trained to use, as learning data, an electrical property… in response to deformation of the flexible material, (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7451, col. 2, section D, paragraph 1, “Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure.”, Park’s “boundary voltage measurements” are the detected “electrical property” acquired by the electrode-based detection arrangement. Park then discloses an “estimation” (inference) component implemented by a “DNN” that is trained to “estimate the conductivity distribution” from the boundary voltage measurements (input) to an inferred output (conductivity/pressure distribution). Thus, Park teaches an estimation unit that inputs the detected electrical property (boundary voltage measurements) to a trained learning model for inference regarding the state of the material/target. )
shape information representing shapes, in the flexible material, of pressure stimuli that impart deformation to the flexible material, (Park, page 7451, col. 2, section D, paragraph 1, “To show the applicability of the proposed model, the piezoresistive tactile sensor was demonstrated to estimate external normal pressure. Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure”,
Park, page 7449, col. 2, paragraph 1, “As shown in Figure 3, the first one is an imposition of conductivity difference on singe element to obtain individual effects of each elements. Next one is a shape of square that contributes on reconstruction of areal distribution. The last one is a selection of random elements. In this method, the random distribution was made as few as possible by inducing each element to be activated the same number of times in all datasets. Different amounts of conductivity differences were given for each distribution to obtain the ability to estimate the magnitude of the force input. As a result, total 76,776 dataset was generated.”, Park’s training framework expressly uses predefined spatial distributions (“preset distribution”) that include a specific area ”whose shape and can be moved, which constitutes “shape information representing shapes … in the flexible material.” Park further ties the conductivity distribution to applied loading by stating the inferred “conductivity distribution … corresponds to the external surface normal pressures,” i.e., pressure stimuli that deform the fabric. Accordingly, Park teaches training that uses learning data including shape-characterizing spatial patterns corresponding to pressure stimuli that impart deformation.)
and to receive the electrical property as input and to output the shape information, (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7449, col. 2, section B, paragraph 2, “The training maps input of 208 voltage measurements to output of n × n conductivity distribution… The output layer represents the reconstructed conductivity and the distribution can be visualized through reshaping”,
Park, page 7449, col. 2, paragraph 1, “As shown in Figure 3, the first one is an imposition of conductivity difference on singe element to obtain individual effects of each elements. Next one is a shape of square that contributes on reconstruction of areal distribution”, Park expressly discloses that the trained DNN receives “boundary voltage measurement” as the model input and produces an output “conductivity distribution” (“mapping boundary voltage measurement to conductivity distribution”), where the conductivity distribution “corresponds to external surface normal pressures,” i.e., the pressure pattern imposed on the tactile fabric. Park further clarifies the input/output relationship structurally: “input of 208 voltage measurements” is mapped to an “output of n × n conductivity distribution,” and the distribution is “visualized through reshaping,” meaning the output is a spatially arranged (image-like) distribution encoding the shape of the pressure/contact region across the material. That spatial output constitutes the claimed “shape information,” and Park’s dataset construction reinforces that the output distribution is shape-representative because it includes explicit geometric patterns such as a “shape of square” used to support “reconstruction of areal distribution,” i.e., reconstruction of a 2D shaped region.)
the electrical property of an estimation target object detected by the detection unit, (Park, page 7448, col. 2, section B, paragraph 1, “The resistance of this fabric sensitively changes from normal pressure and lateral stretch…. In total, 16 electrodes were attached on the boundary of the piezoresistive fabric.”
Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”, Park explains that the fabric’s “resistance… sensitively changes” under “normal pressure” and “lateral stretch,” confirming an “electrical property” that varies with deformation. Park also expressly measures “boundary voltage measurements” by between these electrodes, which constitutes detecting an electrical property between a plurality of detection points in the conductive flexible material as claimed.)
and that estimates shape information of the estimation target object. (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7450, col. 1, section A, paragraph 1, “The reconstruction performance of the proposed DNN model was evaluated using two predefined reference conductivity distributions which are a simple square and a complex geometry… From the forward solving, corresponding boundary voltages were obtained. These boundary voltage was used to reconstruct the conductivity distribution from two different conventional reconstruction methods… As shown in Figure 5 (a) and (c), the proposed model as well as the nonlinear iterative model (PDIPM) successfully reconstructed the reference conductivity distribution compared to the linear model”, Park describes the trained DNN as performing estimation/reconstruction of a spatial distribution from measured electrical data: Park obtains “boundary voltages” and uses them to “reconstruct the conductivity distribution”. For generating a shape, Park evaluates reconstruction on reference distributions having explicit geometric forms e.g. “a simple square and a complex geometry”, and reports that the proposed model “successfully reconstructed the reference conductivity distribution”. Because Park also states the reconstructed conductivity distribution “corresponds to external surface normal pressures,” the reconstructed (estimated) distribution is not merely a scalar pressure value, but a spatial pressure/contact pattern whose geometry (e.g., square vs. complex shape) is the claimed “shape information” of the pressed/loaded target on the material.)
Drimus, in the same field of flexible piezoresistive tactile sensing, teaches the following which Park and Chen fail to teach:
an electrical property that changes chronologically in response to deformation of the flexible material, (Drimus, page 5, section 3.1, teaches that the resistance of the flexible piezoresistive material changes as external force produces strain: “According to the percolation theory [15], the distribution state of the particles, and thereby the resistance, changes as a result of producing strain in the material with external force.” Drimus further teaches that “[t]he piezoresistive rubber shows an electrical resistance that ranges from about 0.5 kΩ in the compressed (on) state to several MΩ in the free (off) state” and evaluates the resistance while force is “linearly increased up to 300 gram-force and then decreased to 0 gram-force.” Drimus additionally states that “the characteristic resistance is not constant over time when a force is applied to the sensor for prolonged durations of time.” Drimus’s squeezing procedure produces dynamic information and “a sequence of measurements.” Accordingly, Drimus teaches a resistance that chronologically changes as external force deforms the flexible piezoresistive material.)
wherein the detection unit is connected to the plurality of detection points disposed at different positions in the conductive flexible material, and detects, in a time series, volume resistance values between a selected pair of detection points selected from the plurality of detection points while the conductive flexible material is deformed by a pressure stimulus, (Drimus, page 6, section 3.2, teaches conductive-thread electrodes arranged as rows on one side of a flexible piezoresistive material and columns on the other side and states: “In this way, by selecting only one column and one row, the information from a single element can be read.” Drimus, page 7, section 3.3.1, further explains that “[t]he rows feed current through the material, which is further conducted to ground over the columns” and that “[t]he material can be modeled as a matrix array of variable resistors, placed at each overlapping of rows and columns.” When one row and one column are selected, “the voltage collected over the Rg resistance will be proportional to the resistance exhibited by the material at their overlapping.” Thus, Drimus detects the resistance exhibited through the body of the piezoresistive material between a selected row-and-column electrode pair, corresponding to the claimed volume-resistance value. Drimus, page 8, sections 3.3.2 and 4, teaches scanning the sensor at 100 frames per second and providing a tactile image every 10 ms. Drimus further teaches actively squeezing an object, whereby “the tactile sensors are stimulated, which results in a sequence of measurements.” Accordingly, Drimus teaches detecting time-series volume-resistance values between selected detection points while a pressure stimulus deforms the conductive flexible material.)
wherein the learning data includes a plurality of data sets, each data set including: (Drimus, page 9, section 4.4, teaches storing multiple time-series training observations: “We use a k nearest neighbors (kNN) classification method to classify the time series resulting from the palpation procedure. A number of training examples are stored for each object. A new observation is compared to the training data…. One palpation procedure results in an observation, z, which consists of μ and σ time series for both fingertip sensors.”, Accordingly, Drimus teaches learning data including a plurality of stored training data sets, with each data set corresponding to a respective palpation procedure and including chronological tactile information obtained during that procedure.)
a time series of volume resistance values measured between a selected pair of detection points when the conductive flexible material is deformed by a corresponding pressure stimulus (Drimus, page 8, sections 3.3.2 and 4, further teaches scanning the resistance-responsive sensor values at 100 frames per second and obtaining a tactile image every 10 ms during a pressure-producing palpation: “By doing so, the tactile sensors are stimulated, which results in a sequence of measurements.” Drimus, page 9, section 4.4, expressly uses those chronological measurements as stored training observations: “We use a k nearest neighbors (kNN) classification method to classify the time series resulting from the palpation procedure. A number of training examples are stored for each object… One palpation procedure results in an observation, z, which consists of μ and σ time series for both fingertip sensors.” Each μ and σ value is calculated from a chronological tactile frame containing the row-and-column resistance-responsive values. Accordingly, Drimus teaches training data that include time series generated from volume-resistance values measured between selected row-and-column detection points while a corresponding squeezing-pressure stimulus deforms the flexible piezoresistive material.)
wherein the learning model has been trained to receive, as input, the time series of volume resistance values included in the data sets (Drimus, page 9, section 4.4, teaches a k-nearest-neighbor learning model that receives a newly acquired time-series observation and compares it to stored time-series training observations: “We use a k nearest neighbors (kNN) classification method to classify the time series resulting from the palpation procedure. A number of training examples are stored for each object. A new observation is compared to the training data.” Drimus further teaches that the model calculates the distance directly between the time series: “In order to calculate the distance between the time series, we use the Dynamic Time Warping algorithm.” Drimus identifies each model observation as a time series derived from the chronological tactile frames: “One palpation procedure results in an observation, z, which consists of μ and σ time series for both fingertip sensors.” The chronological tactile frames contain the volume-resistance-responsive values measured at the selected row-and-column intersections. Drimus’s disclosed implementation calculates μ and σ time-series features from each chronological matrix before classification. Drimus therefore teaches receiving, as model input, time-series information generated directly from the volume-resistance values included in the stored training data sets. Further, Drimus emphasizes that its classification procedure “takes advantage of the complete time series,” because the dynamic information permits more elaborate discrimination than static tactile information.)
the time series of volume resistance values detected by the detection unit with respect to the estimation target object. (Drimus teaches positioning a target object between the gripper jaws, squeezing the target object, detecting the resulting chronological resistance-responsive tactile measurements, and supplying the resulting new observation to the classifier. Drimus, page 8, section 4.1, states: “During the experiments, the objects were manually placed in between the gripper jaws. The palpation procedure started by closing the gripper’s fingers until contact was established.” Drimus further teaches that squeezing stimulates the tactile sensors and “results in a sequence of measurements” and that “[t]his sequence is then used to recognize which object is palpated.” Drimus, page 9, section 4.4, expressly states: “A new observation is compared to the training data.” The new target-object observation comprises time-series information generated from the selected row-and-column volume-resistance-responsive measurements. Accordingly, Drimus teaches detecting, with respect to an estimation target object, the target-object time series that is supplied as the new input observation to the learning model.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the tactile-sensing system of Park in accordance with Drimus. Park teaches a flexible piezoresistive tactile sensor that sequentially obtains electrical measurements from plural electrodes and applies the measurements to a trained DNN to reconstruct a spatial conductivity distribution corresponding to external pressure. Drimus teaches that the resistance of a flexible piezoresistive material changes as external force produces strain and that chronological measurements acquired during deformation provide useful dynamic information. In particular, Drimus explains that a static tactile reading is insufficient for its classification task and that “[i]t is necessary to look at the dynamic sensory patterns resulting from the motor action.” It would therefore have been obvious to retain Park’s sequential electrical measurements as chronological measurement data during deformation, as taught by Drimus, so that Park’s learning model could account for the dynamic electrical response of the flexible material rather than only an isolated measurement. Such a modification would have predictably improved the ability of Park’s tactile-sensing system to distinguish pressure-induced deformation patterns and would have involved using Drimus’s known time-series tactile-sensing technique for its established purpose, without changing Park’s principle of operation.
Chen in the same field of machine learning, teaches the following which Park and Drimus fails to teach:
wherein the conductive flexible material is a material having an electrical property that changes in response to deformation, (Chen teaches a conductive polyurethane sponge whose electrical resistance changes in response to compression deformation. Chen page 1, section 1, states:
“Piezoresistive pressure sensors can convert pressure changes into resistance changes.”
Chen page 6, section 3.3, further teaches that:
“When the GO/PPy@PU conductive sponge was compressed, the bending of the skeleton caused tension on the GO/PPy conductive layers.”
Chen, page 7, paragraph 1, explains that compression changes the conductive paths and resistance of the sponge:
“Moreover, the compression of the sponge under small deformation would lead to the rapid change of the contact between the conductive skeleton of GO/PPy and the contact area, which would cause the breakage-recovery of local conductive path.”
Accordingly, Chen teaches a conductive flexible material having an electrical resistance that changes in response to deformation. )
wherein the electrical property of the conductive flexible material is volume resistance, (Chen, page 7, paragraph 4, expressly characterizes the electrical property of the conductive sponge as volume resistance:
“The volume resistance and volume parameters of the conductive sponge can be described by the Equation (3): R =σm L/S where R represents resistance of the conductive sponge, L and S reveals thickness and cross-sectional area of conductive sponge.”
Chen, in the same section, further distinguishes the resistance in the uncompressed and compressed states:
“where R_0, R_P respectively represents the resistance of the conductive sponge in the initial state and an arbitrary compressed state, L_0, L_P respectively represents the thickness of the conductive sponge in the initial state and an arbitrary compressed state.”
Thus, Chen expressly teaches that the deformation-responsive electrical property is volume resistance.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to characterize the pressure-responsive electrical property measured in Park and Drimus using Chen’s volume-resistance measurement. Both Park and Chen concern flexible piezoresistive pressure sensors in which deformation changes the material’s electrical resistance. Park applies a known constant current and measures voltage between selected electrode pairs. Under the known relationship R=V/I, the resistance corresponding to each selected pair is directly determinable. Chen expressly teaches representing the deformation-sensitive property as volume resistance based on the material’s resistance and dimensions. Employing Chen’s volume-resistance representation in Park would therefore have constituted the use of a known electrical characterization technique for its established purpose of quantifying pressure-induced deformation of a conductive flexible material.
Liu in the same field of machine learning, taches the following which Park, Drimus, and Chen fails to teach:
having a predetermined shape, and (Liu, page 142, section IV.A, teaches intentionally producing tactile observations using four predetermined physical contact geometries: “In this paper, four different contact shapes have been used for recognition, including edge, sphere, ring and rectangle. For each shape, 40 tests were carried out repetitively.” Liu’s Table I further specifies the predetermined geometries and dimensions of the physical contact objects, including a right-angle edge, a sphere having an 8 mm radius, a ring having specified inner and outer radii, and a rectangle having a specified length. When each shaped contact object presses against the tactile sensor, that physical contact produces the corresponding pressure stimulus. Accordingly, Liu teaches that the corresponding pressure stimulus has a predetermined shape.)
shape information representing the shape of the corresponding pressure stimulus, (Liu, page 139, section III, explains that the tactile data are generated when a physical object contacts and presses against the tactile sensing elements: “When a group of pressure sensing elements is in contact with an object, the output from the tactile sensing system to the algorithm is in the form of a pressure map (PM) as indicated in Fig. 1(b).”
Liu, page 142, section A, assigns each tactile test to one of four expressly identified contact shapes: “In this paper, four different contact shapes have been used for recognition, including edge, sphere, ring and rectangle.”
Liu, page 142 identifies the corresponding shape information in Table I as:
“Edge 1 Right angle edge”“Sphere 2 8 mm in radius”“Ring 3 Out radius = 10 mm, inner radius = 4 mm”“Rectangle 4 Length = 10 mm”, The edge, sphere, ring, and rectangle are the physical contact geometries that press against the tactile sensor and produce the corresponding pressure stimuli and pressure maps. The assigned class identifiers and geometrical descriptions therefore constitute shape information representing the shape of the corresponding pressure stimulus.)
and to output the shape information corresponding to the volume resistance values included in the data sets, and (Liu, page 141, section III.D, teaches training a neural-network classifier to output the contact-shape class corresponding to the tactile input: “To use the extracted feature from the pressure map for object classification, a neural network has been developed.” Liu further teaches that the neural network includes “512 input nodes, 10 hidden nodes and 4 output nodes; i.e. a 512-10-4 neural network structure.” Liu, page 142, section IV.A, identifies the four output classes as edge, sphere, ring, and rectangle and states: “It can be seen from Fig. 8 that using the extracted feature vector, the 3-layer neural network can be efficiently trained for classifying different contact shape.” Thus, Liu teaches a trained learning model that outputs the shape class corresponding to the tactile values supplied to the model. In the proposed combination, those tactile input values are the chronological volume-resistance-responsive values taught by Drimus. Accordingly, Liu teaches outputting shape information corresponding to the volume-resistance data sets used to train the combined learning model.)
wherein the estimation unit estimates the shape information of the estimation target object by inputting, to the learning model, (Liu’s Abstract teaches applying tactile information obtained from a contacted target object to a trained neural-network classifier: “To recognize different contact shape from a pressure map, a neural network classifier is developed and uses the feature vector as inputs.” Liu, page 139, section III, further states: “The developed algorithm uses a 3-layer neural network to classify the contact shape of an object based on the characteristics of the PM.” Liu reports that the trained neural network achieved an average identification accuracy of 91.07%. Accordingly, Liu teaches estimating shape information of a contacted target object by inputting tactile information obtained for that target object into a trained learning model.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Liu’s predetermined contact shapes and corresponding shape labels as the target information associated with the time-series tactile measurements taught by Chen and the pair-selected measurements taught by Park and Drimus. Park already trains a learning model using preset spatial distributions, including “a shape of square,” and reconstructs arbitrary pressure distributions. Liu teaches that directly labeling tactile pressure observations by predetermined contact shape permits a learning model to output useful contact-shape information. Applying Liu’s known shape-labeling technique to the dynamic measurement sets of Park and Chen would have predictably produced training data in which each time series is associated with the known shape of the pressure stimulus that generated the series.
Claim 4: Park, Drimus, Chen and Liu teaches the limitations of claim 1, Chen further teaches:
PNG
media_image1.png
384
594
media_image1.png
Greyscale
Figure 2 of Chen
The estimation device of claim 1, wherein the flexible material is a material in which conductivity is imparted to a urethane material having a structure having a fibrous skeleton or a structure having a plurality of microscopic air bubbles scattered inside. (Chen, abstract, “In this work, polyurethane sponge is employed as the structural substrate of the sensor. Graphene oxide (GO) and polypyrrole (PPy) are alternately coated on the sponge fiber skeleton by charge layer-by-layer assembly (LBL) to form a multilayer composite conductive layer to prepare the piezoresistive sensors.”,
Chen, page 6, section 3.3, further teaches that the polyurethane sponge has a three-dimensional network formed by a fiber skeleton: “The original PU sponge had a 3D porous network structure, which was integrally connected by a randomly distributed fiber skeleton.”, Thus, Chen teaches imparting conductivity to a urethane material having a fibrous-skeleton structure. Chen need not additionally teach the alternative limitation concerning microscopic air bubbles because claim 4 recites the fibrous-skeleton structure and the microscopic-air-bubble structure in the alternative.)
Claim 6: Claim 6 recites limitations substantially similar to claims 1, as such a similar analysis applies. Claim 6 also recites the following additional limitations for consideration:
An estimation method, comprising, by a computer: (Park page 7448, col. 1, paragraph 3, “This paper presents a novel nonlinear EIT algorithm using Deep Neural Network (DNN) approach to improve sensing performance of tactile sensing applications. The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures.”, Park teaches a conductive-fabric tactile-sensing device having electronic measurement equipment and a trained deep neural network used to estimate an applied pressure distribution.
Park, page 7451, col. 1, section D, paragraph 1, “To show the applicability of the proposed model, the piezoresistive tactile sensor was demonstrated to estimate external normal pressure. Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure.” Park further demonstrates the trained DNN in an actual tactile-sensing device, thus, Park teaches an estimation device containing a tactile detection arrangement and a trained model that estimates pressure-related spatial information.)
acquiring, from a detection unit that detects an electrical property between a plurality of detection points in a conductive flexible material, the electrical property; and (Park, page 7448, section B, “To implement EIT-based tactile sensors, conductive medium and boundary electrodes are required. In this study, a stretchable piezoresistive fabric (COM-14112, Eeonyx, USA) is used to ensure large-scale manufacturability for whole-body tactile sensing in the future. The resistance of this fabric sensitively changes from normal pressure and lateral stretch… After that, conductive threads are stitched to the piezoresistive fabric as boundary electrodes…. In total, 16 electrodes were attached on the boundary of the piezoresistive fabric.”; Park, page 7449, section C, “Implementing EIT method requires electronic equipment to inject currents and measure voltages from the boundary electrodes.”, Park teaches a stretchable piezoresistive fabric having a plurality of boundary electrodes and electronic equipment that obtains electrical measurements between the electrodes. Accordingly, Park’s electronic equipment corresponds to the claimed detection unit, the sixteen boundary electrodes correspond to the plurality of detection points, and the piezoresistive fabric corresponds to the conductive flexible material.)
and inputting, to a learning model that has been trained to use, as learning data, an electrical property… that changes in response to deformation of the flexible material, (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7451, col. 2, section D, paragraph 1, “Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure.”, Park’s “boundary voltage measurements” are the detected “electrical property” acquired by the electrode-based detection arrangement. Park then discloses an “estimation” (inference) component implemented by a “DNN” that is trained to “estimate the conductivity distribution” from the boundary voltage measurements (input) to an inferred output (conductivity/pressure distribution). Thus, Park teaches an estimation unit that inputs the detected electrical property (boundary voltage measurements) to a trained learning model for inference regarding the state of the material/target. )
and shape information of pressure stimuli that impart deformation to the flexible material, (Park, page 7451, col. 2, section D, paragraph 1, “To show the applicability of the proposed model, the piezoresistive tactile sensor was demonstrated to estimate external normal pressure. Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure”,
Park, page 7449, col. 2, paragraph 1, “As shown in Figure 3, the first one is an imposition of conductivity difference on singe element to obtain individual effects of each elements. Next one is a shape of square that contributes on reconstruction of areal distribution. The last one is a selection of random elements. In this method, the random distribution was made as few as possible by inducing each element to be activated the same number of times in all datasets. Different amounts of conductivity differences were given for each distribution to obtain the ability to estimate the magnitude of the force input. As a result, total 76,776 dataset was generated.”, Park’s training framework expressly uses predefined spatial distributions (“preset distribution”) that include a specific area ”whose shape and can be moved, which constitutes “shape information representing shapes … in the flexible material.” Park further ties the conductivity distribution to applied loading by stating the inferred “conductivity distribution … corresponds to the external surface normal pressures,” i.e., pressure stimuli that deform the fabric. Accordingly, Park teaches training that uses learning data including shape-characterizing spatial patterns corresponding to pressure stimuli that impart deformation.)
and to receive the electrical property as input and to output the shape information, (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7449, col. 2, section B, paragraph 2, “The training maps input of 208 voltage measurements to output of n × n conductivity distribution… The output layer represents the reconstructed conductivity and the distribution can be visualized through reshaping”,
Park, page 7449, col. 2, paragraph 1, “As shown in Figure 3, the first one is an imposition of conductivity difference on singe element to obtain individual effects of each elements. Next one is a shape of square that contributes on reconstruction of areal distribution”, Park expressly discloses that the trained DNN receives “boundary voltage measurement” as the model input and produces an output “conductivity distribution” (“mapping boundary voltage measurement to conductivity distribution”), where the conductivity distribution “corresponds to external surface normal pressures,” i.e., the pressure pattern imposed on the tactile fabric. Park further clarifies the input/output relationship structurally: “input of 208 voltage measurements” is mapped to an “output of n × n conductivity distribution,” and the distribution is “visualized through reshaping,” meaning the output is a spatially arranged (image-like) distribution encoding the shape of the pressure/contact region across the material. That spatial output constitutes the claimed “shape information,” and Park’s dataset construction reinforces that the output distribution is shape-representative because it includes explicit geometric patterns such as a “shape of square” used to support “reconstruction of areal distribution,” i.e., reconstruction of a 2D shaped region.)
the acquired electrical property of an estimation target object, (Park, page 7448, col. 2, section B, paragraph 1, “The resistance of this fabric sensitively changes from normal pressure and lateral stretch…. In total, 16 electrodes were attached on the boundary of the piezoresistive fabric.”
Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”, Park explains that the fabric’s “resistance… sensitively changes” under “normal pressure” and “lateral stretch,” confirming an “electrical property” that varies with deformation. Park also expressly measures “boundary voltage measurements” by between these electrodes, which constitutes detecting an electrical property between a plurality of detection points in the conductive flexible material as claimed.)
and estimating shape information of the estimation target object. (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7450, col. 1, section A, paragraph 1, “The reconstruction performance of the proposed DNN model was evaluated using two predefined reference conductivity distributions which are a simple square and a complex geometry… From the forward solving, corresponding boundary voltages were obtained. These boundary voltage was used to reconstruct the conductivity distribution from two different conventional reconstruction methods… As shown in Figure 5 (a) and (c), the proposed model as well as the nonlinear iterative model (PDIPM) successfully reconstructed the reference conductivity distribution compared to the linear model”, Park describes the trained DNN as performing estimation/reconstruction of a spatial distribution from measured electrical data: Park obtains “boundary voltages” and uses them to “reconstruct the conductivity distribution”. For generating a shape, Park evaluates reconstruction on reference distributions having explicit geometric forms e.g. “a simple square and a complex geometry”, and reports that the proposed model “successfully reconstructed the reference conductivity distribution”. Because Park also states the reconstructed conductivity distribution “corresponds to external surface normal pressures,” the reconstructed (estimated) distribution is not merely a scalar pressure value, but a spatial pressure/contact pattern whose geometry (e.g., square vs. complex shape) is the claimed “shape information” of the pressed/loaded target on the material.)
Drimus further teaches:
an electrical property that changes chronologically in response to deformation of the flexible material, (Drimus, page 5, section 3.1, teaches that the resistance of the flexible piezoresistive material changes as external force produces strain: “According to the percolation theory [15], the distribution state of the particles, and thereby the resistance, changes as a result of producing strain in the material with external force.” Drimus further teaches that “[t]he piezoresistive rubber shows an electrical resistance that ranges from about 0.5 kΩ in the compressed (on) state to several MΩ in the free (off) state” and evaluates the resistance while force is “linearly increased up to 300 gram-force and then decreased to 0 gram-force.” Drimus additionally states that “the characteristic resistance is not constant over time when a force is applied to the sensor for prolonged durations of time.” Drimus’s squeezing procedure produces dynamic information and “a sequence of measurements.” Accordingly, Drimus teaches a resistance that chronologically changes as external force deforms the flexible piezoresistive material.)
Claim 7: Claim 7 recites limitations substantially similar to claims 1, as such a similar analysis applies. Claim 7 also recites the following additional limitations for consideration:
A non-transitory computer-readable medium storing a program for causing a computer to (Reyes, page 651, col. 2, section F.2, “The models in Table III were trained at the CPU level, except for the optimal spatio-temporal RSM model that was trained at the GPU level. Here longer training times are observed for the SVM models with the reconstructed image features. This effect is caused mainly due to the high dimensionality of the feature vector since the execution time is strongly dependent on the number of features and increases in parallel with the memory consumption of each training sample.”, it is interpreted by the examiner that the use of a processing unit (CPU) and memory encompasses the non-transitory computer-readable medium storing a program for causing the computer to perform, as recited in the claim.)
perform processing, the processing comprising: acquiring, from a detection unit that detects an electrical property between a plurality of detection points in a conductive flexible material, the electrical property; (Park, page 7448, col. 2, section B, paragraph 1, “The resistance of this fabric sensitively changes from normal pressure and lateral stretch…. In total, 16 electrodes were attached on the boundary of the piezoresistive fabric.”
Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”, Park describes a sensing arrangement in which a conductive, stretchable/piezoresistive fabric (i.e., a “conductive flexible material”) is provided with “16 electrodes” positioned “on the boundary,” which are discrete “detection points.” Park further explains that the fabric’s “resistance… sensitively changes” under “normal pressure” and “lateral stretch,” confirming an “electrical property” that varies with deformation. Park also expressly measures “boundary voltage measurements” between these electrodes, which constitutes detecting an electrical property between a plurality of detection points in the conductive flexible material as claimed.)
and inputting, to a learning model that has been trained to use, as learning data, an electrical property… that changes in response to deformation of the flexible material, (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7451, col. 2, section D, paragraph 1, “Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure.”, Park’s “boundary voltage measurements” are the detected “electrical property” acquired by the electrode-based detection arrangement. Park then discloses an “estimation” (inference) component implemented by a “DNN” that is trained to “estimate the conductivity distribution” from the boundary voltage measurements (input) to an inferred output (conductivity/pressure distribution). Thus, Park teaches an estimation unit that inputs the detected electrical property (boundary voltage measurements) to a trained learning model for inference regarding the state of the material/target. )
and shape information of pressure stimuli that impart deformation to the flexible material, (Park, page 7451, col. 2, section D, paragraph 1, “To show the applicability of the proposed model, the piezoresistive tactile sensor was demonstrated to estimate external normal pressure. Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure”,
Park, page 7449, col. 2, paragraph 1, “As shown in Figure 3, the first one is an imposition of conductivity difference on singe element to obtain individual effects of each elements. Next one is a shape of square that contributes on reconstruction of areal distribution. The last one is a selection of random elements. In this method, the random distribution was made as few as possible by inducing each element to be activated the same number of times in all datasets. Different amounts of conductivity differences were given for each distribution to obtain the ability to estimate the magnitude of the force input. As a result, total 76,776 dataset was generated.”, Park’s training framework expressly uses predefined spatial distributions (“preset distribution”) that include a specific area ”whose shape and can be moved, which constitutes “shape information representing shapes … in the flexible material.” Park further ties the conductivity distribution to applied loading by stating the inferred “conductivity distribution … corresponds to the external surface normal pressures,” i.e., pressure stimuli that deform the fabric. Accordingly, Park teaches training that uses learning data including shape-characterizing spatial patterns corresponding to pressure stimuli that impart deformation.)
and to receive the electrical property as input and to output the shape information, (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7449, col. 2, section B, paragraph 2, “The training maps input of 208 voltage measurements to output of n × n conductivity distribution… The output layer represents the reconstructed conductivity and the distribution can be visualized through reshaping”,
Park, page 7449, col. 2, paragraph 1, “As shown in Figure 3, the first one is an imposition of conductivity difference on singe element to obtain individual effects of each elements. Next one is a shape of square that contributes on reconstruction of areal distribution”, Park expressly discloses that the trained DNN receives “boundary voltage measurement” as the model input and produces an output “conductivity distribution” (“mapping boundary voltage measurement to conductivity distribution”), where the conductivity distribution “corresponds to external surface normal pressures,” i.e., the pressure pattern imposed on the tactile fabric. Park further clarifies the input/output relationship structurally: “input of 208 voltage measurements” is mapped to an “output of n × n conductivity distribution,” and the distribution is “visualized through reshaping,” meaning the output is a spatially arranged (image-like) distribution encoding the shape of the pressure/contact region across the material. That spatial output constitutes the claimed “shape information,” and Park’s dataset construction reinforces that the output distribution is shape-representative because it includes explicit geometric patterns such as a “shape of square” used to support “reconstruction of areal distribution,” i.e., reconstruction of a 2D shaped region.)
the acquired electrical property of an estimation target object, (Park, page 7448, col. 2, section B, paragraph 1, “The resistance of this fabric sensitively changes from normal pressure and lateral stretch…. In total, 16 electrodes were attached on the boundary of the piezoresistive fabric.”
Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”, Park explains that the fabric’s “resistance… sensitively changes” under “normal pressure” and “lateral stretch,” confirming an “electrical property” that varies with deformation. Park also expressly measures “boundary voltage measurements” by between these electrodes, which constitutes detecting an electrical property between a plurality of detection points in the conductive flexible material as claimed.)
and estimating shape information of the estimation target object. (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7450, col. 1, section A, paragraph 1, “The reconstruction performance of the proposed DNN model was evaluated using two predefined reference conductivity distributions which are a simple square and a complex geometry… From the forward solving, corresponding boundary voltages were obtained. These boundary voltage was used to reconstruct the conductivity distribution from two different conventional reconstruction methods… As shown in Figure 5 (a) and (c), the proposed model as well as the nonlinear iterative model (PDIPM) successfully reconstructed the reference conductivity distribution compared to the linear model”, Park describes the trained DNN as performing estimation/reconstruction of a spatial distribution from measured electrical data: Park obtains “boundary voltages” and uses them to “reconstruct the conductivity distribution”. For generating a shape, Park evaluates reconstruction on reference distributions having explicit geometric forms e.g. “a simple square and a complex geometry”, and reports that the proposed model “successfully reconstructed the reference conductivity distribution”. Because Park also states the reconstructed conductivity distribution “corresponds to external surface normal pressures,” the reconstructed (estimated) distribution is not merely a scalar pressure value, but a spatial pressure/contact pattern whose geometry (e.g., square vs. complex shape) is the claimed “shape information” of the pressed/loaded target on the material.)
Claim 8: Claim 8 recites limitations substantially similar to claims 1, as such a similar analysis applies. Claim 8 also recites the following additional limitations for consideration:
A learning model generation device, comprising: an acquisition unit that acquires, from a detection unit that detects an electrical property between a plurality of detection points in a conductive flexible material, the electrical property, (Park, page 7448, col. 2, section B, paragraph 1, “The resistance of this fabric sensitively changes from normal pressure and lateral stretch…. In total, 16 electrodes were attached on the boundary of the piezoresistive fabric.”
Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”, Park describes a sensing arrangement in which a conductive, stretchable/piezoresistive fabric (i.e., a “conductive flexible material”) is provided with “16 electrodes” positioned “on the boundary,” which are discrete “detection points.” Park further explains that the fabric’s “resistance… sensitively changes” under “normal pressure” and “lateral stretch,” confirming an “electrical property” that varies with deformation. Park also expressly measures “boundary voltage measurements” between these electrodes, which constitutes detecting an electrical property between a plurality of detection points in the conductive flexible material as claimed.)
and acquires shape information of pressure stimuli that impart deformation to the flexible material; (Park, page 7451, col. 2, section D, paragraph 1, “To show the applicability of the proposed model, the piezoresistive tactile sensor was demonstrated to estimate external normal pressure. Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure”,
Park, page 7449, col. 2, paragraph 1, “As shown in Figure 3, the first one is an imposition of conductivity difference on singe element to obtain individual effects of each elements. Next one is a shape of square that contributes on reconstruction of areal distribution. The last one is a selection of random elements. In this method, the random distribution was made as few as possible by inducing each element to be activated the same number of times in all datasets. Different amounts of conductivity differences were given for each distribution to obtain the ability to estimate the magnitude of the force input. As a result, total 76,776 dataset was generated.”, Park’s training framework expressly uses predefined spatial distributions (“preset distribution”) that include a specific area ”whose shape and can be moved, which constitutes “shape information representing shapes … in the flexible material.” Park further ties the conductivity distribution to applied loading by stating the inferred “conductivity distribution … corresponds to the external surface normal pressures,” i.e., pressure stimuli that deform the fabric. Accordingly, Park teaches training that uses learning data including shape-characterizing spatial patterns corresponding to pressure stimuli that impart deformation.)
and a learning model generation unit that generates, based on acquisition results of the acquisition unit, a learning model that has been trained to receive, as input, an electrical property… that changes in response to deformation of the flexible material, (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7451, col. 2, section D, paragraph 1, “Using the piezoresistive tactile sensor and the electronic equipment, the real boundary voltages were measured. The boundary voltages were applied to the trained DNN model to estimate the conductivity distribution, which corresponds to the external normal pressure.”, Park’s “boundary voltage measurements” are the detected “electrical property” acquired by the electrode-based detection arrangement. Park then discloses an “estimation” (inference) component implemented by a “DNN” that is trained to “estimate the conductivity distribution” from the boundary voltage measurements (input) to an inferred output (conductivity/pressure distribution). Thus, Park teaches an estimation unit that inputs the detected electrical property (boundary voltage measurements) to a trained learning model for inference regarding the state of the material/target. )
and to output shape information of a target object. (Park, page 7448, col 1, paragraph 3, “The deep neural network is trained from nonlinear EIT dataset and it is used to mapping boundary voltage measurement to conductivity distribution, which corresponds to external surface normal pressures (See Figure 1 (b)).”,
Park, page 7450, col. 1, section A, paragraph 1, “The reconstruction performance of the proposed DNN model was evaluated using two predefined reference conductivity distributions which are a simple square and a complex geometry… From the forward solving, corresponding boundary voltages were obtained. These boundary voltage was used to reconstruct the conductivity distribution from two different conventional reconstruction methods… As shown in Figure 5 (a) and (c), the proposed model as well as the nonlinear iterative model (PDIPM) successfully reconstructed the reference conductivity distribution compared to the linear model”, Park describes the trained DNN as performing estimation/reconstruction of a spatial distribution from measured electrical data: Park obtains “boundary voltages” and uses them to “reconstruct the conductivity distribution”. For generating a shape, Park evaluates reconstruction on reference distributions having explicit geometric forms e.g. “a simple square and a complex geometry”, and reports that the proposed model “successfully reconstructed the reference conductivity distribution”. Because Park also states the reconstructed conductivity distribution “corresponds to external surface normal pressures,” the reconstructed (estimated) distribution is not merely a scalar pressure value, but a spatial pressure/contact pattern whose geometry (e.g., square vs. complex shape) is the claimed “shape information” of the pressed/loaded target on the material.)
Claims 5 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Park in view of Drimus and in further view of Chen, Liu and Nakajima et al., (Nakajima, K., Hauser, H., Li, T., & Pfeifer, R. (2015). Information processing via physical soft body. Scientific reports, 5(1), 10487.), hereafter referred to as Nakajima.
Claim 5: Park, Drimus, Chen and Liu teaches the limitations of claim 1, Nakajima, in the same field of electrical analysis over flexible materials, teaches the following which Park, Drimus, Chen and Liu fail to teach:
The estimation device of claim 1, wherein the learning model is a model generated by training the model using, with the flexible material as a reservoir, a network obtained by reservoir computing using the reservoir. (Nakajima, figure 1, “Platform setup for a soft silicone arm and schematics showing the information processing scheme using the arm… Schematics expressing an analogy between a conventional reservoir computing system and our system. In a conventional reservoir system, randomly coupled abstract computational units are used for the reservoir, whereas our system exploits a physical reservoir whose units are sensors that are coupled through a soft silicone material.”,
Nakajima, page 2, paragraph 4, “By generating passive body dynamics resulting from the interaction between the water and the soft silicone material29,30, we will show that the sensory time series reflected in the body dynamics can be used to emulate the desired nonlinear dynamical systems, which are often targeted with a recurrent neural network learning or reservoir computing approach… For this purpose, we first need to define how to provide inputs I t( ) to the system and how to generate corresponding outputs O t( + 1). In this study, we apply the motor command as an input, and the output is generated by a weighted sum of the 10 sensory values and a constant valued bias set to 1.0 Fig. 1(a)”, Nakajima explicitly frames the disclosed system as “reservoir computing,” where the system “exploits a physical reservoir that are coupled through a soft silicone material” (a flexible body serving as the reservoir) training/deriving the readout by collecting “target outputs over time,” and generating outputs via a “weighted sum of the sensory values,” which is interpreted by the examiner to be reservoir-computing approach (fixed reservoir dynamics + trained network output). Therefore, Nakajima teaches generating a learning model by training a reservoir-computing network that uses a flexible physical material as the reservoir. )
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use the conductive flexible material of the Park, Drimus, Chen and Liu sensor as the physical reservoir of a reservoir-computing network, as taught by Nakajima. Both systems exploit nonlinear, time-dependent physical responses of flexible materials and embedded or connected sensors. Nakajima expressly teaches that the particular physical implementation of the reservoir is not controlling, provided that the system exhibits the necessary nonlinear dynamics and fading memory. Using the pressure-responsive flexible material itself as the reservoir would have predictably allowed the material’s physical dynamics to participate in processing the tactile time series, while training the output readout to produce the desired pressure-shape information. (Nakajima, page 3, figure 1, “our system exploits a physical reservoir whose units are sensors that are coupled through a soft silicone material.”)
Claim 15 recites limitations substantially similar to claim 5, as such a similar analysis applies.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Yao, A., Yang, C. L., Seo, J. K., & Soleimani, M. (2013). EIT‐Based Fabric Pressure Sensing. Computational and mathematical methods in medicine, 2013(1), 405325.
Duan, X., Taurand, S., & Soleimani, M. (2019). Artificial skin through super-sensing method and electrical impedance data from conductive fabric with aid of deep learning. Scientific reports, 9(1), 8831.
Liu, S., Cao, R., Huang, Y., Ouypornkochagorn, T., & Jia, J. (2020). Time sequence learning for electrical impedance tomography using Bayesian spatiotemporal priors. IEEE Transactions on Instrumentation and Measurement, 69(9), 6045-6057.
US20190227667A1 - Touch-sensing system
US10605680B2 - Devices for static and dynamic body measurements
US20210373707A1 - Touch-sensing system including a touch-sensitive paper
US20110029470A1 - Systems, methods, and apparatus for reconstruction of 3-d object morphology, position, orientation and texture using an array of tactile sensors
US 2019/0227667 A1
Nakamoto, H., Fukui, W., Kobayashi, F., Kojima, F., Imamura, N., & Shirasawa, H. (2009, November). Shape classification based on tactile information by Universal Robot Hand. In 2009 35th Annual Conference of IEEE Industrial Electronics (pp. 2360-2365). IEEE.
US 2009/0133508 A1
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HYUNGJUN B YI whose telephone number is (703)756-4799. The examiner can normally be reached M-F 9-5.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Usmaan Saeed can be reached on (571) 272-4046. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/H.B.Y./Examiner, Art Unit 2146
/DANIEL T PELLETT/Primary Examiner, Art Unit 2121
Prosecution Insights