Data-driven game role action automatic optimization method and system

Abstract

The invention provides a data-driven game role action automatic optimization method and system, and relates to the technical field of data processing.The method comprises the steps that scene data of a reference action space body is collected according to real-time running graph coordinates of game characters, and environmental geometric parameters and environmental physical attributes are obtained; constructing an environment constraint vector; receiving a user real-time operation flow to carry out standard action flow matching; executing constraint adaptive deformation, and outputting an optimized action flow; and driving the action rendering of the game character by adopting the optimized action flow, and outputting a visual role scene action. The technical problem that game role actions in the prior art are generally controlled through a predefined animation library or animation sequence and cannot be adjusted in real time according to environment changes, and therefore the fluency of a game and the immersion of players are affected is solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a data-driven method and system for automatically optimizing game character actions. Background Technology

[0002] With the rapid development of modern game technology, the performance of game characters' movements has become one of the key factors affecting game immersion and player experience. In order to make the behavior of game characters more natural, realistic and interactive, game developers usually need to develop complex animation systems and physics engines to drive character movements.

[0003] In traditional games, character movements are usually controlled by predefined animation libraries or sequences. These animation libraries or sequences are static, and the character's behavior and reactions are preset and cannot be adjusted in real time according to changes in the environment. If the character enters a complex or irregular environment, such as uneven ground or a narrow space, the character's movements will become unnatural or impossible to execute. For example, the character may get stuck, jump at an inappropriate height, or collide with obstacles, which will seriously affect the smoothness of the game and the player's immersion. Summary of the Invention

[0004] This application provides a data-driven method and system for automatically optimizing game character movements, aiming to solve the technical problem that existing technologies typically control game character movements through predefined animation libraries or animation sequences, which cannot be adjusted in real time according to environmental changes, thus affecting the smoothness of the game and the player's immersion.

[0005] The first aspect disclosed in this application provides a data-driven method for automatically optimizing game character actions. The method includes: collecting scene data of a baseline action space volume based on the real-time map coordinates of the game character to obtain environmental geometric parameters and environmental physical properties; constructing an environmental constraint vector based on the environmental geometric parameters and environmental physical properties; receiving a real-time user operation flow and matching it with a standard action flow; performing constraint-adaptive deformation on the standard action flow based on the environmental constraint vector to output an optimized action flow; and using the optimized action flow to drive the action rendering of the game character to output a visualized character scene action.

[0006] The second aspect disclosed in this application provides a data-driven automatic optimization system for game character actions. This system is used in the aforementioned data-driven automatic optimization method for game character actions. The system includes: a data acquisition module for acquiring scene data of a baseline action space volume based on the real-time map coordinates of the game character, obtaining environmental geometric parameters and environmental physical properties; a constraint vector construction module for constructing environmental constraint vectors based on the environmental geometric parameters and environmental physical properties; an action flow matching module for receiving real-time user operation flow and matching it to a standard action flow; an adaptive deformation module for performing constraint adaptive deformation on the standard action flow based on the environmental constraint vectors, outputting an optimized action flow; and an action rendering module for using the optimized action flow to drive the action rendering of the game character, outputting visualized character scene actions.

[0007] One or more technical solutions provided in this application have at least the following beneficial effects: By collecting scene data based on the real-time coordinates of game characters, environmental geometric parameters and physical properties are obtained. This process allows the game character's behavior to reflect the state of its environment in real time, resulting in more optimized and adaptable character actions. By analyzing environmental geometric parameters and physical properties, environmental constraint vectors are constructed, effectively transforming environmental factors into constraints on character movement, thus ensuring that character actions do not collide or violate physical laws in complex environments. By receiving the user's real-time operation stream and matching it with the standard action stream in the game engine, user input is transformed into preset standard actions, ensuring that the player's input is correctly mapped to the character's actions, avoiding abnormal actions caused by input errors or delays. Based on the environmental constraint vectors, the standard action stream is constrained and adaptively deformed, outputting an optimized action stream, allowing the character's actions to be dynamically adjusted according to specific environmental conditions, optimizing character behavior. The optimized action stream drives the character's action rendering, outputting visualized character scene actions, enhancing the realism of the interaction between the character and the environment in the game, making the action performance more natural and realistic, and enhancing the player's immersion.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] Figure 1 A schematic diagram of the data-driven automatic optimization method for game character actions provided in this application embodiment.

[0010] Figure 2A schematic diagram of the structure of a data-driven automatic optimization system for game character actions provided in this application embodiment.

[0011] Figure labeling: Data acquisition module 10, constraint vector construction module 20, motion flow matching module 30, adaptive deformation module 40, motion rendering module 50. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] Example 1, as Figure 1 As shown in the embodiment of this application, a data-driven automatic optimization method for game character actions is provided, the method comprising: Based on the real-time map coordinates of the game character, scene data of the baseline action space volume is collected to obtain environmental geometric parameters and environmental physical properties.

[0014] In the virtual environment, whenever a game character moves on the map, the game engine continuously tracks and records the character's position, obtaining real-time map coordinates that reflect the character's precise location on the map. Based on these real-time map coordinates, surrounding scene data is collected, including the environment's geometry and physical properties. Environmental geometry parameters refer to the geometric structure of objects in the game scene, such as terrain undulations, walls, and obstacles. The game engine obtains this data through collision detection and physics simulation. Environmental physical properties refer to the physical characteristics of objects in the environment, such as ground friction, material elasticity, and rigidity. These properties affect the game character's movement, such as whether they can jump, slide, or climb.

[0015] An environmental constraint vector is constructed based on the environmental geometric parameters and environmental physical properties.

[0016] Environmental constraint vectors are mathematical representations constructed based on environmental geometric parameters and physical properties. They define the constraints and rules that a character must follow when moving within the environment, including radial distance vectors, vertical distance vectors, local ground property vectors, and local environmental property vectors. The combination of these vectors provides the foundation for subsequent motion optimization, ensuring that the character's actions do not violate environmental physical rules. For example, when a character moves within an environment, it needs to follow the shape and position of obstacles in the scene and avoid passing through objects.

[0017] Receive real-time user action streams and match them with standard action streams.

[0018] During gameplay, players control their characters using a keyboard, mouse, gamepad, or other controllers. These actions are transmitted to the game engine in real-time, generating a motion flow, or user input data stream. A standard motion flow refers to a predefined, standardized sequence of character actions in the game, such as walking, running, jumping, and attacking. By analyzing the player's input motion flow and matching it with standard motion templates from a predefined motion library, the matching templates are smoothly stitched together to generate the standard motion flow. This ensures that transitions between actions are seamless and without noticeable abruptness or discontinuity. For example, the transition animation from walking to running needs to be smooth to avoid a stiff appearance.

[0019] Based on the environmental constraint vector, the standard action flow is subjected to constraint adaptive deformation, and an optimized action flow is output.

[0020] Standard motion flow is merely an idealized sequence of actions, without considering the specific physical constraints and geometric limitations of the environment. Applying constraint-adaptive deformation to standard motion flow essentially involves adjusting the execution of actions using environmental constraint vectors to ensure the character's movements conform to the actual physical conditions of the environment. For example, when encountering obstacles, the character will reduce stride length or change jump height; if the character encounters uneven ground while running or jumping, they will automatically adjust their body posture to maintain balance. The resulting optimized motion flow, compared to standard motion flow, better reflects the actual physical limitations and geometric constraints of the current game environment.

[0021] The optimized motion flow is used to drive the motion rendering of the game character, and outputs visual character and scene actions.

[0022] Motion rendering refers to applying optimized motion flow to a character's model, presenting the character's dynamic movement. Based on this optimized motion flow, the game engine renders the character's specific behavior within the game scene using the character's skeletal structure or physical model. Through motion rendering, the character's movements are ultimately transformed into visual images displayed to the player. This rendering includes not only the character's movements themselves but also the character's interactions with the environment, such as stepping on different ground materials or reacting to collisions with obstacles; all of these affect the character's behavior. Ultimately, players will see a physically accurate, fluidly moving character behaving in the game scene, enhancing the user's visual experience.

[0023] Furthermore, it also includes: The game map is aggregated to obtain a typical game environment; during the process of the game character executing the full motion library driven by the typical game environment, motion capture tracking is performed to obtain full-scene motion capture data; a motion capture coordinate system is constructed with the character's pelvis as the origin; the full-scene motion capture data is superimposed in the motion capture coordinate system space to construct the reference motion space volume.

[0024] The game map contains different terrains, buildings, obstacles, platforms and other elements. During the environmental scene aggregation process, representative environmental features are selected, such as wide flat land, complex city streets, and rugged mountains. The result of aggregation is to integrate the physical attributes and geometric features of these different environments into a unified typical game environment for subsequent action analysis and optimization.

[0025] The game character's motion library contains various actions, such as walking, running, jumping, attacking, crawling, and swimming. These actions cover a wide range of behaviors the character will exhibit in the game. To ensure that the character can naturally perform these actions in the game, high-quality data collection and analysis of these actions are necessary. To ensure high-fidelity motion data, motion capture technology is used to track each of the character's movements. Motion capture records the character's body movement trajectory using dedicated cameras and sensors, converting it into digital data. During the character's execution of the entire motion library, motion capture equipment tracks the character's limb movements in real time and records this data. The resulting full-scene motion capture data provides data support for subsequent motion optimization and the construction of a baseline motion space.

[0026] The pelvis was chosen as the origin of the coordinate system because it is the most stable part of the human body, essentially bearing the character's center of gravity and being less affected by the movements of other parts. By establishing a motion capture coordinate system centered on the pelvis, covering the entire body, this system describes how the character moves their upper and lower limbs, originating from the pelvis. Each data point in the motion capture coordinate system represents the position and posture of the pelvis relative to the global coordinate system when the character performs a certain action. In this way, the movements of other limbs can also be deduced from the relative position of the pelvis.

[0027] During motion capture data acquisition, every action of the character is performed throughout the entire scene. Therefore, motion capture data involves various action information of the character at different times and in different environments. This data is acquired by motion capture equipment and includes every action and posture of the character in a typical game environment. This data is mapped to the previously established motion capture coordinate system, and the motion capture data from different actions are aligned in the same coordinate system by overlapping to ensure data consistency.

[0028] The baseline motion space volume is a spatial model composed of overlapping and aggregating motion capture datasets of various motion types. This space volume not only shows the changes in the character's position in different actions, but also takes into account the character's reaction in different environments. It represents all possible actions of the character in a typical game environment and their spatial constraints.

[0029] Furthermore, the full-scene motion capture data is superimposed in the motion capture coordinate system space to construct the reference motion space volume, including: The motion capture data of the entire scene is aggregated according to the action type to obtain multiple motion capture datasets of various action types; a multi-scale convex hull granularity is predefined according to the action amplitude attributes of the various action types; after aligning and overlapping the multiple motion capture datasets to the motion capture coordinate system, convex hull calculation is performed according to the multi-scale convex hull granularity to obtain multiple hierarchical convex hull meshes; the reference motion space volume is obtained by fusing the multiple hierarchical convex hull meshes through voxel overlay.

[0030] In the game, characters perform various types of actions, such as walking, running, jumping, attacking, and turning. Each type of action has different characteristics and behavioral patterns, and the amplitude, speed, and rhythm of the action also vary. The motion capture data of the entire scene is grouped according to different action types. Each action type (such as walking, running, jumping, etc.) forms an independent motion capture dataset. Each motion capture dataset contains the performance of the character when performing the action under different environmental conditions, serving as the basis for subsequently constructing the baseline motion space.

[0031] The range of motion attribute for each action type describes the range, speed, and amplitude of joint or bone movement when a character performs that action. For example, walking has a relatively small range of motion, a stable gait, and a small range of joint movement; running has a larger range of motion, a faster gait, and a wider range of leg and arm movement; and jumping is characterized by a larger vertical displacement.

[0032] A convex hull is a geometric concept that represents the smallest convex polyhedron containing a set of points. In other words, it is the smallest area that encloses all points and can be viewed as the outer shell surrounding the motion capture data set. To address the diversity of different actions, multiple convex hull granularities are predefined for each action type. These granularities define how the motion dataset is enclosed at different scales. For example, at a small granularity, only the spatial extent when the character performs fine, small movements is calculated, while at a large granularity, the spatial extent when the character performs large, sweeping movements is considered.

[0033] Multi-scale granularity can be dynamically adjusted according to the complexity and amplitude of the action. More complex actions use larger granularity to ensure that the range of action can be fully represented. By adjusting the granularity, the spatial range of different action types can be represented more accurately, avoiding loss of calculation accuracy when the action amplitude is small or large.

[0034] Multiple motion capture datasets are aligned and overlapped onto the motion capture coordinate system. The position of each motion capture dataset is corrected according to the position and posture of the pelvis to ensure that all data are processed under the same reference frame.

[0035] Based on the characteristics of each action type (such as the amplitude of movement) and a predefined granularity, convex hull calculation is performed on each motion capture dataset. This calculation generates the outer geometry that encloses the motion data, representing the spatial extent of the action. For each action type, a hierarchical convex hull is generated, with each level corresponding to a bounding volume of different granularities. Smaller granularity convex hulls describe small-range movements, while larger granularity convex hulls describe large-range movements. The hierarchical convex hull mesh combines multiple convex hulls of different scales to form a multi-level mesh structure. Each layer of the convex hull mesh represents the spatial distribution from smallest to largest when a character performs a certain type of action. These convex hull meshes provide accurate spatial data for subsequent motion optimization, capturing the actual behavioral performance of characters performing different actions in the environment.

[0036] A voxel, short for volume pixel, is similar to a pixel in a two-dimensional image. It represents a volume unit in three-dimensional space. In this step, voxels are used to convert the spatial data of multiple convex hull meshes into a unified three-dimensional spatial representation. After generating multiple hierarchical convex hull meshes, these meshes are stacked using voxels. This means that the convex hulls at different levels are combined to form a complete and continuous three-dimensional space. By converting multiple hierarchical convex hull meshes into voxel data, the spatial volume of character movements can be modeled in greater detail. This voxel data describes the positional changes of different parts of the character's body in three-dimensional space when performing specific actions. Finally, through voxel stacking, a baseline action space is obtained. This space represents a comprehensive spatial model of all action types, including the range of motion and constraints of the character when performing different actions. The baseline action space is the basis for subsequent optimization and action adaptation adjustments, providing an accurate spatial reference for the character's movements in different environments.

[0037] Furthermore, receiving real-time user action streams and matching them with standard action streams includes: Align the user's real-time operation flow with the game engine clock to obtain an operation instruction sequence; extract the instruction timestamp sequence from the operation instruction sequence; traverse the instruction timestamp sequence to extract accidental touch signals to obtain a continuous input sequence; encode the continuous input sequence to obtain an operation feature vector; use the operation feature vector to match and call a standard action template sequence in a pre-classified action template library, and then generate the standard action flow by smoothly splicing the standard action template sequence.

[0038] Users interact with the game through input devices such as keyboards, mice, and gamepads. These actions include moving, jumping, attacking, and using items. User input is collected by sensors and drivers, but it is usually asynchronous, meaning it is not synchronized with the game engine's main clock.

[0039] The game engine clock controls the real-time smoothness of the game. It serves as the time basis for various events, animations, and physics simulations within the game. To ensure that user actions are consistent with the game engine's timeline, the user's input flow needs to be aligned with the engine's clock. This synchronizes each user action with the corresponding game engine time, avoiding experience issues caused by input delays or asynchrony. The aligned input flow generates a sequence of operation instructions, each corresponding to a specific game moment, including each user input action and its corresponding time point.

[0040] Each operation instruction is timestamped when it is entered by the user, indicating the exact moment the operation occurred. By extracting these timestamps from the sequence of operation instructions, it is possible to clearly know when each user operation occurred. The instruction timestamp sequence is a list of timestamps arranged in chronological order, with each timestamp corresponding to an operation instruction.

[0041] In actual gameplay, player input is often imperfect and may result in accidental touches, such as incorrect button presses or excessive controller input. These accidental touch signals can affect the character's movement, causing unnatural or disjointed actions in the game. Accidental touch signals manifest as input intervals that are too short or too long, or the input direction not matching the character's intended action.

[0042] Traverse the command timestamp sequence, analyze the input signals, identify accidental touch signals that do not conform to normal operation logic, and after identifying accidental touch signals, remove these unnecessary or unnatural inputs to obtain a continuous input sequence. This sequence contains the player's valid and continuous input commands, which is the basic data for subsequent action flow matching and optimization.

[0043] The continuous input sequence is transformed into a digitized operation feature vector. This operation feature vector is a vector representation obtained by encoding the continuous input sequence; each input signal is quantized and converted into a numerical vector. Each operation feature vector contains multiple elements, including: direction information (e.g., forward, backward, left, right); action type (e.g., jump, attack, run); and time features (e.g., input interval, operation frequency). These elements collectively constitute the feature representation of the input, allowing the system to recognize the player's operational intentions.

[0044] The character's actions in the game are pre-designed and stored in a pre-categorized action template library. This library contains various types of action templates, such as running, jumping, attacking, crawling, etc. Each action template is a standardized sequence of actions that represents the character's actions under specific conditions.

[0045] Based on the obtained operation feature vector, the most matching action template is searched in the pre-classified action template library. This matching process is performed by calculating the similarity between the operation feature vector and various types of actions in the action template, using methods such as Euclidean distance and cosine similarity to measure the matching degree between the feature vector and the template. After finding the most matching action template, the corresponding standard action template sequence is called. These template sequences contain each stage of the corresponding action, such as the start, intermediate, and end stages of the action.

[0046] If the user's input involves multiple actions, such as switching from running to jumping, the sequence of multiple standard action templates is smoothly stitched together to ensure a seamless transition between actions without any breaks or abrupt changes. Smooth stitching is achieved by adjusting the timestamps of the action sequences or using interpolation methods, making the connection between multiple action templates more natural.

[0047] Ultimately, the smoothly spliced ​​sequence of motion templates forms a complete standard motion flow, which is used to drive the character's animation rendering in the game, making the character's actions consistent with the player's actions.

[0048] Furthermore, it also includes: The theoretical input timing and theoretical hardware operation parameter sequence of the continuous input sequence are obtained interactively; the operation time error is calculated by comparing the theoretical input timing and the actual input timing of the continuous input sequence; the operation space error is calculated by comparing the theoretical hardware operation parameter sequence and the actual hardware operation parameter sequence of the continuous input sequence; based on the operation time error and the operation space error, the operation tolerance of the continuous input sequence is corrected to obtain a corrected input sequence; the standard action template sequence is time-smoothly spliced ​​based on the corrected input sequence to generate the standard action flow.

[0049] Theoretical input timing refers to the ideal situation where a user should input according to a predetermined time interval or rhythm when performing an action. For example, in a game, after a player presses a button, it is expected that an action will be triggered within a certain time, such as 0.5 seconds after pressing the button.

[0050] Hardware operation parameter sequences describe the response characteristics of user hardware devices, including but not limited to the strength and duration of input signals, key release delays, etc., such as key press duration, mouse movement speed, and gamepad vibration intensity. Theoretical hardware operation parameter sequences are preset parameter sequences based on the compliant behavior of user operations, representing the response that the hardware device should provide under normal circumstances. These theoretical parameter sequences are predicted based on the design of the game engine and the standard response of the hardware device.

[0051] In practice, user input is often affected by factors such as latency, fatigue, and device responsiveness. Actual input timing refers to the time sequence of when a user presses or releases a button during actual gameplay, which may differ from the expected theoretical input timing. Operational timing error is the difference between the theoretical and actual input timing. This difference can be calculated and quantified as an error value by comparing the timestamps of each operation command. Comparison methods can include using the difference between the two timestamps or measuring the accuracy of the input by calculating the standard deviation of the error. This timing error can lead to choppy character movements in the game or actions that do not match the player's intentions. For example, a player might want to perform an action quickly, but due to input latency, the action execution time may deviate, thus affecting the gaming experience.

[0052] Similar to input timing, hardware operating parameters can also vary. The actual hardware operating parameter sequence refers to the response parameters when a user actually uses the hardware device, while the theoretical hardware operating parameter sequence is a preset parameter sequence based on the ideal response the device should have. Operational space error refers to the difference between actual and theoretical hardware operating parameters. These differences manifest in: button pressure (e.g., the player pressing a button too lightly or too heavily, leading to inaccurate input response in the game); mouse precision (the player's mouse movement speed and precision do not meet expectations); and joystick response (the game controller's joystick does not provide correct spatial feedback at its intended position). By comparing the actual hardware response with the theoretical response, the operational space error can be calculated, and Euclidean distance or other spatial metrics can be used for comparison.

[0053] Operation tolerance correction is the process of adjusting the user input sequence to correct inaccurate inputs caused by time and spatial errors back to the expected standard input. For time errors, the input order can be adjusted by offsetting or interpolating timestamps to make it closer to the theoretical timing. For example, if the user presses a key too early or too late, it can be adjusted to the correct time. For spatial errors, hardware input parameters can be normalized or interpolated. For example, if the joystick input value exceeds the expected range, it can be stretched or compressed according to theoretical values ​​to make the actual input consistent with the expected input. After tolerance correction, the resulting corrected input sequence will be closer to the theoretical input sequence and accurately reflect the player's intention. This corrected input sequence will provide more accurate operation data for subsequent motion flow generation and rendering.

[0054] Applying a modified input sequence to a standard action template sequence, temporal smoothing splicing refers to smoothly transitioning multiple standard action templates according to the modified input sequence. This avoids jumps or unnatural transitions between actions. For example, if a player's input consists of multiple consecutive actions such as walking, jumping, and attacking, temporal smoothing splicing ensures a smoother transition between these actions in time, avoiding disjointed pauses or abrupt transitions. The temporally smoothed standard action template sequence forms a complete standard action flow. This action flow describes the precise actions the character should perform based on the player's input, and due to the modified input sequence, the character's actions become more aligned with the player's intentions.

[0055] Furthermore, constructing an environmental constraint vector based on the environmental geometric parameters and environmental physical properties includes: Using the real-time map coordinates as the action origin, obstacle distances are extracted by traversing the environmental geometric parameters to generate a radial distance vector; a vertical distance vector is extracted from the environmental geometric parameters based on the real-time map coordinates; the environmental physical properties are horizontally decomposed to obtain a local ground property vector and a local environmental property vector; the radial distance vector, vertical distance vector, local ground property vector, and local environmental property vector are encapsulated to obtain the environmental constraint vector.

[0056] Real-time map coordinates refer to the current coordinates of a game character in a virtual environment. They represent the character's position on the map. The real-time map coordinates serve as the origin of the action. In other words, the character's position is used as a reference point for calculating geometric information such as obstacles and distances in the environment.

[0057] Environmental geometry parameters include the position, shape, and size of obstacles in the game world. By traversing the environmental geometry parameters, information about all obstacles in the environment is collected. These obstacles include walls, rocks, buildings, and any other objects that may affect the character's actions.

[0058] Based on real-time map coordinates, the distance between the character and surrounding obstacles is calculated using methods such as Euclidean distance or Manhattan distance, and represented as a vector called the radial distance vector. This vector represents the relative position and distance from the character's position to various obstacles in the environment.

[0059] Vertical distance refers to the distance between a character and objects in the environment, such as the ground, platforms, and rooftops, in the vertical direction. The vertical direction is related to the direction of gravity and represents the character's proximity to other objects in the vertical direction. Based on the character's real-time map coordinates, the system iterates through environmental geometry parameters to extract the vertical distance from the character to objects in the environment. This vertical distance reflects the relationship between the character and various objects in the environment in the vertical direction. Based on the extracted vertical distances, a vertical distance vector is constructed to describe the distance between the character and different objects in the vertical direction, used to determine whether actions such as jumping or crouching are necessary.

[0060] Environmental physical properties include ground material, friction, slope, and gravity. These properties directly affect a character's movement behavior. For example, running on different surfaces will result in different speeds, turning radii, and jump heights. Horizontal hierarchical decomposition refers to processing environmental physical properties by breaking them down into different levels. Specifically, it involves classifying objects based on their different physical characteristics in the environment, decomposing them into local ground property vectors and local environmental property vectors.

[0061] The local ground property vector describes the physical characteristics of the ground in the area where the character is located. It mainly includes: ground friction, which affects the character's movement speed and turning stability on the ground; slope, which describes the inclination of the ground and affects the character's movement trajectory and pace; and ground material, such as sand, grass, ice, etc., which determines the character's movement performance.

[0062] The local environment property vector describes the physical properties of other objects or obstacles in the environment surrounding the character. For example: the hardness of objects, such as stone walls or steel, will affect the character's actions if the character comes into contact with them, such as obstruction or bouncing; the influence of gravity, which will affect the character's jumping and falling speed; and fluid properties, such as the viscosity and buoyancy of water, which will also affect the character's actions if the character enters water.

[0063] The radial distance vector, vertical distance vector, local ground property vector, and local environment property vector are encapsulated to represent various constraints on character actions in the game world. The resulting environmental constraint vector contains all the physical and geometric constraints that the character faces in the current environment, which can effectively describe the character's movement restrictions in the environment. Based on these constraints, the character's action execution can be adjusted, such as avoiding obstacles, adjusting jump height, and adapting to ground slope.

[0064] Furthermore, based on the environmental constraint vector, constraint-adaptive deformation is performed on the standard action flow to output an optimized action flow, including: The standard action flow is pre-performed to locate multiple key spatiotemporal nodes; using the environmental constraint vector, constraint-adaptive deformation correction is performed at the multiple key spatiotemporal nodes to obtain multiple deformation-optimized action segments; the multiple deformation-optimized action segments are overridden and inserted into the standard action flow to output the optimized action flow.

[0065] Trajectory pre-playing refers to simulating and calculating standard motion flows to predict their performance in a real-world environment, such as the character's displacement path and speed changes. Key spatiotemporal nodes refer to the character's state at crucial moments or locations within their motion flow during trajectory pre-playing. These nodes are turning points or important moments in the action, such as the start of a jump, the end of an attack, or a change of direction. These nodes determine the dividing or changing points of the character's actions and are crucial for subsequent optimization and deformation.

[0066] Constraint-adaptive deformation correction refers to adjusting key spatiotemporal nodes in the standard motion flow based on the physical and geometric information provided in the environmental constraint vector. By analyzing the environmental characteristics of each key spatiotemporal node, such as the presence of obstacles, ground slope, and objects around the character, adaptive adjustments are made to the motion to generate deformation-optimized motion fragments. These optimized fragments are motion parts modified according to the actual conditions in the environment. For example, if the character needs to jump and there are obstacles around, the jump height or angle is adjusted to prevent the character from colliding with the obstacles. In this way, each motion fragment becomes more in line with the actual needs of the current environment, making the character's motion flow more reasonable and fluid.

[0067] Multiple deformable, optimized motion fragments are sequentially inserted into the original standard motion flow. These optimized fragments cover parts of the standard motion flow that conflict with or are incompatible with the environment. This means that some actions in the original standard motion flow are replaced with actions adjusted according to environmental characteristics. This overlay insertion makes the character's movements more adaptable to changes in the game world without altering the overall structure of the motion flow. In this way, it ensures that the character's movements both conform to user input and effectively cope with complex environmental changes, such as obstacles and ground slopes. The motion flow after inserting the deformable, optimized fragments is called the optimized motion flow. The optimized motion flow is more precise and fluid than the standard motion flow, and can better drive the character's behavior in complex environments.

[0068] Furthermore, using the aforementioned environmental constraint vector, constraint-adaptive deformation correction is performed at multiple key spatiotemporal nodes to obtain multiple deformation optimization action segments, including: Based on the radial distance vector, a horizontal amplitude convergence analysis is performed to generate a horizontal convergence coefficient; based on the vertical distance vector, a vertical attitude adaptation analysis is performed to generate a bending factor; based on the local ground property vector, a motion inertia correction analysis is performed to output an inertia correction amount; based on the local environmental property vector, a role confidence offset analysis is performed to generate an equivalent offset force; using the horizontal convergence coefficient, bending factor, inertia correction amount, and equivalent offset force, a constraint-adaptive deformation joint correction is performed at multiple key spatiotemporal nodes to obtain the multiple deformation-optimized motion segments.

[0069] The radial distance vector provides information about the relative distance between the character's current position and surrounding obstacles. Horizontal amplitude convergence analysis refers to analyzing the relative position between the character and obstacles when the character performs horizontal movements (such as walking or running) and determining whether adjustments to the character's movement amplitude are necessary. This is because if the character approaches an obstacle during movement, a collision will occur, requiring adjustments to the movement amplitude, such as shortening strides or changing direction. Based on the relative distance between the character and obstacles, a horizontal convergence coefficient is calculated. This coefficient describes the degree to which the character's movement amplitude needs to be reduced or converged when approaching an obstacle; a larger value indicates a stronger convergence effect, meaning a smaller movement amplitude. The horizontal convergence coefficient can be calculated through mathematical operations on the radial distance vector, for example, using linear or nonlinear functions based on distance.

[0070] The vertical distance vector describes the distance between a character and vertical objects in the environment. Vertical pose adaptation analysis focuses on changes in the character's posture, especially how to adjust their vertical posture when the character needs to face the ground, jump, or crouch. For example, when moving in a low-ceilinged space, the character needs to adjust their body height and lower their posture to avoid collisions. The crouching factor is a parameter calculated based on the vertical distance vector. It represents the degree to which the character needs to change their posture under different vertical spatial constraints. For example, the crouching factor increases when the character approaches a low-ceilinged space or needs to crouch, indicating that the character needs to lower their body height more. The crouching factor can be quantified by the change in distance, generated using the square of the distance or linear scaling.

[0071] The local ground property vector includes the physical characteristics of the ground in the character's area, such as ground friction, slope, and material. Motion inertia correction analysis adjusts the character's inertia based on the interaction between the character and the ground's physical properties. For example, when a character runs on a slippery surface (such as ice), their inertia increases, causing them to slide too far; while on a rough surface, inertia is suppressed by friction, slowing the character down. Based on factors such as the friction coefficient and slope in the local ground property vector, an inertia correction is calculated, reflecting changes in the character's inertia, and used to adjust the character's speed, acceleration, and trajectory. For example, the lower the ground friction coefficient, the larger the inertia correction, indicating that the character's movement is more affected by inertia, requiring adjustments to the character's speed and turning radius.

[0072] The local environment property vector includes the physical properties of other objects in the environment, such as hardness, elasticity, and hydrodynamic characteristics. Confidence offset analysis aims to evaluate a character's response when interacting with environmental objects. For example, when a character pushes open a door, the door's hardness and elasticity will affect the force the character applies to the door and the direction of the push, indicating the interaction force between the character and the object. Based on the physical properties in the local environment property vector, an equivalent offset force is calculated, representing the force and offset of the character when interacting with environmental objects. This offset force takes into account the physical properties of the environmental objects and their impact on the character, such as the force the character uses when pushing an obstacle.

[0073] At multiple calculated key spatiotemporal nodes, deformation corrections are performed based on the horizontal convergence coefficient, body flexion factor, inertia correction, and equivalent offset force. Specifically, the horizontal motion amplitude of the character is adjusted based on the horizontal convergence coefficient to ensure collisions with obstacles are avoided; the vertical posture of the character is adjusted based on the body flexion factor to ensure collisions with low-lying objects are avoided; the velocity and acceleration of the character are adjusted based on the inertia correction to ensure reasonable and smooth movement under different ground conditions; and the character's motion response is adjusted based on the equivalent offset force according to the interaction between the character and environmental objects to ensure the realism of physical interaction. After the above corrections, a deformation-optimized motion segment is generated for each key spatiotemporal node. These segments are optimized motion segments under physical constraints.

[0074] Furthermore, during the process of using the optimized motion flow to drive the motion rendering of the game character, scene physical interaction feedback rendering is executed simultaneously to output visual scene interaction actions.

[0075] Scene physics-based interactive feedback rendering refers to simultaneously rendering the physical interactions between the character and its surrounding environment while rendering the character's actions. Examples include the contact between the character and the ground when jumping, the character's reaction to collisions with obstacles, and the friction effects when the character walks on different surfaces. These interactive effects are calculated in real-time based on the character's actions and the physical properties of the environment, and then rendered onto the game screen. This process involves calculations by the physics engine to ensure that the interaction between the character and the environment appears realistic. Ultimately, a visual scene interactive animation is generated and output—an animation effect of the character's interaction with the environment. This effect demonstrates how the character moves naturally and smoothly in the game based on optimized motion flow, while the physical feedback in the environment is rendered synchronously, providing a more immersive experience.

[0076] Example 2, based on the same inventive concept as the data-driven automatic optimization method for game character actions in the foregoing examples, such as... Figure 2 As shown in the figure, this application embodiment provides a data-driven automatic optimization system for game character movements, the system comprising: The data acquisition module 10 is used to acquire scene data of the baseline action space volume based on the real-time running coordinates of the game character, and obtain environmental geometric parameters and environmental physical properties; the constraint vector construction module 20 is used to construct environmental constraint vectors based on the environmental geometric parameters and environmental physical properties; the action flow matching module 30 is used to receive the user's real-time operation flow and perform standard action flow matching; the adaptive deformation module 40 is used to perform constraint adaptive deformation on the standard action flow based on the environmental constraint vectors and output an optimized action flow; the action rendering module 50 is used to drive the action rendering of the game character using the optimized action flow and output visualized character scene actions.

[0077] Furthermore, the data acquisition module 10 is used to perform the following operation steps: The game map is aggregated to obtain a typical game environment; during the process of the game character executing the full motion library driven by the typical game environment, motion capture tracking is performed to obtain full-scene motion capture data; a motion capture coordinate system is constructed with the character's pelvis as the origin; the full-scene motion capture data is superimposed in the motion capture coordinate system space to construct the reference motion space volume.

[0078] Furthermore, the data acquisition module 10 is used to perform the following operation steps: The motion capture data of the entire scene is aggregated according to the action type to obtain multiple motion capture datasets of various action types; a multi-scale convex hull granularity is predefined according to the action amplitude attributes of the various action types; after aligning and overlapping the multiple motion capture datasets to the motion capture coordinate system, convex hull calculation is performed according to the multi-scale convex hull granularity to obtain multiple hierarchical convex hull meshes; the reference motion space volume is obtained by fusing the multiple hierarchical convex hull meshes through voxel overlay.

[0079] Furthermore, the action flow matching module 30 is used to perform the following operation steps: Align the user's real-time operation flow with the game engine clock to obtain an operation instruction sequence; extract the instruction timestamp sequence from the operation instruction sequence; traverse the instruction timestamp sequence to extract accidental touch signals to obtain a continuous input sequence; encode the continuous input sequence to obtain an operation feature vector; use the operation feature vector to match and call a standard action template sequence in a pre-classified action template library, and then generate the standard action flow by smoothly splicing the standard action template sequence.

[0080] Furthermore, the action flow matching module 30 is used to perform the following operation steps: The theoretical input timing and theoretical hardware operation parameter sequence of the continuous input sequence are obtained interactively; the operation time error is calculated by comparing the theoretical input timing and the actual input timing of the continuous input sequence; the operation space error is calculated by comparing the theoretical hardware operation parameter sequence and the actual hardware operation parameter sequence of the continuous input sequence; based on the operation time error and the operation space error, the operation tolerance of the continuous input sequence is corrected to obtain a corrected input sequence; the standard action template sequence is time-smoothly spliced ​​based on the corrected input sequence to generate the standard action flow.

[0081] Furthermore, the constraint vector construction module 20 is used to perform the following operation steps: Using the real-time map coordinates as the action origin, obstacle distances are extracted by traversing the environmental geometric parameters to generate a radial distance vector; a vertical distance vector is extracted from the environmental geometric parameters based on the real-time map coordinates; the environmental physical properties are horizontally decomposed to obtain a local ground property vector and a local environmental property vector; the radial distance vector, vertical distance vector, local ground property vector, and local environmental property vector are encapsulated to obtain the environmental constraint vector.

[0082] Furthermore, the adaptive deformation module 40 is used to perform the following operation steps: The standard action flow is pre-performed to locate multiple key spatiotemporal nodes; using the environmental constraint vector, constraint-adaptive deformation correction is performed at the multiple key spatiotemporal nodes to obtain multiple deformation-optimized action segments; the multiple deformation-optimized action segments are overridden and inserted into the standard action flow to output the optimized action flow.

[0083] Furthermore, the adaptive deformation module 40 is used to perform the following operation steps: Based on the radial distance vector, a horizontal amplitude convergence analysis is performed to generate a horizontal convergence coefficient; based on the vertical distance vector, a vertical attitude adaptation analysis is performed to generate a bending factor; based on the local ground property vector, a motion inertia correction analysis is performed to output an inertia correction amount; based on the local environmental property vector, a role confidence offset analysis is performed to generate an equivalent offset force; using the horizontal convergence coefficient, bending factor, inertia correction amount, and equivalent offset force, a constraint-adaptive deformation joint correction is performed at multiple key spatiotemporal nodes to obtain the multiple deformation-optimized motion segments.

[0084] Furthermore, during the process of using the optimized motion flow to drive the motion rendering of the game character, scene physical interaction feedback rendering is executed simultaneously to output visual scene interaction actions.

[0085] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A data-driven method for automatically optimizing game character movements, characterized in that, The method includes: Based on the real-time map coordinates of the game character, scene data of the baseline action space volume is collected to obtain environmental geometric parameters and environmental physical properties; An environmental constraint vector is constructed based on the environmental geometric parameters and environmental physical properties; Receive real-time user action streams and match them with standard action streams; Based on the environmental constraint vector, the standard motion flow is subjected to constraint-adaptive deformation, and an optimized motion flow is output. The optimized motion flow is used to drive the motion rendering of the game character, and outputs visual character and scene actions.

2. The data-driven automatic optimization method for game character actions as described in claim 1, characterized in that, Also includes: The game map is aggregated to obtain typical game environments; During the process of driving the game character to execute the full motion library in the typical game environment, motion capture tracking is performed to obtain full-scene motion capture data; A motion capture coordinate system is constructed with the character's pelvis as the origin; The motion capture data of the entire scene is superimposed in the motion capture coordinate system space to construct the reference motion space volume.

3. The data-driven automatic optimization method for game character actions as described in claim 2, characterized in that, The motion capture data of the entire scene is superimposed in the motion capture coordinate system space to construct the reference motion space volume, including: The motion capture data of the entire scene is aggregated according to the action type to obtain multiple motion capture datasets for various action types; Multi-scale convex hull granularity is predefined based on the motion amplitude attributes of the various motion types; After aligning and overlapping the multiple motion capture datasets onto the motion capture coordinate system, the motion class convex hull is calculated based on the multi-scale convex hull granularity to obtain multiple hierarchical convex hull meshes. The reference motion space volume is obtained by fusing the multiple hierarchical convex hull meshes through voxel overlay.

4. The data-driven automatic optimization method for game character actions as described in claim 1, characterized in that, Receive real-time user action streams and match them with standard action streams, including: Align the user's real-time action stream with the game engine clock to obtain the action command sequence; Extract the instruction timestamp sequence from the operation instruction sequence; By traversing the instruction timestamp sequence to extract accidental touch signals, a continuous input sequence is obtained; The continuous input sequence is encoded to obtain an operational feature vector; After matching and calling the standard action template sequence in the pre-classified action template library using the operation feature vector, the standard action stream is generated by smoothly splicing the standard action template sequence.

5. The data-driven automatic optimization method for game character actions as described in claim 4, characterized in that, Also includes: The theoretical input timing and theoretical hardware operation parameter sequence of the continuous input sequence are obtained interactively. The operation time error is calculated by comparing the theoretical input timing with the actual input timing of the continuous input sequence. The operational space error is calculated by comparing the theoretical hardware operation parameter sequence with the actual hardware operation parameter sequence of the continuous input sequence. Based on the operation time error and operation space error, the continuous input sequence is corrected for operation tolerance to obtain a corrected input sequence; The standard action template sequence is spliced ​​sequentially based on the corrected input sequence to generate the standard action stream.

6. The data-driven automatic optimization method for game character actions as described in claim 1, characterized in that, An environmental constraint vector is constructed based on the environmental geometric parameters and environmental physical properties, including: Using the real-time map coordinates as the action origin, the obstacle distance is extracted by traversing the environmental geometric parameters to generate a radial distance vector; The vertical distance vector is extracted from the environmental geometric parameters based on the real-time map coordinates. The environmental physical properties are decomposed horizontally to obtain local ground property vectors and local environmental property vectors; The environmental constraint vector is obtained by encapsulating the radial distance vector, vertical distance vector, local ground property vector, and local environmental property vector.

7. The data-driven automatic optimization method for game character actions as described in claim 6, characterized in that, Based on the environmental constraint vector, the standard action flow undergoes constraint-adaptive deformation to output an optimized action flow, including: The execution trajectory of the standard action flow is pre-rehearsed to locate multiple key spatiotemporal nodes; Using the aforementioned environmental constraint vector, constraint-adaptive deformation correction is performed at multiple key spatiotemporal nodes to obtain multiple deformation optimization action segments; The multiple deformation-optimized motion fragments are overridden and inserted into the standard motion stream, and the optimized motion stream is output.

8. The data-driven automatic optimization method for game character actions as described in claim 7, characterized in that, Using the aforementioned environmental constraint vector, constraint-adaptive deformation correction is performed at multiple key spatiotemporal nodes to obtain multiple deformation optimization action segments, including: Based on the radial distance vector, a horizontal amplitude convergence analysis of the action is performed to generate horizontal convergence coefficients; Based on the vertical distance vector, a vertical posture adaptation analysis is performed to generate a bending factor. Based on the local ground property vector, an inertial correction analysis is performed, and the inertial correction amount is output. Based on the local environmental property vector, a role confidence offset analysis is performed to generate an equivalent offset force; Using the horizontal convergence coefficient, body bending factor, inertia correction amount and equivalent offset force, constraint-adaptive joint deformation correction is performed at multiple key spatiotemporal nodes to obtain the multiple deformation optimization action segments.

9. The data-driven automatic optimization method for game character actions as described in claim 1, characterized in that, During the process of driving the game character's motion rendering using the optimized motion flow, scene physical interaction feedback rendering is executed synchronously to output visual scene interaction actions.

10. A data-driven automatic optimization system for game character movements, characterized in that: The system for implementing the data-driven automatic optimization method for game character actions according to any one of claims 1-9, the system comprising: The data acquisition module is used to collect scene data of the baseline action space volume based on the real-time running coordinates of the game character, and obtain environmental geometric parameters and environmental physical properties. The constraint vector construction module is used to construct environmental constraint vectors based on the environmental geometric parameters and environmental physical properties. The action flow matching module is used to receive real-time user action flows and perform standard action flow matching. An adaptive deformation module is used to perform constraint adaptive deformation on the standard motion flow based on the environmental constraint vector and output an optimized motion flow. The motion rendering module is used to drive the motion rendering of the game character using the optimized motion flow, and output the visual character scene motion.

Images