Animation generation processing method and device and related equipment

Abstract

The invention provides an animation generation processing method and device and related equipment, and the method comprises the steps: obtaining the semantic data of a 3D model, and the semantic data comprises the dismounting sequence description information, geometric information and material parameter information of parts in the 3D model; and based on the dismounting sequence description information, the geometric information and the material parameter information, carrying out motion decision in the dismounting process of the parts in the 3D model to obtain a first dismounting animation path of the parts in the 3D model.

Description

Technical Field

[0001] This application relates to the field of virtual reality technology, and in particular to an animation generation and processing method, apparatus and related equipment. Background Technology

[0002] With the rapid development of virtual reality (VR) technology, virtual simulation animation has been widely used. For example, virtual simulation animation for 3D model disassembly scenarios can be applied to virtual simulation training scenarios, including industrial maintenance guidance such as standardized process demonstrations of equipment disassembly and troubleshooting, equipment operation training such as virtual simulation of dangerous operations (chemical handling, high-altitude operations), and emergency drills, physical simulation of scenarios such as daily fire escape and hazardous chemical leaks.

[0003] Currently, virtual simulation animations for 3D model disassembly scenarios are usually generated manually or semi-automatically. They are highly rule-dependent, require manual pre-definition of disassembly and assembly steps, and are difficult to adapt to the physical simulation of complex assembly relationships. Furthermore, the virtual simulation animation generation operation is cumbersome, with low automation. It requires adjusting the animation path of the parts frame by frame or setting key frames, which is time-consuming and relies on human experience. Summary of the Invention

[0004] This application provides an animation generation processing method, apparatus, and related equipment, which can solve the technical problems in related technologies where virtual simulation animation generation requires manual / semi-automatic operation, resulting in difficulty in adapting to the physical simulation of complex assembly relationships, cumbersome operation, time consumption, and reliance on human experience.

[0005] In a first aspect, embodiments of this application provide an animation generation processing method, the method comprising:

[0006] Obtain semantic data of the 3D model, including the disassembly and assembly sequence description information, geometric information and material parameter information of the components in the 3D model;

[0007] Based on the disassembly and assembly sequence description information, geometric information, and material parameter information, motion decisions are made during the disassembly and assembly process of the components in the 3D model to obtain the first disassembly and assembly animation path of the components in the 3D model.

[0008] Secondly, embodiments of this application provide an animation generation and processing apparatus, the apparatus comprising:

[0009] The acquisition module is used to acquire semantic data of the 3D model, including the disassembly and assembly sequence description information, geometric information and material parameter information of the parts in the 3D model;

[0010] The decision module is used to make motion decisions during the disassembly and assembly process of the components in the 3D model based on the disassembly and assembly sequence description information, geometric information and material parameter information, so as to obtain the first disassembly and assembly animation path of the components in the 3D model.

[0011] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, it implements the steps of the animation generation processing method as described in the first aspect.

[0012] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the animation generation processing method as described in the first aspect.

[0013] Fifthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the animation generation processing method as described in the first aspect.

[0014] In this embodiment, semantic data of a 3D model is acquired, including disassembly / assembly sequence descriptions, geometric information, and material parameters of the components in the 3D model. Motion decisions are made during the disassembly / assembly process of the components in the 3D model based on these information, resulting in a first disassembly / assembly animation path for the components. This allows for automated motion decisions during the disassembly / assembly process of components in the 3D model, based on the disassembly / assembly sequence descriptions, geometric information, and material parameters. This enables adaptive physical simulation of complex assembly relationships, avoids manual operation, and improves operational efficiency. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of an animation generation processing method provided in an embodiment of this application;

[0017] Figure 2 This is a flowchart illustrating the multimodal semantic matching process in an embodiment of this application;

[0018] Figure 3 This is a schematic diagram of the process for constructing the component assembly / disassembly relationship diagram in an embodiment of this application;

[0019] Figure 4 This is a schematic diagram illustrating the principle of determining the first disassembly / assembly animation path through reinforcement learning strategy design in the embodiments of this application;

[0020] Figure 5 This is a schematic diagram illustrating the principle of trajectory replanning based on the PPO algorithm in the embodiments of this application;

[0021] Figure 6 This is a schematic flowchart of an animation generation processing method provided in a specific example of an embodiment of this application.

[0022] Figure 7 This is a schematic diagram of the structure of an animation generation and processing device provided in an embodiment of this application;

[0023] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Virtual simulation animation generation technologies for 3D model disassembly scenarios mainly fall into four categories: manual / semi-automatic generation based on Computer-Aided Design (CAD) software, procedural generation based on semantic rules, 3D animation visualization enhancement technologies, and virtual simulation and interactive tools. These technologies achieve component disassembly animations through geometric modeling, physical constraint simulation, or interactive design, but they generally suffer from low automation, insufficient physical plausibility, and weak cross-device generalization capabilities.

[0026] For example, manual / semi-automatic generation based on CAD software utilizes the built-in disassembly view function of CAD software (such as desktop content creation systems like SOLIDWORKS Composer and Tianhe Software's CAD platform (THCAD)) to generate disassembly animations by manually adjusting component displacement paths or preset hierarchical structures. For instance, SOLIDWORKS' spherical disassembly algorithm automatically separates components or avoids clipping by moving parts frame-by-frame and adding disassembly lines and label annotations. This method is cumbersome, requiring frame-by-frame adjustment of component positions, relying on manual experience to handle complex assemblies (such as multi-stage gear linkages), and prone to component interference or path conflicts. It lacks dynamic optimization; the generated animation only displays the static disassembly sequence and cannot detect physical collisions or dynamically adjust trajectories (such as component movement under gravity) in real time.

[0027] Semantic rule-based procedural generation defines assembly and disassembly logic through configuration files or knowledge graphs, and combines it with a physics engine to simulate motion trajectories. For example, it reads the component interaction characteristics in the equipment model configuration file (such as screws requiring the shell to be disassembled first) and generates animation paths that conform to the operation sequence. This method is highly rule-dependent, requires manual pre-definition of assembly and disassembly steps (such as the component hierarchy in coal preparation equipment disassembly diagram software), and is difficult to adapt to complex assembly relationships (such as multi-level linkage of non-standard equipment). Physical simulation simplification is difficult, and collision detection often uses Oriented Bounding Box (OBB) simplified algorithms, which cannot handle real physical interactions (such as the friction force during gear meshing).

[0028] 3D animation visualization enhancement technology combines cross-sectional views, perspective effects, and dynamic disassembly to present assembly logic (such as Maiwei Animation Technology). For example, it uses high-precision modeling to display gear transmission paths, overlaying dynamic backgrounds and lighting effects to enhance visual impact. However, this method has a long production cycle, requires manual keyframe setting to control component displacement / rotation, and cannot automatically adapt to different models (e.g., an animation script for a car gearbox needs to be redesigned). It lacks physical verification, focusing on visual effects while ignoring the rationality of component movement (e.g., trajectory deviations caused by inertia in high-speed rotating components are not simulated).

[0029] Virtual simulation and interactive tools utilize VR / AR technology to build immersive disassembly and assembly training systems (such as the Oubier liquid chromatography simulation software), supporting stylus operation and semi-transparent structural display. For example, they can create 1:1 high-precision models to recreate the internal structure of equipment and observe fluid paths in perspective mode. However, this method is costly to develop, requiring high-precision modeling and real-time rendering support (such as pharmaceutical equipment disassembly simulation systems), making it difficult for small and medium-sized teams to afford the hardware and data annotation costs. Path generation is rigid, relying on preset scripts (such as decomposition process animations) and cannot dynamically optimize the disassembly and assembly sequence based on user actions.

[0030] It can be seen that the relevant technologies have the following technical drawbacks:

[0031] 1. Generated manually or semi-automatically using modeling software.

[0032] The operation is cumbersome and has a low degree of automation. It requires adjusting the displacement path of components frame by frame or setting keyframes. For example, the animation module of ProE requires manually defining the start / end posture, which is time-consuming and relies on human experience. For complex assemblies (such as car transmissions), component interference or path conflicts are prone to occur. Dynamic optimization capabilities are lacking. The generated animation only shows the preset decomposition order and cannot dynamically adjust the motion trajectory according to real-time operations (such as user intervention causing component lag). Furthermore, there is a lack of physical rationality verification (such as the influence of gravity and inertia on component motion is not included in the calculation).

[0033] 2. Semantic rule-based procedural generation.

[0034] The rigid rule configuration, relying on manually predefined logic, necessitates the manual definition of assembly and disassembly steps through configuration files or knowledge graphs (e.g., the hierarchical relationships of components in coal preparation equipment disassembly diagram software). This fails to adapt to the non-standard linkage relationships of complex assemblies (e.g., dynamic adjustments to multi-stage gear transmission paths). For instance, if the equipment structure undergoes minor changes (e.g., the addition of a snap-fit ​​structure), the rule file must be reconfigured, resulting in insufficient flexibility. Furthermore, the simulation accuracy of physical constraints is insufficient. Collision detection relies on simplified algorithms such as bounding boxes, making it difficult to handle realistic physical interactions (e.g., friction during component disassembly, gear meshing inertia), causing the component motion trajectories in the animation to deviate from actual physical laws (e.g., clipping occurs in high-speed rotating components due to neglecting inertia).

[0035] 3. Combining physics engine with animation tools.

[0036] Parameter adjustment is complex, relying on preset scripts and requiring manual adjustment of physics engine parameters (such as gravity, drag, and collision bounce coefficients) to optimize effects. Furthermore, parameter settings demand a high level of user experience (e.g., in After Effects, repeated trial and error adjustments to the shatter effect are necessary). Semantic-driven design is disconnected from physical constraints; related solutions fail to achieve real-time collaboration between semantic logic (such as assembly / disassembly order) and the physics engine, potentially leading to violations of operational procedures during animation generation (e.g., logical errors such as component damage due to improper screw removal following the assembly manual's order).

[0037] This application aims to address the aforementioned technical shortcomings by resolving issues such as inaccurate semantic parsing, lack of physical logic, and insufficient personalized adaptation in traditional virtual simulation animations. It can automatically make motion decisions during the assembly and disassembly of components in a 3D model, thereby generating virtual simulation animations. It mainly involves technologies such as VR, human-computer interaction, physical simulation, and reinforcement learning.

[0038] See Figure 1 , Figure 1This is a flowchart of an animation generation processing method provided in an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0039] Step 101: Obtain semantic data of the 3D model, the semantic data including the disassembly and assembly sequence description information, geometric information and material parameter information of the components in the 3D model;

[0040] Step 102: Based on the disassembly and assembly sequence description information, geometric information and material parameter information, make motion decisions during the disassembly and assembly process of the components in the 3D model to obtain the first disassembly and assembly animation path of the components in the 3D model.

[0041] In step 101, a 3D model and its semantic data can be input. The semantic data can include assembly manual text and material parameter information. The assembly manual text can include disassembly and assembly sequence description information and geometric information. The geometric information can include geometric attribute information and geometric connection relationships. The material parameter information can include the material of the component and the friction coefficient of the corresponding material, such as metal or plastic.

[0042] The disassembly and assembly sequence description information can be obtained by extracting keywords (such as "priority disassembly") from the assembly manual text using the BERT model.

[0043] In step 102, motion decisions during the assembly and disassembly process of components in the 3D model can be made based on the disassembly and assembly sequence description information, geometric information, and material parameter information through a deep learning model or a large model, so as to obtain the first assembly and disassembly animation path of the components in the 3D model.

[0044] Taking a deep learning model as an example, this deep learning model needs to be pre-trained. It can be trained based on data from multiple 3D models. These 3D models typically cover a large number of parts and various disassembly and assembly methods. The geometric information and material parameters of the parts in different 3D models can be different. Correspondingly, the disassembly and assembly sequence description information, geometric information, and material parameter information of the parts in the 3D models can be input into the deep learning model. The deep learning model can make motion decisions during the disassembly and assembly process of the parts in the 3D model and obtain the disassembly and assembly animation path of each part in the 3D model. The generated disassembly and assembly animation path can be compared with the ground truth path to obtain the loss value, which is fed back to the deep learning model to adjust its parameters so that it can generate more accurate disassembly and assembly animation paths.

[0045] Once training is complete, the disassembly and assembly sequence descriptions, geometric information, and material parameters of the components in the 3D model can be input into the deep learning model, which will output the first disassembly and assembly animation paths of each component in the 3D model. Thus, the disassembly and assembly animations of the 3D model can be integrated based on the first disassembly and assembly animation paths of each component.

[0046] In some embodiments, step 102 specifically includes:

[0047] Semantic matching is performed based on the disassembly and assembly sequence description information and the geometric information to generate disassembly and assembly priority weight information of the components in the 3D model;

[0048] Based on the geometric information and the material parameter information, determine the physical motion simulation parameters of the component assembly and disassembly process in the 3D model;

[0049] Based on the disassembly and assembly priority weight information and physical motion simulation parameters, motion decisions are made during the disassembly and assembly process of the components in the 3D model to obtain the first disassembly and assembly animation path of the components in the 3D model.

[0050] In this way, based on the disassembly / assembly sequence description information and geometric information, the disassembly / assembly priority weight information of components in the 3D model can be automatically extracted. Combined with the physical motion simulation parameters during the disassembly / assembly process of the components in the 3D model, motion decisions during the disassembly / assembly process can be automated. This allows for adaptive physical simulation of complex assembly relationships, avoiding manual operation and improving operational efficiency. Furthermore, combining the disassembly / assembly priority weight information with the physical motion simulation parameters during the disassembly / assembly process of the components in the 3D model allows for more flexible and accurate motion decisions during the disassembly / assembly process, improving the accuracy of the first disassembly / assembly animation path planning for the components in the 3D model, thus generating more vivid disassembly / assembly animations that conform to the laws of object motion.

[0051] The assembly / disassembly priority weight information can include the priority weight of the assembly / disassembly instructions for each component in the 3D model. The higher the priority weight of the assembly / disassembly instruction, the earlier the component should be disassembled. In other words, the assembly / disassembly priority weight information can characterize the assembly / disassembly steps of the components in the 3D model.

[0052] Semantic matching can be achieved through multimodal semantic alignment, i.e., text-geometric feature alignment, to generate assembly and disassembly priority weight information for components in the 3D model.

[0053] In some embodiments, a multimodal alignment of knowledge graphs and rule engines can be employed. A knowledge graph of equipment assembly and disassembly rules can be constructed based on expert experience (e.g., "shell disassembly precedes gear disassembly"). A rule engine (e.g., Drools) can then parse the assembly manual text to directly generate priority weights, rather than relying on neural networks to learn assembly relationships. This approach sacrifices the ability to dynamically adapt to non-standard equipment, but through flexible expansion of the manually created rule base, it can cover common scenarios. Geometric feature alignment of the 3D model employs traditional feature engineering (e.g., extracting statistical indicators of contact surface curvature and volume) combined with text keyword matching (e.g., term frequency and inverse document frequency (TF-IDF)) to generate priority weights.

[0054] In some embodiments, Figure 2 This is a flowchart illustrating the multimodal semantic matching process in an embodiment of this application, as shown below. Figure 2 As shown, you can input assembly manual text, and use the BERT model to extract keywords (such as "priority disassembly") from the assembly manual text to obtain disassembly and assembly sequence description information, as well as extract the geometric information of the parts in the 3D model from the assembly manual text. Semantic features of different modalities can be extracted based on the disassembly and assembly sequence description information and geometric information, respectively. Then, based on the semantic features of different modalities, semantic matching is performed through the BERT model to generate disassembly and assembly priority weight information of the parts in the 3D model.

[0055] Among them, such as Figure 2 As shown, the disassembly and assembly sequence description information can be obtained through text embedding to obtain text features, while the geometric information of the parts in the 3D model can be obtained through feature extraction to obtain geometric features that characterize the 3D model. These geometric features may include the part volume, contact surface curvature, and material friction coefficient (μ), etc. The geometric features of the geometric embedding and the text features of the text embedding can be concatenated, and based on the concatenated features, the priority weight of the disassembly and assembly instructions of the parts in the 3D model can be output through a normalization function (e.g., "shell disassembly" weight = 0.92).

[0056] The semantic matching formula of the BERT model is as follows:

[0057]

[0058] in, The text embedding vectors (text features, such as the semantic encoding of "prioritize disassembly") of the assembly manual text are extracted using the BERT model. : Embedded vectors of geometric features of the 3D model (such as component volume, contact surface curvature). Vector concatenation operation merges textual and geometric features. Learnable parameter matrices and bias terms are used to align the semantic spaces of different modalities. : Normalization function, outputs the priority weight of disassembly and assembly instructions (e.g., "shell disassembly" weight = 0.92).

[0059] In this step, textual instructions (such as "prioritize disassembly") can be aligned with the geometric information of the 3D model (such as contact surface curvature and volume) to generate disassembly / assembly priority weight information (e.g., shell disassembly weight = 0.92). Specifically, the semantic vector of the assembly manual text can be extracted using the BERT model, concatenated with the geometric feature vector of the 3D model, and then the priority weight of the disassembly / assembly instructions is generated using Softmax. In this way, by fusing multimodal features, the problem that traditional rule-based systems cannot handle non-standard equipment can be solved.

[0060] In some embodiments, the physical motion simulation parameters may include physical collision result parameters.

[0061] In some embodiments, the physical collision result parameters can be obtained through the physics engine part of the collision detection module, using discrete-time step collision detection combined with a bounding box simplified model. For example, the Separated Axis Theorem (SAT) is used to detect discrete collisions of the bounding box of the component, and a rough penetration depth is calculated by projection. This method reduces the detection accuracy (error > 3mm) and the realism of the physical simulation, but it has higher computational efficiency and is suitable for scenarios with stringent real-time requirements but high tolerance for accuracy.

[0062] In some embodiments, determining the physical motion simulation parameters of the component assembly / disassembly process in the 3D model based on the geometric information and the material parameter information specifically includes:

[0063] Based on the geometric information and the material parameter information, the continuous physical collision situation during the disassembly and assembly process of the components in the 3D model is detected, and the physical collision result parameters during the disassembly and assembly process of the components in the 3D model are obtained.

[0064] The physical motion simulation parameters include the physical collision result parameters.

[0065] The physical collision result parameters may include the number of collisions, minimum penetration distance, and direction.

[0066] In related technologies, using bounding box (AABB / OBB) simplified geometric models for collision detection results in low accuracy (error > 5mm) and easy to miss complex contact surfaces (such as gear meshing surfaces), making complex contact surfaces prone to distortion. In this step, a continuous collision detection algorithm can be used to replace the bounding box simplified model. Combined with the material properties of the components (such as the friction coefficient), the physical parameters can be optimized to solve the problem of trajectory deviating from the real physical laws.

[0067] In some embodiments, geometric information and material parameters can be used as input parameters for a continuous collision detection algorithm to determine the continuous physical collisions during the assembly and disassembly of components in a 3D model. Continuous collision detection can accurately determine the collision time point by solving a cubic polynomial distance function of the point-to-surface motion trajectory; the GJK-EPA algorithm calculates the penetration depth and correction direction, improving detection accuracy to an error of <1mm, and is suitable for irregular geometries (such as threads and curved surfaces).

[0068] In some embodiments, the step of detecting continuous physical collisions during the assembly and disassembly of components in the 3D model based on the geometric information and the material parameter information, and obtaining physical collision result parameters during the assembly and disassembly of components in the 3D model, includes:

[0069] Based on the geometric information, using the components in the 3D model as nodes and the connection relationships between different components as edges, a graph neural network is used to construct a component assembly / disassembly relationship graph of the 3D model; wherein, the component assembly / disassembly relationship graph is used to indicate the topological relationships between different components;

[0070] Based on the component assembly / disassembly relationship diagram and the material parameter information, the continuous physical collision situation during the component assembly / disassembly process in the 3D model is detected, and the physical collision result parameters during the component assembly / disassembly process in the 3D model are obtained.

[0071] In some embodiments, the geometric information of the components in the 3D model can be used for graph neural network modeling, and graph convolutional networks (GCNs) can be used to construct a graph of component assembly and disassembly relationships. Figure 3 This is a flowchart illustrating the process of constructing a component assembly / disassembly relationship diagram in an embodiment of this application. The components in the 3D model may include screws and gears, and the geometric information may include screw geometric parameters and gear topological relationships, such as... Figure 3 As shown, taking a component node with input screw geometry parameters and gear topology as an example, it is first modeled using a Graph Neural Network (GNN). Nodes represent components (such as screws and gears), and edges represent connections (threads, meshing). The modeling formula for the Graph Neural Network (GNN) is as follows:

[0072]

[0073] in, :node In the The feature vector of a layer represents the geometric and functional attributes of a component (such as the thread parameters of a screw or the module of a gear). :node The set of neighbors (e.g., the neighbors of a screw node may include connecting parts such as the shell and gears). Normalization coefficient, usually taken from the nodes The square root of the degree is used to balance the weights of different nodes. : No. The layer can learn weight matrix to capture assembly dependencies between components (such as threaded connections needing to mesh before gears). Nonlinear activation functions (such as ReLU) enhance the model's ability to express complex assembly logic.

[0074] Graph Neural Network (GNN) modeling formulas dynamically construct component assembly / disassembly relationship graphs by aggregating features of neighboring nodes, overcoming the limitations of traditional rule bases. They can output component assembly / disassembly relationship graphs (nodes = components, edges = connection types), such as node features: screw thread parameters, gear module; edge weights: thread connection dependency strength (e.g., "screw-casing" weight = 0.9). This provides input for continuous collision detection.

[0075] Subsequently, continuous collision detection can be performed based on the component assembly / disassembly diagram and material parameter information. Specifically, the component assembly / disassembly diagram can be used to determine the moving points and motion planes during the component assembly / disassembly process in the 3D model. Furthermore, based on the material parameter information, continuous collision detection between the moving points and motion planes during the component assembly / disassembly process in the 3D model can be calculated. The purpose is to determine the collision between the moving points and the motion planes within a given time interval. Whether a collision occurs within the system. The material parameters of the components can affect the linear velocity of the moving point and the plane of motion.

[0076] Assumption point and plane The trajectory of its movement is as follows:

[0077] The equation of motion for the point is: (0≤t≤1), where, This is the initial position. It is linear velocity.

[0078] The equations of planar motion are: ,in, It is the initial position of the plane. It is linear velocity. It is a rotation matrix. It is a plane normal vector.

[0079] The distance function is: point to plane The distance can be expressed as: Substituting into the equations of motion and expanding them into a cubic polynomial, we get: .

[0080] Where, n(t): the normal vector of the plane, Nonlinear motion terms caused by rotation (such as the product of angular velocity and position shift). , , It is determined by the linear velocity, the initial position difference, and the change of the normal vector.

[0081] Yes, according to Cartesian notation rules, it is possible to quickly determine the case without precisely solving the cubic equation. Does a real root exist inside (i.e., does a collision occur)? Finally, the extended polytope collision detection algorithm (Gilbert Johnson Keerthi Expanding Polytope Algorithm, GJK-EPA) is used to determine whether the objects overlap. After a collision occurs, the minimum penetration distance and direction are calculated by approximating the contact surface with an extended polytope.

[0082] In related technologies, ignoring material differences (such as the coefficient of friction between metal and plastic) leads to distortion in physical simulations (e.g., plastic gears are misjudged as having a metal friction response). Alternatively, even if material differences are considered, the material response is usually hard-coded into the physical equations using a fixed coefficient of friction (e.g., μ=0.25 for metal and μ=0.15 for plastic), resulting in decreased accuracy.

[0083] In some embodiments, the physical motion simulation parameters may further include rigid body dynamic parameters. The step of determining the physical motion simulation parameters during the assembly and disassembly of components in the 3D model based on the geometric information and the material parameter information further includes:

[0084] Based on the material parameter information, rigid body dynamic parameters for the assembly and disassembly of components in the 3D model are constructed.

[0085] Rigid body dynamics modeling can be performed using a physical dynamics model. This model employs a modified Newton-Euler equation and incorporates material parameters such as the material friction coefficient. Compared to simple disassembly and explosion simulations in related technologies, this embodiment can support differentiated physical responses in metals and plastics. The rigid body dynamics friction torque formula is as follows: .

[0086] in, The friction coefficient of the component material (e.g., μ=0.3 for metal, μ=0.2 for plastic) can be loaded from the material database. : Normal force at the contact surface (calculated in real time by the rigid body dynamics engine). : Contact point radius (such as the equivalent radius of the gear meshing surface). The angular velocity of a component determines the direction of friction.

[0087] In this way, material parameters can be dynamically loaded, and the material friction coefficients (such as μ=0.3 for metal and μ=0.2 for plastic) can be read from the database and embedded into the rigid body dynamics equations. This allows for differentiated physical responses, such as requiring greater torque to disassemble metal parts and allowing plastic parts to slide quickly, thus improving the physical realism of the animation.

[0088] In this embodiment, an AI model automatically extracts semantic logic (such as the disassembly and assembly sequence) from the assembly manual text, and combines this with a rigid body dynamics engine to optimize the motion trajectory in real time. This avoids the rigidity caused by manually predefined rules and achieves dynamic coordination between the disassembly and assembly logic and physical simulation. Furthermore, the collision time interval can be quickly determined by solving the cubic polynomial distance function of the point-to-surface motion trajectory, and the minimum penetration distance can be calculated using the GJK-EPA algorithm. In this case, the parameter μ in the friction torque formula is dynamically loaded from the material parameter information. Thus, by using a continuous collision detection algorithm (cubic polynomial root-finding to determine the collision time point) and the GJK-EPA algorithm (calculating the penetration direction and depth), combined with the material friction coefficient (μ), the physical response can be optimized.

[0089] Specifically, motion decisions during the assembly and disassembly process of components in a 3D model can be made based on assembly and disassembly priority weights and physical motion simulation parameters. The assembly and disassembly priority weights influence the assembly and disassembly steps of the components in the 3D model, while the physical motion simulation parameters affect the physical response during the assembly and disassembly process. By combining the assembly and disassembly steps and the physical response, the first assembly and disassembly animation path of the component in the 3D model can be obtained, and the assembly and disassembly animation of that component can be rendered and displayed accordingly. This allows for automated motion decisions during the assembly and disassembly process of components in the 3D model, enabling adaptive physical simulation of complex assembly relationships, avoiding manual operation, and improving operational efficiency.

[0090] In some embodiments, motion decisions during the assembly and disassembly process of components in the 3D model can be made directly based on disassembly and assembly priority weight information and physical motion simulation parameters.

[0091] In some embodiments, the step of making motion decisions during the disassembly and assembly process of components in the 3D model based on the disassembly and assembly priority weight information and physical motion simulation parameters, and obtaining the first disassembly and assembly animation path of the components in the 3D model, specifically includes:

[0092] Based on the disassembly and assembly priority weight information, the target component to be disassembled or assembled in the 3D model is determined.

[0093] Based on the physical motion simulation parameters during the disassembly and assembly process of the target component, the decision-driving information is determined;

[0094] Based on the decision-driven information, the driving strategy network generates motion decisions during the disassembly and assembly process of the target component, thereby obtaining the first disassembly and assembly animation path of the target component.

[0095] In this embodiment, reinforcement learning strategies can be designed to achieve real-time determination and optimization of the assembly / disassembly animation path. Figure 4 This is a schematic diagram illustrating the principle of determining the first disassembly / assembly animation path through reinforcement learning strategy design in an embodiment of this application, as shown below. Figure 4 As shown, a multi-objective reward function is designed (e.g., reward function R = 0.6·collision penalty + 0.3·time efficiency + 0.1·stress constraint). By fusing semantic priority (i.e., disassembly and assembly priority weight information), physical collision result parameters (e.g., number of collisions), and rigid body dynamic parameters (as rigid body dynamic constraints), decision-driving information is obtained. This drives the policy network to generate motion decisions for component disassembly and assembly (e.g., direction of movement, speed), realizing dynamic coordination between semantic logic and physical laws at the action level, ensuring that the disassembly animation conforms to both operational specifications and physical simulation realism.

[0096] The specific implementation process is as follows:

[0097] First, based on the disassembly / assembly priority weight information, the target component to be disassembled / assembled in the 3D model is determined, and the state space is defined, which may include:

[0098] Component location ( ): The displacement vector of a component in a three-dimensional coordinate system (such as the coordinates of the starting point for screw removal).

[0099] Component attitude ( : The rotational state of a component as represented by a quaternion or Euler angle (such as the meshing angle of a gear).

[0100] linear velocity ( ): Instantaneous moving speed of the component on the XYZ axes (unit: m / s).

[0101] The physical meaning of the state space is to fully describe the motion state of the components and provide dynamic input for path planning.

[0102] Next, a reward function is designed. Based on the physical motion simulation parameters during the assembly and disassembly of the target component, decision-driven information is obtained. The reward function can integrate physical constraints and operational efficiency objectives, as shown in the following formula:

[0103]

[0104] Among them, the collision penalty item is Its function is to encourage the agent (i.e., the AI ​​model designed with reinforcement learning strategies) to reduce interference in the movement of parts. The number of collisions during the assembly and disassembly of the target component is determined by physical collision result parameters. : Maximum permissible collision threshold (dynamically calculated based on material hardness, e.g., N=2 for metal parts).

[0105] The time efficiency item is Its purpose is to improve disassembly and assembly efficiency while ensuring safety. The time consumed during the disassembly and assembly of the target component. : Preset maximum allowed time.

[0106] Stress constraint terms are It is determined by the rigid body dynamics parameters.

[0107] Correspondingly, motion decisions during the disassembly and assembly process of the target component can be made based on decision-driven information and the motion state space of the target component, thus obtaining the first disassembly and assembly animation path of the target component. For example, when the disassembly and assembly animation path decision is correct, a decision reward will be given, and the corresponding decision-driven information will indicate a relatively large value. At this time, based on the motion state space of the target component, a motion decision to maintain the disassembly and assembly process of the target component can be made, thus obtaining the first disassembly and assembly animation path of the target component.

[0108] When the disassembly and assembly animation path decision is incorrect, a decision penalty will be imposed. The reward function will contain a penalty term, which will make the decision-driven information relatively small. At this time, the motion decision needs to be adjusted based on the motion state space of the target component so that the policy network can make the correct decision and obtain the first disassembly and assembly animation path of the target component.

[0109] In related technologies, reliance on manually predefined rule bases makes it difficult to adapt to non-standard equipment (such as special fixtures) and dynamic scenarios (such as users temporarily adjusting paths). Furthermore, at the system architecture level, these technologies separate semantic logic and physical simulation into independent modules, replacing dynamic collaboration mechanisms with asynchronous interface calls. For example, after the semantic module generates an initial path, it submits it to the physics engine for verification. If a collision is detected, it returns for readjustment, forming a loosely coupled "generation-verification-iteration" process, rather than real-time collaborative optimization. While this method reduces system complexity, asynchronous iteration and module decoupling increase latency and weaken the synergistic effect between semantics and physics, thus introducing iteration latency (>200ms), making it difficult to support high real-time requirements.

[0110] In some embodiments, after generating motion decisions for the target component's assembly / disassembly process based on the decision-driven information and obtaining the first assembly / disassembly animation path for the target component, the method further includes:

[0111] If the first disassembly and assembly animation path planning is found to be unreasonable based on the physical motion simulation parameters, the motion decision during the disassembly and assembly of the target component is corrected based on the target information to obtain the second disassembly and assembly animation path of the target component.

[0112] The target information includes strategy network parameters and physical motion simulation parameters. The strategy network parameters are used to characterize the decision rules for motion decisions in the first disassembly and assembly animation path.

[0113] In related technologies, user commands cannot intervene in animation generation in real time, resulting in low operational efficiency. Alternatively, offline pre-computation and heuristic algorithms can be used. For example, a genetic algorithm can be used to generate multiple candidate paths offline, and a better solution can be selected through a fitness function (including collision count, time efficiency, etc.). The pre-computation results can be directly called at runtime. User interaction is simplified to manually adjusting the pre-stored path library (such as the order of dragging and dropping parts) and verifying feasibility through a physics engine, rather than dynamically injecting multimodal commands into the reward function. The offline pre-computation and manual adjustment approach sacrifices real-time correction capabilities and cannot handle dynamic scenes.

[0114] In some embodiments, the target information further includes user-input disassembly / assembly limitation instructions, and the method further includes:

[0115] Based on the disassembly and assembly limitation instructions, adjust the decision-driven information;

[0116] The motion decision correction during the disassembly and assembly process of the target component based on target information, to obtain the second disassembly and assembly animation path of the target component, includes:

[0117] Based on the decision-driven information, the strategy network parameters, and the physical motion simulation parameters, the motion decision is corrected during the disassembly and assembly process of the target component to obtain the second disassembly and assembly animation path of the target component.

[0118] In this embodiment, when a conflict is detected in the disassembly and assembly animation path, such as a component collision, exceeding the stress threshold, or deviating from physical laws, trajectory replanning based on the Proximal Policy Optimization (PPO) algorithm can be initiated in real time.

[0119] In some embodiments, Figure 5 This is a schematic diagram illustrating the principle of trajectory replanning based on the PPO algorithm in an embodiment of this application, as shown below. Figure 5 As shown, its inputs include policy network parameters (decision rules from motion decisions in the first disassembly / assembly animation path), physical collision result parameters (such as penetration direction and depth), and disassembly / assembly constraint instructions (which can be user multimodal instructions, such as the voice command "avoid tool contact with the tubing"). The clip function can be used to limit the policy update amplitude, generating a smooth and physically compliant new trajectory (such as a gradual change in screw unscrewing angle) under delay constraints (such as <50ms), thus obtaining the second disassembly / assembly animation path for the target component.

[0120] In some embodiments, the disassembly / assembly restriction instruction can be a single interaction mode (e.g., supporting only text instructions), which directly modifies the priority of the rule base by parsing keywords (e.g., "disassemble the shell first"), rather than integrating multimodal signals (e.g., voice). In some embodiments, the disassembly / assembly restriction instruction can be a multimodal instruction, that is, an instruction that integrates multiple different modalities (e.g., text, voice, and gestures).

[0121] In some embodiments, disassembly and assembly constraint instructions can directly intervene in the policy network by dynamically modifying the reward function (e.g., increasing the weight of the collision penalty term), thereby adjusting the decision-driven information, such as user input "avoid tool contact with pipeline" → which is converted into a constraint term. And add it to the reward function; the assembly manual text description "the circuit board must be kept level" → generate an attitude penalty term. This is then incorporated into the reward function, and combined with policy network parameters and physical motion simulation parameters to achieve collaborative decision-making between humans and AI. The final output is a seamless, efficient, and compliant disassembly animation, completely replacing the traditional manual trajectory adjustment process. This allows users to dynamically adjust the disassembly and assembly process using multimodal commands (such as voice and gestures), while the AI ​​model corrects the animation path in real time based on semantic and physical constraints, reducing human intervention and improving operational efficiency.

[0122] Its core objective is to generate new trajectories that conform to physical laws under delay constraints (such as <50ms), replacing the traditional manual adjustment process. The real-time trajectory correction formula is as follows:

[0123]

[0124] in, Policy network parameters represent the decision rules of the AI ​​model (such as choosing the direction / speed of movement). The probability ratio between the old and new strategies measures the magnitude of strategy updates and prevents excessively large policy changes. The advantage function (estimated using the Generalized Advantage Estimation (GAE) algorithm) quantifies the superiority or inferiority of the current action compared to the average performance. :limit The fluctuation range forces the strategy to update smoothly and limits it. The fluctuation range is forced to be updated smoothly by the strategy.

[0125] when Exceeding When the clip function is used, it is truncated to the interval boundary. The min function selects a smaller gradient update amount to avoid policy oscillation and ensure the continuity of actions during trajectory correction (such as the gradual change of screw rotation angle) to prevent secondary collisions or stress over-limits caused by drastic single-step adjustments (such as sudden gear misalignment).

[0126] In this embodiment, a multi-objective reward function can be designed to fuse semantic priority, collision penalty, and stress constraint, driving the policy network to generate motion decisions during the component assembly and disassembly process in the 3D model. The policy network can be intervened in dynamically by modifying the reward function weights through user commands (such as voice commands). Furthermore, trajectory replanning based on the PPO algorithm is initiated in real time. The policy network update can use a Clip function to limit the probability ratio fluctuation range, ensuring smooth trajectory adjustment to generate the optimal trajectory. This supports dynamic obstacle avoidance (such as avoiding oil pipes) and smooth adjustment (such as gradually unscrewing screws).

[0127] In some embodiments, the corrected motion decision can be fed back to the collision detection module and the physical dynamics model in real time to update the physical motion simulation parameters, thereby forming a closed-loop optimization, realizing the dynamic coordination of physical laws at the action level, and ensuring that the disassembly animation conforms to both the operation specifications and the physical simulation reality.

[0128] In other words, the embodiments of this application can dynamically inject instructions, converting speech (such as "keep level") into reward function constraints; a closed-loop feedback link corrects trajectory parameters and updates the physics engine (friction coefficient, collision range) and semantic model (priority weight) in real time, forming a "generation-detection-correction" self-optimization loop.

[0129] In this embodiment, a component assembly / disassembly relationship graph (e.g., nodes = component attributes, edges = connection strength) is constructed using a graph convolutional network, dynamically fusing the semantics of the assembly manual text with the simulation results of the physics engine. The collaboration between semantic logic and physical simulation is achieved through a reinforcement learning strategy, with the reward function including collision penalty terms, time efficiency terms, and stress constraint terms. Thus, semantic logic (e.g., assembly / disassembly order) can be extracted from the assembly manual text using a graph neural network, dynamically interacting with the physics engine (rigid body dynamics, continuous collision detection) to correct the motion trajectory of components during assembly / disassembly in the 3D model in real time, overcoming the limitations of traditional predefined rule bases.

[0130] Figure 6 This is a schematic diagram of the flowchart of an animation generation processing method provided in a specific example of an embodiment of this application, such as... Figure 6 As shown, it includes the following steps:

[0131] Step 1: Input the 3D model and semantic data; the semantic data includes assembly manual text and material parameter information, and the assembly manual text includes disassembly and assembly sequence description information and geometric information;

[0132] Step 2: Graph neural network modeling to construct a component assembly / disassembly relationship diagram of the 3D model, which serves as the input for continuous collision detection in Step 5;

[0133] Step 3: Input the assembly manual text from Step 1, and output the disassembly and assembly priority weight information through text-geometric alignment;

[0134] Step 4: Input the material parameter information from Step 1 and perform rigid body dynamics modeling;

[0135] Step 5: Continuous collision detection;

[0136] Step 6: Reinforcement learning strategy design. Based on the disassembly and assembly priority weight information and physical motion simulation parameters, the motion decision of the component disassembly and assembly process in the 3D model is made through the policy network.

[0137] Step 7: Based on the policy network parameters from Step 6 and user instructions, correct motion decisions for real-time trajectory correction;

[0138] Step 8: Output the disassembly animation of the 3D model.

[0139] In addition, the corrected motion decisions can be fed back to the collision detection module and the physical dynamics model in real time to update the physical motion simulation parameters, thereby forming a closed-loop optimization.

[0140] This application's embodiments, through deep integration of semantic logic reasoning and physical law simulation, construct a highly efficient, accurate, and highly automated intelligent toolchain, bringing multi-dimensional business transformation to fields such as industrial manufacturing, education and training, product design, and after-sales service. Its core value lies in three dimensions: reducing total lifecycle costs, improving operational efficiency and security, and expanding new application scenarios.

[0141] In the field of industrial equipment maintenance and training, traditional training models relying on human experience and physical prototypes are costly and time-consuming. However, the embodiments of this application utilize AI to automatically generate high-precision disassembly animations, replacing manual production of instructional videos and on-site guidance, saving over 90% of training costs. Simultaneously, real-time interactive animations guide trainees step-by-step, significantly reducing the risk of operational errors and shortening the training cycle by over 50%. For example, the complex disassembly and assembly process of aero-engines can be accurately simulated using virtual animation to precisely simulate key parameters such as screw torque and gear meshing angles, avoiding equipment damage due to operational errors, and reducing the cost of a single training session from tens of thousands of US dollars to thousands of US dollars.

[0142] In virtual product design and intelligent manufacturing scenarios, this application's embodiments accelerate product iteration and fault response through a dynamic collaborative mechanism. During the design phase, disassembly and assembly feasibility can be quickly verified, design flaws such as interference or tool inaccessibility can be identified, reducing the number of physical prototype tests and compressing the R&D cycle by 30%-50%. The cost of a single virtual verification is only 10%-20% of that of physical testing. In intelligent factory operation and maintenance, real-time generated maintenance animations can guide engineers to quickly troubleshoot equipment faults. Combined with augmented reality (AR) glasses, remote expert collaboration can be achieved, reducing downtime from hours to minutes, potentially saving millions of dollars in losses per production line annually. For example, disassembly animations of key components of semiconductor wafer equipment can be pushed to on-site engineers in real time, overlaid with AR annotations indicating the operating sequence and torque thresholds, ensuring both improved maintenance efficiency and safety.

[0143] In the after-sales support and education markets, this application's embodiments replace traditional text and image manuals with interactive 3D animations, significantly reducing the operational threshold for customers. Medical device manufacturers can transform the maintenance process of Magnetic Resonance Imaging (MRI) equipment into multilingual animations, supporting dynamic adjustment of the viewing angle via voice commands, enabling non-technical personnel to complete basic maintenance and reducing after-sales support requests by more than 60%. In the education field, AI-generated disassembly and assembly animations can be seamlessly integrated into VR training platforms, allowing students to disassemble gearboxes using virtual tools via gestures, immersing themselves in learning mechanical principles and promoting the transition of Science, Technology, Engineering, and Mathematics (STEM) education from theory to practice. Furthermore, the technology licensing model can provide semantic-physical collaborative engines to industrial software vendors or customize solutions for specific industries (such as automotive and energy).

[0144] See Figure 7 , Figure 7 This is a schematic diagram of the structure of an animation generation and processing device provided in an embodiment of this application, such as... Figure 7 As shown, the animation generation processing device 700 includes:

[0145] The acquisition module 701 is used to acquire semantic data of the 3D model, the semantic data including the disassembly and assembly sequence description information, geometric information and material parameter information of the parts in the 3D model;

[0146] The decision module 702 is used to make motion decisions during the disassembly and assembly process of the components in the 3D model based on the disassembly and assembly sequence description information, geometric information and material parameter information, so as to obtain the first disassembly and assembly animation path of the components in the 3D model.

[0147] Optionally, the decision module 702 includes:

[0148] A semantic matching unit is used to perform semantic matching based on the disassembly and assembly sequence description information and the geometric information to generate disassembly and assembly priority weight information of the components in the 3D model.

[0149] The determining unit is used to determine the physical motion simulation parameters of the component assembly and disassembly process in the 3D model based on the geometric information and the material parameter information.

[0150] The motion decision unit is used to make motion decisions during the disassembly and assembly process of the components in the 3D model based on the disassembly and assembly priority weight information and physical motion simulation parameters, so as to obtain the first disassembly and assembly animation path of the components in the 3D model.

[0151] Optionally, the motion decision unit is specifically used for:

[0152] Based on the disassembly and assembly priority weight information, the target component to be disassembled or assembled in the 3D model is determined.

[0153] Based on the physical motion simulation parameters during the disassembly and assembly process of the target component, the decision-driving information is determined;

[0154] Based on the decision-driven information, the driving strategy network generates motion decisions during the disassembly and assembly process of the target component, thereby obtaining the first disassembly and assembly animation path of the target component.

[0155] Optionally, the device further includes:

[0156] The decision correction module is used to correct the motion decision during the disassembly and assembly of the target component based on the target information when the first disassembly and assembly animation path planning is found to be unreasonable based on the physical motion simulation parameters, so as to obtain the second disassembly and assembly animation path of the target component.

[0157] The target information includes strategy network parameters and physical motion simulation parameters. The strategy network parameters are used to characterize the decision rules for motion decisions in the first disassembly and assembly animation path.

[0158] Optionally, the target information further includes user-input disassembly / assembly limitation instructions, and the device further includes:

[0159] An adjustment module is used to adjust the decision-driven information based on the disassembly and assembly limitation instructions;

[0160] The decision correction module is specifically used to correct the motion decision during the disassembly and assembly process of the target component based on the decision-driven information, the strategy network parameters, and the physical motion simulation parameters, so as to obtain the second disassembly and assembly animation path of the target component.

[0161] Optionally, the determining unit is specifically used for:

[0162] Based on the geometric information and the material parameter information, the continuous physical collision situation during the disassembly and assembly process of the components in the 3D model is detected, and the physical collision result parameters during the disassembly and assembly process of the components in the 3D model are obtained.

[0163] The physical motion simulation parameters include the physical collision result parameters.

[0164] Optionally, the determining unit is specifically used for:

[0165] Based on the geometric information, using the components in the 3D model as nodes and the connection relationships between different components as edges, a graph neural network is used to construct a component assembly / disassembly relationship graph of the 3D model; wherein, the component assembly / disassembly relationship graph is used to indicate the topological relationships between different components;

[0166] Based on the component assembly / disassembly relationship diagram and the material parameter information, the continuous physical collision situation during the component assembly / disassembly process in the 3D model is detected, and the physical collision result parameters during the component assembly / disassembly process in the 3D model are obtained.

[0167] Optionally, the physical motion simulation parameters further include rigid body dynamic parameters, and the determining unit is further used for:

[0168] Based on the material parameter information, rigid body dynamic parameters for the assembly and disassembly of components in the 3D model are constructed.

[0169] The animation generation and processing device 700 can implement all the processes implemented in the above-described animation generation and processing method embodiments and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0170] See Figure 8 The figure shows a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 8 As shown, the electronic device 800 includes: a processor 801, a memory 802, a user interface 803, and a bus interface 804.

[0171] Processor 801 is used to read the program from memory 802 and execute the following procedures:

[0172] Obtain semantic data of the 3D model, including the disassembly and assembly sequence description information, geometric information and material parameter information of the components in the 3D model;

[0173] Based on the disassembly and assembly sequence description information, geometric information, and material parameter information, motion decisions are made during the disassembly and assembly process of the components in the 3D model to obtain the first disassembly and assembly animation path of the components in the 3D model.

[0174] exist Figure 8 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 801 and memory represented by memory 802 together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 804 provides an interface. For different user devices, user interface 803 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0175] The processor 801 is responsible for managing the bus architecture and general processing, while the memory 802 can store the data used by the processor 801 when performing operations.

[0176] In some embodiments, the processor 801 is further configured to:

[0177] Semantic matching is performed based on the disassembly and assembly sequence description information and the geometric information to generate disassembly and assembly priority weight information of the components in the 3D model;

[0178] Based on the geometric information and the material parameter information, determine the physical motion simulation parameters of the component assembly and disassembly process in the 3D model;

[0179] Based on the disassembly and assembly priority weight information and physical motion simulation parameters, motion decisions are made during the disassembly and assembly process of the components in the 3D model to obtain the first disassembly and assembly animation path of the components in the 3D model.

[0180] In some embodiments, the processor 801 is further configured to:

[0181] Based on the disassembly and assembly priority weight information, the target component to be disassembled or assembled in the 3D model is determined.

[0182] Based on the physical motion simulation parameters during the disassembly and assembly process of the target component, the decision-driving information is determined;

[0183] Based on the decision-driven information, the driving strategy network generates motion decisions during the disassembly and assembly process of the target component, thereby obtaining the first disassembly and assembly animation path of the target component.

[0184] In some embodiments, the processor 801 is further configured to:

[0185] If the first disassembly and assembly animation path planning is found to be unreasonable based on the physical motion simulation parameters, the motion decision during the disassembly and assembly of the target component is corrected based on the target information to obtain the second disassembly and assembly animation path of the target component.

[0186] The target information includes strategy network parameters and physical motion simulation parameters. The strategy network parameters are used to characterize the decision rules for motion decisions in the first disassembly and assembly animation path.

[0187] In some embodiments, the target information further includes user-input disassembly / assembly limitation instructions, and the processor 801 is further configured to:

[0188] Based on the disassembly and assembly limitation instructions, adjust the decision-driven information;

[0189] Based on the decision-driven information, the strategy network parameters, and the physical motion simulation parameters, the motion decision is corrected during the disassembly and assembly process of the target component to obtain the second disassembly and assembly animation path of the target component.

[0190] In some embodiments, the processor 801 is further configured to:

[0191] Based on the geometric information and the material parameter information, the continuous physical collision situation during the disassembly and assembly process of the components in the 3D model is detected, and the physical collision result parameters during the disassembly and assembly process of the components in the 3D model are obtained.

[0192] The physical motion simulation parameters include the physical collision result parameters.

[0193] In some embodiments, the processor 801 is further configured to:

[0194] Based on the geometric information, using the components in the 3D model as nodes and the connection relationships between different components as edges, a graph neural network is used to construct a component assembly / disassembly relationship graph of the 3D model; wherein, the component assembly / disassembly relationship graph is used to indicate the topological relationships between different components;

[0195] Based on the component assembly / disassembly relationship diagram and the material parameter information, the continuous physical collision situation during the component assembly / disassembly process in the 3D model is detected, and the physical collision result parameters during the component assembly / disassembly process in the 3D model are obtained.

[0196] In some embodiments, the physical motion simulation parameters further include rigid body dynamic parameters, and the processor 801 is further configured to:

[0197] Based on the material parameter information, rigid body dynamic parameters for the assembly and disassembly of components in the 3D model are constructed.

[0198] Preferably, the present invention also provides an electronic device 800, including a processor 801, a memory 802, and a computer program stored in the memory 802 and executable on the processor 801. When the computer program is executed by the processor 801, it implements the various processes of the above-described animation generation processing method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0199] This invention also provides a readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described animation generation method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0200] This application also provides a computer program product, including computer instructions. When executed by a processor, the computer instructions implement the various processes of the above-described animation generation processing method embodiments and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0201] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0202] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0203] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0204] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0205] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0206] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0207] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An animation generation and processing method, characterized in that, The method includes: Obtain semantic data of the 3D model, including the disassembly and assembly sequence description information, geometric information and material parameter information of the components in the 3D model; Based on the disassembly and assembly sequence description information, geometric information, and material parameter information, motion decisions are made during the disassembly and assembly process of the components in the 3D model to obtain the first disassembly and assembly animation path of the components in the 3D model.

2. The method according to claim 1, characterized in that, The motion decision-making process for the disassembly and assembly of components in the 3D model based on the disassembly and assembly sequence description information, geometric information, and material parameter information, to obtain the first disassembly and assembly animation path of the components in the 3D model, includes: Semantic matching is performed based on the disassembly and assembly sequence description information and the geometric information to generate disassembly and assembly priority weight information of the components in the 3D model; Based on the geometric information and the material parameter information, determine the physical motion simulation parameters of the component assembly and disassembly process in the 3D model; Based on the disassembly and assembly priority weight information and physical motion simulation parameters, motion decisions are made during the disassembly and assembly process of the components in the 3D model to obtain the first disassembly and assembly animation path of the components in the 3D model.

3. The method according to claim 2, characterized in that, The motion decision-making process for the component disassembly and assembly process in the 3D model based on the disassembly and assembly priority weight information and physical motion simulation parameters, to obtain the first disassembly and assembly animation path of the component in the 3D model, includes: Based on the disassembly and assembly priority weight information, the target component to be disassembled or assembled in the 3D model is determined. Based on the physical motion simulation parameters during the disassembly and assembly process of the target component, the decision-driving information is determined; Based on the decision-driven information, the driving strategy network generates motion decisions during the disassembly and assembly process of the target component, thereby obtaining the first disassembly and assembly animation path of the target component.

4. The method according to claim 3, characterized in that, After generating motion decisions for the target component's assembly and disassembly process based on the decision-driven information and obtaining the first assembly and disassembly animation path for the target component, the method further includes: If the first disassembly and assembly animation path planning is found to be unreasonable based on the physical motion simulation parameters, the motion decision during the disassembly and assembly of the target component is corrected based on the target information to obtain the second disassembly and assembly animation path of the target component. The target information includes strategy network parameters and physical motion simulation parameters. The strategy network parameters are used to characterize the decision rules for motion decisions in the first disassembly and assembly animation path.

5. The method according to claim 4, characterized in that, The target information also includes user-input disassembly and assembly limitations instructions, and the method further includes: Based on the disassembly and assembly limitation instructions, adjust the decision-driven information; The motion decision correction during the disassembly and assembly process of the target component based on target information, to obtain the second disassembly and assembly animation path of the target component, includes: Based on the decision-driven information, the strategy network parameters, and the physical motion simulation parameters, the motion decision is corrected during the disassembly and assembly process of the target component to obtain the second disassembly and assembly animation path of the target component.

6. The method according to claim 2, characterized in that, The determination of physical motion simulation parameters during the assembly and disassembly of components in the 3D model based on the geometric information and material parameter information includes: Based on the geometric information and the material parameter information, the continuous physical collision situation during the disassembly and assembly process of the components in the 3D model is detected, and the physical collision result parameters during the disassembly and assembly process of the components in the 3D model are obtained. The physical motion simulation parameters include the physical collision result parameters.

7. The method according to claim 6, characterized in that, Based on the geometric information and the material parameter information, the continuous physical collisions during the assembly and disassembly of components in the 3D model are detected to obtain the physical collision result parameters during the assembly and disassembly of components in the 3D model, including: Based on the geometric information, using the components in the 3D model as nodes and the connection relationships between different components as edges, a graph neural network is used to construct a component assembly / disassembly relationship graph of the 3D model; wherein, the component assembly / disassembly relationship graph is used to indicate the topological relationships between different components; Based on the component assembly / disassembly relationship diagram and the material parameter information, the continuous physical collision situation during the component assembly / disassembly process in the 3D model is detected, and the physical collision result parameters during the component assembly / disassembly process in the 3D model are obtained.

8. The method according to claim 6, characterized in that, The physical motion simulation parameters also include rigid body dynamic parameters. The determination of the physical motion simulation parameters during the assembly and disassembly of components in the 3D model based on the geometric information and the material parameter information further includes: Based on the material parameter information, rigid body dynamic parameters for the assembly and disassembly of components in the 3D model are constructed.

9. An animation generation and processing device, characterized in that, The device includes: The acquisition module is used to acquire semantic data of the 3D model, including the disassembly and assembly sequence description information, geometric information and material parameter information of the parts in the 3D model; The decision module is used to make motion decisions during the disassembly and assembly process of the components in the 3D model based on the disassembly and assembly sequence description information, geometric information and material parameter information, so as to obtain the first disassembly and assembly animation path of the components in the 3D model.

10. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the animation generation processing method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the animation generation processing method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the animation generation processing method as described in any one of claims 1 to 8.

Images