Generalized contour data processing method, object cognition method and device

Abstract

The invention provides a generalized contour data processing method, an object cognition method and equipment. The method is applied to robot sensing technology. The method comprises the steps that feature learning is carried out according to generalized contour data of a target object, generalized contour codes are obtained, the generalized contour data and the generalized contour codes both comprise generalized contour features of a plurality of points, and the number of points in the generalized contour codes is smaller than the number of points in the generalized contour data; the generalized contour feature number in the generalized contour coding is greater than the generalized contour feature number in the generalized contour data; and decoding the generalized contour code to obtain generalized contour data of the target object. According to the method, the generalized contour data processing efficiency and quality are remarkably improved, and efficient generalized contour construction driven by a small amount of data is achieved.

Description

Technical Field

[0001] The present invention relates to the field of robot perception technology, and in particular to a generalized contour data processing method, an object recognition method and a device. Background Art

[0002] A robot's object recognition ability is a crucial foundation for intelligent manipulation. Currently, both vision and touch provide a certain level of object recognition. Vision can capture the geometric outline of an object in visible conditions, but it cannot provide information related to forces (such as inertia, stiffness, and friction). Tactile sensing can provide force-related information, but currently lacks a corresponding modeling method. The acquisition of tactile data relies on direct interaction with the object, potentially causing damage to the robot, sensor, or target object. This also results in high data acquisition costs, making efficient data utilization crucial. Summary of the Invention

[0003] In view of this, on the one hand, the present invention provides a generalized contour data processing method based on generalized contour coding, including: performing feature learning according to the generalized contour data of the target object to obtain the generalized contour coding, the generalized contour data and the generalized contour coding both include generalized contour features of several points, and the number of points in the generalized contour coding is less than the number of points in the generalized contour data, and the number of generalized contour features in the generalized contour coding is greater than the number of generalized contour features in the generalized contour data; decoding the generalized contour coding to obtain the generalized contour data of the target object.

[0004] Optionally, feature learning is performed based on the generalized contour data of the target object to obtain a generalized contour code, including: sampling the farthest point of the generalized contour data to obtain a preset number of first representative points, the preset number being less than the number of points in the generalized contour data; based on each first representative point, screening the points in the generalized contour data that are within the first representative point and multiple preset neighborhoods to obtain the neighborhood points of each first representative point; performing feature fusion on the generalized contour features of the neighborhood points of each first representative point to obtain fused neighborhood features of a preset number of first representative points; performing dimensionality transformation on the fused neighborhood features of each first representative point using a multi-layer perceptron to obtain target learning features of a preset number of first representative points; and performing feature splicing on the generalized contour features and target learning features in the preset number of first representative points to obtain a generalized contour code.

[0005] Optionally, the preset neighborhood is a spherical area determined by a neighborhood radius.

[0006] The second aspect of the present invention provides an object recognition method based on robot touch, including: obtaining generalized contour data and tactile data of a target object; using any generalized contour data processing method based on generalized contour coding to perform feature learning on the generalized contour data to obtain a generalized contour coding; using a tactile coding model to perform feature learning on the tactile data to obtain a tactile coding; using a generalized contour diffusion model to perform noise extraction on the generalized contour coding and tactile data with random noise added to obtain noise, and based on the noise, performing noise removal on the generalized contour coding with random noise added to obtain a target generalized contour coding; decoding the target generalized contour coding to obtain the true generalized contour data of the target object.

[0007] Optionally, before using the generalized contour diffusion model to extract noise from the generalized contour code and tactile data added with random noise, it also includes: generating random noise and time step with the same shape as the generalized contour code based on the generalized contour code; adding the random noise to the generalized contour code to obtain the generalized contour code added with random noise.

[0008] Optionally, a generalized contour diffusion model is used to extract noise from the generalized contour code and tactile data added with random noise to obtain noise, including: feature splicing of the generalized contour code and tactile data added with random noise to obtain fused features; and iteratively extracting noise from the fused features based on the time step using the generalized contour diffusion model to obtain noise.

[0009] Optionally, obtaining the generalized contour data of the target object includes: using a visual collector to collect a point cloud of the target object; sampling based on the point cloud to obtain a second representative point; obtaining displacement information of applying a force to the target object at the second representative point to cause the target object to displace; obtaining proportional relationship information between force and displacement based on the applied force and the corresponding displacement information; and obtaining the generalized contour data of the target object based on the point cloud and the proportional relationship information between force and displacement.

[0010] Optionally, the force applied to the target object at the second representative point is applied along a normal direction at the second representative point.

[0011] Optionally, obtaining tactile data of the target object includes: controlling the robot end operating device to move to the target position based on a preset position and a preset direction; controlling the robot end operating device to grasp the target object with a preset load; adjusting the position and load of the robot end operating device, and collecting tactile data of the target object during the adjustment process.

[0012] Optionally, the load of the robot's end operating device is adjusted, including: increasing the preset load until the preset load reaches a preset maximum load, and stopping increasing the preset load; or, collecting the position of the target object during the process of increasing the preset load; judging whether the position of the target object has changed; if the position of the target object has not changed, continuing to increase the preset load; if the position of the target object has changed and the displacement reaches a preset displacement, stopping increasing the preset load.

[0013] Optionally, adjusting the position of the robot end operating device includes: keeping the preset load unchanged, and adjusting the position and orientation of the robot end operating device.

[0014] A third aspect of the present invention provides an object recognition device based on robot tactile perception, which includes: a processor and a memory connected to the processor; wherein the memory stores instructions that can be executed by the processor, and the instructions are executed by the processor to enable the processor to execute the above-mentioned object recognition method based on robot tactile perception.

[0015] The generalized contour data processing method based on generalized contour coding of the present invention performs feature learning on the generalized contour data of the target object to generate a generalized contour code with fewer points but richer generalized contour features. This code is then decoded to obtain the generalized contour data of the target object. This embodiment requires only a small amount of original generalized contour data. Through feature learning and compression enhancement, a high-information-density code is generated, reducing the data volume while preserving core contour features. The method then rapidly decodes and reconstructs generalized contour data consistent with the true generalized contour. This significantly improves the efficiency and quality of generalized contour data processing, enabling efficient generalized contour construction driven by a small amount of data.

[0016] The present invention's robot tactile object recognition method obtains a small amount of generalized contours and tactile data of a target object as basic input. It then uses a generalized contour encoding model to perform feature learning and compression on the contour data, retaining core key features. It then learns and encodes newly added fusion features to significantly reduce the data volume. The tactile encoding model then extracts and compresses the tactile data, simultaneously reducing the scale of the tactile information. The generalized contour diffusion model then adds noise and removes potential offsets from the learned data, enhancing the model's adaptability to complex distributions. The generalized contour diffusion model simultaneously integrates generalized contour encoding and tactile encoding, collaboratively mapping geometric structure and contact state information into a generalized contour space. Finally, the generalized contour encoder decodes the data to obtain the true generalized contour, reconstructing a high-precision generalized contour that both contains the object's complete shape and retains contact details. Furthermore, a complete generalized contour consistent with the true generalized contour is reconstructed from a small amount of raw data. This enables efficient acquisition of a large number of generalized contours driven by a small amount of input data, providing a cost-effective generalized contour input solution for robot tactile cognition and improving efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 Flowchart of a generalized profile data processing method based on generalized profile coding in an embodiment of the present invention; Figure 2 4 is a structural diagram of a generalized contour coding model in an embodiment of the present invention; Figure 3 4 is a flow chart of a method for object recognition based on robot tactile perception in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0021] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components; wireless connections or wired connections. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0022] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0023] like Figure 1 As shown, an embodiment of the present invention provides a generalized profile data processing method based on generalized profile coding, which is executed by an electronic device such as a computer or a server, and specifically includes: S11, feature learning is performed based on the generalized contour data of the target object to obtain the generalized contour coding, the generalized contour data and the generalized contour coding both include generalized contour features of several points, and the number of points in the generalized contour coding is less than the number of points in the generalized contour data, and the number of generalized contour features in the generalized contour coding is greater than the number of generalized contour features in the generalized contour data.

[0024] In this embodiment, the target object is any object for which contour data is required (e.g., a person or object). Force response testing equipment is used to obtain generalized contour data, including point cloud data or force response data at point cloud coordinates, to characterize the target object's geometric contour morphology, mechanical response distribution characteristics, and structural spatial properties. Feature learning is performed on the generalized contour data of the target object, ultimately generating a generalized contour code with fewer points but richer generalized contour features.

[0025] S12, decoding the generalized outline code to obtain generalized outline data of the target object.

[0026] Generalized contour data refers to the spatial coordinate positions of several points on the surface of the target object (such as three-dimensional coordinates (x, y, z)), which are used to accurately describe the geometric shape of the target object and the force response data at the point cloud coordinates.

[0027] The contour data processing method based on generalized contour coding provided by the embodiment of the present invention is essentially a contour data processing algorithm implemented by a generalized contour coding model. In addition to using common unsupervised dimensionality reduction technologies such as autoencoders or variational autoencoders, the generalized contour coding model can also adopt an encoder-decoder architecture, specifically Figure 2 As shown, the generalized contour encoder 101 includes multiple downsampling modules, such as Figure 2 The downsampling modules 1 to m in the embodiment are used to perform feature learning based on the generalized profile data of the target object to obtain the generalized profile code. The generalized profile decoder 102 includes multiple upsampling modules, such as Figure 2In the upsampling module 1 to upsampling module m, multiple upsampling modules are used to decode the generalized profile code to obtain the generalized profile data of the target object. The downsampling module needs to maintain the meaning of the generalized profile features in the architecture, while the upsampling module does not need to maintain the meaning of the generalized profile features in the architecture and can be implemented using a multilayer perceptron.

[0028] In this embodiment, the generalized contour coding model can be trained in advance using the generalized contour training data. During training, the generalized contour training data is used as input and the generalized contour data is used as the target. The key contour features are extracted and compressed into generalized contour codes through multiple downsampling modules of the generalized contour encoder, and then decoded and restored into generalized contours through multiple upsampling modules of the generalized contour decoder. The training process uses the generalized contour reconstruction error as the loss function, and optimizes the encoder and decoder parameters through back propagation, so that the model can finally efficiently reconstruct the target generalized contour data from the generalized contour data.

[0029] like Figure 2 As shown in the figure, the steps for processing the generalized contour data using the trained generalized contour encoding model are as follows: Input the generalized contour data of the target object. The generalized contour data format is [B, C, N], where B is the batch size of the input data, C is the number of generalized contour features (including features from the generalized contour data and contact data), and N is the number of points. The generalized contour data is then transformed using the downsampling module in the generalized contour encoder, resulting in data format [B, C + D_1, N_1], where C is the number of generalized contour features (including features from the generalized contour data and contact data), D_1 is the number of features learned by the model, and N_1 is the number of points after downsampling, which is less than N. As needed, more downsampling modules are then added to continue the transformation process, resulting in data formats [B, C + D_2, N_2], [B, C + D_3, N_3], and so on, [B, C + D_m, N_m]. Overall, the number of learned features increases while the number of points decreases. Finally, multiple upsampling modules in the generalized contour decoder are used to transform the data in the format of [B, C+D_m, N_m] to data in the format of [B, C+D_m-1, N_m-1], ..., and finally to data in the format of [B, C, N], that is, the generalized contour data of the target object.

[0030] This embodiment performs feature learning on the generalized contour data of the target object to generate a generalized contour code with fewer points but richer generalized contour features. This code is then decoded to obtain the generalized contour data of the target object. This embodiment requires only a small amount of original generalized contour data. Through feature learning and compression enhancement, a high-information-density code is generated, reducing the data volume while preserving core contour features. This code is then quickly decoded and reconstructed to produce generalized contour data consistent with the true generalized contour. This significantly improves the efficiency and quality of generalized contour data processing, enabling efficient generalized contour construction driven by a small amount of data.

[0031] In some optional implementations of this embodiment, step S11 performs feature learning based on the generalized outline data of the target object to obtain the generalized outline code, specifically including: S111 , performing farthest point sampling on the generalized contour data to obtain a preset number of first representative points, where the preset number is smaller than the number of midpoints in the generalized contour data.

[0032] In specific implementation, each downsampling module is used to learn features from the generalized contour data. Taking one of the downsampling modules as an example, based on the preset number N' of points required for sampling, the longest sampling point is used to find N' points from the generalized contour data according to the spatial coordinates of the points in the input generalized contour data of the shape [B, C + D, N], and obtain the first representative points of the preset number N', where N' < N and D represents the new features learned by each downsampling module. If the first downsampling module is used as an example, the input shape is [B, C, N], that is, D = 0.

[0033] S112 , based on each first representative point, screening points in the generalized contour data that are within the first representative point and a plurality of preset neighborhoods to obtain neighborhood points of each first representative point.

[0034] K areas are specified in advance. For each of the n' first representative points, it is necessary to find points within the k neighborhoods of the first representative point, and finally obtain the neighborhood points of each first representative point.

[0035] S113 , fusing the generalized contour features of the neighborhood points of each first representative point to obtain fused neighborhood features of a preset number of first representative points.

[0036] The feature fusion method can be to average, maximize or minimize the generalized contour features of the neighborhood points corresponding to each first representative point, or other complex combination methods. In this case, the format of the fused neighborhood feature data is [B, k(C+D), N'].

[0037] S114, using a multi-layer perceptron to perform dimension transformation on the fused neighborhood features of each first representative point, to obtain target learning features of a preset number of first representative points.

[0038] The fusion neighborhood features of each first representative point are dimensionally transformed, and the target learning feature format obtained is [B, d', n'].

[0039] S115 , performing feature concatenation on the generalized contour features and the target learning features in a preset number of first representative points to obtain a generalized contour code.

[0040] The C generalized contour features originally in the generalized contour data in step S111 are combined with the d' generalized contour features learned in step S114 to obtain the generalized contour encoding, which has the format of [B, C+d', n'].

[0041] This embodiment uses farthest point sampling to filter out a preset number of key representative points (fewer points than the original data) from the original generalized contour data, reducing redundant points while retaining the core contour position information. It then performs neighborhood point screening and feature fusion based on these representative points, integrating the generalized contour features of the local area and enhancing the ability to characterize details. It then learns new fused features through dimensional transformation using a multi-layer perceptron, concatenating the original features with clear physical meanings with the newly added fused features to generate a generalized contour code with higher information density. This process significantly reduces the data size (the reduction in the number of points far exceeds the increase in features) while also learning the optimal new features, fully leveraging the optimal information compression capabilities of the deep learning encoder. Ultimately, the generalized contour code can still fully retain the core contour information of the target object with a smaller amount of data, laying the foundation for subsequent efficient decoding and reconstruction of high-precision generalized contour data.

[0042] In some optional implementations of this embodiment, the preset neighborhood in step S112 is a spherical area determined by a neighborhood radius.

[0043] For each of the n' first representative points, it is necessary to form multiple spherical areas with the first representative point as the center and the radius of each area, that is, one first representative point corresponds to k spherical areas, and then find the original point in each spherical area to obtain a neighborhood point of the first representative point. The same method can be used to find the neighborhood points of the remaining first representative points.

[0044] This embodiment sets the preset neighborhood as a spherical area determined by the neighborhood radius, and collects the original points inside the multiple spherical areas corresponding to each first representative point as neighborhood points. This can capture local spatial features more flexibly and comprehensively, enhance the ability to collect neighborhood information at different positions, and thus improve the richness of feature fusion and the model's representation effect on contour details.

[0045] like Figure 3As shown, an embodiment of the present invention provides an object recognition method based on robot tactile perception, which is executed by an electronic device such as a computer or a server, and specifically includes: S21, obtaining generalized contour data and tactile data of the target object.

[0046] Force response testing equipment is used to collect generalized contour data of the target object, and a robot end effector is used to collect tactile data of the target object. Generalized contour data includes point cloud data of the target object or force response data at the point cloud coordinates. This data is used to characterize the geometric contour morphology, mechanical response distribution characteristics, and structural spatial characteristics of the target object. Tactile data includes parameters such as the position coordinates of the end effector during adjustment, the magnitude and direction of the applied force, the displacement change of the contact point, pressure distribution characteristics, and deformation response under contact. This data is used to characterize the target object's contact mechanical properties (such as stiffness and softness), the fine structure of the local geometric contour (such as curvature characteristics), and the surface pressure distribution pattern. Data acquisition is completed only with contact information from a small number of discrete points or local areas. This step does not require high-density, full-coverage acquisition methods, significantly reducing data acquisition costs and time.

[0047] S22, using the generalized contour data processing method based on generalized contour coding described in any of the above embodiments to perform feature learning on the generalized contour data to obtain generalized contour coding.

[0048] That is, according to step S11, feature learning is performed on the generalized contour data, that is, feature learning is achieved through the generalized contour encoder of the generalized contour coding model to obtain the generalized contour coding of the target object. For the specific processing process, please refer to the relevant description of the above embodiment, which will not be repeated here.

[0049] S23, using a tactile coding model to perform feature learning on the tactile data to obtain tactile coding.

[0050] The tactile data is learned and encoded using a tactile coding model, which is an existing supervised encoder. Alternatively, common unsupervised dimensionality reduction techniques such as autoencoders and variational autoencoders are used to represent the tactile data with a vector to obtain tactile coding, thereby compressing the originally high-dimensional and sparse tactile data into a compact low-dimensional vector representation, ultimately achieving a significant reduction in the amount of data.

[0051] S24, using a generalized contour diffusion model to extract noise from the generalized contour code added with random noise and the tactile data to obtain noise, and based on the noise, performing noise removal on the generalized contour code added with random noise to obtain a target generalized contour code.

[0052] In this embodiment, random noise and extracted noise do not refer to physical noise, but to random offsets added to the generalized profile diffusion model. This process of adding and removing noise can help the generalized profile diffusion model learn the potential structure and distribution of data.

[0053] S25, decoding the target generalized outline code to obtain real generalized outline data of the target object.

[0054] The generalized contour encoder of the trained generalized contour encoding model is used to decode the target generalized contour code to obtain the true generalized contour of the target object. Specifically, the generalized contour encoder uses multiple upsampling modules to gradually restore the "compressed information" in the low-dimensional generalized contour code to the specific coordinates of the high-dimensional generalized contour. Ultimately, generalized contour data is reconstructed that is highly consistent with the true contour of the target object in terms of geometry and mechanical response distribution.

[0055] This embodiment obtains a small amount of generalized contour and tactile data of a target object as basic input. It then uses a generalized contour encoding model to learn and compress the contour data, retaining core key features. It then learns and encodes newly fused features, significantly reducing the data volume. The tactile encoding model then extracts and compresses the tactile data, simultaneously reducing the scale of the tactile information. The generalized contour diffusion model then adds noise and removes potential offsets from the learned data, enhancing the model's adaptability to complex distributions. The generalized contour diffusion model simultaneously integrates generalized contour encoding and tactile encoding, collaboratively mapping geometric structure and contact state information into a generalized contour space. Finally, the generalized contour encoder decodes the data to obtain the true generalized contour, reconstructing a high-precision generalized contour that both captures the object's complete shape and preserves contact details. Furthermore, a complete generalized contour consistent with the true generalized contour is reconstructed from a small amount of raw data. This enables efficient acquisition of a large number of generalized contours driven by a small amount of input data, providing a cost-effective generalized contour input solution for robot tactile cognition and improving efficiency.

[0056] In some optional implementations of this embodiment, before performing noise extraction on the generalized contour code and tactile data to which random noise is added using the generalized contour diffusion model in step S24, the object recognition method based on robot tactile sense further includes: generating random noise and time steps having the same shape as the generalized contour code based on the generalized contour code.

[0057] Generates random offsets (such as coordinate offsets and feature offsets) based on the shape encoded by the generalized contour. This process does not depend on the actual time, and the offset generation logic is controlled only by the time step (number of iterations).

[0058] Adding random noise to the generalized contour code results in a generalized contour code with added random noise.

[0059] The generated random offsets are gradually added to the generalized profile code at each time step. For example, a small offset is added once at time step t=1, and twice (or with a stronger offset) at time step t=2. Each time step represents a "noise diffusion" operation, which iteratively destroys the structure of the original code, simulating the cumulative perturbations of data in a complex environment.

[0060] This embodiment uses time-step-controlled offset addition to enable the subsequent generalized contour diffusion model to focus on core contour information (rather than local perturbations) while also enhancing the ability to reconstruct the original data through inverse denoising. Ultimately, the model can more accurately decode the target generalized contour from a small number of generalized contour encodings.

[0061] In some optional implementations of this embodiment, in step S24, noise extraction is performed on the generalized contour code and tactile data to which random noise is added using a generalized contour diffusion model to obtain noise, specifically including: S241, performing feature concatenation on the generalized contour code with random noise added and the tactile data to obtain fused features.

[0062] The generalized contour encoding with random offsets is concatenated with the tactile data at the feature level to form a fused feature. This step fuses visual geometric information with tactile mechanical information, allowing the model to utilize global perturbation information from both "shape" and "contact state," providing more comprehensive input for subsequent offset extraction.

[0063] S242, using a generalized contour diffusion model based on the time step, iteratively extracting noise from the fused features to obtain noise.

[0064] Using time steps as the control unit (each time step represents a noise diffusion operation), the fused features are iterated over multiple rounds. The generalized contour diffusion model gradually separates and extracts the offset components at each time step, ultimately obtaining a pure offset signal.

[0065] This embodiment uses time-step-controlled iterative extraction of offsets. The generalized contour diffusion model simultaneously leverages global geometric and mechanical information and accurately separates offset components in stages. This model can more accurately recover the original, unoffset generalized contour encoding from the offset generalized contour encoding, providing more realistic input for subsequent generalized contour data reconstruction, significantly improving the robustness and accuracy of object shape perception in robot tactile cognition.

[0066] In some optional implementations of this embodiment, obtaining the generalized outline data of the target object in step S21 specifically includes: S211a, using a visual collector to collect point clouds of the target object.

[0067] A visual capture device (such as a 3D scanner) collects a series of spatial points of the target object, for example, 2048 points in the shape of (x, y, z). The collected point cloud data is also called outline data. During the acquisition, the target object can be placed in some way, such as freely placed on a flat surface, suspended by elastic strings, etc.

[0068] S212a: Sampling is performed based on the point cloud to obtain a second representative point.

[0069] A typical method of selecting the second representative point is to sample the farthest point to reduce the amount of data. For the specific method, please refer to step S11.

[0070] S213a: Acquire displacement information of applying a force to the target object at the second representative point to cause the target object to be displaced.

[0071] At each second representative point, a force is applied along that point, and the displacement of the target object is measured. The force can be applied along the normal direction at the second representative point. The direction of the local normal can be determined by fitting a plane to points "near" the second representative point, with the plane's normal being the normal direction of the second representative point.

[0072] S214a, obtaining proportional relationship information between force and displacement based on the applied force and the corresponding displacement information.

[0073] For example, when an object is suspended with an elastic string, the force and displacement should be in direct proportion. In this case, the proportional coefficient is extracted as the proportional relationship information between the force and the displacement.

[0074] For example, when a target object is placed freely on a plane, it should have no displacement (or very small displacement) when the force is less than a certain value, and a linear relationship when the force is greater than a certain value. In this case, the force limit and the proportional coefficient of the linear relationship are extracted as information about the proportional relationship between force and displacement.

[0075] In this step, a high-density array sensor can also be installed at the end of the force-applying device to apply force and measure the displacement and force distribution to obtain information on the proportional relationship between force and displacement.

[0076] S215a, obtaining generalized contour data of the target object based on the point cloud and the proportional relationship information between force and displacement.

[0077] This embodiment uses a visual collector to collect point cloud data of the target object as basic contour data. This data is further reduced by sampling the farthest point. Force is then applied at a small number of secondary representative points, and displacement is measured. The force direction is determined by combining local normal fitting. Mechanical characteristics are extracted by analyzing the relationship between force and displacement. Finally, the point cloud and the proportional relationship between force and displacement are used as generalized contour data. This process requires only a small amount of raw collected data to generate high-information generalized contour data, reducing data acquisition costs.

[0078] In some optional implementations of this embodiment, obtaining tactile data of the target object in step S21 specifically includes: S211b, based on the preset position and preset direction, controlling the robot end operating device to move to the target position.

[0079] Use an end-operating device such as a robotic arm or other tactile device to move to a target position near the target object. The preset position and preset direction can use the existing grasping decision algorithm to first find a rough graspable position and direction based on vision, and then randomly fine-tune near the position, or it can be determined manually.

[0080] S212b, controlling the robot end operating device to grasp the target object with a preset load.

[0081] The preset load can grip the target object with a small force to prevent the shape of the target object from being damaged.

[0082] S213b, adjusting the position and load of the robot's end operating device, and collecting tactile data of the target object during the adjustment process.

[0083] Continuously adjust the grasping load or the position of the end-operating device, and collect in real time the contact status information of the robot's end-operating device when grasping the target object during the adjustment process, such as contact pressure distribution (force value at each contact point), contact force pressure change rate, contact area ratio, etc.

[0084] This embodiment first controls the robot's end effector to move to a preset position, applying a light load and gently grasping the object to prevent deformation. It then dynamically adjusts the gripping load or position of the end effector to collect tactile data from the contact during this adjustment process. This process requires only a few grip adjustments to acquire key tactile information reflecting the object's contact characteristics, reducing data acquisition complexity while providing the core input for describing the contact state in subsequent generalized contour encoding and tactile encoding models. This allows the models to more accurately learn the coupled relationship between geometry and mechanics, ultimately reconstructing extended contour information that encompasses both shape and contact details.

[0085] In some optional implementations of this embodiment, the load of the robot end effector is adjusted in step S213b, specifically in two ways: First way: Increase the preset load until the preset load reaches the preset maximum load, and then stop increasing the preset load.

[0086] Second way: Collect the position of the target object during the process of increasing the preset load.

[0087] Determine whether the position of the target object has changed.

[0088] If the position of the target object does not change, continue to increase the preset load.

[0089] If the position of the target object changes and the displacement reaches the preset displacement, the increase of the preset load is stopped.

[0090] The above two methods start with a preset load (small load) and gradually increase the load until a certain criterion is met, such as reaching a preset maximum load or the position of the target object changes and reaches a certain displacement when the load is increased, then the load increase can be stopped.

[0091] This embodiment achieves dual optimization for accurate tactile data collection and object protection through two load adjustment methods: starting with a small load and gradually increasing it to avoid deformation or damage to the object caused by sudden force application; and triggering a stop by monitoring position changes to accurately capture transitions in contact state, ensuring that the collected tactile data covers contact characteristics under different loads. These two methods increase the diversity of tactile data, including contact details under different loads, and provide more comprehensive input for subsequent model learning of the coupled relationship between geometry and mechanics.

[0092] In some optional implementations of this embodiment, the position of the robot end effector is adjusted in step S213b, specifically including: Keeping the preset load unchanged, adjust the position and orientation of the robot's end effector.

[0093] The position and orientation of the robot's end effector can be adjusted randomly or fixedly.

[0094] This embodiment achieves diverse tactile data collection and comprehensive coverage of contact scenarios by maintaining a constant preset grip load and adjusting the position and orientation of the robot's end effector. Random adjustment covers contact states at various positions and angles on the target object, avoiding localized data bias caused by fixed positions. Fixed adjustment allows for targeted collection of contact details at specific key locations (such as edges and depressions). Combining these two approaches allows tactile data to contain richer contextual information, reflecting the mechanical properties of the object at various contact locations and providing more comprehensive input for subsequent model learning of the coupling relationship between geometry and mechanics.

[0095] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0096] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0097] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0099] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will readily appreciate that other variations or modifications based on the above descriptions are possible. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A generalized profile data processing method based on generalized profile coding, characterized in that: include: Performing feature learning based on generalized outline data of a target object to obtain a generalized outline code, wherein both the generalized outline data and the generalized outline code include generalized outline features of a plurality of points, the number of points in the generalized outline code is less than the number of points in the generalized outline data, and the number of generalized outline features in the generalized outline code is greater than the number of generalized outline features in the generalized outline data; The generalized outline code is decoded to obtain the generalized outline data of the target object.

2. The method according to claim 1, characterized in that The step of performing feature learning based on the generalized profile data of the target object to obtain the generalized profile code includes: Performing farthest point sampling on the generalized profile data to obtain a preset number of first representative points, where the preset number is less than the number of midpoints in the generalized profile data; Based on each first representative point, screening points in the generalized contour data that are within the first representative point and a plurality of preset neighborhoods to obtain neighborhood points of each first representative point; Performing feature fusion on the generalized contour features of the neighborhood points of each first representative point to obtain fused neighborhood features of a preset number of first representative points; Performing dimension transformation on the fused neighborhood features of each first representative point using a multi-layer perceptron to obtain target learning features of a preset number of first representative points; The generalized contour features in the preset number of first representative points and the target learning features are feature spliced ​​to obtain the generalized contour code.

3. The method according to claim 2, characterized in that The preset neighborhood is a spherical area determined by the neighborhood radius.

4. A method for object recognition based on robot tactile perception, characterized in that: include: Acquiring generalized contour data and tactile data of a target object; Using the method according to any one of claims 1 to 3, performing feature learning on the generalized profile data to obtain a generalized profile code; performing feature learning on the tactile data using a tactile coding model to obtain tactile coding; performing noise extraction on the generalized contour code added with random noise and the tactile data using a generalized contour diffusion model to obtain noise, and performing noise removal on the generalized contour code added with random noise based on the noise to obtain a target generalized contour code; The target generalized outline code is decoded to obtain the real generalized outline data of the target object.

5. The method according to claim 4, characterized in that Before performing noise extraction on the generalized contour code added with random noise and the tactile data using the generalized contour diffusion model, the method further includes: generating random noise and a time step having the same shape as the generalized contour code based on the generalized contour code; The random noise is added to the generalized contour code to obtain the generalized contour code with the random noise added.

6. The method according to claim 5, characterized in that The step of extracting noise from the generalized contour code added with random noise and the tactile data using the generalized contour diffusion model to obtain noise includes: Perform feature concatenation on the generalized contour code with random noise added and the tactile data to obtain a fusion feature; The noise is obtained by performing iterative noise extraction on the fusion feature based on the time step using a generalized contour diffusion model.

7. The method according to claim 4, characterized in that Obtain generalized outline data of the target object, including: Collecting a point cloud of the target object using a visual collector; Sampling based on the point cloud to obtain a second representative point; acquiring displacement information of applying a force to the target object at the second representative point to cause the target object to be displaced; Obtaining proportional relationship information between force and displacement based on the applied force and the corresponding displacement information; The generalized contour data of the target object is obtained based on the point cloud and the proportional relationship information of the force and displacement.

8. The method according to claim 7, characterized in that The force applied to the target object at the second representative point is applied along a normal direction at the second representative point.

9. The method according to claim 4, characterized in that Get the tactile data of the target object, including: Based on the preset position and preset direction, control the robot end effector to move to the target position; Controlling the robot end effector to grasp the target object with a preset load; The position and load of the robot end operating device are adjusted, and tactile data of the target object during the adjustment process is collected.

10. The method according to claim 9, characterized in that Adjusting the load of the robot end operating device includes: increasing the preset load until the preset load reaches a preset maximum load, and then stopping increasing the preset load; or collecting the position of the target object during the process of increasing the preset load; Determining whether the position of the target object changes; If the position of the target object does not change, continue to increase the preset load; If the position of the target object changes and the displacement reaches a preset displacement, the increase of the preset load is stopped.

11. The method according to claim 10, characterized in that Adjusting the position of the robot end operating device includes: Keeping the preset load unchanged, the position and orientation of the robot end operating device are adjusted.

12. An object recognition device based on robot tactile perception, characterized in that: include: A processor and a memory connected to the processor; wherein the memory stores instructions that can be executed by the processor, and the instructions are executed by the processor to enable the processor to perform the method according to any one of claims 1 to 11.

Images