A skeleton material selection method, system, device and medium for a humanoid robot

Abstract

The application discloses a skeleton material selection method, system, device and medium for a humanoid robot, wherein the method acquires process hard constraint conditions, a plurality of voxel grid units of a skeleton of the humanoid robot, and a high-frequency load slice of each voxel grid unit; according to the process hard constraint conditions, all the voxel grid units and the high-frequency load slice, a multi-mode semantic vector generation process is performed to obtain a semantic vector matrix; material matching is performed on the semantic vector matrix to obtain a grid material table, and the grid material table records a plurality of target candidate materials corresponding to each voxel grid unit; according to the grid material table, voxel optimization is performed on all the voxel grid units to obtain a skeleton material selection result of the humanoid robot. The method can effectively improve the material screening efficiency of skeleton material selection, reduce the material selection time consumption and improve the skeleton material selection effect. The application relates to the technical field of humanoid robots.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of humanoid robots, and in particular to a skeleton material selection method, system, device and medium for a humanoid robot. BACKGROUND

[0002] With the entry of lightweight high-strength materials such as carbon fiber composite materials, magnesium-aluminum alloy low-pressure castings, and 3D printed titanium grids into the humanoid robot industry, it is often necessary to mix and arrange multiple materials to obtain the skeleton of the humanoid robot, and to adjust the partition wall thickness of each different skeleton. The selection of materials and thickness of the skeleton of the humanoid robot has become one of the key concerns of relevant practitioners.

[0003] Currently, the related art usually relies on finite element analysis (FEA) and empirical formula to realize material selection and thickness selection of the skeleton of the humanoid robot. This material selection method has low efficiency, takes a long time, and is prone to extreme solutions such as "excessive strength of the skeleton, high weight" or "low cost of the skeleton but difficult to process". The effect of skeleton material selection is not satisfactory.

[0004] Therefore, the problems of the related art still need to be solved and optimized. SUMMARY

[0005] The purpose of the present application is to at least partially solve one of the technical problems in the related art.

[0006] To this end, one purpose of an embodiment of the present application is to provide a skeleton material selection method, system, device and medium for a humanoid robot, wherein the method can effectively improve the material selection efficiency of the skeleton material selection, reduce the time consumption of the material selection, and improve the effect of the skeleton material selection.

[0007] In order to achieve the above technical purpose, the technical solutions adopted by the embodiments of the present application include:

[0008] In a first aspect, the present application provides a skeleton material selection method for a humanoid robot, comprising:

[0009] obtaining process hard constraint conditions, a plurality of voxel grid units of a humanoid robot skeleton, and a high-frequency load slice of each voxel grid unit;

[0010] According to the process hard constraint conditions, all the voxel grid units and the high-frequency load slice, a multi-modal semantic vector generation process is performed to obtain a semantic vector matrix, each matrix row of the semantic vector matrix corresponding to a voxel grid unit;

[0011] performing material matching on the semantic vector matrix to obtain a grid material table, the grid material table recording a plurality of target candidate materials corresponding to each voxel grid unit;

[0012] According to the mesh material table, voxel optimization is performed on all the voxel grid units to obtain the skeleton material selection result of the humanoid robot.

[0013] In addition, the method according to the above-mentioned embodiment of the application can further have the following additional technical features:

[0014] Further, in an embodiment of the application, the multi-modal semantic vector generation processing according to the process hard constraint condition, all the voxel grid units and the high-frequency load slice includes:

[0015] geometric feature extraction is performed on the voxel grid units to obtain a geometric feature vector;

[0016] dynamic feature coding is performed on the high-frequency load slice to obtain a load feature vector;

[0017] vector mapping is performed on the process hard constraint condition to obtain a process feature vector;

[0018] time delay alignment is performed on the load feature vector according to the geometric feature vector to obtain the aligned geometric feature vector and the load feature vector

[0019] attention fusion is performed on the aligned load feature vector and the geometric feature vector according to the process feature vector to obtain the semantic vector matrix.

[0020] Further, in an embodiment of the application, the material matching of the semantic vector matrix to obtain the mesh material table includes:

[0021] acquiring a material knowledge graph or a target screening condition;

[0022] vector matching is performed on the material knowledge graph according to each matrix row in the semantic vector matrix to obtain an initial candidate material set of each voxel grid unit;

[0023] condition screening and table construction processing are performed on the initial candidate material set according to the target screening condition to obtain the mesh material table.

[0024] Further, in an embodiment of the application, the voxel optimization of all the voxel grid units according to the mesh material table to obtain the skeleton material selection result of the humanoid robot includes:

[0025] According to the mesh material table, all the voxel mesh units are double-gene coded to obtain a first coding gene and a second coding gene of each voxel mesh unit, the first coding gene is used to represent a material index corresponding to the voxel mesh unit, and the second coding gene is used to represent a wall thickness index corresponding to the first coding gene;

[0026] According to all the first coding genes and the second coding genes, all the voxel mesh units are population parallel optimized to obtain a population optimization solution of each voxel mesh unit;

[0027] According to all the population optimization solutions, the skeleton material selection result of the humanoid robot is obtained.

[0028] Further, in an embodiment of the present application, the population parallel optimization of all the voxel mesh units according to all the first coding genes and the second coding genes to obtain a population optimization solution of each voxel mesh unit comprises:

[0029] A soft constraint condition is obtained;

[0030] A current population corresponding to the voxel mesh unit is obtained, the current population comprises a plurality of first population individuals, each first population individual comprises a third coding gene and a fourth coding gene, the third coding gene is the first coding gene or an updated first coding gene, and the fourth coding gene is the second coding gene or an updated second coding gene;

[0031] Parallel crossover and mutation are performed on all the first population individuals to obtain a plurality of second population individuals and a plurality of target fitnesses of each second population individual;

[0032] According to the soft constraint condition and all the target fitnesses, all the second population individuals are constraint optimized to obtain the population optimization solution.

[0033] Further, in an embodiment of the present application, the parallel crossover and mutation of all the first population individuals to obtain a plurality of second population individuals and a plurality of target fitnesses of each second population individual comprises:

[0034] An individual fitness function is obtained;

[0035] Cell region stress data and cell region stress data of the current population are obtained;

[0036] According to the unit region stress data, the unit region stress data and all the target fitness, the first population individual is carried out parallel cross variation and mutation to obtain a second population individual, the positive mutation probability of the third coding gene of the first population individual is positively correlated with the unit region stress data, and the reverse mutation probability of the fourth coding gene of the first population individual is positively correlated with the unit region stress data.

[0037] According to the individual fitness function, the second population individual is subjected to fitness analysis to obtain the target fitness.

[0038] Further, in an embodiment of the present application, the method further comprises:

[0039] The skeleton material selection result is subjected to super-limit prediction and screening update to obtain a screened and updated skeleton material selection result.

[0040] The screened and updated skeleton material selection result is subjected to finite element optimization to obtain an optimized skeleton material selection result.

[0041] In a second aspect, an embodiment of the present application provides a skeleton material selection system of a humanoid robot, comprising:

[0042] A first processing unit is configured to obtain process hard constraint conditions, a plurality of voxel grid units of a skeleton of a humanoid robot and high-frequency load slices of each of the voxel grid units;

[0043] A second processing unit is configured to perform multi-modal semantic vector generation processing according to the process hard constraint conditions, all the voxel grid units and the high-frequency load slices to obtain a semantic vector matrix, each matrix row of the semantic vector matrix corresponding to one of the voxel grid units;

[0044] A third processing unit is configured to perform material matching on the semantic vector matrix to obtain a grid material table, the grid material table recording a plurality of target candidate materials corresponding to each voxel grid unit;

[0045] A fourth processing unit is configured to perform voxel optimization on all the voxel grid units according to the grid material table to obtain a skeleton material selection result of the humanoid robot.

[0046] In a third aspect, an embodiment of the present application further provides an electronic device, comprising:

[0047] At least one processor;

[0048] At least one memory configured to store at least one program;

[0049] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0050] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a processor executable program. The processor executable program, when executed by a processor, is configured to implement the method described above.

[0051] The advantages and beneficial effects of the present application will be partially given in the following description, partially will become apparent from the following description, or will be understood by the practice of the present application:

[0052] The method disclosed by the embodiments of the present application comprises the following steps: acquiring process hard constraint conditions, a plurality of voxel grid units of a humanoid robot skeleton, and a high-frequency load slice of each voxel grid unit; performing multi-modal semantic vector generation processing according to the process hard constraint conditions, all the voxel grid units, and the high-frequency load slice to obtain a semantic vector matrix, each matrix row of the semantic vector matrix corresponding to one voxel grid unit; performing material matching on the semantic vector matrix to obtain a grid material table, the grid material table recording a plurality of target candidate materials corresponding to each voxel grid unit; and performing voxel optimization on all the voxel grid units according to the grid material table to obtain a skeleton material selection result of the humanoid robot. The method generates a semantic vector matrix with unified feature space through multi-modal semantic production on the process hard constraint conditions, the voxel grid units, and the high-frequency load slice, so that the material matching of the semantic vector matrix can consider the strength, stiffness and other properties of the voxel grid units, material fatigue under the high-frequency load slice, and manufacturability of the process hard constraint conditions in the unified feature space, which can effectively avoid the case of extreme scheme in skeleton material selection and improve the skeleton material selection effect. In addition, the method determines a preliminary grid material table through material matching of the semantic vector matrix, and then performs voxel optimization on the voxel grid units through the grid material table, which can effectively improve the material screening efficiency of skeleton material selection and reduce the time consumption of material selection. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following introduces the drawings of the related technical solutions in the embodiments of the present application or the prior art. It should be understood that the drawings in the following introduction are only for the convenience of expressing part of the embodiments of the technical solutions of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the premise.

[0054] Figure 1 A flowchart of a skeleton material selection method of a humanoid robot is provided for the embodiments of the present application.

[0055] Figure 2A framework schematic diagram of a skeleton material selection system of a humanoid robot is provided for an embodiment of the present application.

[0056] Figure 3 A structural schematic diagram of an electronic device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0057] Embodiments of the present application are described in detail below with reference to examples shown in the attached drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the attached drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application. For the step numbers in the following embodiments, they are only set for the convenience of explanation and description, and the order between the steps is not limited in any way, and the execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0059] In order to imitate the movement mechanism of human beings, the skeleton of a humanoid robot usually adopts a bionic structure such as "long bone type arm, hollow spine, foot ankle rolling", which needs to provide sufficient torque bearing capacity and requires the overall mass to be as light as possible to reduce driving power consumption and gait impact. With the entry of lightweight high-strength materials such as carbon fiber composite material, magnesium-aluminum alloy low-pressure castings, 3D printed titanium grid into the humanoid robot industry, it is often necessary to mix and arrange multiple materials to obtain the skeleton of the humanoid robot, and to adjust the partition wall thickness of each different skeleton. However, the elastic modulus, fatigue limit, density and cost of each material differ significantly, coupled with the dynamic load of the humanoid robot during walking, carrying and drop buffering, which changes with the joint angle, resulting in a multi-objective coupling of "lightweight, strength, safety factor, manufacturing cost".

[0060] At present, the related technologies usually realize material selection and thickness selection of a humanoid robot skeleton by means of finite element analysis (FEA) and empirical formula. Specifically, an engineer manually divides limb bones, a spine, an ankle joint and the like in a CAD model, and then replaces materials and adjusts wall thickness one by one according to specific stiffness or yield strength, and verifies whether stress and deformation meet the standards by using finite element software. However, due to factors such as high coupling between geometry modification and material replacement, any time an arbitrary load curve or joint travel of the skeleton is updated, the whole FEA is triggered, which takes a long time. In addition, this method often lacks unified semantic consideration of elastic modulus-fatigue limit-manufacturing feasibility-cost, resulting in low efficiency of candidate material selection, and extreme schemes such as "excessive strength and high weight" or "low cost but difficult to process" are prone to occur, and the effect of skeleton material selection is not satisfactory.

[0061] In addition, there are some related technologies that infer material arrangement and wall thickness setting of a robot skeleton by means of a large language model / multi-modal model. However, this method mainly focuses on single material topology reduction or static load conditions, and cannot well adapt to high-frequency dynamic working conditions such as "hollow shear of long bone type arm, multi-axis bending and torsion of spine, and rolling impact of foot and ankle" of a humanoid robot in actual application, and the effect of skeleton material selection is not satisfactory, the efficiency of skeleton material selection is not high, and the applicable range of skeleton material selection is limited.

[0062] It should be noted that the above-mentioned related technologies are only used to assist in understanding the technical solutions of the present application, and do not mean that they belong to the published prior art.

[0063] Therefore, the embodiments of the present application provide a skeleton material selection method, system, device and medium for a humanoid robot, wherein the method produces a semantic vector matrix with unified characteristics by means of multi-modal semantic production on process hard constraint conditions, voxel grid units and high-frequency load slices, so that the subsequent material matching of the semantic vector matrix can consider the strength, stiffness and other properties of the voxel grid units, material fatigue under the high-frequency load slices and manufacturability of the process hard constraint conditions in the unified characteristic space, which can effectively avoid the occurrence of extreme schemes in skeleton material selection and improve the effect of skeleton material selection.

[0064] In addition, the method can reduce the situation of FEA replacing materials one by one and adjusting the wall thickness, and is beneficial to improving the material screening efficiency, by determining the preliminary grid material table through material matching of the semantic vector matrix, specifically by vector matching based on the material knowledge graph, and by condition screening of the target screening condition voxel grid unit. In addition, the method can effectively improve the effect and application range of the skeleton material selection, by adjusting at least one of the material and the wall thickness under the consideration of dynamic load conditions, by performing voxel optimization on the voxel grid unit through the grid material table, specifically by considering the first encoding gene and the second encoding gene obtained by high-frequency load slicing to perform population optimization on each voxel grid unit. In addition, the method can make each voxel grid unit converge faster to obtain a population optimization solution, by positively correlating the forward mutation probability of the third encoding gene with stress and negatively correlating the reverse mutation probability of the fourth encoding gene with stress when performing parallel crossover mutation on the population individuals, which is beneficial to improving the skeleton material selection efficiency.

[0065] Reference Figure 1 In the embodiments of the present application, a skeleton material selection method of a humanoid robot comprises:

[0066] Step 110, obtaining process hard constraint conditions, a plurality of voxel grid units of a skeleton of a humanoid robot, and high-frequency load slices of each voxel grid unit;

[0067] In the embodiments of the present application, the process hard constraint conditions can be hard constraint conditions such as minimum wall thickness, tool diameter, upper limit of layup angle, and manufacturing cost in the manufacturing process. The voxel grid unit can first be obtained by obtaining a three-dimensional CAD model of the skeleton of the humanoid robot and a full-body kinematic load curve, splitting the skeleton into four-level elements of volume, surface, edge, and vertex through geometric processing, and then realizing fine meshing of the skeleton based on a certain voxel size, so as to obtain the voxel grid unit of the skeleton of the humanoid robot.

[0068] It can be understood that the high-frequency load slice of each voxel grid unit can be determined based on six-degree-of-freedom torque-velocity-acceleration time series curves obtained under actual working conditions or simulation working conditions such as "arm swinging, falling buffer, and object carrying" of the humanoid robot. Specifically, the torque-velocity-acceleration time series curves can be sampled at a sampling frequency of 1 kHz, so as to obtain a frequency-spectrum-time tensor for characterizing dynamic impact and high-frequency vibration of the humanoid robot. Then, based on the position of each voxel grid unit, the high-frequency load slice of the frequency-spectrum-time tensor at the corresponding position is determined.

[0069] Step 120, performing a multi-modal semantic vector generation process according to the process hard constraint conditions, all the voxel grid units, and the high-frequency load slices, to obtain a semantic vector matrix, each matrix row of the semantic vector matrix corresponding to one of the voxel grid units;

[0070] In the embodiment of the present application, the process hard constraint condition, all the voxel grid units and the high-frequency load slice can be subjected to vector generation and cross-modal fusion through a multi-modal semantic vector generation process, so as to obtain a semantic vector matrix, which includes a plurality of matrix rows, and each matrix row records a comprehensive stress image of a corresponding voxel grid unit under the load and process condition.

[0071] In some embodiments, the multi-modal semantic vector generation process is performed according to the process hard constraint condition, all the voxel grid units and the high-frequency load slice, and a semantic vector matrix is obtained, including:

[0072] geometric feature extraction is performed on the voxel grid units to obtain a geometric feature vector;

[0073] dynamic feature coding is performed on the high-frequency load slice to obtain a load feature vector;

[0074] vector mapping is performed on the process hard constraint condition to obtain a process feature vector;

[0075] time delay alignment is performed on the load feature vector according to the geometric feature vector, to obtain the aligned geometric feature vector and the load feature vector

[0076] attention fusion is performed on the aligned load feature vector and the geometric feature vector according to the process feature vector, to obtain the semantic vector matrix.

[0077] In the embodiment of the present application, for any voxel grid unit, the local curvature and cavity depth and other geometric features of the voxel grid unit can be extracted through a Patch-ViT network to obtain a geometric feature vector; the dynamic features of the voxel grid unit in the impact-fatigue angle can be obtained by performing fast Fourier transform on the high-frequency load slice and inputting the transformed features into a GRU neural network for time series modeling, which is recorded as a load feature vector; and the process hard constraint condition can be mapped into a multi-dimensional vector through a natural language model (such as a Bert model), which is recorded as a process feature vector.

[0078] It can be understood that after obtaining the load feature vector and the geometric feature vector, the geometric feature vector and the load feature vector can be subjected to time series feature alignment, so as to obtain the aligned geometric feature vector and the load feature vector. There are various time delay alignment methods, for example, the geometric feature vector and the load feature vector can be automatically aligned to the same time axis based on a differentiable displacement parameter.

[0079] It should be noted that the attention fusion can be that the process feature vector, the aligned load feature vector and the geometric feature vector are jointly input into the cross-modal Transformer network, and the input feature vectors are fused through a plurality of layers of multi-head attention layers to obtain a semantic vector matrix.

[0080] In step 130, the semantic vector matrix is subjected to material matching to obtain a mesh material table recording a plurality of target candidate materials corresponding to each voxel mesh element.

[0081] In the embodiment of the present application, a plurality of materials suitable for each voxel mesh element of the humanoid robot skeleton can be queried based on the semantic vector matrix of the humanoid robot skeleton, denoted as target candidate materials, and a mesh material table is generated based on all the determined target candidate materials.

[0082] In some embodiments, the material matching of the semantic vector matrix to obtain the mesh material table comprises:

[0083] Obtaining a material knowledge graph or a target screening condition;

[0084] According to each matrix row in the semantic vector matrix, vector matching is performed on the material knowledge graph to obtain an initial candidate material set of each voxel mesh element.

[0085] According to the target screening condition, the initial candidate material set is subjected to conditional screening and table construction processing to obtain the mesh material table.

[0086] In the embodiment of the present application, a material knowledge graph stored in a cloud database can be obtained, the material knowledge graph comprising a plurality of material nodes, each material node being used to indicate a corresponding material and an attribute label of the material, the attribute label specifically comprising a density, an elastic modulus, a fatigue limit, a unit price, a processability index and the like of the material, the present application examples are only for illustration, and do not make any limitation on the attribute label. The target screening condition can be a rule condition such as an inventory unit price and a material processing mode, wherein the specific material processing mode can be any one of a directed energy deposition (DED) additive processing, a laser powder bed fusion (LPBF) additive processing and a mechanical processing.

[0087] It can be understood that for any one matrix row in the semantic vector matrix, similarity calculation can be performed on each material node vector in the material knowledge graph, and a plurality of materials corresponding to the material node vectors with high similarity are determined as initial candidate materials, and according to the corresponding mapping relationship between the matrix row and the voxel mesh element, all the initial candidate materials are determined as the initial candidate material set of the corresponding voxel mesh element.

[0088] It should be noted that for any initial candidate material set of a voxel mesh element, a secondary screening can be performed on all initial candidate materials in the initial candidate material set based on one or more rules in the target screening criteria. This yields several initial candidate materials that meet the target screening criteria, denoted as target candidate materials. The content of other voxel mesh elements is similar and can be easily deduced. After obtaining the target candidate materials for each voxel mesh element, a mesh material table recording the target candidate materials for each voxel mesh element can be generated.

[0089] Step 140: Based on the mesh material table, perform voxel optimization on all the voxel mesh units to obtain the skeleton material selection results of the humanoid robot.

[0090] In this embodiment of the application, voxel optimization can be based on a mesh material table to optimize the material to be used for each voxel mesh cell and the material wall thickness, thereby obtaining the material selection result for the humanoid robot's skeleton.

[0091] In some embodiments, the step of performing voxel optimization on all voxel mesh cells according to the mesh material table to obtain the skeleton material selection result of the humanoid robot includes:

[0092] According to the mesh material table, all the voxel mesh units are encoded using dual genes to obtain a first coding gene and a second coding gene for each voxel mesh unit. The first coding gene is used to characterize the material index corresponding to the voxel mesh unit, and the second coding gene is used to characterize the wall thickness index corresponding to the first coding gene.

[0093] In the embodiments of this application, for any voxel grid cell, the first coding gene and the second coding gene of the voxel grid cell can be obtained by encoding. The first coding gene is specifically the material index of the voxel grid cell, which is used to indicate the material selected by the voxel grid cell; while the second coding gene is specifically the wall thickness traceability of the selected material, which is used to indicate the wall thickness of the selected material at the location of the voxel grid cell.

[0094] Based on all the first coding genes and the second coding genes, population parallel optimization is performed on all the voxel grid cells to obtain the population optimization solution for each voxel grid cell;

[0095] Further, the step of performing population-parallel optimization on all voxel grid cells based on all the first coding genes and the second coding genes to obtain the population optimization solution for each voxel grid cell includes:

[0096] Obtain soft constraints;

[0097] acquire a current population corresponding to the voxel grid unit, the current population comprising a plurality of first population individuals, each of the first population individuals comprising a third encoding gene and a fourth encoding gene, the third encoding gene being the first encoding gene or an updated first encoding gene, and the fourth encoding gene being the second encoding gene or an updated second encoding gene;

[0098] In the embodiments of the present application, the soft constraint condition can be a manufacturability probability, a predicted processing time, or the like. For any voxel grid unit, the voxel grid unit can be optimized in parallel by using an improved genetic algorithm to obtain an optimized solution of the population of each voxel grid unit.

[0099] Specifically, for one-time population optimization of any voxel grid unit, a current population corresponding to the voxel grid unit can be acquired first. If the current population optimization is the first population optimization, the current population can be an initialized population generated based on the first encoding gene and the second encoding gene of the voxel grid unit, the third encoding gene of the first population individual being the first encoding gene, and the fourth encoding gene of the first population individual being the second encoding gene.

[0100] Alternatively, if the current population optimization is the second or more population optimization, the first population individual of the current population can be a third population individual, which can be the second population individual optimized by constraint in the last population optimization; the third encoding gene of the first population individual can be the third encoding gene after the crossover and / or mutation in the last population optimization, i.e., the updated first encoding gene; and the fourth encoding gene of the first population individual can be the fourth encoding gene after the crossover and / or mutation in the last population optimization, i.e., the updated second encoding gene.

[0101] parallel crossover and mutation of all the first population individuals to obtain a plurality of second population individuals and a plurality of target fitnesses of each of the second population individuals;

[0102] Further, the parallel crossover and mutation of all the first population individuals to obtain a plurality of second population individuals and a plurality of target fitnesses of each of the second population individuals comprises:

[0103] acquiring an individual fitness function;

[0104] acquiring cell region stress data and cell region stress data of the current population;

[0105] According to the unit region stress data, the unit region stress data and all the target fitness, the first population individual is carried out parallel cross variation to obtain the second population individual, the positive mutation probability of the third coding gene of the first population individual is positively correlated with the unit region stress data, and the reverse mutation probability of the fourth coding gene of the first population individual is positively correlated with the unit region stress data.

[0106] According to the individual fitness function, the second population individual is subjected to fitness analysis to obtain the target fitness.

[0107] In the embodiments of the present application, for any one first population individual of the current population, the individual target function can include at least one of the target functions of the minimum overall weight, the minimum maximum strain and the minimum comprehensive cost; the unit region stress data can be the stress data of the voxel grid unit corresponding to the current population, the unit region stress data can be the stress data of the voxel grid unit corresponding to the current population; and the positive mutation probability of the third coding gene can be the mutation probability of the third coding gene for material upgrading, and the reverse mutation probability of the fourth coding gene can be the mutation probability of the fourth coding gene for wall thickness reduction.

[0108] It can be understood that parallel cross variation can be K-point cross and simulated binary mutation performed by CPU on the third coding gene and / or the fourth coding gene of each first population individual, so that the third coding gene and / or the fourth coding gene of the first population individual is changed, and the second population individual is obtained. And the target fitness of the first population individual or the second population individual is calculated based on the individual fitness function by GPU, which can include at least one of the overall weight fitness, the maximum strain fitness, the comprehensive cost fitness and the like.

[0109] It should be noted that the positive mutation probability of the third coding gene is positively correlated with the unit region stress data, and the reverse mutation probability of the fourth coding gene is positively correlated with the unit region stress data, which can make the humanoid robot skeleton preferentially try to reduce the wall thickness in the region with small stress, and preferentially try to upgrade the material in the region with stress concentration, which is beneficial to the population of each voxel grid unit in the humanoid robot skeleton to converge more quickly to the effective solution, reduce the required convergence time, improve the population optimization efficiency, and further improve the skeleton material selection efficiency.

[0110] According to the soft constraint condition and all the target fitness, all the second population individuals are subjected to constraint optimization to obtain the population optimization solution.

[0111] In the embodiments of the present application, for any one second population individual, the constraint optimization can be to obtain a fitness threshold corresponding to the second population individual, and to predict the manufacturability prediction value and the predicted processing time of the second population individual through the inference model deployed in the cloud; then, the target fitness of the second population individual is verified through the fitness threshold, and the manufacturability prediction value and the predicted processing time of the second population individual are verified through the soft constraint condition.

[0112] Specifically, for any one second population individual, if the second population individual satisfies the fitness condition and the soft constraint condition at the same time, the second population individual can be determined as a third population individual; or, if the second population individual does not satisfy the fitness condition and the soft constraint condition at the same time, the second population individual can be discarded. Then, based on all the retained third population individuals, a population optimization solution of the current population in the current population optimization process is determined, which is used to indicate the final selected material index and wall thickness value of the corresponding voxel grid cell, as well as the overall category, maximum hard label and cost index of each voxel grid cell, etc.

[0113] According to all the population optimization solutions, the skeleton material selection result of the humanoid robot is obtained.

[0114] In the embodiments of the present application, after obtaining the population optimization solution corresponding to each voxel grid cell, the skeleton material selection result of the humanoid robot can be generated based on the population optimization solution corresponding to each voxel grid cell.

[0115] In some embodiments, the method further comprises:

[0116] Performing over-limit prediction and screening update on the skeleton material selection result to obtain a screening updated skeleton material selection result;

[0117] Performing finite element optimization on the screening updated skeleton material selection result to obtain an optimized skeleton material selection result.

[0118] In the embodiments of the present application, the over-limit prediction can be performed on the skeleton material selection result through a stress over-limit prediction model, and the specific stress over-limit prediction model can be a CNN convolutional neural network, etc., so as to obtain the prediction score of each voxel grid cell in the skeleton material selection result in terms of whether the stress is over-limit; then, the screening update can be to perform adaptive grid encryption optimization on the material selection and / or material wall thickness of a number of voxel grid cells with high prediction scores, and to retain the data information of the material selection and material wall thickness of the remaining voxel grid cells, so as to obtain the screening updated skeleton material selection result.

[0119] It can be understood that the finite element optimization can first generate the load boundary conditions of the humanoid robot skeleton under the conditions of arm swinging, fall buffering, carrying static load and the like, and input the generated load boundary conditions and the updated skeleton material selection result to the GPU solver for FEA finite element analysis and optimization, so as to obtain the optimized skeleton material selection result.

[0120] Hereinafter, a skeleton material selection system of a humanoid robot according to an embodiment of the present application is described in detail with reference to the accompanying drawings.

[0121] With reference to Figure 2 The skeleton material selection system of the humanoid robot according to the embodiment of the present application comprises:

[0122] The first processing unit 101 is configured to obtain process hard constraint conditions, a plurality of voxel grid units of a humanoid robot skeleton, and a high-frequency load slice of each voxel grid unit.

[0123] The second processing unit 102 is configured to perform multi-modal semantic vector generation processing according to the process hard constraint conditions, all the voxel grid units and the high-frequency load slice, to obtain a semantic vector matrix, each matrix row of the semantic vector matrix corresponding to one of the voxel grid units.

[0124] The third processing unit 103 is configured to perform material matching on the semantic vector matrix to obtain a grid material table, the grid material table recording a plurality of target candidate materials corresponding to each voxel grid unit.

[0125] The fourth processing unit 104 is configured to perform voxel optimization on all the voxel grid units according to the grid material table to obtain a skeleton material selection result of the humanoid robot.

[0126] It can be understood that the contents in the above method embodiments are applicable to the system embodiments, the system embodiments specifically realize the same functions as the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0127] With reference to Figure 3 The embodiment of the present application further provides an electronic device, comprising:

[0128] At least one processor 201;

[0129] At least one memory 202 for storing at least one program;

[0130] When the at least one program is executed by the at least one processor 201, the at least one processor 201 implements the above method embodiments.

[0131] Similarly, it can be understood that the contents in the above method embodiments are applicable to the present device embodiments, the present device embodiments specifically implement the same functions as the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0132] The embodiment of the present application further provides a computer readable storage medium, wherein a program executable by the processor 201 is stored, and the program executable by the processor 201 is used for implementing the above method embodiments when executed by the processor 201.

[0133] Similarly, the contents in the above method embodiments are applicable to the present computer readable storage medium embodiments, the present computer readable storage medium embodiments specifically implement the same functions as the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0134] The embodiment of the present application further provides a computer program product, comprising a computer program, which is executed by a processor to implement the steps in the above method embodiments.

[0135] Those skilled in the art can understand that all or part of the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the computer program can include the flow of the above-mentioned embodiment methods.

[0136] In some alternative embodiments, the functions / operations mentioned in the block diagram can not occur in the order mentioned in the operation diagram. For example, depending on the functions / operations involved, two blocks shown in succession can actually be executed substantially simultaneously or the blocks can sometimes be executed in reverse order. In addition, the embodiments presented and described in the flowcharts of the present application are provided by way of example, and the purpose is to provide a more comprehensive understanding of the technology. The disclosed method is not limited to the operations and logical flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of larger operations are independently executed.

[0137] Furthermore, although the present application is described in the context of functional modules, it is understood that one or more of the functions and / or features can be integrated in a single physical device and / or software module, or one or more functions and / or features can be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary to an understanding of the present application. Rather, the actual implementation is within the routine skill of engineers familiar with the property, function and internal relationships of the various functional modules disclosed herein. Accordingly, the present application is not limited to the specific details of the functional modules described herein. Rather, it is understood that one of ordinary skill in the art is able to practice the application as claimed without undue experimentation having regard to the property, function and internal relationships of the various functional modules disclosed herein. It is also understood that the specific concepts disclosed are merely illustrative and that the scope of the present application is determined by the appended claims and their equivalents.

[0138] If the functions are implemented in software, the functions can be stored in or implemented as one or more computer program products. The computer program product can be stored in a computer readable medium, which can include, but is not limited to, RAM, ROM, electrically programmable read only memory (EPROM or EEPROM), flash memory, or a tangible computer diskette, such as a compact disc (CD) or DVD, etc. When the computer program product is implemented as one or more computer program products, the computer program product can be executed by one or more processors.

[0139] The logic and / or steps represented in the flowcharts and / or otherwise described herein, for example, can be embodied in non-transitory computer-readable media, which can be executed by an instruction execution system, apparatus, or device such as a computer-based system, processor, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this specification, a "computer-readable medium" can be any means that can contain, store, communicate, propagate or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0140] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0141] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above described embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art, or combinations thereof, can be used: a discrete logic circuit having logic gates for implementing logic functions upon data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), and / or the like.

[0142] In the above description of the present specification, the description referring to the terms "one embodiment", "another embodiment", or "certain embodiments" or the like means that a specific feature, structure, material or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. The illustrative expressions of the above terms do not necessarily refer to the same embodiment or example in the present specification. Also, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in an appropriate manner.

[0143] Although the embodiments of the present application have been shown and described, it would be appreciated by those skilled in the art that changes, modifications, alternatives and variations of these embodiments can be made without departing from the principles and spirit of the application, and the scope of the present application is defined by the claims and their equivalents.

[0144] The above is a specific description of the preferred embodiments of the present application, but the present application is not limited to the embodiments, and those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the present application, and these equivalent modifications or replacements are included in the scope defined by the claims of the present application.

Claims

1. A method for selecting materials for the skeleton of a humanoid robot, characterized in that, include: Obtain the hard constraints of the process, several voxel mesh units of the humanoid robot skeleton, and high-frequency load slices of each voxel mesh unit; Based on the process hard constraints, all the voxel mesh elements and the high-frequency load slice, a multi-mode semantic vector generation process is performed to obtain a semantic vector matrix, where each row of the semantic vector matrix corresponds to one voxel mesh element. Material matching is performed on the semantic vector matrix to obtain a mesh material table, which records several target candidate materials corresponding to each voxel mesh cell; Based on the mesh material table, voxel optimization is performed on all the voxel mesh units to obtain the skeleton material selection results of the humanoid robot; The step of performing voxel optimization on all voxel mesh cells according to the mesh material table to obtain the skeleton material selection results of the humanoid robot includes: According to the mesh material table, all the voxel mesh units are encoded using dual genes to obtain a first coding gene and a second coding gene for each voxel mesh unit. The first coding gene is used to characterize the material index corresponding to the voxel mesh unit, and the second coding gene is used to characterize the wall thickness index corresponding to the first coding gene. Based on all the first coding genes and the second coding genes, population parallel optimization is performed on all the voxel grid cells to obtain the population optimization solution for each voxel grid cell; Based on all the population optimization solutions, the results of the skeleton material selection for the humanoid robot are obtained.

2. The method according to claim 1, characterized in that, The process involves generating a multi-modal semantic vector matrix based on the hard constraints of the process, all the voxel mesh elements, and the high-frequency load slices, resulting in the following: Geometric features are extracted from the voxel mesh units to obtain geometric feature vectors; Dynamic feature encoding is performed on the high-frequency load slice to obtain the load feature vector; The process hard constraints are vector-mapped to obtain the process feature vector; Based on the geometric feature vector, the load feature vector is time-delay aligned to obtain the aligned geometric feature vector and the load feature vector. The semantic vector matrix is ​​obtained by performing attention fusion on the process feature vector, the aligned load feature vector, and the geometric feature vector.

3. The method according to claim 1, characterized in that, The step of performing material matching on the semantic vector matrix to obtain a mesh material table includes: Obtain a materials knowledge graph or target filtering criteria; Based on each row of the semantic vector matrix, vector matching is performed on the material knowledge graph to obtain an initial candidate material set for each voxel grid cell; Based on the target screening criteria, the initial candidate material set is subjected to conditional screening and table construction processing to obtain the grid material table.

4. The method according to claim 1, characterized in that, The step of performing population-parallel optimization on all voxel grid cells based on all the first coding genes and the second coding genes to obtain the population optimization solution for each voxel grid cell includes: Obtain soft constraints; Obtain the current population corresponding to the voxel grid unit. The current population includes a plurality of first population individuals. Each first population individual includes a third coding gene and a fourth coding gene. The third coding gene is the first coding gene or an updated first coding gene. The fourth coding gene is the second coding gene or an updated second coding gene. Parallel crossover mutation is performed on all individuals in the first population to obtain several individuals in the second population, and several target fitness values ​​for each individual in the second population. Based on the soft constraints and all the target fitnesss, constrained optimization is performed on all individuals in the second population to obtain the population optimization solution.

5. The method according to claim 4, characterized in that, The process of performing parallel crossover mutation on all individuals in the first population to obtain several individuals in the second population, and several target fitness values ​​for each individual in the second population, includes: Obtain the individual fitness function; Obtain the force data and stress data of the unit region of the current population; Based on the force data of the unit region, the stress data of the unit region, and the fitness of all the targets, the first population individuals are subjected to parallel crossover mutation to obtain the second population individuals. The positive mutation probability of the third coding gene of the first population individuals is positively correlated with the stress data of the unit region, and the negative mutation probability of the fourth coding gene of the first population individuals is positively correlated with the force data of the unit region. Based on the individual fitness function, fitness analysis is performed on the individuals in the second population to obtain the target fitness.

6. The method according to claim 1, characterized in that, The method further includes: The skeleton material selection results are subjected to over-limit prediction and screening update to obtain the updated skeleton material selection results. Finite element method optimization is performed on the updated skeleton material selection results to obtain optimized skeleton material selection results.

7. A skeleton material selection system for a humanoid robot, characterized in that, include: The first processing unit is used to acquire the process hard constraints, a number of voxel mesh units of the humanoid robot skeleton, and a high-frequency load slice of each voxel mesh unit. The second processing unit is used to perform multi-mode semantic vector generation processing based on the process hard constraints, all the voxel mesh elements and the high-frequency load slice, to obtain a semantic vector matrix, wherein each row of the semantic vector matrix corresponds to one voxel mesh element. The third processing unit is used to perform material matching on the semantic vector matrix to obtain a mesh material table, which records several target candidate materials corresponding to each voxel mesh cell. The fourth processing unit is used to perform voxel optimization on all the voxel mesh units according to the mesh material table to obtain the skeleton material selection result of the humanoid robot. The step of performing voxel optimization on all voxel mesh cells according to the mesh material table to obtain the skeleton material selection results of the humanoid robot includes: According to the mesh material table, all the voxel mesh units are encoded using dual genes to obtain a first coding gene and a second coding gene for each voxel mesh unit. The first coding gene is used to characterize the material index corresponding to the voxel mesh unit, and the second coding gene is used to characterize the wall thickness index corresponding to the first coding gene. Based on all the first coding genes and the second coding genes, population parallel optimization is performed on all the voxel grid cells to obtain the population optimization solution for each voxel grid cell; Based on all the population optimization solutions, the results of the skeleton material selection for the humanoid robot are obtained.

8. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method as described in any one of claims 1-6.

9. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to implement the method as described in any one of claims 1-6.

Images