WO2023214602A1 - 훈련된 신경망에 기반한 피격 반응 생성 방법 및 컴퓨터 판독가능 저장 매체 - Google Patents
훈련된 신경망에 기반한 피격 반응 생성 방법 및 컴퓨터 판독가능 저장 매체 Download PDFInfo
- Publication number
- WO2023214602A1 WO2023214602A1 PCT/KR2022/006495 KR2022006495W WO2023214602A1 WO 2023214602 A1 WO2023214602 A1 WO 2023214602A1 KR 2022006495 W KR2022006495 W KR 2022006495W WO 2023214602 A1 WO2023214602 A1 WO 2023214602A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- hit
- neural network
- data
- response
- electronic device
- Prior art date
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 219
- 238000006243 chemical reaction Methods 0.000 title claims abstract description 72
- 238000000034 method Methods 0.000 title claims description 43
- 238000012549 training Methods 0.000 claims abstract description 88
- 238000005070 sampling Methods 0.000 claims abstract description 56
- 230000004044 response Effects 0.000 claims description 299
- 230000033001 locomotion Effects 0.000 claims description 85
- 230000008859 change Effects 0.000 claims description 44
- 230000008569 process Effects 0.000 claims description 7
- 230000009471 action Effects 0.000 abstract description 18
- 230000003044 adaptive effect Effects 0.000 description 41
- 238000012545 processing Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 10
- 230000005540 biological transmission Effects 0.000 description 5
- 238000004590 computer program Methods 0.000 description 5
- 230000014509 gene expression Effects 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 4
- 238000005859 coupling reaction Methods 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 4
- 210000000225 synapse Anatomy 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000007704 transition Effects 0.000 description 3
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 2
- 230000009118 appropriate response Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000008649 adaptation response Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000000740 bleeding effect Effects 0.000 description 1
- 230000003920 cognitive function Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007717 exclusion Effects 0.000 description 1
- 230000008921 facial expression Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000001151 other effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035939 shock Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/55—Controlling game characters or game objects based on the game progress
- A63F13/56—Computing the motion of game characters with respect to other game characters, game objects or elements of the game scene, e.g. for simulating the behaviour of a group of virtual soldiers or for path finding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/55—Controlling game characters or game objects based on the game progress
- A63F13/57—Simulating properties, behaviour or motion of objects in the game world, e.g. computing tyre load in a car race game
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/60—Generating or modifying game content before or while executing the game program, e.g. authoring tools specially adapted for game development or game-integrated level editor
- A63F13/67—Generating or modifying game content before or while executing the game program, e.g. authoring tools specially adapted for game development or game-integrated level editor adaptively or by learning from player actions, e.g. skill level adjustment or by storing successful combat sequences for re-use
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/80—Special adaptations for executing a specific game genre or game mode
- A63F13/837—Shooting of targets
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T13/00—Animation
- G06T13/20—3D [Three Dimensional] animation
- G06T13/40—3D [Three Dimensional] animation of characters, e.g. humans, animals or virtual beings
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T19/00—Manipulating 3D models or images for computer graphics
- G06T19/20—Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
Definitions
- the descriptions below relate to a method for generating a shot response based on a trained neural network and a computer readable storage medium.
- reaction to being hit In order to improve the visual quality of battles in the game, the quality of the response to being hit, which is the reaction to an attack, is important. However, research on reaction to being hit was insufficient. In most games, response to being hit is implemented in the form of outputting a random hit animation according to set conditions. Movements resulting from being hit are affected by relatively diverse conditions compared to other general movements. For example, the movement following a hit is affected by not only spatial conditions but also physical conditions.
- a neural network may refer to a model that has the ability to solve problems by changing the coupling strength of synapses based on training nodes that form a network through the coupling of synapses.
- the neural network can be trained through supervised learning or unsupervised learning.
- a computer readable storage medium stores one or more programs, and when the one or more programs are executed by a processor of an electronic device, the first program for learning the operations of an object Provide first training data to a neural network, identify second training data by performing data sampling on the hit response data, and train the second neural network to learn the hit response of the object. It may include instructions that provide data and cause the electronic device to obtain a result of a hit response of the object when the object is hit based on the output of the first neural network and the output of the second neural network.
- a method executed in an electronic device includes providing first training data to a first neural network for learning movements of an object and performing data sampling on hit response data, 2 An operation of identifying training data and an operation of providing the second training data to a second neural network for learning the response of the object to being hit, and based on the output of the first neural network and the output of the second neural network, the object When being hit, it may include an operation of obtaining the result of the hit response of the object.
- the apparatus and method according to embodiments of the present disclosure can efficiently train a neural network for hit response and provide various hit responses by sampling the generated hit response data and identifying training data.
- FIG. 1 shows an example of a hit-reaction according to embodiments.
- FIG. 2 shows an example of a functional configuration of an electronic device according to embodiments.
- FIG. 3 illustrates an operation flow of an electronic device for deriving a result of a hit response based on an adaptive hit response neural network according to embodiments.
- FIG. 4 illustrates an operation flow of an electronic device for deriving a result of a hit response based on an adaptive hit response neural network according to embodiments.
- FIG. 5 illustrates an operational flow of an electronic device for learning an adaptive hit response neural network according to embodiments.
- Figure 6 shows an example of generating response data based on raw data according to an embodiment.
- FIG. 7A shows an example of sampling of hit response data according to embodiments.
- FIG. 7B shows an example of sampling based on change in hit response data according to embodiments.
- FIG. 7C shows an example of initial posture-based sampling of hit response data according to embodiments.
- Figure 8 shows an example of an adaptive hit response neural network according to embodiments.
- FIG. 10 shows an example of comparison according to whether or not to sample hit response data according to embodiments.
- Figure 11 shows an example of performance improvement of an adaptive hit response neural network according to embodiments.
- the present disclosure relates to an apparatus and method for generating an appropriate attack response operation according to various attack conditions in real time in a wireless communication system. Specifically, the present disclosure describes a technique for efficiently outputting reaction actions to being hit through a neural network for learning hit reaction data.
- expressions of more or less may be used to determine whether a specific condition is satisfied or fulfilled, but this is only a description for expressing an example and does not exclude descriptions of more or less. It's not. Conditions written as ‘more than’ can be replaced with ‘more than’, conditions written as ‘less than’ can be replaced with ‘less than’, and conditions written as ‘more than and less than’ can be replaced with ‘greater than and less than’.
- variables related to movement e.g., target object, method of being hit, means of being hit, part of the hit
- components of the device e.g., neural network, generator
- first, second, first or second can modify the corresponding components regardless of order or importance, and are only used to distinguish one component from another component.
- the components are not limited.
- a component e.g., a first component
- another component e.g., a second component
- An element may be connected directly to the other component, or may be connected through another component (eg, a third component).
- module used in this disclosure includes a unit comprised of hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit, for example.
- a module may be an integrated part, a minimum unit that performs one or more functions, or a part thereof.
- a module may be comprised of an application-specific integrated circuit (ASIC).
- ASIC application-specific integrated circuit
- Electronic devices may be of various types. Electronic devices may include, for example, portable communication devices (e.g., smartphones), computer devices, portable multimedia devices, portable medical devices, cameras, wearable devices, or home appliances. Electronic devices according to embodiments of the present disclosure are not limited to the above-mentioned devices.
- Embodiments of the present disclosure include software (e.g., one or more instructions stored in a storage medium (e.g., internal memory or external memory) that can be read by a machine (e.g., electronic device 210). : It can be implemented as a program).
- the processor of the device may call at least one instruction among one or more instructions stored from a storage medium and execute it. This allows the device to be operated to perform at least one function according to the at least one instruction called.
- the one or more instructions may include code generated by a compiler or code that can be executed by an interpreter.
- a storage medium that can be read by a device may be provided in the form of a non-transitory storage medium.
- 'non-transitory' only means that the storage medium is a tangible device and does not contain signals (e.g. electromagnetic waves), and this term refers to cases where data is semi-permanently stored in the storage medium. There is no distinction between temporary storage cases.
- FIG. 1 shows an example of a hit-reaction according to embodiments.
- terms are defined to describe a neural network for learning a response to being hit according to the present disclosure and procedures for obtaining results for a response to being hit through the neural network.
- the reaction to being hit means the reaction of the object corresponding to the hit 102 when the object 101 is hit 102.
- the means of attack may be a gun, sword, cannon, arm, or foot.
- an object 101 is being hit 102 by a gun.
- the reaction of the object 101 in response to gunfire may be referred to as a shot reaction.
- a response to being hit may mean a reaction action when an object 101 in a virtual environment (eg, a game) is attacked by another object by being hit 102 .
- response to attacks is required to be natural (plausibility).
- the present disclosure describes learning about being hit by an object 101, a neural network for the learning, and an apparatus and method using the neural network.
- the state of the object 101 may change depending on the conditions of being hit.
- the state of the object 101 may include the location of the object 101.
- the location of the object 101 may change from a first location to a second location based on an attack 102 by another object.
- the state of the object 101 may include the direction of movement.
- the direction of the object 101 may change from the first direction to the second direction.
- the state of the object 101 may include the posture of the object 101.
- the posture of the object 101 may change from the first posture to the second posture based on the attack of the other object.
- the attack 102 of the other object may be used in the virtual environment to change at least one of the position, direction, or posture of the object 101.
- the state of the object 101 may include external factors such as the facial expression of the object 101, whether the object 101 is bleeding, and a change in the clothing of the object 101.
- Object 101 may have multiple states.
- the plurality of states may be represented within the virtual environment through the motion of the object 101.
- the motion of the object 101 means that the state of the object 101 changes.
- the states that constitute the motion of the object 101 are required to be more numerous and accurate.
- the current state of the object 101 is the first state.
- the object 101 may be attacked.
- object 101 may transition from the first state to a second state through at least one third state.
- the motion of transitioning from the first state to the second state through at least one third state is referred to as a hit response.
- the response to being hit may include a walking state as the first state, a state with the shoulder tilted as the third state, and a state lying on the floor as the second state.
- the third state can be variously composed of one or more states depending on the degree of distortion and pushing of the shoulder.
- the reaction to being hit may vary depending on the conditions of being hit.
- the hit condition may include an attack condition.
- the attack condition may include an attack means.
- an object can be attacked with a gun or sword.
- object 101 may be attacked from a body such as a hand or a foot.
- an adaptive response to being hit is required depending on what object is being attacked.
- the attack condition may include attack strength.
- the response to being hit may vary depending on whether the hit is performed with a force greater than the reference value for determining a fall.
- the hit condition may include an object condition.
- Object conditions may include the hit area, hit range, and initial posture of the object.
- the response to being hit when another object attacks the shoulder of the object 101 (102) may be different from the response to being hit when another object attacks the waist of the object 101 (102).
- the response to being hit when another object attacks one shoulder of the object 101 (102) may be different from the response to being hit when another object attacks both shoulders of the object 101 (102).
- the response to being hit when attacking 102 with a force that does not knock off the shoulder of the object 101 and the response to being hit with a force that causes the object 101 to fall over are different. You can.
- the response to being hit when another object attacks (102) the standing object (101) may be different from the response to being hit when another object attacks (102) the running object (101).
- a neural network may refer to a model that has the ability to solve problems by changing the coupling strength of synapses based on training nodes that form a network through the coupling of synapses.
- the neural network can be trained through supervised learning or unsupervised learning.
- the supervised learning may mean learning performed by providing a label (or correct answer). Because the supervised learning requires the label, it may require fewer resources compared to the unsupervised learning to evaluate the reliability of output data derived from the neural network. On the other hand, since the supervised learning requires the label, it may require resources (eg, time resources) to obtain the label.
- the unsupervised learning may mean learning performed without labels. Since the unsupervised learning does not require the label, it may not require resources to obtain the label. On the other hand, because the unsupervised learning does not require the label, it may require more resources than the supervised learning to evaluate the reliability of output data derived from the neural network.
- the present disclosure focuses on the response to being hit rather than the overall motion of the object, and proposes output of the response to being hit using a neural network to separately learn only the response to being hit.
- FIG. 2 components of an electronic device for implementing the neural network of the present disclosure will be described through FIG. 2.
- FIG. 2 shows an example of a functional configuration of an electronic device 210 according to embodiments.
- the configuration illustrated in FIG. 2 may be understood as a configuration of the electronic device 210 for outputting a response of an object when the object is hit, as mentioned in FIG. 1 .
- Terms such as 'jdeungbu' and 'jeotgi' used hereinafter refer to a unit that processes at least one function or operation, and may be implemented as hardware, software, or a combination of hardware and software.
- the electronic device 210 may include a memory 230, a processor 250, and a transceiver 270.
- the memory 230 may store data such as basic programs, application programs, and setting information for operation of the electronic device 210.
- the memory 230 may be referred to by other terms having the same technical meaning, such as a storage unit or a storage medium.
- the memory 230 may be comprised of volatile memory, non-volatile memory, or a combination of volatile memory and non-volatile memory. Additionally, the memory 230 may provide stored data upon request from the processor 250.
- the memory 230 may store attack response data for operations and learning for the electronic device 210 to drive an adaptive attack response neural network according to embodiments.
- Embodiments of the present disclosure include software (e.g., one or more instructions stored in a storage medium (e.g., internal memory or external memory) that can be read by a machine (e.g., electronic device 210). : It can be implemented as a program).
- a processor e.g., processor 250
- a device e.g., electronic device 210
- the one or more instructions may include code generated by a compiler or code that can be executed by an interpreter.
- a storage medium that can be read by a device may be provided in the form of a non-transitory storage medium.
- 'non-transitory' only means that the storage medium is a tangible device and does not contain signals (e.g. electromagnetic waves), and this term refers to cases where data is semi-permanently stored in the storage medium. There is no distinction between temporary storage cases.
- the processor 250 may control overall operations of the electronic device 210. For example, the processor 250 can write and read data into the memory 230. Additionally, the processor 250 may transmit and receive signals through the transceiver 270. Additionally, the processor 250 can perform protocol stack functions required by communication standards. To this end, the processor 250 may include at least one sub-processor. Depending on embodiments, the processor 250 may be configured to enable the electronic device 210 to perform an output operation according to an adaptive hit response neural network and a learning operation of the neural network according to embodiments.
- the transceiver 270 can perform functions for transmitting and receiving signals in a wired communication environment.
- the transceiver 270 may include a wired interface for controlling direct connection between devices through a transmission medium (e.g., copper wire, optical fiber).
- a transmission medium e.g., copper wire, optical fiber
- the transceiver 270 may transmit an electrical signal to another device through a copper wire or perform conversion between an electrical signal and an optical signal.
- the transceiver 270 may perform functions for transmitting and receiving signals through a wireless channel. For example, the transceiver 270 may perform a conversion function between a baseband signal and a bit stream according to the physical layer standard of the system. For example, when transmitting data, the transceiver 270 may generate complex symbols by encoding and modulating the transmission bit stream. Additionally, when receiving data, the transceiver 270 can restore the received bit stream through demodulation and decoding of the baseband signal. Additionally, the transceiver 270 may up-convert a baseband signal into a radio frequency (RF) band signal and transmit it through an antenna, and down-convert the RF band signal received through the antenna into a baseband signal. To this end, the transceiver 270 may include a transmission filter, a reception filter, an amplifier, a mixer, an oscillator, a digital-to-analog converter (DAC), an analog-to-digital converter (ADC), etc.
- RF radio frequency
- the transceiver 270 may include multiple transmission and reception paths. Furthermore, the transceiver 270 may include at least one antenna array comprised of multiple antenna elements. In terms of hardware, the transceiver 270 may be composed of a digital unit and an analog unit, and the analog unit is composed of a number of sub-units depending on operating power, operating frequency, etc. It can be.
- Transceiver 270 can transmit and receive signals as described above. Accordingly, all or part of the transceiver 270 may be referred to as a 'transmitter', 'receiver', or 'transceiver'. Additionally, in the following description, transmission and reception performed through a wireless channel may be used to mean that processing as described above is performed by the transceiver 270.
- the configuration of the electronic device 210 shown in FIG. 2 is only an example, and examples of electronic devices that perform various embodiments of the present disclosure are not limited to the configuration shown in FIG. 2 . That is, some configurations may be added, deleted, or changed according to various embodiments. For example, in the case of a device that performs both learning about the response to being hit and deriving results about the response to being hit within the electronic device 210, the transceiver 270 may be omitted from the electronic device 210.
- a neural network and a set of parameters related to the neural network may be stored in the memory 230 of the electronic device 210 according to one embodiment.
- a neural network is a recognition model implemented in software or hardware that imitates the computational ability of a biological system using a large number of artificial neurons (or nodes). Neural networks can perform human cognitive functions, learning processes, or training through artificial neurons. Parameters related to a neural network may represent, for example, a plurality of nodes included in the neural network and/or weights assigned to connections between the plurality of nodes.
- processor 250 may train a neural network.
- the neural network may be trained through unsupervised learning.
- processor 250 may provide input data to a neural network to train the neural network.
- the input data may be training data generated through sampling reaction data.
- a neural network may include multiple layers.
- a neural network may include an input layer, one or more hidden layers, and an output layer. Signals generated at each node in the input layer based on the input data may be transmitted from the input layer to one or more hidden layers.
- the output layer may obtain output data of the neural network based on one or more signals received from one or more hidden layers.
- the input layer, one or more hidden layers, and the output layer may include a plurality of nodes.
- One or more hidden layers may be a convolution filter or a fully connected layer in a CNN (convolution neural network), or may be various types of filters or layers connected based on specified functions or characteristics. there is.
- one or more hidden layers may be a layer based on a recurrent neural network (RNN) whose output value is re-input to the hidden layer at the current time.
- RNN recurrent neural network
- one or more hidden layers may be configured in plural, and may form a deep neural network. For example, training a neural network that includes one or more hidden layers that form at least part of a deep neural network may be referred to as deep learning.
- a node included in one or more hidden layers may be referred to as a hidden node.
- Nodes included in the input layer and one or more hidden layers may be connected to each other through a connection line with a connection weight, and nodes included in one or more hidden layers and an output layer may also be connected to each other through a connection line with a connection weight.
- Tuning and/or training a neural network may mean changing the connection weights between nodes included within each of the layers included within the neural network (e.g., an input layer, one or more hidden layers, and an output layer). For example, tuning or training of a neural network may be performed based on unsupervised learning.
- the method according to the embodiments of the present disclosure may be included and provided in a computer program product.
- Computer program products are commodities and can be traded between sellers and buyers.
- a computer program product may be distributed in the form of a machine-readable storage medium (e.g. compact disc read only memory (CD-ROM)), or through an application store (e.g. Play Store), or on two user devices (e.g. : Smartphones) can be distributed (e.g. downloaded or uploaded) directly or online.
- a portion of the computer program product may be at least temporarily stored or temporarily created in a machine-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or a relay server.
- Figure 3 shows an example of learning an adaptive hit reaction according to embodiments.
- the operations of FIG. 3 are described as being performed by the processor 250 of the electronic device 210 of FIG. 2, but embodiments of the present disclosure are not limited thereto.
- some of the operations described below are performed by the processor 250 of the electronic device 210, and other operation(s) are performed by the transceiver 270 of the electronic device 210. It can be performed through an external device (e.g., OTA (over the air) server).
- some of the operations described later may be implemented in advance in the memory 230 of the electronic device 210, and some other operation(s) may be implemented in advance in the processor (s) of the electronic device 210. 250).
- learning of an adaptive hit response can be performed through generating a hit response, data sampling, and learning a hit response.
- the processor 250 may obtain raw data 301.
- the processor 250 may obtain raw data 301 from the memory 230.
- the processor 250 may obtain raw data 301 from an external server.
- Raw data 301 may refer to all data related to an object.
- raw data 301 may include motion capture data.
- the processor 250 may generate a hit response.
- the instructions performed by the processor 250 to generate a hit response or the operation of the processor 250 to generate a hit response can be understood as a function of the hit response generator 303.
- the hit response generator 303 may generate a hit response based on the raw data 301. In order to learn data that has the characteristic of greatly increasing the diversity of motion at the moment of being hit, hit reaction data is required.
- the hit response generator 303 may generate various hit responses to obtain the data.
- the hit response generator 303 can generate various hit response data under given hit conditions through the raw data 301.
- the hit response generator 303 may output hit response data based on raw data 301, which is motion data.
- raw data 301 may include motion data according to a designated attack angle.
- the raw data 301 may include motion data according to a designated target area (eg, head).
- the hit response generator 303 may generate a hit response based on a recognition response mechanism.
- the hit response generator 303 can implement motion using physical simulation during a defined time after impact, and find and output a suitable motion from a motion database after the defined time.
- raw data may be motion capture data of the head being struck.
- the shot angle (eg, vertical), and the shot part (eg, head) are input data
- the shot response generator 303 may generate shot response data based on the input data.
- the hit response generator 303 may generate hit reaction data based on the instantaneous hit response and the subsequent hit reaction.
- the hit response generator 303 may generate hit response data by connecting the instantaneous hit response when the head moves by force during a defined time and the subsequent hit response through motion matching after the instant hit response.
- data on reaction to being hit may be a movement in which the head falls back first in an instantaneous reaction state, and then falls while struggling with one's hands to keep one's balance in a subsequent reaction state.
- the shot response generator 303 may generate a shot response for continuous shot conditions based on shot response data for discrete shot conditions. For example, when the hit response generator 303 only has hit reaction data for the condition of hitting the shoulder with a first force and hit reaction data for the condition of hitting the shoulder with a second force, Hit response data can be generated for the condition of hitting with a third force between the first force and the second force. For example, if there is only hit response data for the condition of attacking the shoulder with a sword with a first length and hit response data for the condition of attacking the shoulder with a sword with a second length, the first length and the second length It is possible to generate hit response data for the conditions of attacking with a sword of the third length between the lengths.
- hit reaction data for the condition of attacking the first part of the head For example, if there is only hit reaction data for the condition of attacking the first part of the head and hit reaction data for the condition of attacking the second part of the head, there is a third part between the first part and the second part. It is possible to generate hit reaction data for the conditions of attacking.
- Data sampling 305 refers to the process of identifying training data 307 from hit response data.
- the burden on the adaptive hit response neural network 309 increases.
- the processor 250 can exclude hit response data that does not meet at least one standard from the training data 307 in order to efficiently learn the adaptive hit response neural network 309.
- data sampling 507 may be performed through an algorithm to exclude hit reaction data for at least one of the initial posture, attack site, attack range, and/or attack intensity that has a change amount below a reference value. .
- the adaptive hit response neural network 309 may be composed of at least one neural network.
- the adaptive hit response neural network 309 may be referred to as a diversity-adaptive refinement network.
- the adaptive hit response neural network 309 may be composed of an object motion learning neural network 311 and a hit response learning neural network 313.
- the object motion learning neural network 311 is a neural network for learning the motion of an object.
- the object motion learning neural network 311 may be referred to as a first subneural network, a first neural network, or a regular subnetwork.
- the hit response learning neural network 313 is a neural network for learning the hit response.
- a neural network with this structure can output more accurate and rich response to being hit.
- the adaptive hit response neural network 309 has a gating module that combines the output results of the object motion learning neural network 311 and the hit response learning neural network 313. More may be added.
- FIG. 4 illustrates an operation flow of an electronic device for deriving a result of a hit response based on an adaptive hit response neural network according to embodiments.
- the electronic device exemplifies the electronic device 210 of FIG. 2 . At least some of the operations in FIG. 4 may be executed by the processor 250 of the electronic device 210.
- raw data refers to data about the operation of an object.
- raw data may include motion capture data.
- the raw data may be motion capture data of an object being hit on the shoulder and falling.
- the electronic device 210 may perform learning of an adaptive hit response neural network. Neural network training based on raw data may include the content shown in FIG. 5. For example, the electronic device 210 generates at least one hit response data based on raw data (505). Thereafter, the electronic device 210 may generate training data through a process of sampling the hit response data (509) and learn a hit response learning neural network based on the training data (511).
- the electronic device 210 may input a shot condition to an adaptive hit response neural network.
- the electronic device 210 may provide the hit condition as input data to the learned neural network (405) and obtain the result of the object's hit response based on the neural network output (407).
- the hit condition may be information about attack conditions, object conditions, etc.
- a hit condition may be input to the neural network.
- the hit condition which is input data
- the processor 250 may analyze the motion and extract information about the hitting conditions. If the input data is a motion of shooting an object in a standing state from the front perpendicular to the plane forming the human body, the processor 250 may extract that the initial posture is a standing state. Additionally, the processor 250 can extract that the method of attack is shooting and that the angle of attack is forward and perpendicular to the plane forming the human body.
- the electronic device 210 may obtain a shot response result from the adaptive hit response neural network.
- the adaptive hit response neural network may be composed of at least one neural network.
- the adaptive hit response neural network 309 may be composed of an object motion learning neural network 311 and a hit response learning neural network 313.
- the adaptive hit response neural network 309 determines the hit response result in the action 407 by considering the weight assigned to the output of the object motion learning neural network 311 and the output of the hit response learning neural network 313. You can.
- the weight assignment may be performed based on the combining gating module 805.
- Gating module 805 may be a neural network.
- the gating module 805 may be a neural network that receives a hit condition as input data and outputs weights assigned to the output of the object motion learning neural network 311 and the output of the hit response learning neural network 313.
- the gating module 805 may determine weights assigned to the output of the object motion learning neural network 311 and the output of the striking response learning neural network 313 based on the striking intensity. For example, the stronger the intensity of the attack, the higher the weight assigned to the output of the attack response learning neural network 313 can be determined.
- the gating module may assign a higher weight to the output of the object motion learning neural network 311 than the output of the hitting response learning neural network 313 as the intensity of the striking is lower. In this case, the response to being hit is that the person does not fall even if the shoulder is attacked, and the size of the struggling motion may be reduced.
- the output result of the hit response learning neural network can be considered to obtain the hit response result only when there is an attack motion of another object.
- the output results of the hit response learning neural network may not be considered.
- the output value of the hit response learning neural network may not be considered when generating the object's response.
- operations 401, 403, 405, and 407 are shown to be performed sequentially, but embodiments of the present disclosure are not limited thereto. According to one embodiment, operations 401 and 403 for learning the reaction to being hit may be performed in parallel with operations 405 and 407 for outputting the results for the reaction to being hit. According to another embodiment, operations 401 and 403 are performed periodically, but operations 405 and 407 may be performed only when an object is hit.
- FIG. 5 illustrates an operational flow of an electronic device for learning an adaptive hit response neural network according to embodiments.
- the electronic device exemplifies the electronic device 210 of FIG. 2 .
- At least some of the operations in FIG. 5 may be executed by the processor 250 of the electronic device 210.
- the operations in FIG. 5 can be understood as specific operations of learning the adaptive hit response neural network in operation 403 of FIG. 4 .
- the electronic device 210 may provide first training data to an object motion learning neural network for learning motions of an object.
- An object motion learning neural network may refer to a neural network for learning all motions of an object.
- the first training data may include data related to the operations of the object, such as movement of the object, state of the object, and environment of the object, as well as data on the response to being hit.
- the electronic device 210 may learn an object motion learning neural network using first training data.
- the first training data may be general motion capture data.
- the hit response learning neural network may be divided into an object motion learning neural network and a hit response learning neural network.
- an object motion learning neural network for learning all motions of an object can be trained using first training data.
- the electronic device 210 may generate various hit response data based on raw data.
- the hit reaction data may include instantaneous hit reaction data and subsequent hit reaction data based on the recognition reaction mechanism.
- the instantaneous response to being hit may mean a response to being hit according to the force applied to the object.
- the follow-up response to being hit may mean a response to being hit according to an input from an object.
- the electronic device 210 may generate second training data by performing data sampling on response data to being hit.
- the electronic device 210 may perform data sampling according to embodiments to efficiently extract training data.
- data sampling may be performed through an algorithm for excluding hit response data with a change amount less than a reference value.
- data sampling may be performed through an algorithm to exclude shot response data for a shot area with a change amount below a reference value. If there is already a lot of reaction data where the shot area is the shoulder in the training data, the shot reaction data where the hit area is the shoulder may no longer be added to the training data.
- data sampling may be performed through an algorithm for excluding hit response data for a shot range that has a change amount below a reference value.
- the range of attack may vary depending on the length of the sword. If the difference between the length of the sword in the secured hit response data and the length of the sword in the target hit response is less than or equal to the reference value, the target hit response data may not be added to the training data. For example, data sampling may be performed through an algorithm for excluding hit response data for the hit area and hit intensity that have a change amount below a reference value. For example, if the training data contains at least one hit response data for attacking the shoulder of an object in a standing state, train the hit response data for attacking the shoulder of an object in a standing state with a hit intensity below the reference value. It may not be added to the data.
- the electronic device 210 may train a hit response learning neural network using second training data.
- the second training data may be different from the first training data.
- the second training data may be data sampled from hit response data. Efficient learning is possible by extracting and sampling only data related to response to being hit.
- operations 501 to 511 are shown as being performed sequentially, but embodiments of the present disclosure are not limited thereto.
- operations 501 and 503 for learning an object motion learning neural network are operations 505, 507, 509, and 511 for learning a hit response learning neural network. can be performed in parallel.
- Hit reaction data refers to information related to the reaction of an object when the object is hit.
- the electronic device 210 may generate hit reaction data.
- the state before being hit by an object may be referred to as an idle state 601. Afterwards, from the point when the object is hit (603), the state of the object transitions from the idle state (601) to the reactive state. After the reactive state, the object may transition back to the idle state 609.
- the reaction state may include an instantaneous reaction state (605) and a subsequent reaction state (607).
- the distinction between the instantaneous response state 605 and the subsequent response state 607 may be based on the recognition response mechanism.
- the recognition response mechanism is the principle that when a living organism receives an unexpected and sudden shock, it responds after a delay time of about 100 to 200 ms.
- the instantaneous reaction state 605 may be a state in which an action by physical force occurs for a certain period of time immediately after being hit (603).
- the above operation may be referred to as an instantaneous striking operation.
- the subsequent reaction state 607 may be a state in which an action based on the will and characteristics of the living entity occurs after the instantaneous reaction state 605.
- the operation may be referred to as a follow-up shooting operation.
- the hit response data may include data related to the above-described instantaneous hit action and subsequent hit action.
- a hit reaction may be generated through a motion matching algorithm based on motion data.
- the hit response generator 303 of the electronic device 210 may find a motion corresponding to being hit from a motion database and output the motion.
- reaction data may include the head falling back first in the instantaneous reaction state 605 and falling while struggling with the hands to maintain balance in the subsequent reaction state 607. You can.
- the attack response generator 303 can generate various attack response data.
- FIG. 7A shows an example 700 of sampling of hit response data according to embodiments.
- Data sampling refers to the process of extracting some data from the entire data. Resource efficiency can be increased by learning some of the extracted data instead of learning all the data. In order to output more accurate results, selection of some data that meets the purpose is required.
- the electronic device 210 may identify data for learning instead of all data on response to being hit through sampling.
- the electronic device 210 may generate hit response data 701 for efficient learning of the adaptive hit response neural network. Thereafter, the electronic device 210 may acquire training data 703 through sampling of the hit response data 701.
- the electronic device 210 may perform sampling based on whether the amount of change in the initial posture of the object after being hit is greater than or equal to a reference value. For example, the electronic device 210 may perform sampling to exclude hit response data in which the amount of change in the initial posture of the object when hit is less than a reference value. If the amount of change in the initial posture is less than the standard value, the effectiveness of the data for learning response to being hit can be considered relatively low.
- the electronic device 210 may perform sampling based on whether the amount of change in the attack range is greater than or equal to a reference value. For example, the electronic device 210 may perform sampling to exclude hit response data in which the change in the object's attack range is less than the reference value. If the attack range change is less than the standard value, the effectiveness of the data for learning response to being hit can be considered relatively low.
- the electronic device 210 may perform sampling based on whether the amount of change in the attack area is greater than or equal to a reference value. For example, the electronic device 210 may perform sampling to exclude attack response data in which the change in the object's attack area is less than a reference value. If the amount of change in the attack area is less than the standard value, the effectiveness of the data for learning response to being hit can be considered relatively low.
- the electronic device 210 may perform sampling based on whether the amount of change in response to being hit is greater than or equal to a reference value. For example, the electronic device 210 may perform sampling to exclude hit response data in which the change in the object's hit response is less than a reference value. If the amount of change in response to being hit is less than the standard value, the effectiveness of the data for learning response to being hit can be considered relatively low.
- the electronic device 210 may perform sampling and obtain training data 703 based on parameters related to at least one of the initial posture of being hit, attack range, attack area, and response to being hit.
- the loss of the neural network output results can be reduced.
- a decrease in the loss value of the neural network output result may mean that the accuracy of the hit response output by the neural network increases.
- FIG. 7B illustrates an example 710 of sampling based on change in response to being hit data according to embodiments.
- FIG. 7B an example of a hit response for an algorithm that excludes hit response data with a change amount below a reference value is shown among the sampling methods for hit response data.
- the electronic device 210 may exclude hit response data with a change amount below a reference value from the training data to improve the efficiency of neural network learning.
- the operation state 711, the operation state 713, and the operation state 715 are hit response data for an action of attacking the chin.
- the reaction to being hit may include an action in which the object's chin turns backwards and the object falls.
- the electronic device 101 has secured response data for the jaw attack, the response corresponding to the operating state 711, 713, and 715 has a change amount below the reference value. Therefore, the operation state 711, operation state 713, and operation state 715 can be excluded from the training data. If the amount of change in the response to being hit is less than the standard value, the validity of the data for learning the response to being hit can be considered relatively low.
- FIG. 7C shows an example 720 of initial posture-based sampling of hit response data according to embodiments.
- FIG. 7C an example of an initial posture for an algorithm that excludes hit response data for an initial posture with a change amount below a reference value is shown among the sampling methods for hit response data.
- the operation state 721, operation state 723, operation state 725, operation state 727, and operation state 729 which are the states immediately before being hit by the object, are shown in FIG. 7C.
- the operation state 721 is a running and turning posture
- the operation state 723 is a running posture
- the operation state 725 is a shoulder shrug posture
- the operation state 727 is a squatting posture.
- the operating state 729 may be a standing position.
- the electronic device 210 may exclude from the training data the hit reaction data in which the posture immediately before being hit has a change amount below the reference value. For example, since there is a lot of response data in the training data in which the posture immediately before being shot is a standing posture, the operating state 729 in which the posture immediately before being shot is a standing posture may be excluded from the training data. If the reference value is set high because the training data contains a lot of attack response data in which the posture immediately before being shot is a standing posture, the motion state 725 may be treated as a standing posture, and the motion state 725 may be excluded from the training data. If the reference value is set low, the operating state 725 may be treated as not being a standing posture and included in the training data.
- the validity of the data for learning response to being hit can be considered relatively low.
- Figure 8 shows an example 800 of an adaptive hit response neural network according to embodiments.
- the hit response neural network is a neural network for generating the hit response.
- Figure 8 illustrates the structure of the adaptive hit response neural network 309 of Figure 3.
- sub-neural networks of the hit response neural network are described.
- the adaptive hit response neural network 800 may include an object motion learning neural network 801 and a hit response learning neural network 803.
- the electronic device 210 may train an object motion learning neural network 801 for learning all motions of an object using first training data (503).
- the electronic device 210 may train the hit response learning neural network 803 for learning the hit response of an object using second training data (511).
- the second training data may be obtained by sampling (507) the hit response data (505) generated based on raw data.
- the electronic device 210 can efficiently learn the response to being hit.
- the first training data may be general motion capture data as well as response data to being hit.
- the adaptive hit response neural network 800 may include a gating module 805.
- the gating module 805 may be a module for determining the final output for the hit response based on the output of the object motion learning neural network 801 and the output of the hit response learning neural network 803.
- the adaptive hit response neural network 800 includes a first output of the object motion learning neural network 801, a first weight assigned to the first output, a second output of the hit response learning neural network 803, A hit response result may be generated by considering the second weight assigned to the second output.
- allocation of the first weight eg, a value greater than 0
- the allocation of the second weight eg, a value greater than 0
- the gating module 805 may receive a hit condition as input data and provide weights assigned to the output of the object motion learning neural network 803 and the output of the hit response learning neural network 801 to the adaptive hit response neural network 800. there is.
- the gating module 805 may determine weights assigned to the output of the object motion learning neural network 801 and the output of the hitting response learning neural network 803 based on the intensity of the hit. For example, the stronger the intensity of the attack, the higher the weight assigned to the output of the attack response learning neural network 803 can be determined.
- the gating module 805 may assign a higher weight to the output of the object motion learning neural network 801 than the output of the hitting response learning neural network 803 as the intensity of the striking is lower. In this case, the response to being hit is that the person does not fall even if the shoulder is attacked, and the size of the struggling motion may be reduced.
- the adaptive hit response neural network 800 may include an object motion learning neural network 801 and a hit response learning neural network 803.
- the electronic device 210 may train the object motion learning neural network 801 for learning all motions of an object using first training data (eg, operation 503 of FIG. 5).
- the electronic device 210 may train the hit response learning neural network 803 for learning the hit response of an object using second training data (e.g., operation 511 of FIG. 5 ).
- the second training data may be obtained through sampling from hit response data 505 generated based on raw data.
- the first training data may include general motion capture data as well as hit reaction data. Meanwhile, the second training data may be data sampled from hit response data.
- Figure 9 shows examples 910 of response to being hit according to embodiments.
- the response to being hit may vary depending on the hit and attack conditions.
- Attack conditions may include attack means.
- Figure 9 shows examples of responses to being hit according to attack means among attack conditions.
- the operation state 901 represents the state of an object being hit by kickboxing.
- the operation state 903 represents the state of an object being hit by a sword.
- the operation state 905 represents the state of an object being hit by a gun. It can be confirmed that various response to being hit can be obtained by using the present invention.
- Example Data size(s) Action types Gun 2, 484 idle, walk, hit Kickboxing 256 idle, walk, run, turn, dodge, punch, kick, hit Sword 2, 021 idle, walk, run, turn, swing, hit
- Table 1 shows various action types and action response data sizes according to attack conditions (example).
- An object hit by a gun shows three types of response to being hit.
- Objects hit by kickboxing show eight types of hit reactions.
- An object hit with a sword shows six types of response to being hit.
- the types of responses to being hit vary depending on the attack method.
- Table 1 it can be seen that the size of the hit response data varies for each attack method.
- FIG. 10 illustrates an example 1000 of comparison depending on whether or not to sample hit response data according to embodiments.
- a graph 1001 represents the average loss value of a neural network when hit when a data sampling method for hit response data is not performed.
- the graph 1003 shows the average loss value of the neural network when hit when a data sampling method is performed on hit reaction data.
- each point represents the average loss value of the network for 10 frames after a shot occurs in a specific shot posture. The lighter the color of the area, the relatively greater the loss. A large loss means that the learning of the neural network was relatively insufficient.
- the average loss value of the neural network is relatively small compared to the case where data sampling is not performed. Accordingly, the learning effect of the adaptive hit response neural network of the present disclosure can be maximized through data sampling described in FIGS. 7A to 7C.
- FIG. 11 shows an example 1100 of performance improvement of an adaptive hit response neural network according to embodiments.
- a shot response in a situation using an adaptive shot response neural network according to embodiments is compared with a shot response in a situation in which the adaptive shot response neural network is not used.
- the operating state 1101 is a state before being hit based on a single neural network
- the operating state 1103 is a state before being hit based on an adaptive hit response neural network
- the operating state 1105 is a state before being hit based on a single neural network.
- the operation state 1107 is a state after being hit based on an adaptive hit response neural network.
- the posture before and after the attack did not differ significantly.
- the action state 1107 in which the neural network structure consists of a hit response learning neural network and an object motion learning neural network, the difference in the details of the posture before and after the attack can be confirmed. This means that the neural network learned better due to changes in the neural network structure according to one embodiment.
- the processor can efficiently generate a hit action with limited resources. Another problem is that it is difficult to synthesize attack response behavior under arbitrary conditions from attack response data under limited conditions.
- the present invention can generate appropriate response actions to being hit according to arbitrary attack conditions.
- the processor can synthesize reaction actions to secondary or multiple hits that are difficult to synthesize through motion capture.
- the present invention can improve consistency and richness in generated imagery.
- a computer readable storage medium stores one or more programs, and when the one or more programs are executed by a processor of an electronic device, the operations of the object are performed.
- the one or more programs include additional instructions that, when executed by a processor of the electronic device, cause the electronic device to generate the hit response data, where the hit response data is applied to the object. It may include instructions that cause the electronic device to include instantaneous hit response data according to force and subsequent hit response data according to the input of the object.
- the data is based on an algorithm for excluding hit response data with a change amount below a reference value in order to identify the second training data. and instructions that cause the electronic device to perform sampling.
- the amount of change includes the amount of change in at least one of the initial posture, the area hit, the posture immediately before being hit, the range of the shot, and the intensity of the shot. May contain instructions that cause device operation.
- the initial posture, attack site, attack range, and/or attack having a change amount below a reference value are used to identify the second training data. and instructions that cause the electronic device to perform data sampling based on an algorithm for excluding hit response data for at least one of the intensities.
- the result of the hit response of the object is may include instructions that cause the electronic device to acquire.
- the second neural network may be activated when an attack motion of another object is detected in the second training data.
- the second neural network may be activated when the intensity of being hit in the second training data is greater than or equal to a reference value.
- the method of an electronic device includes providing first training data to a first neural network for learning the movements of an object and performing data sampling on the hit response data. , an operation of identifying second training data and an operation of providing the second training data to a second neural network for learning a response to being hit by the object, and based on the output of the first neural network and the output of the second neural network,
- the operation may include obtaining a result of the object's hit response.
- a method of an electronic device includes an operation of generating the hit reaction data, wherein the hit reaction data includes instantaneous hit reaction data according to a force applied to the object and an input of the object. It may include actions generated based on subsequent hit reaction data.
- the operation of identifying the second training data may include performing the data sampling based on an algorithm for excluding hit reaction data with a change amount less than the reference value. .
- the amount of change may include the amount of change in at least one of the initial posture, the area hit, the posture immediately before being hit, the range of the shot, and the intensity of the shot.
- the operation of identifying the second training data includes receiving reaction data for at least one of an initial posture, attack part, attack range, and/or attack intensity with a change amount of less than or equal to a reference value.
- An operation of performing the data sampling based on an algorithm for exclusion may be included.
- the method of the electronic device includes an operation of obtaining a result of a hit response of the object when the object is hit based on the output of the first neural network and the output of the second neural network. can do.
- the second neural network may be activated when an attack motion of another object is detected in the second training data.
- the second neural network may be activated when the intensity of being hit in the second training data is greater than or equal to a reference value.
- the device described above may be implemented with hardware components, software components, and/or a combination of hardware components and software components.
- the devices and components described in the embodiments include a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), and a programmable logic unit (PLU).
- ALU arithmetic logic unit
- FPGA field programmable gate array
- PLU programmable logic unit
- It may be implemented using one or more general-purpose or special-purpose computers, such as a logic unit, microprocessor, or any other device capable of executing and responding to instructions.
- the processing device may execute an operating system (OS) and one or more software applications running on the operating system. Additionally, a processing device may access, store, manipulate, process, and generate data in response to the execution of software.
- OS operating system
- a processing device may access, store, manipulate, process, and generate data in response to the execution of software.
- a single processing device may be described as being used; however, those skilled in the art will understand that a processing device includes multiple processing elements and/or multiple types of processing elements. It can be seen that it may include.
- a processing device may include a plurality of processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are possible.
- Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing unit to operate as desired, or may be processed independently or collectively. You can command the device.
- the software and/or data may be embodied in any type of machine, component, physical device, computer storage medium or device for the purpose of being interpreted by or providing instructions or data to the processing device. there is.
- Software may be distributed over networked computer systems and stored or executed in a distributed manner.
- Software and data may be stored on one or more computer-readable recording media.
- the method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium.
- the medium may continuously store a computer-executable program, or temporarily store it for execution or download.
- the medium may be a variety of recording or storage means in the form of a single or several pieces of hardware combined. It is not limited to a medium directly connected to a computer system and may be distributed over a network. Examples of media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, And there may be something configured to store program instructions, including ROM, RAM, flash memory, etc. Additionally, examples of other media include recording or storage media managed by app stores that distribute applications, sites or servers that supply or distribute various other software, etc.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Architecture (AREA)
- Computer Graphics (AREA)
- Computer Hardware Design (AREA)
- Image Analysis (AREA)
Abstract
Description
Example | Data size(s) | Action types |
Gun | 2, 484 | idle, walk, hit |
Kickboxing | 256 | idle, walk, run, turn, dodge, punch, kick, hit |
Sword | 2, 021 | idle, walk, run, turn, swing, hit |
Claims (14)
- 하나 이상의 프로그램들을 저장하는 컴퓨터 판독가능 저장 매체(computer readable storage medium)에 있어서,상기 하나 이상의 프로그램들은, 전자 장치의 프로세서에 의해 실행될 시에, 객체의 동작들의 학습을 위한 제1 신경망에게 제1 훈련 데이터를 제공하고,피격 반응 데이터에 대한 데이터 샘플링(data sampling)을 수행함으로써, 제2 훈련 데이터를 식별하고,상기 객체의 피격 반응의 학습을 위한 제2 신경망에게 상기 제2 훈련 데이터를 제공하고,상기 제1 신경망의 출력 및 상기 제2 신경망의 출력에 기반하여 상기 객체가 피격되는 경우, 상기 객체의 피격 반응 결과를 획득하도록,상기 전자 장치를 야기하는 인스트럭션들을 포함하는, 컴퓨터 판독가능 저장매체.
- 청구항 1에 있어서,상기 하나 이상의 프로그램들은, 상기 전자 장치의 프로세서에 의해 실행될 시에,상기 피격 반응 데이터를 생성하도록 상기 전자 장치를 야기하는 추가 인스트럭션들을 포함하고,상기 피격 반응 데이터는 상기 객체에 적용되는 힘에 따른 순간 피격 반응 데이터와 상기 객체의 입력에 따른 후속 피격 반응 데이터를 포함하는, 컴퓨터 판독가능 저장매체.
- 청구항 1에 있어서,상기 하나 이상의 프로그램들은, 상기 전자 장치의 프로세서에 의해 실행될 시에,상기 제2 훈련 데이터를 식별하기 위해서 피격 반응 데이터가 기준 값 이하의 변화량을 갖는 피격 반응 데이터를 제외하기 위한 알고리즘에 기반하여 상기 데이터 샘플링을 수행하도록 상기 전자 장치를 야기하는 인스트럭션들을 포함하는, 컴퓨터 판독가능 저장매체.
- 청구항 1에 있어서,상기 하나 이상의 프로그램들은, 상기 전자 장치의 프로세서에 의해 실행될 시에,상기 제2 훈련 데이터를 식별하기 위해서 기준 값 이하의 변화량을 갖는 초기 자세, 공격 부위, 공격 범위, 및/또는 공격 세기 중 적어도 하나에 대한 피격 반응 데이터를 제외하기 위한 알고리즘에 기반하여 데이터 샘플링을 수행하는 과정을 포함하도록, 상기 전자 장치를 야기하는 인스트럭션들을 포함하는, 컴퓨터 판독가능 저장매체.
- 청구항 1에 있어서,상기 하나 이상의 프로그램들은, 상기 전자 장치의 프로세서에 의해 실행될 시에,상기 제1 신경망의 출력에 할당되는 가중치와 상기 제2 신경망의 출력에 할당되는 가중치를 결정하는 게이팅 모듈에 기반하여 상기 객체가 피격되는 경우, 상기 객체의 피격 반응 결과를 획득하도록, 상기 전자 장치를 야기하는 인스트럭션들을 포함하는, 컴퓨터 판독가능 저장매체.
- 청구항 1에 있어서,상기 제2 신경망은, 상기 제2 훈련 데이터에서 다른 객체의 공격 동작이 검출되는 경우, 활성화되는, 컴퓨터 판독가능 저장매체.
- 청구항 1에 있어서, 상기 제2 신경망은, 상기 제2 훈련 데이터에서 피격 세기가 기준 값 이상인 경우, 활성화되는, 컴퓨터 판독가능 저장매체.
- 전자 장치(electronic device)에 의해 수행되는 방법에 있어서,객체의 동작들의 학습을 위한 제1 신경망에게 제1 훈련 데이터를 제공하는 동작과,피격 반응 데이터에 대한 데이터 샘플링(data sampling)을 수행함으로써, 제2 훈련 데이터를 식별하는 동작과,상기 객체의 피격 반응의 학습을 위한 제2 신경망에게 상기 제2 훈련 데이터를 제공하는 동작과,상기 제1 신경망의 출력 및 상기 제2 신경망의 출력에 기반하여 상기 객체가 피격되는 경우, 상기 객체의 피격 반응 결과를 획득하는 동작을 포함하는 방법.
- 청구항 8에 있어서,상기 피격 반응 데이터를 생성하는 동작을 포함하고,상기 피격 반응 데이터는 상기 객체에 적용되는 힘에 따른 순간 피격 반응 데이터와 상기 객체의 입력에 따른 후속 피격 반응 데이터에 기반하여 생성되는 방법.
- 청구항 8에 있어서, 상기 제2 훈련 데이터를 식별하는 동작은,기준 값 이하의 변화량을 갖는 피격 반응 데이터를 제외하기 위한 알고리즘에 기반하여 상기 데이터 샘플링을 수행하는 동작을 포함하는 방법.
- 청구항 8에 있어서, 상기 제2 훈련 데이터를 식별하는 동작은,기준 값 이하의 변화량을 갖는 초기 자세, 공격 부위, 공격 범위, 및/또는 공격 세기 중 적어도 하나에 대한 피격 반응 데이터를 제외하기 위한 알고리즘에 기반하여 상기 데이터 샘플링을 수행하는 동작을 포함하는 방법.
- 청구항 8에 있어서,상기 제1 신경망의 출력 및 상기 제2 신경망의 출력에 기반하여 상기 객체가 피격되는 경우, 상기 객체의 피격 반응 결과를 획득하는 동작을 포함하는 방법.
- 청구항 8에 있어서,상기 제2 신경망은, 상기 제2 훈련 데이터에서 다른 객체의 공격 동작이 검출되는 경우, 활성화되는 방법.
- 청구항 8에 있어서,상기 제2 신경망은, 상기 제2 훈련 데이터에서 피격 세기가 기준 값 이상인 경우, 활성화되는 방법.
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020247034938A KR20240169010A (ko) | 2022-05-06 | 2022-05-06 | 훈련된 신경망에 기반한 피격 반응 생성 방법 및 컴퓨터 판독가능 저장 매체 |
PCT/KR2022/006495 WO2023214602A1 (ko) | 2022-05-06 | 2022-05-06 | 훈련된 신경망에 기반한 피격 반응 생성 방법 및 컴퓨터 판독가능 저장 매체 |
US18/936,335 US20250061331A1 (en) | 2022-05-06 | 2024-11-04 | Method and computer-readable storage medium for generating hit reaction on basis of trained neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/KR2022/006495 WO2023214602A1 (ko) | 2022-05-06 | 2022-05-06 | 훈련된 신경망에 기반한 피격 반응 생성 방법 및 컴퓨터 판독가능 저장 매체 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US18/936,335 Continuation US20250061331A1 (en) | 2022-05-06 | 2024-11-04 | Method and computer-readable storage medium for generating hit reaction on basis of trained neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023214602A1 true WO2023214602A1 (ko) | 2023-11-09 |
Family
ID=88646526
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/KR2022/006495 WO2023214602A1 (ko) | 2022-05-06 | 2022-05-06 | 훈련된 신경망에 기반한 피격 반응 생성 방법 및 컴퓨터 판독가능 저장 매체 |
Country Status (3)
Country | Link |
---|---|
US (1) | US20250061331A1 (ko) |
KR (1) | KR20240169010A (ko) |
WO (1) | WO2023214602A1 (ko) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR102173452B1 (ko) * | 2019-07-05 | 2020-11-03 | 넷마블 주식회사 | 데이터 처리 방법 |
KR20210053739A (ko) * | 2019-11-04 | 2021-05-12 | 주식회사 마인즈랩 | 게임 플레이 콘텐츠 제작 장치 |
KR102344804B1 (ko) * | 2021-10-19 | 2021-12-29 | 주식회사 큐로드 | 인공지능 기반 모니터링 기술을 이용한 유저 피드백 정보 관리방법 |
JP2022509485A (ja) * | 2018-10-31 | 2022-01-20 | 株式会社ソニー・インタラクティブエンタテインメント | クロスドメインバッチ正規化を使用したニューラルネットワークにおけるドメイン適応のためのシステム及び方法 |
JP2022510793A (ja) * | 2018-12-14 | 2022-01-28 | バルブ コーポレーション | ビデオゲーム状態を動的に制御するためのプレイヤーバイオフィードバック |
-
2022
- 2022-05-06 KR KR1020247034938A patent/KR20240169010A/ko active Pending
- 2022-05-06 WO PCT/KR2022/006495 patent/WO2023214602A1/ko active Application Filing
-
2024
- 2024-11-04 US US18/936,335 patent/US20250061331A1/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2022509485A (ja) * | 2018-10-31 | 2022-01-20 | 株式会社ソニー・インタラクティブエンタテインメント | クロスドメインバッチ正規化を使用したニューラルネットワークにおけるドメイン適応のためのシステム及び方法 |
JP2022510793A (ja) * | 2018-12-14 | 2022-01-28 | バルブ コーポレーション | ビデオゲーム状態を動的に制御するためのプレイヤーバイオフィードバック |
KR102173452B1 (ko) * | 2019-07-05 | 2020-11-03 | 넷마블 주식회사 | 데이터 처리 방법 |
KR20210053739A (ko) * | 2019-11-04 | 2021-05-12 | 주식회사 마인즈랩 | 게임 플레이 콘텐츠 제작 장치 |
KR102344804B1 (ko) * | 2021-10-19 | 2021-12-29 | 주식회사 큐로드 | 인공지능 기반 모니터링 기술을 이용한 유저 피드백 정보 관리방법 |
Also Published As
Publication number | Publication date |
---|---|
US20250061331A1 (en) | 2025-02-20 |
KR20240169010A (ko) | 2024-12-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109499068B (zh) | 对象的控制方法和装置、存储介质、电子装置 | |
Zhao et al. | Deep reinforcement learning with experience replay based on SARSA | |
WO2022050792A1 (ko) | 테니스 자율훈련 시스템 | |
US9855497B2 (en) | Techniques for providing non-verbal speech recognition in an immersive playtime environment | |
US11836841B2 (en) | Animation video processing method and apparatus, electronic device, and storage medium | |
CN110149553B (zh) | 图像的处理方法和装置、存储介质及电子装置 | |
EP3938065A1 (en) | Systems and methods for training an artificial intelligence model for competition matches | |
CN109960545B (zh) | 虚拟对象控制方法、系统、装置、介质及电子设备 | |
WO2024075958A1 (ko) | 인공지능을 이용한 사용자 수련 수준에 따른 태권도 겨루기 평가 방법 | |
CN110989839B (zh) | 人机对战的系统和方法 | |
EP3186599B1 (en) | Feedback provision system | |
WO2024139307A1 (zh) | 图像的文本纠错方法、装置、电子设备及存储介质 | |
WO2023214602A1 (ko) | 훈련된 신경망에 기반한 피격 반응 생성 방법 및 컴퓨터 판독가능 저장 매체 | |
Yeo et al. | Augmented learning for sports using wearable head-worn and wrist-worn devices | |
CN111273953A (zh) | 模型处理方法、装置、终端及存储介质 | |
US20230360442A1 (en) | Virtual reality user health monitoring | |
CN106390444A (zh) | 球类游戏系统及其控制方法 | |
WO2025023669A1 (ko) | 인공지능을 이용한 음성 인식 기반 태권도 동작 평가 방법 및 장치 | |
WO2023068832A1 (ko) | 사용자 자세에 대한 평가 정보를 제공하기 위한 전자 장치 및 이의 제어 방법 | |
WO2022215874A9 (ko) | 리그 오브 레전드 게임의 전장 내의 챔피언을 자동으로 플레이하는 봇의 행동을 결정하는 방법 및 이를 수행하는 컴퓨팅 시스템 | |
CN113599802B (zh) | 数据处理方法、装置和系统 | |
US20210365122A1 (en) | Device and method for generating haptic data, and device and method for providing haptic effect | |
US20240001246A1 (en) | Online gaming anti-cheat system | |
Morris et al. | Interaction and Learning in a Humanoid Robot Magic Performance. | |
WO2023243754A1 (ko) | 지정된 모션이 캡쳐된 시점을 추출하는 전자 장치 및 방법 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22940843 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 20247034938 Country of ref document: KR Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1020247034938 Country of ref document: KR |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 22940843 Country of ref document: EP Kind code of ref document: A1 |