Physics-based character simulation method and apparatus

Abstract

A physics-based character simulation apparatus according to an embodiment of the disclosed invention may comprise: a motion embedding model that embeds a motion sequence in a latent space and includes a low-level policy model for reproducing motion data of a character in a physical environment and a high-level policy model for setting a motion goal of the character; a first skill learning module that learns a dribble policy of the character; a second skill learning module that learns a trap policy of the character; a third skill learning module that learns a move policy of the character; and a fourth skill learning module that learns a kick policy of the character.

Description

Method and device for simulating physically based characters

[0001] The present invention relates to a method and device for simulating a physically based character, and more particularly, to a technology capable of implementing a physically natural transition based on motion embedding in character simulation.

[0002] With the advancement of computer graphics and the gaming industry in recent years, the demand for character control technology in virtual environments is increasing.

[0003] Especially in sports games and simulations, character movements must be implemented similarly to actual human movements, and various animation and physics-based simulation technologies are utilized for this purpose.

[0004] The existing keyframe-based animation method has a structure that simply plays back a set motion, and has the limitation of being difficult to flexibly respond to user input.

[0005] To overcome these limitations, technologies that control character movements using physics-based simulation have recently been gaining attention.

[0006] Physics-based control generates motions that reflect actual physical characteristics such as the character's joints, mass, friction, and ground reaction force, enabling more natural and realistic animation.

[0007] However, these physics-based methods are difficult to control in real time, and require a lot of computational resources and complex tuning to design them to accurately perform specific actions (e.g., kicking, dribbling, etc.).

[0008] Real sports movements involve a variety of skills, such as dribbling, trapping, passing, and shooting, each of which requires specialized movement patterns and control methods.

[0009] Therefore, there are technical limitations in implementing all skills into a single integrated control model, and to address this, the need for a hierarchical structure that designs specialized control models for each skill and links them with the integrated control model of the upper layer is emerging.

[0010] A method and device for simulating a physics-based character according to one embodiment of the disclosed invention is an invention created to solve the problems of the above-described prior art, and more specifically, a method and device for simulating a physics-based character can be provided that can implement a physically natural and visually smooth transition even in the transition section between individual movements of a character by combining physics-based control, motion embedding, and a skill-specific model.

[0011] A simulation device for a physically based character according to one embodiment of the disclosed invention may include a motion embedding model including a low-level policy model for embedding a motion sequence in a latent space and reproducing motion data of a character in a physical environment and a high-level policy model for setting a motion goal of the character, a first skill learning module for learning a dribble policy of the character, a second skill learning module for learning a trap policy of the character, a third skill learning module for learning a move policy of the character, and a fourth skill learning module for learning a kick policy of the character.

[0012] The above-described physics-based character simulation device further includes a processor including the motion embedding model, the first skill learning module, the second skill learning module, the third skill learning module, and the fourth skill learning module, wherein the processor can sample the states of the character and the ball to construct a buffer of the dribble policy, a buffer of the trap policy, and a buffer of the movement policy.

[0013] The above first skill learning module can receive a target dribbling speed as input information.

[0014] The first skill learning module may perform a reward based on a first element that induces the ball to move at the target dribbling speed, a second element that causes the distance between the character and the ball to become closer, and a third element that induces the character to move at the target dribbling speed.

[0015] The above first skill learning module can initialize the character based on the buffer of the movement policy and the buffer of the trap policy.

[0016] The above second skill learning module can receive information as input information regarding which part of the character the ball will be received.

[0017] The second skill learning module may perform compensation so that the distance between the ball and the part where the character receives the ball becomes closer before the character and the ball collide, and so that the relative speed of the ball and the character becomes smaller after the character and the ball collide.

[0018] The second skill learning module can initialize the character based on the buffer of the movement policy.

[0019] The above third skill learning module can receive the target movement speed and target gaze direction of the character as input information.

[0020] The third skill learning module can store latent variables for specific motions of the character in advance in a buffer, calculate the similarity between the stored latent variables and latent variables generated during learning, and reflect the result in a reward.

[0021] The above fourth skill learning module can receive a target kick speed of the ball as input information and perform compensation so that the speed of the ball becomes closer to the target kick speed.

[0022] A method for simulating a physics-based character according to one embodiment of the disclosed invention may include a step of learning a plurality of skill policies based on state initialization, a step of inputting latent variables output from the plurality of skill policies into a low-level policy of a motion embedding model, and a step of performing a transition between each skill when a predefined transition condition is achieved.

[0023] The step of learning multiple skill policies based on the above state initialization may include a step of learning a dribble policy of the character, a step of learning a trap policy of the character, and a step of learning a kick policy of the character.

[0024] The step of learning multiple skill policies based on the above state initialization can be used to sample the states of the character and the ball to construct a buffer of the dribble policy and a buffer of the trap policy, which can then be used for compensation.

[0025] A simulation device for a physically based character according to one embodiment of the disclosed invention includes a motion embedding model that embeds a motion sequence in a latent space and includes a low-level policy model that reproduces motion data of a character in a physical environment and a high-level policy model that sets a motion goal of the character, and a plurality of skill learning modules that learn skill policies of the character, wherein the plurality of skill learning modules may include a first skill learning module that learns a dribble policy of the character based on state initialization, a second skill learning module that learns a trap policy of the character, a fourth skill learning module that learns a kick policy of the character, and a third skill learning module that learns a move policy of the character based on a similarity between a latent variable stored in advance for a specific motion and a latent variable generated during learning.

[0026] A method and device for simulating a physically based character according to one embodiment of the disclosed invention have the advantage of being able to implement physically natural and visually smooth transitions even in transition sections between individual movements of the character.

[0027] In addition, the method and device for simulating a physics-based character according to one embodiment of the disclosed invention have the advantage of being able to hierarchically configure and integrate a physics-based motion embedding model and a model specialized for each skill.

[0028] FIG. 1 is a schematic diagram of a simulation device for a physics-based character according to one embodiment of the disclosed invention.

[0029] FIG. 2 is a flowchart illustrating a method for simulating a physically based character according to one embodiment of the disclosed invention.

[0030] FIG. 3 is a diagram illustrating elements used for compensation of a dribbling policy in a method and device for simulating a physics-based character according to one embodiment of the disclosed invention.

[0031] FIG. 4 is a diagram illustrating a process of learning a dribbling policy in a method and device for simulating a physics-based character according to one embodiment of the disclosed invention.

[0032] FIG. 5 is a diagram illustrating elements used for compensation of a trap policy in a method and device for simulating a physics-based character according to one embodiment of the disclosed invention, comparing them before and after a collision between a ball and a character.

[0033] FIG. 6 is a diagram showing the interrelationship between multiple models specialized for multiple skill policies and low-level policies in a method and device for simulating a physically based character according to one embodiment of the disclosed invention.

[0034] FIG. 7 is a diagram showing conditions under which transitions between multiple skills occur in a method and device for simulating a physically based character according to one embodiment of the disclosed invention.

[0035] The embodiments described in this specification and the configurations illustrated in the drawings are merely preferred examples of the disclosed invention, and there may be various modified examples that can replace the embodiments and drawings of this specification at the time of filing of this application.

[0036] Additionally, the same reference numbers or symbols presented in each drawing of this specification represent parts or components that perform substantially the same function.

[0037] Additionally, the terminology used herein is for the purpose of describing embodiments and is not intended to limit and / or restrict the disclosed invention. Singular expressions include plural expressions unless the context clearly dictates otherwise.

[0038] In this specification, terms such as “include” or “have” are intended to specify the presence of a feature, number, step, operation, component, part or combination thereof described in the specification, but do not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

[0039] Additionally, terms including ordinal numbers such as “first,” “second,” etc., used herein may be used to describe various components, but the components are not limited by the terms, and the terms are used only for the purpose of distinguishing one component from another.

[0040] For example, without departing from the scope of the present invention, a first component could be referred to as a second component, and similarly, a second component could also be referred to as a first component. The term "and / or" includes any combination of a plurality of related listed items or any one of a plurality of related listed items.

[0041] Hereinafter, embodiments according to the present invention will be described in detail with reference to the attached drawings.

[0042]

[0043] FIG. 1 is a schematic diagram illustrating a simulation device for a physically based character according to one embodiment of the disclosed invention. FIG. 2 is a flowchart illustrating a simulation method for a physically based character according to one embodiment of the disclosed invention. FIG. 3 is a diagram illustrating elements used for compensating for a dribble policy in a simulation method and device for a physically based character according to one embodiment of the disclosed invention. FIG. 4 is a diagram illustrating a process for learning a dribble policy in a simulation method and device for a physically based character according to one embodiment of the disclosed invention. FIG. 5 is a diagram illustrating elements used for compensating for a trap policy before and after a collision between a ball and a character in a simulation method and device for a physically based character according to one embodiment of the disclosed invention. FIG. 6 is a diagram illustrating the interrelationship between several models specialized for a plurality of skill policies and low-level policies in a simulation method and device for a physically based character according to one embodiment of the disclosed invention.

[0044] Referring to FIG. 1, a simulation device (100) of a physics-based character according to one embodiment of the disclosed invention may include a processor (110) and a memory (120).

[0045] Specifically, the components of the processor (110) of the simulation device (100) of a physically based character according to one embodiment of the disclosed invention may be representations of different functions performed by the processor (110) according to control commands provided by the program code stored in the simulation device (100) of a physically based character according to one embodiment of the disclosed invention.

[0046] Additionally, the processor (110) and components of the processor (110) can control the simulation device (100) of a physical-based character to perform steps included in a method for simulating a physical-based character according to an embodiment of the disclosed invention, which will be described later.

[0047] At this time, the processor (110) and components of the processor (110) may be implemented to execute commands according to the code of the operating system included in the memo and the code of at least one program.

[0048] In addition, the processor (110) of the simulation device (100) of a physics-based character according to one embodiment of the disclosed invention may include a first skill learning module (111), a second skill learning module (112), a third skill learning module (113), a fourth skill learning module (114), and a motion embedding model (115).

[0049] Specifically, the motion embedding model (115) of the simulation device (100) of a physically based character according to one embodiment of the disclosed invention may include a low-level policy model that embeds a motion sequence in a latent space and reproduces motion data of the character in a physical environment, and a high-level policy model that sets a motion goal of the character.

[0050] More specifically, a high-level policy of a simulation device (100) of a physics-based character according to one embodiment of the disclosed invention determines the movement purpose or behavioral intention of the character, and means a control policy that influences the operation method or operation type selection of a low-level policy.

[0051] The low-level policy model and high-level policy model of this motion embedding model (115) form a hierarchical control structure, where the high-level policy is responsible for strategic judgment and setting behavioral intentions, and the low-level policy plays a role in controlling detailed actions corresponding to the intentions in a physics-based simulation environment.

[0052] Additionally, the first skill learning module (111) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can learn the dribble policy of the character.

[0053] Specifically, the first skill learning module (111) of the simulation device (100) of a physics-based character according to one embodiment of the disclosed invention can receive a target dribbling speed as input information.

[0054] Referring to FIGS. 1 to 3, the first skill learning module (111) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention includes a first element (r) that induces the ball to move at a target dribbling speed. t ball-vel), the second element (r) that brings the character and the ball closer t ball-root-pos), a third element (r) that induces the character to move at the target dribble speed t Compensation can be performed based on the ball-vel.

[0055] Referring to FIGS. 1 to 4, a first skill learning module (111) of a simulation device (100) of a physics-based character according to one embodiment of the disclosed invention can perform an agile transition based on state initialization.

[0056] For example, skill transitions based on state initialization can be performed by saving the state values ​​of the ball and character in a buffer while playing other previously learned skill policies, and then initializing the episode of a newly learned skill based on the saved buffer.

[0057] Specifically, the processor (110) of the simulation device (100) of a physics-based character according to one embodiment of the disclosed invention can sample the states of the character and the ball to construct a buffer of a dribble policy, a buffer of a trap policy, and a buffer of a movement policy.

[0058] Accordingly, the first skill learning module (111) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can initialize the character based on the buffer of the movement policy and the buffer of the trap policy.

[0059] In addition, the first skill learning module (111) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can immediately end the episode and start the next episode when 10 seconds have passed since the episode started or when the ball and the character move away by more than 3 meters.

[0060] Additionally, the second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can learn the trap policy of the character.

[0061] More specifically, the second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can receive information as input information regarding which part of the character will receive the ball.

[0062] Referring to FIGS. 1 and 5, the second skill learning module (112) of the simulation device (100) of a physics-based character according to one embodiment of the disclosed invention can perform compensation for learning by distinguishing between the case (a) before the collision between the character and the ball and the case (b) after the collision between the character and the ball.

[0063] Specifically, the second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can perform compensation so that the distance between the ball and the part of the character that receives the ball becomes closer before the character and the ball collide.

[0064] The second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can perform compensation so that the relative speed of the ball and the character decreases after the character and the ball collide.

[0065] A second skill learning module (112) of a physics-based character simulation device (100) according to one embodiment of the disclosed invention can perform an agile transition based on state initialization.

[0066] For example, the second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can initialize the character based on a buffer of a movement policy.

[0067] The second skill learning module (112) of the simulation device (100) of a physics-based character according to one embodiment of the disclosed invention can randomly determine at least one piece of information among the launch angle, speed, and arrival position of the ball in the case of a lob pass episode, and calculate the initial position of the ball based on this to initialize the ball.

[0068] In addition, the second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can randomly determine at least one piece of information among the launch angle, speed, and arrival location of the ball in the case of a ground pass episode and initialize the ball based on this.

[0069] The second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can end the episode in the case of a lob pass episode when the ball touches the ground before the collision between the ball and the character.

[0070] Additionally, the second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can end the episode in the case of a ground pass episode when the ball passes the character before the ball and the character collide.

[0071] In addition, the second skill learning module (112) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can end the episode in common in the lob pass episode and the ground pass episode when the ball collides with the character and 1 / 6 second has passed, or when the ball collides with an arm or hand.

[0072] Additionally, the third skill learning module (113) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can learn the character's move policy.

[0073] Specifically, the third skill learning module (113) of the simulation device (100) of a physics-based character according to one embodiment of the disclosed invention can receive the target movement speed and target gaze direction of the character as input information.

[0074] More specifically, the third skill learning module (113) of the simulation device (100) of a physics-based character according to one embodiment of the disclosed invention can store latent variables for specific motions of the character in advance in a buffer, and calculate the similarity between the stored latent variables and latent variables generated during learning and reflect it in the reward.

[0075] Additionally, the fourth skill learning module (114) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can learn the kick policy of the character.

[0076] Specifically, the fourth skill learning module (114) of the simulation device (100) of a physics-based character according to one embodiment of the disclosed invention can receive a target kick speed of a ball as input information and perform compensation so that the speed of the ball becomes closer to the target kick speed.

[0077] Additionally, the fourth skill learning module (114) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can perform skill transition based on state initialization.

[0078] More specifically, the fourth skill learning module (114) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can initialize the character based on a buffer of basic postures and dribbling policies.

[0079] The fourth skill learning module (114) of the physics-based character simulation device (100) according to one embodiment of the disclosed invention can end the episode when 1 / 3 second has passed since the ball and the character's feet collide or when the ball and the character's feet do not collide for 3 seconds.

[0080] Referring to FIG. 2, a method for simulating a physics-based character according to one embodiment of the disclosed invention may include a step (S110) of learning a plurality of skill policies based on state initialization or similarity of latent variables.

[0081] For example, a method for simulating a physics-based character according to one embodiment of the disclosed invention may include a step of learning a plurality of skill policies based on state initialization and performing transitions between skills.

[0082] Additionally, a method for simulating a physics-based character according to one embodiment of the disclosed invention may include a step of learning a plurality of skill policies based on the similarity of latent variables and performing transitions between skills.

[0083] For example, a step of learning multiple skill policies based on state initialization may include a step of learning a character's dribbling policy, a step of learning a character's trap policy, and a step of learning a character's kick policy.

[0084] Additionally, the step of learning multiple skill policies based on the similarity of latent variables may include a step of learning a movement policy of the character.

[0085] Specifically, the step of learning the character's dribbling policy may include a step of receiving a target dribbling speed as input information.

[0086] For example, a step of learning a character's dribbling policy may reward based on a first factor that induces the ball to move at a target dribbling speed, a second factor that induces the distance between the character and the ball to become closer, and a third factor that induces the character to move at the target dribbling speed.

[0087] In addition, the skill transition based on the state initialization described above can be performed by saving the state values ​​of the ball and the character in a buffer while playing other previously learned skill policies, and initializing the episode of the newly learned skill based on the saved buffer.

[0088] Specifically, a method for simulating a physics-based character according to one embodiment of the disclosed invention may include a step of constructing a buffer of a dribble policy, a buffer of a trap policy, and a buffer of a movement policy by sampling the states of a character and a ball.

[0089] Accordingly, the step of learning the dribbling policy of the character may include the step of initializing the character based on a buffer of the movement policy and a buffer of the trap policy.

[0090] Additionally, the step of learning the character's dribbling policy may include a step of immediately ending the episode and starting the next episode if 10 seconds have passed since the episode started or if the ball and the character move more than 3 meters apart.

[0091] Additionally, the step of learning the character's trap policy may include a step of receiving information as input information about which part of the character the ball should be received.

[0092] Thereafter, the step of learning the trap policy of the character may include a step of performing a reward for learning by distinguishing between the case (a) before the collision between the character and the ball and the case (b) after the collision between the character and the ball.

[0093] Specifically, the step of learning the character's trap policy may include a step of performing a reward to bring the distance between the ball and the part of the character that will receive the ball closer before the character collides with the ball.

[0094] Additionally, the step of learning the character's trap policy may include a step of performing compensation so that the relative speed of the ball and the character decreases after the character and the ball collide.

[0095] The step of learning the character's trap policy may include performing agile transitions based on state initialization.

[0096] For example, learning a character's trap policy may include initializing the character based on a buffer of movement policies.

[0097] The step of learning the character's trap policy may include a step of randomly determining at least one piece of information among the launch angle, speed, and arrival location of the ball in the case of a lob pass episode, and calculating the initial location of the ball based on the information, thereby initializing the ball.

[0098] Additionally, the step of learning the character's trap policy may include a step of randomly determining at least one piece of information among the launch angle, speed, and arrival location of the ball for a ground pass episode and initializing the ball based on this.

[0099] Subsequently, the step of learning the character's trap policy may include, for a lob pass episode, terminating the episode if the ball touches the ground before the collision between the ball and the character.

[0100] Additionally, the step of learning the character's trap policy may include, for a ground pass episode, terminating the episode if the ball passes the character before the ball and the character collide.

[0101] Additionally, the step of learning the character's trap policy may include a step common to both the lob pass episode and the ground pass episode, such as terminating the episode when the ball collides with the character and 1 / 6 of a second has passed, or when the ball collides with an arm or hand.

[0102] The step of learning the character's movement policy may include a step of receiving the character's target movement speed and target gaze direction as input information.

[0103] More specifically, the step of learning the movement policy of the character may include a step of pre-storing latent variables for specific motions of the character in a buffer, calculating the similarity between the stored latent variables and latent variables generated during learning, and reflecting the result in a reward.

[0104] Additionally, the step of learning the character's kick policy may include a step of receiving a target kick speed of the ball as input information and performing a reward so that the speed of the ball becomes closer to the target kick speed.

[0105] Additionally, the step of learning the character's kick policy may include a step of performing skill transitions based on state initialization.

[0106] More specifically, the step of learning the character's kicking policy may include the step of initializing the character based on a buffer of base stances and dribbling policies.

[0107] The step of learning the character's kick policy may include terminating the episode after 1 / 3 of a second has passed since the ball and the character's foot collided, or if the ball and the character's foot do not collide for 3 seconds.

[0108] In addition, referring to FIGS. 2 and 6, a method for simulating a physically based character according to one embodiment of the disclosed invention may include a step (S120) of inputting latent variables output from a plurality of skill policies into a low-level policy of a motion embedding model.

[0109] Specifically, the first skill learning module (111), the second skill learning module (112), the third skill learning module (113), and the fourth skill learning module (114) that train multiple skill policies output latent variables z, and the output latent variables z can be used as input information of a low-level policy model.

[0110] More specifically, a method for simulating a physics-based character according to one embodiment of the disclosed invention may include a step in which, when a low-level policy model receives a specific z value together with a current character state and outputs a low-level action to a physics simulation, the simulated character performs an action corresponding to the z value identical to the action of the dataset.

[0111] Additionally, a method for simulating a physics-based character according to one embodiment of the disclosed invention may include a step (S130) of performing a transition between each skill when a predefined transition condition is met. Further details regarding this will be described later in the description of FIG. 6.

[0112] FIG. 7 is a diagram showing conditions under which transitions between multiple skills occur in a method and device for simulating a physically based character according to one embodiment of the disclosed invention.

[0113] Referring to FIGS. 2 and 7, a method for simulating a physics-based character according to one embodiment of the disclosed invention may include a step (S130) of performing a transition between each skill when a predefined transition condition is achieved.

[0114] More specifically, the transition from a move skill to a trap skill occurs when the user inputs a trapping start command and the ball approaches the character.

[0115] So while the ball is in the air, the user can use the movement policy to position the character *?* and initiate the trap command at the desired moment to transition control to the trap policy.

[0116] Additionally, the transition from a move skill to a dribble skill occurs when the horizontal distance between the ball and the character's root is less than 2m and the ball is moving towards the character (i.e., the distance between the two characters decreases).

[0117] Therefore, when the user moves the character towards the ball, the character can perform dribbling skills more naturally as it gets closer to the ball.

[0118] Additionally, the transition from dribble skill to kick skill occurs when the user inputs the kick start command.

[0119] Additionally, the transition from dribble skill to move skill occurs when the horizontal distance between the ball and the character exceeds 2m.

[0120] Additionally, the transition from trap skill to dribble skill occurs when the ball and the character collide.

[0121] Additionally, the transition from a trap skill to a move skill occurs when the user inputs the trapping end command or when the ball moves away from the character.

[0122] Additionally, the transition from a kick skill to a move skill occurs when the ball and the character collide or the horizontal distance between them exceeds 2m.

[0123] Finally, the transition from the kick skill to the dribble skill occurs when the user inputs the kick end command.

[0124] Therefore, the method and device for simulating a physically based character according to one embodiment of the disclosed invention have the advantage of being able to implement physically natural and visually smooth transitions even in the transition section between individual movements of the character.

[0125] In addition, the method and device for simulating a physics-based character according to one embodiment of the disclosed invention have the advantage of being able to hierarchically configure and integrate a physics-based motion embedding model and a model specialized for each skill.

[0126] The devices described above may be implemented as hardware components, software components, and / or a combination of hardware components and software components. For example, the devices and components described in the embodiments may be implemented using one or more general-purpose computers or special-purpose computers, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing instructions and responding to them. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. Furthermore, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing device is sometimes described as being used singly; however, those skilled in the art will appreciate that the processing device may include multiple processing elements and / or multiple types of processing elements. For example, the processing device may include multiple processors, or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are also possible.

[0127] Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing device to perform a desired operation or may independently or collectively command the processing device. The software and / or data may be embodied in any type of machine, component, physical device, virtual equipment, computer storage medium, or device for interpretation by the processing device or for providing instructions or data to the processing device. The software may also be distributed over networked computer systems and stored or executed in a distributed manner. The software and data may be stored on one or more computer-readable recording media.

[0128] The method according to the embodiment may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program commands, data files, data structures, etc., alone or in combination. The program commands recorded on the medium may be those specially designed and configured for the embodiment or may be those known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program commands such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

[0129] Although the embodiments have been described with limited examples and drawings, those skilled in the art will appreciate that various modifications and variations can be made based on the above teachings. For example, appropriate results can be achieved even if the described techniques are performed in a different order than described, and / or components of the described systems, structures, devices, circuits, etc. are combined or combined in a different manner than described, or are replaced or substituted with other components or equivalents. Therefore, other implementations, other embodiments, and equivalents of the claims also fall within the scope of the claims described below.

Claims

1. A motion embedding model including a low-level policy model that embeds motion sequences in a latent space and reproduces motion data of a character in a physical environment, and a high-level policy model that sets the motion goal of the character; A first skill learning module that learns the dribble policy of the above character; A second skill learning module that learns the trap policy of the above character; A third skill learning module that learns the move policy of the above character; and characterized in that it includes a fourth skill learning module that learns the kick policy of the above character; A simulation device for physics-based characters.

2. In paragraph 1, A processor including the motion embedding model, the first skill learning module, the second skill learning module, the third skill learning module, and the fourth skill learning module; The processor is characterized in that it samples the state of the character and the ball to construct a buffer of the dribble policy, a buffer of the trap policy, and a buffer of the movement policy. A simulation device for physics-based characters.

3. In paragraph 2, The above first skill learning module is characterized in that it receives a target dribbling speed as input information. A simulation device for physics-based characters.

4. In paragraph 3, The above first skill learning module is, A method of performing a reward based on a first element that induces the ball to move at the target dribble speed, a second element that causes the distance between the character and the ball to become closer, and a third element that induces the character to move at the target dribble speed, A simulation device for physics-based characters.

5. In paragraph 4, The above first skill learning module is, Characterized in that the above character is initialized based on the buffer of the movement policy and the buffer of the trap policy. A simulation device for physics-based characters.

6. In paragraph 2, The second skill learning module is characterized in that it receives information as input information about which part of the character the ball will be received. A simulation device for physics-based characters.

7. In paragraph 6, The above second skill learning module is, In the case where the character and the ball collide, compensation is performed so that the distance between the ball and the part of the character that will receive the ball becomes closer, and in the case where the character and the ball collide, compensation is performed so that the relative speed of the ball and the character becomes smaller. A simulation device for physics-based characters.

8. In paragraph 7, The above second skill learning module is, characterized in that the above character is initialized based on the buffer of the above movement policy, A simulation device for physics-based characters.

9. In paragraph 1, The third skill learning module is characterized in that it receives the target movement speed and target gaze direction of the character as input information. A simulation device for physics-based characters.

10. In paragraph 9, The above third skill learning module is, A method characterized in that latent variables for specific motions of the character are stored in advance in a buffer, and the similarity between the stored latent variables and latent variables generated during learning is calculated and reflected in the reward. A simulation device for physics-based characters.

11. In paragraph 1, The above 4th skill learning module is, A method characterized in that a target kick speed of a ball is input as input information, and compensation is performed so that the speed of the ball becomes closer to the target kick speed. A simulation device for physics-based characters.

12. Step of learning multiple skill policies based on state initialization; A step of inputting latent variables output from the above multiple skill policies into a low-level policy of the motion embedding model; and characterized in that it comprises a step of performing a transition between each skill when a predefined transition condition is achieved; A method for simulating physically based characters.

13. In paragraph 12, The step of learning multiple skill policies based on the above state initialization is: Steps to learn the character's dribble policy; A step of learning the trap policy of the above character; and characterized in that it comprises a step of learning the kick policy of the above character; A method for simulating physically based characters.

14. In paragraph 12, The step of learning multiple skill policies based on the above state initialization is: A method characterized in that the buffer of the dribble policy and the buffer of the trap policy are constructed by sampling the state of the character and the ball and are utilized for compensation. A method for simulating physically based characters.

15. A motion embedding model including a low-level policy model that embeds motion sequences in a latent space and reproduces the motion data of a character in a physical environment, and a high-level policy model that sets the motion goal of the character; and Includes multiple skill learning modules that learn the skill policies of the above character; The above multiple skill learning modules A first skill learning module that learns the dribble policy of the character based on state initialization, a second skill learning module that learns the trap policy of the character, and a fourth skill learning module that learns the kick policy of the character; and A third skill learning module that learns the move policy of the character based on the similarity between the latent variables stored in advance for a specific motion and the latent variables generated during learning; characterized in that it includes; A simulation device for physics-based characters.

Images