Asset joint construction method based on graph neural network, computer equipment and medium

Abstract

The invention discloses an asset joint construction method based on a graph neural network, computer equipment and a medium, and particularly relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining an asset data file containing a plurality of assets; analyzing the asset data file to obtain graph data corresponding to each asset; obtaining a training data set based on the graph data of all assets, and training a graph neural network model based on the training data set; and constructing joints of the target assets based on to-be-assembled parts of the target assets and the trained graph neural network model. According to the method, the joint definition and construction process of the complex assets is automated, the asset production cycle is greatly shortened, and the labor cost is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to an asset joint construction method, a computer device, and a medium based on a graph neural network. Background Art

[0002] 3D assets refer to various digital resources and elements used for creating and producing 3D content, including but not limited to 3D models, texture maps, animation keyframes, materials, and special effects. These assets record and edit the physical world through 3D visualization technology, and then create and express the content of the virtual world through editing tools. To construct complex movable assets composed of multiple components connected by joints, traditionally it relies on a highly professional and time-consuming manual modeling and assembly process. Each component needs to be manually designed and modeled using professional modeling software, and the joint types, connection points, axes, and motion ranges between components need to be precisely defined one by one. This method not only consumes a large amount of time and labor costs, making it difficult to provide a sufficient variety of asset variants in a short time, thus limiting the training and testing of the generalization ability of agents, but also due to the extremely high requirements for professional modeling, physical engine knowledge, and mechanical design background, manually setting joint parameters is prone to introducing physical inaccuracies or functional errors. For example, improper definition of the joint axis may cause the door to not open normally.

[0003] Although there are some parametric modeling tools that allow generating asset variants by adjusting predefined parameters, such as adjusting the leg length of a chair or the size of a tabletop, the flexibility of these tools is very limited. They can only be modified within the preset parameter range, unable to generate completely new assets with significantly different topological structures, and their models still need to be predefined by experts and do not have the ability to automatically learn new structural patterns from existing data. The current technology has significant deficiencies in automatically and efficiently generating correct joint connection relationships for complex 3D assets, especially in reducing manual intervention and dependence on professional knowledge, ensuring the physical validity and functional rationality of the generated asset connection relationships, and effectively integrating multi-modal information for asset understanding and generation, and thus urgently needs improvement. This application precisely proposes an innovative solution to these technical challenges. Summary of the Invention

[0004] To overcome the above defects, the asset joint construction method, computer device, and medium based on a graph neural network provided by the present invention can solve the technical problems such as high manual dependence and low efficiency when constructing complex movable assets composed of multiple components connected by joints in the prior art.

[0005] In a first aspect, the present invention provides an asset joint construction method based on a graph neural network, including: Obtain an asset data file containing multiple assets; Parse the asset data file to obtain graph data corresponding to each asset, where the nodes of the graph data are used to represent the components that make up the asset, and the edges of the graph data are used to represent the joints connecting the components; Obtain a training data set based on the graph data of all assets, and train a graph neural network model based on the training data set; Construct the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model.

[0006] In some embodiments, the asset data file includes a description file corresponding to each asset, and the parsing of the asset data file to obtain graph data corresponding to each asset includes: For each asset, parse the description file to obtain all the components that make up the asset and the connection relationships between the components; Determine the nodes of the graph data based on the components, and determine the edges of the graph data based on the connection relationships between the components.

[0007] Optionally, the asset data file further includes geometric body files of the components that make up the asset; the parsing of the asset data file to obtain graph data corresponding to each asset further includes: For each asset, extract the geometric features of the components that make up the asset from the description file or from the description file and the geometric body files, and set the features of the nodes corresponding to the components in the graph data according to the geometric features.

[0008] Optionally, the parsing of the asset data file to obtain graph data corresponding to each asset further includes: Extract the semantic categories and / or physical attributes of the components from the description file; for each asset, from the description file or from the Set the features of the nodes corresponding to the components in the graph data according to the geometric features and the semantic categories and / or physical attributes.

[0009] In some embodiments, the parsing of the asset data file to obtain graph data corresponding to each asset further includes: Extract the joint types and / or joint space parameters of the joints connecting the components from the description file; Set the features of the edges corresponding to the joints in the graph data according to the joint types and / or the joint space parameters.

[0010] Further, the components connected by the joints are divided into parent components and child components, and the joint space parameters include at least one of the position, rotation axis, and motion range of the joint in the coordinate system of the parent component.

[0011] In some embodiments, the graph neural network model includes a prediction module. Constructing the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model includes: Inputting the component information for characterizing the components to be assembled of the target asset into the trained graph neural network model to obtain joint information for characterizing the connection relationship between the components to be assembled predicted by the prediction module according to the component information; Constructing the joints of the target asset according to the joint information.

[0012] In some embodiments, the graph neural network model includes a prediction module and a generation module. Constructing the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model includes: Inputting the component information for characterizing the components to be assembled of the target asset into the trained graph neural network model to obtain the graph data generated by the generation module; Wherein, the prediction module is used to predict joint information for characterizing the connection relationship between the components to be assembled according to the component information, and the generation module is used to generate graph data according to the component information and the joint information, and the graph data is used to characterize the target asset with the constructed joints.

[0013] In a second aspect, the present invention provides a computer device, which includes a processor and a memory. The memory is adapted to store multiple program codes, and the program codes are adapted to be loaded and run by the processor to execute the graph neural network-based asset joint construction method according to any one of the technical solutions in the above technical solutions of the graph neural network-based asset joint construction method.

[0014] In a third aspect, the present invention provides a computer-readable storage medium, which stores multiple program codes therein, and the program codes are adapted to be loaded and run by a processor to execute the graph neural network-based asset joint construction method according to any one of the technical solutions in the above technical solutions of the graph neural network-based asset joint construction method.

[0015] One or more of the above technical solutions of the present invention have at least one or more of the following beneficial effects: (1) Significantly improve efficiency: Automate the joint definition and construction process of complex assets, greatly shortening the asset production cycle and labor costs; (2) Enhance diversity: Can generate a large number of assets with topological structure and functional changes, providing a richer and more diverse training and testing environment for the agent, thereby enhancing the generalization ability and robustness of the agent; (3) Lower professional threshold: Reduces the dependence on professional knowledge of manual modeling and joint definition, enabling non-expert users to construct complex movable assets. (4) Improve accuracy and physical rationality: By learning from a large amount of real data, the GNN model can capture the internal physical laws of the connections between components, thereby generating more accurate and physically realistic joint parameters. (5) Support multi-modal information fusion: Can use data of multiple modalities such as the geometric shape, semantics, and physical properties of components as node features to improve the model's understanding and generation capabilities. (6) Promote the development of embodied intelligence: The automated generation of large-scale, high-quality, and interactive assets is a key infrastructure for promoting the development of fields such as embodied intelligence and virtual reality. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Referring to the accompanying drawings, the disclosure of the present invention will become more understandable. It is easily understood by those skilled in the art that these drawings are only for illustrative purposes and are not intended to limit the protection scope of the present invention. In addition, similar numbers in the figures are used to represent similar components, where: Figure 1 is a schematic flowchart of the main steps of a method for constructing asset joints based on a graph neural network provided by an embodiment of the present application; Figure 2 is a Figure 1 specific implementation step flowchart of step S102 provided by an embodiment of the present application; Figure 3 is another Figure 1 specific implementation step flowchart of step S102 provided by an embodiment of the present application; Figure 4 is a schematic diagram of the main structure of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The following describes some embodiments of the present invention with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0018] In the description of the present invention, "module" and "processor" may include hardware, software, or a combination of both. A module may include a hardware circuit, various suitable sensors, communication ports, memory, and may also include a software part, such as program code, or may be a combination of software and hardware. A processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, in hardware, or in a combination of both. A non-transitory computer-readable storage medium includes any suitable medium for storing program code, such as magnetic disks, hard disks, optical disks, flash memories, read-only memories, random access memories, and so on. The term "A and / or B" represents all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one of A or B" or "at least one of A and B" has a meaning similar to "A and / or B", and may include only A, only B, or A and B. The singular terms "a" and "the" may also include the plural form.

[0019] First, the technical terms involved in this application will be explained below.

[0020] Graph Neural Network (GNN): A deep learning model specifically designed to process graph data. Graph data consists of nodes and edges. Nodes usually represent entities (such as people, objects, etc.), and edges represent the relationships between entities. GNN realizes the representation learning of nodes, edges, and the entire graph through mechanisms such as message passing, aggregation operations, and node state updates.

[0021] Asset: An interactive complex three-dimensional object used in a virtual environment (such as a simulator, game). It is usually composed of multiple rigid body parts connected by joints and has a certain hierarchical relationship (such as a microwave oven, a chair, etc.).

[0022] Part / Link / Body: The basic rigid body component unit of an asset, which serves as a node in the graph structure.

[0023] Joint: A mechanical connection that connects two parts in an asset and defines the relative motion constraints between them, and serves as an edge in the graph structure. The joint types may include revolute, prismatic, fixed, etc.

[0024] Graph Data: A data structure composed of nodes (representing entities) and edges (representing the relationships between entities), which can effectively represent the parts of an asset and their connection relationships.

[0025] MJCF (MuJoCo XML Format): The XML format used by the physics simulation engine (MuJoCo) to describe complex multi-body dynamics models, including components, joints, geometry, and physical properties.

[0026] URDF (Unified Robot Description Format): A commonly used XML format in the Robot Operating System (ROS) to describe the kinematic and dynamic models of robots, including links and joints.

[0027] See the appendix Figure 1 , Figure 1 Figure Step S101: Obtain an asset data file containing multiple assets; In this embodiment, the asset data file may include a description file and a geometry file for defining and describing 3D assets. Among them, each asset corresponds to one description file, and each component constituting the asset corresponds to one geometry file. The description file is usually in XML file format, and the geometry file is usually in.obj file format. It should be understood that this step can directly obtain the asset data file from an existing storage file for storing a large number of 3D assets, or generate a large number of 3D assets using existing software and export the corresponding asset data file.

[0028] Exemplarily, the asset data file may be an MJCF file obtained based on the physics simulation engine (MuJoCo), or a URDF file obtained based on the Robot Operating System (ROS), and the geometry file may be an.obj file obtained based on the 3D graphics software (blender).

[0029] Step S102: Parse the asset data file to obtain graph data corresponding to each asset; Among them, the nodes of the graph data are used to represent the components constituting the asset, and the edges of the graph data are used to represent the joints connecting the various components.

[0030] In this embodiment, the smallest constituent unit of an asset is called a component, the connection relationship that connects two components in an asset and defines the relative motion constraint between them is called a joint, and the two components connected by each joint are defined as the parent component and the child component. It should be understood that in a joint connection, the fixed or relatively fixed end is usually called the parent component. In this step, all components that make up each asset and the connection relationships between components can be obtained by parsing the asset data file. For each asset, graph data corresponding to each asset can be obtained by using the components as nodes and determining the edges according to the connection relationships between components. For example, the graph data of each asset can be represented by G=(V, E), where V represents the set of nodes and E represents the set of edges. For two adjacent nodes u and v, the edge between these two nodes can be represented by e=(u,v).

[0031] Exemplarily, the process of parsing the asset data file, that is, parsing the XML file, includes: obtaining components according to the identifier in the XML file or <link> components can be obtained, and the connection relationship between components can be obtained according to the identifier in the XML file <joint / >

[0032] Furthermore, in this step, information such as the geometric features, semantic categories, and physical properties of each component that makes up the asset can also be extracted from the asset data file. The geometric features may include at least one of the geometric shape and size, rotation, texture, bounding box size, volume, centroid, and inertia principal axis corresponding to the component, and the features of the nodes corresponding to each component in the graph data are set according to at least one of the geometric features, the semantic categories, and the physical properties. For example: the geometric features or the semantic categories or the physical properties can be directly set as the features of the nodes, or the fusion features obtained by fusing the geometric features, the semantic categories, and the physical properties can be set as the features of the nodes. It should be understood that the feature fusion can be implemented by using existing feature fusion algorithms.

[0033] Furthermore, in this step, information such as the joint type and joint space parameters of the joints connecting each component can also be extracted from the asset data file. The joint space parameters include at least one of the position, rotation axis, and motion range of the joint in the coordinate system of its parent component, and the features of the edges corresponding to each joint in the graph data are set according to at least one of the joint type and the joint space parameters. For example: any one of the joint type or the joint space parameters can be directly set as the feature of the edge, or the fusion feature obtained by fusing the joint type and the joint space parameters can be set as the feature of the edge.

[0034] ​After the above processing, the graph data obtained in step S102 can represent each asset as a graph structure containing rich node and edge features in a complete manner.

[0035] Step S103: Obtain a training dataset based on the graph data of all assets, and train the graph neural network model based on the training dataset; In this embodiment, the graph neural network model (GNN model) can be one or more of existing neural network structures such as graph convolutional network (GCN), graph attention network (GAT), relational graph network, graph variational autoencoder (Graph VAE), and graph generative adversarial network (Graph GAN). For example, the graph neural network model may include a prediction module implemented by GAT, and the neural network model may also include a generation module implemented by Graph GAN.

[0036] This step trains the GNN model based on the training dataset formed by the graph data of the large-scale assets obtained in step S102. The training objective is to enable the GNN model to accurately predict which joints (i.e., edge prediction) should exist between a given set of components, and predict the specific types and spatial positions of these joints, such as detailed information like precise axes and motion ranges (i.e., edge attribute prediction). In this embodiment, the training method of the model will not be introduced in detail, and the model training can be specifically completed by using the existing training methods of graph neural network models. For example, when the trained neural network model is used in the asset generation task, the GNN model can learn the latent distribution of the asset graph structure, enabling it to generate a brand-new asset graph structure that conforms to physical rules and functional habits from random noise input.

[0037] Step S104: Construct the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model.

[0038] In this embodiment, this step mainly uses the trained GNN model to perform actual asset construction. Combining with the actual application scenario, according to different task types, the GNN model can directly output joint information or directly output graph data. For example, when the user provides a given set of parts to be assembled, for the prediction task, the trained GNN model performs efficient reasoning based on the input parts to be assembled, and the output of the GNN is a detailed instruction on how to connect these parts (i.e., joint information), which can include which parts should be connected, and can also include specific joint types, the exact axis of the joint in three-dimensional space, the connection point positions, and the necessary range of motion, etc. For the direct generation task, the trained GNN model will output a complete and new asset graph structure. In practical applications, the joint information output by the GNN model can also be combined with the known part geometry information (this geometry information can be selected from a predefined part library or generated by other generation methods), and automatically organize and export the asset description file in the standard physical simulation file format (such as MJCF or URDF). The assets generated in this way can be directly loaded into the physical simulation engine for kinematic and dynamic simulations, and support intelligent agents for real interactions.

[0039] The following describes the above steps S102 and S104 in combination with specific implementation manners.

[0040] In some embodiments, the asset data file specifically includes a description file corresponding to each asset. The specific implementation of parsing the asset data file to obtain the graph data corresponding to each asset in step S102 is as Figure 2 shown, including: Step 201: For each asset, parse the description file to obtain all the parts that make up the asset and the connection relationships between the parts; Step 202: Determine the nodes of the graph data based on the parts, and determine the edges of the graph data based on the connection relationships between the parts; Step 203: Extract the geometric features of the parts that make up the asset from the description file, and set the features of the nodes corresponding to the parts in the graph data according to the geometric features.

[0041] In this embodiment, the geometric features extracted from the description file are mainly used to characterize the features of simple type geometries. For example, the position and rotation and other information of a single cuboid, cylinder, sphere and other geometries can be extracted from the description file.

[0042] Step 204: Extract the joint types and / or joint space parameters of the joints connecting the parts from the description file, and set the features of the edges corresponding to the joints in the graph data according to the joint types and / or the joint space parameters.

[0043] In some embodiments, the asset data file specifically includes a description file corresponding to each asset and a geometry file of each component constituting the asset. The specific implementation of parsing the asset data file to obtain the graph data corresponding to each asset in step S102 is as follows Figure 3 shown, including: Step 301: For each asset, parse the description file to obtain all components constituting the asset and the connection relationships between the components; Step 302: Determine the nodes of the graph data based on the components, and determine the edges of the graph data based on the connection relationships between the components; Step 303: Extract the geometric features of each component constituting the asset from the description file and the geometry file, and set the features of the nodes corresponding to each component in the graph data according to the geometric features.

[0044] In this embodiment, the geometric features extracted from the description file may be macroscopic geometric features, which mainly describe information such as the shape (such as cuboid, cylinder, sphere, etc.) and size of an object; the geometric features extracted from the geometry file may be microscopic geometric features. Different from macroscopic geometric features, microscopic geometric features focus on the details and variations of the surface at a very small scale. The microscopic geometric features may be information such as roughness, waviness, texture, etc., and may also be information such as bounding box size, volume, centroid, and principal inertia axes. Optionally, the geometric features extracted from the geometry file may be more abstract shape embedding vectors obtained by processing the original point cloud or voxel data through a dedicated 3D shape encoder (such as PointNet or 3D CNN).

[0045] Step 304: Extract the joint type and / or joint space parameters of the joints connecting the components from the description file, and set the features of the edges corresponding to each joint in the graph data according to the joint type and / or the joint space parameters.

[0046] Optionally, steps 203 and 303 above may also include extracting the semantic category and / or physical attributes of each component from the description file; correspondingly, setting the features of the nodes corresponding to each component in the graph data according to the geometric features may be replaced by: setting the features of the nodes corresponding to each component in the graph data according to the geometric features and the semantic category and / or physical attributes.

[0047] In some embodiments, the graph neural network model includes a prediction module. Building the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model in step S104 may include: Input the component information of the components to be assembled that characterize the target asset into the trained graph neural network model to obtain the joint information characterizing the connection relationship between the components to be assembled predicted by the prediction module according to the component information; Construct joints of the target asset according to the joint information.

[0048] In some embodiments, the graph neural network model includes a prediction module and a generation module. The constructing of the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model in step S104 may include: Input the component information of the components to be assembled that characterize the target asset into the trained graph neural network model to obtain the graph data generated by the generation module; Wherein, the prediction module is used to predict joint information characterizing the connection relationship between the components to be assembled according to the component information, the generation module is used to generate graph data according to the component information and the joint information, and the graph data is used to characterize the target asset including the constructed joints.

[0049] Furthermore, it should be understood that the settings of the various modules mentioned in this application are only for illustrating the functional units of the graph neural network model of the present invention. The physical devices corresponding to these modules can be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only illustrative. Those skilled in the art can understand that the various modules of the model can be adaptively split or combined. Such splitting or combination of specific modules will not cause the technical solution to deviate from the principle of the present invention. Therefore, the technical solutions after splitting or combination will all fall within the protection scope of the present invention.

[0050] Those skilled in the art can understand that all or part of the processes in the method of implementing the above-mentioned embodiment of the present invention can also be completed by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. Wherein, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable storage medium can include: any entity or device, medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code.

[0051] Furthermore, the present invention also provides a computer device. Refer to the appendixFigure 4 , Figure 4 is a schematic diagram of the main structure of a computer device according to an embodiment of the present application. In an embodiment of a computer device according to the present application, the computer device includes a processor 401 and a memory 402. The memory 402 can be configured to store a program for executing the method embodiment of the graph neural network-based asset joint construction method. The processor 401 can be configured to execute the program in the memory. The program includes, but is not limited to, a program for executing the graph neural network-based asset joint construction method of the above method embodiment. For the sake of convenience of description, only the parts related to the embodiments of the present invention are shown. For the specific technical details not disclosed, please refer to the method part of the embodiments of the present invention. The computer device can be a control device formed by various electronic devices.

[0052] In some possible implementation manners of the present application, the computer device may include multiple processors 401 and multiple memories 402. The program for executing the graph neural network-based asset joint construction method of the above method embodiment can be divided into multiple sub-programs. Each sub-program can be loaded and run by the processor 401 respectively to execute different steps of the graph neural network-based asset joint construction method of the above method embodiment. Specifically, each sub-program can be stored in a different memory 402 respectively. Each processor 401 can be configured to execute the program in one or more memories 402 to jointly implement the graph neural network-based asset joint construction method of the above method embodiment, that is, each processor 401 executes different steps of the graph neural network-based asset joint construction method of the above method embodiment respectively to jointly implement the graph neural network-based asset joint construction method of the above method embodiment.

[0053] Furthermore, the present invention also provides a computer-readable storage medium. In an embodiment of a computer-readable storage medium according to the present invention, the computer-readable storage medium can be configured to store a program for executing the graph neural network-based asset joint construction method of the above method embodiment. The program can be loaded and run by the processor to implement the above graph neural network-based asset joint construction method. For the sake of convenience of description, only the parts related to the embodiments of the present invention are shown. For the specific technical details not disclosed, please refer to the method part of the embodiments of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiments of the present invention is a non-transitory computer-readable storage medium.

[0054] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An asset joint construction method based on a graph neural network, characterized in that, Including: Obtain an asset data file containing multiple assets; Parse the asset data file to obtain graph data corresponding to each asset, where the nodes of the graph data are used to represent the components that make up the asset, and the edges of the graph data are used to represent the joints connecting the components; Obtain a training data set based on the graph data of all assets, and train a graph neural network model based on the training data set; Construct the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model.

2. The method for constructing an asset joint based on a graph neural network according to claim 1, wherein The asset data file includes a description file corresponding to each asset. The parsing the asset data file to obtain graph data corresponding to each asset includes: For each asset, parse the description file to obtain all the components that make up the asset and the connection relationships between the components; Determine the nodes of the graph data based on the components, and determine the edges of the graph data based on the connection relationships between the components.

3. The method for constructing an asset joint based on a graph neural network according to claim 2, wherein The asset data file further includes geometric files of the components that make up the asset. The parsing the asset data file to obtain graph data corresponding to each asset further includes: For each asset, extract the geometric features of the components that make up the asset from the description file or from the description file and the geometric files, and set the features of the nodes corresponding to the components in the graph data according to the geometric features.

4. The method for constructing an asset joint based on a graph neural network according to claim 3, wherein The parsing the asset data file to obtain graph data corresponding to each asset further includes: Extract the semantic categories and / or physical properties of the components from the description file; Set the features of the nodes corresponding to the components in the graph data according to the geometric features and the semantic categories and / or physical properties.

5. The method for constructing an asset joint based on a graph neural network according to claim 3 or 4, wherein The parsing the asset data file to obtain graph data corresponding to each asset further includes: Extract the joint types and / or joint space parameters of the joints connecting the components from the description file; Set the features of the edges corresponding to the joints in the graph data according to the joint types and / or the joint space parameters.

6. The method for constructing an asset joint based on a graph neural network according to claim 5, wherein The components connected by the joints are divided into parent components and child components, and the joint space parameters include at least one parameter of the position, rotation axis, and motion range of the joint in the coordinate system where the parent component is located.

7. The method for constructing an asset joint based on a graph neural network according to claim 1, wherein The graph neural network model includes a prediction module. The constructing the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model includes: Input the component information used to represent the components to be assembled of the target asset into the trained graph neural network model to obtain joint information used to represent the connection relationship between the components to be assembled predicted by the prediction module according to the component information; Construct the joints of the target asset according to the joint information.

8. The method for constructing an asset joint based on a graph neural network according to claim 1, wherein The graph neural network model includes a prediction module and a generation module. The constructing the joints of the target asset based on the components to be assembled of the target asset and the trained graph neural network model includes: Input the component information used to represent the components to be assembled of the target asset into the trained graph neural network model to obtain the graph data generated by the generation module; Among them, the prediction module is used to predict joint information for characterizing the connection relationship between the to-be-assembled components according to the component information, and the generation module is used to generate graph data according to the component information and the joint information, and the graph data is used to characterize the target asset including the constructed joints.

9. A computer device, comprising a processor and a memory, the memory being adapted to store a plurality of program codes, characterized in that, The program code is adapted to be loaded and run by the processor to execute the graph neural network-based asset joint construction method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing multiple program codes, characterized in that, The program code is adapted to be loaded and run by the processor to execute the graph neural network-based asset joint construction method according to any one of claims 1 to 8.

Images