A state tree-based virtual object behavior simulation method and system
By using a virtual object behavior simulation method based on state trees and deep learning, the accuracy and scalability issues of traditional simulation techniques in simulating the behavior of group objects are solved, resulting in more realistic and flexible simulation results and reducing development costs.
Patent Information
- Application Number
- CN202610521039.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-20
- Publication Date
- 2026-08-25
AI Technical Summary
Traditional computer simulation technology struggles to accurately reflect the behavior of groups of objects in different scenarios, and its modeling lacks scalability and flexibility.
A virtual object behavior simulation method based on state tree is adopted, combined with deep learning technology. By acquiring the task type and environment information of the virtual object, and using a pre-trained deep learning model and state tree structure, the behavior of the virtual object in the simulation scene is generated.
It improves the realism and credibility of simulation results, enhances scalability and flexibility in the face of new tasks and environmental factors, and reduces the cost and time of simulation development.
Smart Images
Figure CN122633307A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer simulation technology, and in particular relates to a virtual object behavior simulation method and system based on state tree. Background Technology
[0002] Currently, computer simulation technology is widely used in fields such as group behavior simulation. Computer simulation technology can simulate and generate virtual intelligent agents with autonomous or semi-autonomous capabilities, and then simulate various entity behaviors of individuals in a group collaborative environment.
[0003] However, the models generated by traditional computer simulation technology are difficult to accurately reflect the behavior of group objects in different scenarios; at the same time, the scalability and flexibility of the models are also difficult to adapt to constantly changing scenarios. Summary of the Invention
[0004] To address the aforementioned technical problems, this invention proposes a virtual object behavior simulation method and system based on a state tree.
[0005] The first aspect of this invention proposes a virtual object behavior simulation method based on a state tree, the method comprising: Obtain the task type data and behavior data of the virtual object in the simulation scenario; Based on the position of the virtual object in the simulation scene, obtain the environmental information data of the virtual object; Based on the task type, a root node corresponding to the task type is determined from the pre-established state tree; wherein, the state tree includes at least two subtrees, each subtree having a root node and at least one child node belonging to the root node; the root node has a correspondence with the task type, and the child node has a correspondence with the behavior type; The environmental information data and behavior data of the virtual object are input into a pre-trained deep learning model to obtain the behavior type of the virtual object; Based on the behavior type of the virtual object, the corresponding first child node is determined from at least one of the child nodes belonging to the determined root node; Based on the parameter information of the first child node, the behavior of the virtual object in the simulation scene is generated.
[0006] A second aspect of the present invention provides a virtual object behavior simulation system based on a state tree, the system comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to implement the method as described in the first aspect of the present invention.
[0007] In summary, this invention proposes a virtual object behavior simulation method and system based on state trees. By comprehensively and systematically organizing and managing the behavioral states of virtual objects through state trees and deep learning technology, and combining state tree-based search and processing mechanisms, it can more accurately simulate the behavior of virtual objects in complex environments, making their decisions and actions more consistent with reality and improving the realism and credibility of the simulation results. Simultaneously, the state tree-based modeling structure enhances scalability and flexibility in the face of new tasks and environmental factors, reducing the cost and time of simulation development. Attached Figure Description
[0008] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0009] Figure 1 This is a flowchart illustrating a virtual object behavior simulation method based on a state tree, as proposed in an embodiment of the present invention.
[0010] Figure 2 This is a schematic diagram of the structure of a virtual object behavior simulation system based on a state tree, as proposed in an embodiment of the present invention. Detailed Implementation
[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0012] Figure 1 This invention illustrates a virtual object behavior simulation method based on a state tree, as proposed in the first aspect of the present invention. The method includes: Step 101: Obtain the task type data and behavior data of the virtual object in the simulation scene.
[0013] The virtual objects can correspond to individuals in various group activities in a real-world environment. For example, vehicles in a group of vehicles performing a transportation task, or personnel in a group working together to perform a task.
[0014] The task type primarily represents the overall name of the task issued to the virtual object; for example, in competitive gaming environments, task types can include offensive tasks, observation tasks, and defensive tasks. In different industries, task types can be flexibly set according to industry characteristics.
[0015] The behavioral data includes at least one of the following: velocity information data, direction information data, motion posture information data, motion duration information data, or acceleration information data. In an optional embodiment, the behavioral data reflects the current behavior and state of the virtual object when the task type data is acquired. For example, during a continuous simulation, new task type data is generated. At this time, the behavioral data of the virtual object can be directly retrieved. If the task type data is being issued for the first time, the required data of the virtual object can be obtained from a previously collected information database as the initial value of the behavioral data, such as the acceleration information of the virtual object. This information database can be pre-collected and stored for the behavioral information of a type of object to be simulated, so that it can be directly retrieved when simulating that type of object.
[0016] Step 102: Based on the position of the virtual object in the simulation scene, obtain the environmental information data of the virtual object; In one optional embodiment, environmental information data of the virtual object is obtained based on the virtual object's position in the simulation scene, including... Based on the location in the simulation scenario, obtain the corresponding coordinates of the real geographical location; The environmental information data of the virtual object is obtained from the environmental database; wherein, the environmental database stores environmental information data corresponding to the coordinates of the real geographical location that has been collected in advance.
[0017] The environmental information data includes at least one of the following: geographic environmental information data and weather environmental information data. In an optional embodiment, the corresponding geographic environmental information data and weather environmental information data can be retrieved more accurately from the environmental database based on the actual time to be simulated. Alternatively, the environmental database may only store a fixed set of geographic environmental information data and weather environmental information data corresponding to the coordinates of the actual geographical location. The geographic environmental information data and the weather environmental information data can be updated periodically based on a set collection cycle of the real environment. By incorporating environmental information, the behavior of virtual objects can be simulated more realistically.
[0018] Step 103: Based on the task type, determine the root node corresponding to the task type from the pre-established state tree.
[0019] The state tree includes at least two subtrees, each subtree having a root node and at least one child node belonging to the root node; the root node corresponds to the task type, and the child node corresponds to the behavior type.
[0020] In one alternative embodiment, the logical order of all the subtrees in the at least two subtrees is set according to the execution order of the task type corresponding to the root node of the subtree.
[0021] In one optional embodiment, the child nodes belonging to the same root node include at least two levels of child nodes; a lower-level third child node directly belongs to an upper-level second child node; the logical order of the third child nodes at the same level that directly belong to the same second child node is set according to the execution order of the behavior type corresponding to the third child node. Wherein, for any specific child node corresponding to a behavior type, if it can be further subdivided into multiple behavior types, then the child nodes corresponding to the multiple subdivided behavior types belong to that specific child node, and that specific child node is subordinate to the child nodes corresponding to these multiple subdivided behavior types. In this way, by establishing multi-layered child nodes, the finer detail of the action simulation of the virtual object can be improved.
[0022] Step 104: Input the environmental information data and behavior data of the virtual object into a pre-trained deep learning model to obtain the behavior type of the virtual object.
[0023] In this context, "behavior type" refers to the collective term for behavioral actions and behavioral states, which can influence each other. In one optional embodiment, behavioral states can influence behavioral actions by establishing correlations or functional relationships. Overall, behavioral actions and behavioral states, as mutually influencing factors, jointly form the specific behavior of the virtual object in the simulation scenario. For example, when simulating an offensive mission in a live-action combat game, the offensive mission can be categorized into behavioral types such as maneuver behavior, reconnaissance behavior, and attack behavior.
[0024] In this embodiment of the invention, a pre-trained deep learning model is used to estimate the most likely behavior type under specific environmental information data and behavioral data, thereby improving the accuracy of simulation and addressing the low scalability issues caused by traditional simulation methods that rely on mapping relationships to obtain behavior types. The deep learning model can be trained using any general training method, and this embodiment of the invention does not limit this. For example, when simulating an offensive mission in a live-action combat game, if the environmental information data indicates a mountainous, rainy environment, and the behavioral data shows high-speed and prolonged movement, the deep learning model may output the corresponding behavior for performing a maneuvering action.
[0025] In one optional embodiment, the environmental information data and behavioral data of the virtual object further include at least one of the following forms of information data: video information data or image information data. Considering the accuracy requirements of simulation, simple numerical information is not as rich in information as videos and images; therefore, videos and images can be used to record environmental information data and behavioral data.
[0026] The environmental information data and behavioral data of the virtual object are input into a pre-trained deep learning model to obtain the behavior type of the virtual object, including: Environmental features are extracted from the environmental information data of the virtual object using a residual convolutional neural network. Motion features are extracted from the behavioral data of the virtual object using a long short-term memory network. The environmental features and motion features are concatenated and then input into a fully connected layer to obtain the behavior type of the virtual object.
[0027] The structures of residual convolutional neural networks, long short-term memory networks, and fully connected layers can directly use existing empirical models, or the coefficients of these models can be adjusted by obtaining corresponding training data. This embodiment of the invention does not further limit this.
[0028] It should be noted that the combination of the above deep learning models is not fixed. Specific deep learning models can be selected and combined according to the different features that need to be processed.
[0029] Step 105: Based on the behavior type of the virtual object, determine the corresponding first child node among at least one child node belonging to the determined root node; The behavior type can be in a one-to-one correspondence with the child node, or it can be a many-to-one, one-to-many, or many-to-many relationship, so that parallel behavior logic can be established as needed.
[0030] Step 106: Based on the parameter information of the first child node, generate the behavior state of the virtual object in the simulation scene.
[0031] In one optional embodiment, each child node stores corresponding parameter information, which includes at least one of the following: velocity information, direction information, motion posture information, acceleration information, or state information. In one embodiment, the parameter information is defined as behavioral data of the virtual object's to-be-executed behavior.
[0032] In an optional embodiment, the child nodes in the state tree may further have judgment logic and forwarding logic. If the behavioral state of the virtual object in the simulation scenario has been generated, the first child node sends completion information to the upper-level child node to which it belongs. The upper-level child node can determine whether all behavioral states under its corresponding behavioral type have been generated. If they have been generated, it continues to forward the completion information to its upper-level child node; if they have not been fully generated, the upper-level child node determines at least one other child node belonging to it according to its execution logic to generate other behavioral states of the virtual object in the simulation scenario.
[0033] In an alternative embodiment, the state tree can also be extended, including: Obtain node extension information; Based on the extended task type in the node extended information, a new root node corresponding to the extended task type is added to the state tree; Based on the extension behavior type of the extension task type in the node extension information, add the extension child node corresponding to the extension behavior type to the newly added root node; Obtain the extended parameter information of the extended behavior type; Add the extended parameter information to the corresponding extended child node.
[0034] Figure 2 The structure of a state tree-based virtual object behavior simulation system 200 according to a second aspect of the present invention is shown. The system 200 includes: at least one processor 201; and a memory 202 communicatively connected to the at least one processor. The memory 202 stores instructions executable by the at least one processor 201, which, when executed by the at least one processor 201, enable the at least one processor 201 to implement the following method steps: Step 1: Obtain the task type data and behavior data of the virtual object in the simulation scene.
[0035] The virtual objects can correspond to individuals in various group activities in a real-world environment. For example, vehicles in a group of vehicles performing a transportation task, or personnel in a group working together to perform a task.
[0036] The task type primarily represents the overall name of the task issued to the virtual object; for example, in competitive gaming environments, task types can include offensive tasks, observation tasks, and defensive tasks. In different industries, task types can be flexibly set according to industry characteristics.
[0037] The behavioral data includes at least one of the following: velocity information data, direction information data, motion posture information data, motion duration information data, or acceleration information data. In an optional embodiment, the behavioral data reflects the current behavior and state of the virtual object when the task type data is acquired. For example, during a continuous simulation, new task type data is generated. At this time, the behavioral data of the virtual object can be directly retrieved. If the task type data is being issued for the first time, the required data of the virtual object can be obtained from a previously collected information database as the initial value of the behavioral data, such as the acceleration information of the virtual object. This information database can be pre-collected and stored for the behavioral information of a type of object to be simulated, so that it can be directly retrieved when simulating that type of object.
[0038] Step 2: Based on the position of the virtual object in the simulation scene, obtain the environmental information data of the virtual object; In one optional embodiment, environmental information data of the virtual object is obtained based on the virtual object's position in the simulation scene, including... Based on the location in the simulation scenario, obtain the corresponding coordinates of the real geographical location; The environmental information data of the virtual object is obtained from the environmental database; wherein, the environmental database stores environmental information data corresponding to the coordinates of the real geographical location that has been collected in advance.
[0039] The environmental information data includes at least one of the following: geographic environmental information data and weather environmental information data. In an optional embodiment, the corresponding geographic environmental information data and weather environmental information data can be retrieved more accurately from the environmental database based on the actual time to be simulated. Alternatively, the environmental database may only store a fixed set of geographic environmental information data and weather environmental information data corresponding to the coordinates of the actual geographical location. The geographic environmental information data and the weather environmental information data can be updated periodically based on a set collection cycle of the real environment. By incorporating environmental information, the behavior of virtual objects can be simulated more realistically.
[0040] Step 3: Based on the task type, determine the root node corresponding to the task type from the pre-established state tree.
[0041] The state tree includes at least two subtrees, each subtree having a root node and at least one child node belonging to the root node; the root node corresponds to the task type, and the child node corresponds to the behavior type.
[0042] In one alternative embodiment, the logical order of all the subtrees in the at least two subtrees is set according to the execution order of the task type corresponding to the root node of the subtree.
[0043] In one optional embodiment, the child nodes belonging to the same root node include at least two levels of child nodes; a lower-level third child node directly belongs to an upper-level second child node; the logical order of the third child nodes at the same level that directly belong to the same second child node is set according to the execution order of the behavior type corresponding to the third child node. Wherein, for any specific child node corresponding to a behavior type, if it can be further subdivided into multiple behavior types, then the child nodes corresponding to the multiple subdivided behavior types belong to that specific child node, and that specific child node is subordinate to the child nodes corresponding to these multiple subdivided behavior types. In this way, by establishing multi-layered child nodes, the finer detail of the action simulation of the virtual object can be improved.
[0044] Step 4: Input the environmental information data and behavior data of the virtual object into a pre-trained deep learning model to obtain the behavior type of the virtual object.
[0045] In this context, "behavior type" refers to the collective term for behavioral actions and behavioral states, which can influence each other. In one optional embodiment, behavioral states can influence behavioral actions by establishing correlations or functional relationships. Overall, behavioral actions and behavioral states, as mutually influencing factors, jointly form the specific behavior of the virtual object in the simulation scenario. For example, when simulating an offensive mission in a live-action combat game, the offensive mission can be categorized into behavioral types such as maneuver behavior, reconnaissance behavior, and attack behavior.
[0046] In this embodiment of the invention, a pre-trained deep learning model is used to estimate the most likely behavior type under specific environmental information data and behavioral data, thereby improving the accuracy of simulation and addressing the low scalability issues caused by traditional simulation methods that rely on mapping relationships to obtain behavior types. The deep learning model can be trained using any general training method, and this embodiment of the invention does not limit this. For example, when simulating an offensive mission in a live-action combat game, if the environmental information data indicates a mountainous, rainy environment, and the behavioral data shows high-speed and prolonged movement, the deep learning model may output the corresponding behavior for performing a maneuvering action.
[0047] In one optional embodiment, the environmental information data and behavioral data of the virtual object further include at least one of the following forms of information data: video information data or image information data. Considering the accuracy requirements of simulation, simple numerical information is not as rich in information as videos and images; therefore, videos and images can be used to record environmental information data and behavioral data.
[0048] The environmental information data and behavioral data of the virtual object are input into a pre-trained deep learning model to obtain the behavior type of the virtual object, including: Environmental features are extracted from the environmental information data of the virtual object using a residual convolutional neural network. Motion features are extracted from the behavioral data of the virtual object using a long short-term memory network. The environmental features and motion features are concatenated and then input into a fully connected layer to obtain the behavior type of the virtual object.
[0049] It should be noted that the combination of the above deep learning models is not fixed. Specific deep learning models can be selected and combined according to the different features that need to be processed.
[0050] Step 5: Based on the behavior type of the virtual object, determine the corresponding first child node among at least one child node belonging to the determined root node; The behavior type can be in a one-to-one correspondence with the child node, or it can be a many-to-one, one-to-many, or many-to-many relationship, so that parallel behavior logic can be established as needed.
[0051] Step Six: Based on the parameter information of the first child node, generate the behavior state of the virtual object in the simulation scene.
[0052] In one optional embodiment, each child node stores corresponding parameter information, which includes at least one of the following: velocity information, direction information, motion posture information, acceleration information, or state information. In one embodiment, the parameter information is defined as behavioral data of the virtual object's to-be-executed behavior.
[0053] In an optional embodiment, the child nodes in the state tree may further have judgment logic and forwarding logic. If the behavioral state of the virtual object in the simulation scenario has been generated, the first child node sends completion information to the upper-level child node to which it belongs. The upper-level child node can determine whether all behavioral states under its corresponding behavioral type have been generated. If they have been generated, it continues to forward the completion information to its upper-level child node; if they have not been fully generated, the upper-level child node determines at least one other child node belonging to it according to its execution logic to generate other behavioral states of the virtual object in the simulation scenario.
[0054] In an alternative embodiment, the state tree can also be extended, including: Obtain node extension information; Based on the extended task type in the node extended information, a new root node corresponding to the extended task type is added to the state tree; Based on the extension behavior type of the extension task type in the node extension information, add the extension child node corresponding to the extension behavior type to the newly added root node; Obtain the extended parameter information of the extended behavior type; Add the extended parameter information to the corresponding extended child node.
[0055] In summary, the virtual object behavior simulation method and system proposed in this invention, based on state trees, comprehensively and systematically organizes and manages the behavioral states of virtual objects through state trees and deep learning technology. Combined with a state tree-based search and processing mechanism, it can more accurately simulate the behavior of virtual objects in complex environments, making their decisions and actions more consistent with reality and improving the realism and credibility of the simulation results. Furthermore, the state tree-based modeling structure enhances scalability and flexibility in the face of new tasks and environmental factors, reducing the cost and time of simulation development.
[0056] It should be noted that, in the several embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units or steps is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or steps may be combined or integrated into another system or step, 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.
[0057] 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 this embodiment according to actual needs.
[0058] Furthermore, in the various embodiments of the present invention, the functional units or method steps can be integrated into one processing unit or one method step, or each unit or method step can exist separately physically or functionally, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware or in the form of hardware plus software functional units.
[0059] It should be noted that the embodiments provided in this invention are all illustrative, and different embodiments can be arbitrarily and reasonably combined. For the sake of brevity, not all possible combinations of the various technical features in the above embodiments are described; however, as long as such combinations do not contradict each other, they should all be considered to fall within the scope of this specification. Furthermore, it should be understood that the systems and methods disclosed in the embodiments provided in this invention can be implemented in other ways or with modifications. Any substitutions made in hardware or software, or any modifications made without departing from the concept of this invention, are within the protection scope of this invention.
Claims
1. A virtual object behavior simulation method based on a state tree, characterized in that, include: Obtain the task type data and behavior data of the virtual object in the simulation scenario; Based on the position of the virtual object in the simulation scene, obtain the environmental information data of the virtual object; Based on the task type, a root node corresponding to the task type is determined from the pre-established state tree; wherein, the state tree includes at least two subtrees, each subtree having a root node and at least one child node belonging to the root node; the root node has a correspondence with the task type, and the child node has a correspondence with the behavior type; The environmental information data and behavior data of the virtual object are input into a pre-trained deep learning model to obtain the behavior type of the virtual object; Based on the behavior type of the virtual object, a corresponding first child node is determined from at least one of the child nodes belonging to the determined root node; Based on the parameter information of the first child node, the behavior of the virtual object in the simulation scene is generated.
2. The virtual object behavior simulation method based on state tree according to claim 1, characterized in that: The environmental information data includes at least one of the following: geographic environmental information data and weather environmental information data; The behavioral data includes at least one of the following: velocity information data, direction information data, motion posture information data, motion duration information data, or acceleration information data.
3. The virtual object behavior simulation method based on a state tree according to claim 2, characterized in that: Based on the location of the virtual object in the simulation scene, environmental information data of the virtual object is obtained, including... Based on the location in the simulation scenario, obtain the corresponding coordinates of the real geographical location; The environmental information data of the virtual object is obtained from the environmental database; wherein, the environmental database stores environmental information data corresponding to the coordinates of the real geographical location that has been collected in advance.
4. The virtual object behavior simulation method based on a state tree as described in claim 3, characterized in that: The logical order of all the subtrees in the at least two subtrees is set according to the execution order of the task type corresponding to the root node of the subtree.
5. The virtual object behavior simulation method based on a state tree as described in claim 4, characterized in that: The child nodes belonging to the same root node include child nodes at least two levels; A third child node in a lower level is directly subordinate to a second child node in an upper level; The logical order of the third child nodes that are directly subordinate to the same second child node at the same level is set according to the execution order of the behavior type corresponding to the third child node.
6. The virtual object behavior simulation method based on a state tree as described in claim 5, characterized in that: The environmental information data and behavioral data of the virtual object are input into a pre-trained deep learning model to obtain the behavior type of the virtual object, including: Environmental features are extracted from the environmental information data of the virtual object using a residual convolutional neural network. Motion features are extracted from the behavioral data of the virtual object using a long short-term memory network. The environmental features and motion features are concatenated and then input into a fully connected layer to obtain the behavior type of the virtual object.
7. The virtual object behavior simulation method based on a state tree as described in claim 6, characterized in that: The parameter information of the child node includes at least one of the following: state information, direction information, velocity information, acceleration information, or attitude information.
8. The virtual object behavior simulation method based on a state tree as described in claim 7, characterized in that: In response to generating the behavioral state of the virtual object in the simulation scene, the first child node sends completion information to the upper-level child node to which the first child node belongs.
9. The virtual object behavior simulation method based on state tree as described in claim 1, characterized in that: Obtain node extension information; Based on the extended task type in the node extended information, a new root node corresponding to the extended task type is added to the state tree; Based on the extension behavior type of the extension task type in the node extension information, add an extension child node corresponding to the extension behavior type to the newly added root node; Obtain the extended parameter information of the extended behavior type; Add the extended parameter information to the corresponding extended child node.
10. A virtual object behavior simulation system based on a state tree, the system comprising: At least one processor; And a memory communicatively connected to at least one processor; wherein the memory stores instructions executable by at least one processor, the instructions being executed by at least one processor to enable the at least one processor to implement the method as described in any one of claims 1-9.