JP7459328B1 - Game systems, programs, learning devices, and game provision methods - Google Patents

Abstract

[Problem] To provide a game system or the like that allows efficient learning in a fighting game in which the actions of a second character are controlled using learned learning data based on the actions of a first character and related information related to said actions. [Solution] The actions of a first character and related information related to said actions are learned. Based on player information, at least one of the combination of multiple items that make up the related information, the learning frequency, and the weight of each of the multiple items that make up the related information is set before learning. [Selected Figure] Figure 3

Description

The present invention relates to a game system, a program, a learning device, and a game providing method.

Conventionally, there are known games in which a first character and a second character perform offensive and defensive movements in a fighting match in a virtual three-dimensional space.

In recent years, there have been game systems that utilize learning models generated by analyzing and learning data.

For example, in the conventional technology disclosed in Patent Document 1, operation data related to the character's performing technique and screen state data related to the screen display are collected at predetermined timings according to the operation of the input operation unit, and the learning data is collected. The weight of the learning result is optimized by writing into the storage unit and performing deep learning calculation processing based on the operation data and the screen state data. When a first character to be controlled by the player and a second character to be controlled by the computer compete against each other, key data of the inference result is obtained for the second character and added to the key history. Then, the action (or technique) is determined by referring to the key history and command table (see paragraphs 0061 to 0062 and 0077 to 0079 of Patent Document 1, and FIG. 9).

JP 2019-195512 Publication

When multiple characters play a competitive game, the actions of the first character to be operated by the player and related information related to the actions are learned, and the learned data is used to control the actions of the second character. May be inferred. It is preferable that such learning improves efficiency and accuracy.

The present invention has been made in view of the above-mentioned problems, and provides a competitive game in which the actions of a second character are controlled using learned data based on the actions of a first character and related information related to the actions. The purpose is to provide a game system etc. that allows for efficient learning.

(1) The present invention provides
A game system for controlling an action of a second character in a battle game in which a first character, the action of which is controlled based on a player's operation input, and a second character, the second character, are battled against each other, the game system comprising:
a learning unit that learns an action of a first character and related information related to the action;
a behavior control unit that controls a behavior of a second character using the learning data learned by the learning unit;
and a setting unit that sets, based on the player's information, at least one of a combination of multiple items constituting the related information, a learning frequency, and a weight for each of the multiple items constituting the related information before learning.

The present invention also relates to a learning device (terminal device, game device) including the above-mentioned units. The present invention relates to a server device including the above-mentioned parts. The present invention relates to a program that causes a computer to function as each of the above units. The present invention also relates to a computer-readable information storage medium that stores a program that causes the computer to function as each of the above units.

The present invention also provides a server that controls the actions of the second character in a competitive game in which a first character whose actions are controlled based on a player's operation input competes with a second character controlled by a computer. A game providing method for providing information regarding the game from a device to a player's terminal device, the server device transmitting information regarding the player's operation input accepted by the terminal device to the server device, and the server device Regarding the game providing method, including each part.

According to the present invention, at least one of the combination of multiple items constituting related information, the learning frequency, and the weight of each of the multiple items constituting related information is set based on the player's information before learning, allowing efficient learning from an early stage.

"Learning" is, for example, machine learning. Moreover, "learning" may also mean accumulating (memorizing) information.

The "learning data" may be, for example, the learned data itself or a learning model generated by machine learning.

(2) The present invention:
A game system for controlling the actions of a second character in a competitive game in which a first character whose actions are controlled based on a player's operation input competes with a second character controlled by a computer, the game system comprising:
a learning section that learns the actions of the first character and related information related to the actions;
a behavior control unit that controls the behavior of the second character using the learning data learned in the learning unit;
Based on the player's information, at least one of a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information is determined according to the progress of the game. The present invention relates to a game system characterized by including a setting section for making settings.

The present invention also relates to a learning device (terminal device, game device) including the above-mentioned units. The present invention relates to a server device including the above-mentioned parts. The present invention relates to a program that causes a computer to function as each of the above units. The present invention also relates to a computer-readable information storage medium that stores a program that causes the computer to function as each of the above units.

The present invention also provides a server that controls the actions of the second character in a competitive game in which a first character whose actions are controlled based on a player's operation input competes with a second character controlled by a computer. A game providing method for providing information regarding the game from a device to a player's terminal device, the server device transmitting information regarding the player's operation input accepted by the terminal device to the server device, and the server device Regarding the game providing method, including each part.

According to the present invention, based on the player's information and in accordance with the progress of the game, at least the combination of a plurality of items constituting the related information, the learning frequency, and the weight of each of the plurality of items constituting the related information are determined. By setting one, you can study efficiently in real time.

(3) Furthermore, in the game system, learning device, server device, program, information storage medium, and game providing method according to the present invention,
The setting section includes:
Based on the operation input from the player, selection of an item with a higher weight among a plurality of items making up the related information may be accepted. According to the present invention, since the player can select an item with a high weight, it is possible to set a high weight item with emphasis on the player's intention.

(4) Furthermore, in the game system, learning device, server device, program, information storage medium, and game providing method according to the present invention,
The setting section includes:
The priority order of each of the plurality of items of the related information may be set based on the player's operation input. According to the present invention, since the player can set the priority order of each of the plurality of items, the player can set the priority order of each of the plurality of items with emphasis on the player's intention.

(5) Furthermore, in the game system, learning device, server device, program, information storage medium, and game providing method according to the present invention,
The player information may include information regarding the player's play history. According to the present invention, weights can be set in consideration of a player's play history.

(6) Furthermore, in the game system, learning device, server device, program, information storage medium, and game providing method according to the present invention,
The player information may include information regarding the player's past play tendencies. According to the present invention, weights can be set in consideration of the player's past play tendencies.

(7) Furthermore, in the game system, learning device, server device, program, information storage medium, and game providing method according to the present invention,
The player information may include information related to other players determined based on predetermined information about the player. According to the present invention, weights can be set in consideration of information related to other players determined based on predetermined information (for example, level) of the player.

FIG. 1 is a diagram showing an example of the configuration of a game system according to an embodiment of the present invention. FIG. 3 is a diagram showing functional blocks of a server device according to the present embodiment. FIG. 3 is a diagram showing functional blocks of a terminal device according to the present embodiment. FIG. 1 is a diagram showing an example of a controller according to the present embodiment. The figure which shows an example of the game screen of this embodiment. An explanatory diagram of behavior and related information according to the present embodiment. An explanatory diagram for searching for related information according to the present embodiment. FIG. 11 is a diagram showing an example of a command associated with an action according to the embodiment. FIG. 4 is an explanatory diagram of acquisition of behavior information and related information according to the present embodiment. FIG. 3 is a diagram for explaining related information of this embodiment. FIG. 3 is a diagram for explaining a predetermined period of the present embodiment. FIG. 3 is a diagram for explaining a setting screen for a predetermined period according to the present embodiment. FIG. 4 is a diagram for explaining continuous behavior according to the present embodiment. The figure which shows an example of the weight setting screen of this embodiment. The figure which shows an example of the weight setting screen of this embodiment. The figure which shows an example of the weight setting screen of this embodiment. 5 is a flowchart illustrating an example of the flow of processing according to the present embodiment. 5 is a flowchart illustrating an example of the flow of processing according to the present embodiment.

This embodiment will be described below. Note that this embodiment described below does not unduly limit the content of the present invention described in the claims. Furthermore, not all of the configurations described in this embodiment are essential components of the present invention.

[1] Game System First, an overview and a general configuration of the game system of this embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of a system configuration showing the configuration of the game system of this embodiment.

The server device 10 is an information processing device that can provide a given service using a terminal device (game device) 20 that is communicatively connected via the Internet (an example of a network).

The terminal device 20 includes a home video game machine, a personal computer (also referred to as a PC), a smartphone, a mobile phone, a PHS, a computer, a PDA, a portable game machine, an image generating device, an arcade game device, an arcade game device, and a housing. It is an information processing device such as a computer, and is a device that can be connected to the server device 10 via a network such as the Internet (WAN) or LAN. An input section 260 is connected to the terminal device 20, and a display section 290 is also connected thereto.

The terminal device 20 of this embodiment has a learning function (for example, a learning function using machine learning). The terminal device 20 may also be referred to as a learning device. Furthermore, the terminal device 20 has an inference function (prediction function) based on machine learning. The terminal device 20 may also be referred to as an inference device (prediction device).

[1.1] Game system configured in a client-server model As shown in FIG. 1, the game system of this embodiment is configured such that a server device 10 and terminal devices 20 (e.g., terminal devices 20A, 20B, 20C) are connectable to the Internet (an example of a network).

By accessing the server device 10 from the terminal device 20, the player can play the game based on information transmitted from the server device 10 via the Internet.

In particular, in this embodiment, the server device 10, which executes a game in which a first character and a second character play against each other, provides information about the game to the player's terminal device 20. Information about the player's operation input accepted by the terminal device 20 is transmitted to the server device 10. Information about the game is transmitted from the server device 10 to the terminal device 20.

In this embodiment, each game may be provided to the terminal device 20 by one server device 10, or a server system may be constructed by linking a plurality of server devices 10, and each game may be provided to the terminal device 20. Good too.

Note that the server device 10 alone may perform various processes related to the present invention, or the terminal device 20 alone may perform various processes related to the present invention.

[1.2] Terminal Device The game system of this embodiment may be realized by a single terminal device having the functions of the server device 10, that is, a device that operates independently (standalone) without relying on other devices such as a server device.

Further, in this embodiment, the present invention may be implemented only by the terminal device 20 without being connected to the server device 10. For example, it may be a game system realized by a plurality of terminal devices 20 through P2P (peer-to-peer) communication.

The system may be realized by connecting multiple such terminal devices by wire or wirelessly, with one terminal device functioning as a host (server device 10).

Note that the terminal device may be not only a terminal device but also a tablet type information terminal device, a personal computer, or a terminal device (housing) installed in an amusement park.

[1.3] Cloud-based Game System The game system of this embodiment may be a cloud-based game system. For example, in a cloud-based game system, basic processing (including game processing, drawing processing, etc.) is performed only by the server device 10 (cloud device), and the terminal device 20 only controls the display of the processing results of the server device 10 (cloud device). The terminal device 20 transmits information such as operation input (operation information, input information) of the controller to the server device 10. In this case, the terminal device 20 may transmit information other than the operation input.

Furthermore, in this embodiment, the server device 10 executes each function of the processing unit 200 of the terminal device 20 described above and the game program, except for input, and the terminal device 20 executes input and image display by streaming. The above game may be realized by doing the following.

Further, the game system may store information stored in the server device 10 and information stored in the terminal device 20 in a given storage area on the Internet.

[1.4] Terminal device including multiple controllers In the present embodiment, one terminal device 20 may include multiple controllers (input section 260). Note that when a plurality of players play a game using the controllers of each player of one terminal device 20, it is assumed that the plurality of players are realistically close to each other.

[1.5] Example of Social Game Additionally, in this embodiment, the server may function as an SNS server that provides communication-type services. Here, the SNS server may be an information processing device that provides a service that allows communication between multiple players.

In addition, in this embodiment, for example, when functioning as an SNS server, it is possible to provide games known as social games that are executed using the operating environment (API (Application Programming Interface), platform, etc.) of the SNS provided.

Note that, unlike existing online games, social games include games that do not require dedicated client software and can be used only with a web browser and SNS account. Furthermore, this embodiment has a configuration that can provide an online game in which the player can connect to the terminal device 20 of another player via a network and share the same game progress online at the same time.

[1.6] Examples of browser games In particular, in this embodiment, games provided on the web browser of the terminal device 20, such as HTML, FLASH (registered trademark), CGI, PHP, shockwave, Java (registered trademark) applets Browser games (games that can be started simply by opening an installation site with a Web browser) created in various languages such as , JavaScript (registered trademark), etc. may be provided.

The terminal device 20 also has a web browser capable of viewing web pages (data in HTML format). That is, the terminal device 20 has a communication control function for communicating with the server device 10, and a web browser function for controlling display using data received from the server device 10 (web data, data created in HTML format, etc.), as well as transmitting data of player operation to the server device 10, and executes various processes for providing a screen to the player and allowing the player to play the game. However, the terminal device 20 may also acquire game control information provided by the server device 10, execute a predetermined game process, and play a game based on the game process.

Specifically, when the terminal device 20 makes a request to the server device 10 to play a predetermined game, it is connected to the game site of the server device 10 and starts the game. In particular, the terminal device 20 uses the API as necessary to cause the server device 10 functioning as an SNS server to perform predetermined processing, or to perform player information etc. managed by the server device 10 functioning as an SNS server. It may also have a configuration in which the game is executed by causing the player to acquire the following information.

[1.7] Others This embodiment may be configured with one (device, processor), or may be configured with a plurality of (devices, processors). Information (for example, player information, game information, etc.) stored in the storage area (storage unit 170 described later) of the server device 10 is stored in a database (in a broad sense, a storage device) connected via a network (intranet or Internet). , memory). Note that the communication line between the terminal device 20 and the server device 10 may be wired or wireless.

[2] Server Device Next, the server device 10 of this embodiment will be described using FIG. 2. Note that FIG. 2 is a diagram showing functional blocks of the server device 10 of this embodiment. Further, in this embodiment of the present invention, some of the constituent elements (each part) in FIG. 2 may be omitted.

In this embodiment, an input section 160 used for administrator and other input, an information storage medium 180 in which predetermined information is stored, a communication section 196 for communicating with the terminal device 20 and others, and a communication section 196 mainly related to the game to be provided. It includes a processing section 100 that executes processing, and a storage section 170 that stores various data mainly used for games.

The input unit 160 is used by a system administrator or the like to input game-related settings and other necessary settings and data. For example, the input unit 160 of this embodiment is configured with a mouse, a keyboard, or the like.

The information storage medium 180 (a computer-readable medium) stores programs, data, etc., and its functions are realized by an optical disk (CD, DVD), a magneto-optical disk (MO), a magnetic disk, a hard disk, a magnetic tape, or a memory (ROM), etc.

The communication unit 196 performs various controls for communicating with the outside (for example, terminals, other servers, and other network systems), and its functions are performed by hardware such as various processors or communication ASICs. It consists of software, programs, etc.

The storage unit 170 serves as a work area for the processing unit 100, the communication unit 196, and the like, and its functions are configured by RAM (VRAM) and the like. Note that the information stored in the storage unit 170 may be managed in a database.

In addition to the main storage section 171, the storage section 170 of this embodiment includes a game information storage section 174 in which game information indicating information about the game is stored, and a player information storage section (user information storage section) 176. have.

In particular, the game information storage unit 174 stores information on a virtual space (game field) in which the game is played, information on a virtual camera, setting values used in the game, and the like.

Further, the player information storage unit 176 stores player information (user information) for each player (for each user) in association with the player ID (user ID). The player ID is identification information for identifying the player.

The processing unit 100 performs various processes using the main storage unit 171 within the storage unit 170 as a work area. The functions of the processing unit 100 can be realized by hardware such as various processors (CPU, DSP, etc.), ASIC (gate array, etc.), and programs.

The processing unit 100 performs various processes of this embodiment based on programs (data) stored in the information storage medium 180. That is, the information storage medium 180 stores a program for making the computer function as each part of this embodiment (a program for causing the computer to execute the processing of each part).

For example, the processing unit 100 (processor) controls the entire server device 10 based on a program stored in the information storage medium 180, and performs various processes such as controlling the exchange of data etc. between each unit. . Furthermore, processing for providing various services in response to requests from the terminal device 20 is performed.

Note that in this embodiment, the server device 10 may perform some or all of the processing of the processing section 100, or the terminal device 20 may perform some of the processing of the processing section 100.

The processing unit 100 includes a game processing unit 111, a display control unit 112, a reception unit 119, a communication control unit 120, a Web processing unit 121, a management unit 122, and a notification unit 123.

The game processing unit 111 performs various controls for providing the game (for example, a fighting game) of this embodiment. For example, the processing unit 100 works in conjunction with the terminal device 20, and executes game processing of the game provided in this embodiment in each player based on player input input via the terminal device 20.

Further, the processing unit 100 may have a timer function and may be used to synchronize with each terminal device 20 in order to manage the progress of the game. In particular, the current time or a preset time may be output to each section.

The game processing unit 111 executes game processing based on the player's operation input (operation information). For example, the game processing unit 111 may receive the player's operation input (operation information) from the player's terminal device 20 and perform game processing. Furthermore, during online competition, matching processing with each player is performed.

The display control unit 112 controls the screen (image) generated by the server device 10 according to the game situation to be displayed on the display unit 290 of the terminal device 20. In other words, the screen (image) generated by the server device 10 is displayed on the display unit 290 of the terminal device 20 .

The display control unit 112 controls the display of the screen (game images, presentation images) on the player's terminal device 20 for each player.

The accepting unit 119 may accept information regarding the game. For example, the receiving unit 119 may receive and accept the contents of the operation input from the terminal device 20.

Further, the receiving unit 119 may receive player input, information regarding the game, etc. through data transmission and reception by the communication control unit 120.

The communication control unit 120 establishes a connection (session or connection) with the terminal device 20 and performs processing for communicating (sending and receiving) data via the network.

For example, the communication control unit 120 transmits information regarding the game to the player's terminal device 20. Further, the communication control unit 120 receives, from the player's terminal device 20, the player's operation information accepted by the terminal device 20. Further, the communication control unit 120 transmits information regarding the game to the terminal device 20.

The web processing unit 121 functions as a web server. For example, the web processing unit 121 performs a process of transmitting data in response to a request from a web browser installed in the terminal device 20 through a communication protocol such as HTTP (Hypertext Transfer Protocol), and a process of receiving data transmitted by the web browser of the terminal device 20.

The management unit 122 manages game information and player information associated with player identification information for each player.

The notification unit 123 also notifies the player of given information. Note that "notifying" means notifying the player (the player's terminal device 20) of information. Note that "notifying" may also be interpreted as providing, presenting, displaying, or transmitting.

The notification unit 123 may perform transmission control so as to notify the screen on the terminal device 20. That is, the notification unit 123 generates display control information for the screen (image), and controls the generated display control information to be displayed on the player's terminal device 20 (for example, the display unit 290 of the terminal device 20).

Note that the server device 10 may have all or part of the functions of the terminal device 20 described below.

[3] Terminal Device Next, the terminal device 20 of this embodiment will be described using FIG. 3. Note that FIG. 3 is an example of a functional block diagram showing the configuration of the terminal device 20 in this embodiment. Further, the terminal device 20 of this embodiment may have a configuration in which some of the components (each unit) in FIG. 3 are omitted.

The information storage medium 280 (computer readable medium) stores programs, data, etc., and its functions include optical disks (CD, DVD), magneto-optical disks (MO), magnetic disks, hard disks, and magnetic tapes. , or a memory (ROM).

The input unit 260 includes buttons and levers through which operation inputs (operation signals) can be input. FIG. 4 is an external view of the controller 30 as an example of the input section 260. Note that the controller 30 may be the same controller used for arcade games at game centers.

The terminal device 20 reads the game program or game information from the storage unit 270 or the information storage medium 280, or the controller in order to make the player enjoy the fighting game while viewing the game screen displayed on the display unit 290. The fighting game is executed based on the operation input etc. input from 30.

The controller 30 is equipped with a cross-shaped direction switch 31 having four push points, and four push buttons 33 (33a to 33d) arranged in a cross shape (in four directions: up, down, left, and right).

Further, an analog joystick (hereinafter abbreviated as "joystick") 32 and a joystick 34 are attached to the controller 30 so as to be tiltable.

The direction switch 31, the joystick 32, the push button 33, and the joystick 34 are used for operations for instructing the moving direction of the character or specifying the type of technique. Specifically, the direction switch 31 has push points in four directions (up, down, left, and right), and by pressing any of the push points in the up, down, left, and right directions, input for instructing any of the up, down, left, and right directions is performed. It is something. Furthermore, the joystick 32 and the joystick 34 are arranged in four directions: up, down, left and right, and diagonally upper right (or upper right), diagonally lower right (or lower right), diagonally lower left (or lower left), and diagonally upper left, which are in the middle of the four directions. Inputs can be made in eight directions, including the four directions (or upper left).

When specifying eight directions using the direction switch 31, press the top and right touch points at the same time to select "diagonally up right", press the right and bottom touch points simultaneously to select "diagonally right down", and press the right and bottom touch points simultaneously to select "diagonally right down" It is also possible to instruct the respective inputs (in four diagonal directions), such as "diagonally down left" by pressing the touch points at the same time, and "diagonally up left" by pushing the left and upper touch points simultaneously.

The pushbuttons 33a to 33d may be marked with symbols such as △〇×□ on their upper surfaces. The pushbuttons 33a to 33d are used to instruct the character's actions (eg, attack) in the game. For example, in this embodiment, the functions of right punch, right kick, left kick, and left punch are assigned to the pushbuttons 33a to 33d. For example, when playing a game, the player operates the direction switch 31 or joystick 32 with the thumb of the left hand, and operates the push buttons 33a to 33d or the joystick 34 with the thumb of the right hand.

Note that the input unit 260 may be realized by a touch panel, a touch panel type display, or the like. That is, the input unit 260 includes a detection unit 262 that can detect two-dimensional designated position coordinates (x, y) on the screen where an image is displayed. For example, the input unit 260 includes a detection unit 262 that can detect two-dimensional contact position coordinates (x, y) in a contact detection area (touch panel). Note that the touch operation on the display screen (hereinafter referred to as the "touch panel" except in special cases) may be performed using a fingertip, or may be performed using an input device such as a touch pen. .

Further, the input unit 260 may include a keyboard, a steering wheel, a microphone, an acceleration sensor, and the like.

The memory unit 270 serves as a work area for the memory unit 200 and the communication unit 296, and its functions can be realized by a RAM (VRAM) or the like.

3 is a diagram illustrating an example of a storage area of a storage unit 270. FIG. The storage unit 270 of this embodiment includes a main storage unit 271, an image buffer 272, a game information storage unit 274, a player information storage unit 276, a learning storage unit 278, and a learning model storage unit 279. Note that some of these may be omitted, or the storage unit 170 of the server device 10 or a given storage device (storage area on the cloud) connected via a network may constitute part of the storage section. Good too.

Main storage section 271 is used as a work area. The image buffer 272 stores image data (display image) generated by the image generation unit 230. Note that image data generated by the server device 10 may be received and the image data may be stored in the image buffer 272.

Game information is stored in the game information storage section 274. Note that the terminal device 20 may receive game information from the server device 10 and store the game information in the game information storage section 274.

Game information includes command information and character information. The game information stored in the game information storage unit 274 may include character behavior information and related information related to the behavior information. Further, the game information stored in the game information storage unit 274 may include information on the game field (game space), information on the virtual camera, and other setting values used in the game.

Command information is prepared for each character. Command information is predefined key data. In other words, the command information indicates the key to be inputted from the controller 30 (if there are multiple keys to be inputted, the order in which the keys are inputted), which corresponds to the action.

Character information is information regarding a character. For example, the information includes parameters of the character (parameters such as physical strength, attack power, defense power, etc.), items (weapons) possessed by the character, level (grade, rank), and the like. The level indicates the skill level and strength of the character. In this embodiment, the level is sequentially updated according to game results.

The player information storage section 276 stores player information. Note that the terminal device 20 may receive player information corresponding to the player ID from the server device 10 and store it in the player information storage section 276.

For example, the player information storage unit 276 stores the player's level, play history, play tendency, information on other players related to the player, and the like.

The player information may include the operation input (operation information) of the player. Furthermore, the player information may be a concept including not only the player information associated with the player, but also information on the player's game media (e.g., game media such as characters, cards, and items owned by the player), etc.

The learning storage unit 278 stores information (for example, learning data, teacher data, weight data, various parameters) for learning (for example, machine learning).

Machine learning is a learning method that aims to acquire latent patterns from input and output values. Neural networks (deep learning) are one framework used in machine learning. A neural network is expressed using a mathematical model (function) called an artificial neuron. Machine learning in this embodiment is not limited to neural networks, and may be other algorithms (logistic regression, decision tree, boosting, random forest, support vector machine, etc.).

The learning model storage unit 279 stores learning models. A learning model is a formula (function) that expresses the correlation and tendency between input data and output data. The learning model storage unit 279 stores a learning model that has been subjected to machine learning in the learning unit 218.

Regarding the information storage medium 280, the information storage medium 280 can store programs for causing a computer to function as each part of this embodiment (programs for causing a computer to execute the processing of each part). Note that the processing unit 200 performs various processes of this embodiment based on the programs (data) stored in the information storage medium 280, as described below.

The display unit 290 outputs the image generated according to the present embodiment, and its function can be realized by a CRT, LCD, touch panel display, HMD (head mounted display), or the like.

In particular, in this embodiment, the display section 290 also functions as an input section 260 through which the player operates the game by using a touch panel display. Here, as the touch panel, for example, a resistive film type (4-wire type, 5-wire type), a capacitance type, an electromagnetic induction type, an ultrasonic surface acoustic wave type, an infrared scanning type, or the like can be used.

The sound output unit 292 outputs the sound generated according to this embodiment, and its function can be realized by a speaker, headphones, or the like.

The communication unit 296 performs various controls for communicating with the outside (for example, the server device 10 and other terminal devices 20), and its functions are performed by hardware such as various processors or communication ASICs. This can be achieved by using a program, etc.

Note that the terminal device 20 receives programs and data stored in the information storage medium 180 and the storage unit 170 of the server device 10 for making the computer function as each part of the present embodiment via the network. The programs and data may be stored in the information storage medium 280 or the storage unit 270. A case where the terminal device 20 functions by receiving programs and data in this manner can also be included within the scope of the present invention.

The processing unit 200 (processor) performs processing such as game processing, image generation processing, sound generation processing, etc. in conjunction with the server device 10 based on input data and programs from the input unit 260.

In particular, in this embodiment, the game processing includes processing for starting the game when game start conditions are met, processing for advancing the game, processing for arranging objects such as characters and enemy characters, and processing for displaying objects. , processing for calculating game results, or processing for ending the game when game ending conditions are met.

Further, the processing unit 200 performs game processing (for example, processing of a competitive game, etc.) based on the player's operation input input through the input unit 260. Further, the processing unit 200 may be executed in conjunction with the server device 10, or a part or all of it may be formed in the server device 10.

Furthermore, the processing unit 200 performs various processes using the storage unit 270 as a work area. The functions of the processing unit 200 can be realized by hardware such as various processors (CPU, DSP, etc.), ASIC (gate array, etc.), and programs.

The processing section 200 includes a communication control section 211, an object space setting section 212, a movement/motion control section 213, an operation input reception section 214, a game control section 215, a virtual camera control section 216, an action control section 217, a learning section 218, and a setting section. section 219 , determination section 220 , search section 221 , display control section 222 , change section 223 , image generation section 230 , and sound processing section 240 . Note that a configuration may be adopted in which some of these are omitted.

The communication control unit 211 performs processing for transmitting and receiving data with each of the server devices 10. The communication control unit 211 also performs processing for storing data received from the server device 10 in the storage unit 270, processing for analyzing the received data, and other control processing related to transmission and reception of data.

The communication control unit 211 transmits information from the server device 10 or other terminal devices 20 when executing a competitive game based on information received from the server device 10 or when competing against other players via a network. It receives various types of information and transmits them to the server device 10 and other terminal devices 20.

The communication control unit 211 may be configured to communicate with another terminal device 20 or the server device 10 when it receives an operation input from the player to start communication.

Furthermore, the communication control unit 211 performs a process of mutually transmitting and receiving data at a predetermined cycle from the time a communication connection is established with another terminal device 20 or the server device 10 until the communication connection is disconnected.

The communication control unit 211 performs a process of transmitting player identification information and operation input to the server device 10 and receiving data (player's web page, screen, etc.) from the server device 10 .

The communication control unit 211 may transmit and receive data to and from the server device 10 at a predetermined interval, or may transmit and receive data to and from the server device 10 when an operation input is received from the input unit 260. In particular, the communication control unit 211 of this embodiment may perform a process of receiving given screen information (e.g., screen information, etc.) from the server device 10.

Furthermore, in the case of a network system composed of multiple terminal devices 20, the communication control unit 211 may perform communication control using a peer-to-peer (so-called P2P) method in which an online game is played while sending and receiving data between the multiple terminal devices 20, or may perform communication control using a client-server method in which each terminal device 20 plays an online game while sending and receiving data (information) via the server device 10.
In the game system of this embodiment, data may be transmitted and received not only by wired communication but also by wireless communication.

The object space setting unit 212 performs processing for arranging objects in an object space (virtual three-dimensional space). For example, a process is performed to form a game space (for example, a stage for characters to fight) in a virtual three-dimensional virtual space. In addition to characters, display objects such as buildings, stadiums, cars, trees, pillars, walls, and maps (terrain) are placed in the object space. Here, the object space is a virtual game space, and is a space in which objects are arranged in three-dimensional coordinates (X, Y, Z), such as a world coordinate system or a virtual camera coordinate system, for example.

For example, the object space setting unit 212 arranges an object (an object composed of primitives such as a polygon, a free-form surface, or a subdivision surface) in the world coordinate system. Also, for example, you can determine the position and rotation angle (synonymous with orientation and direction) of an object in the world coordinate system, and add the rotation angle (rotation around the X, Y, and Z axes) to that position (X, Y, Z). position the object at an angle).

The object space setting unit 212 may perform a process of arranging a special command input instruction object, marker, or other instruction object on the screen (two-dimensional image, screen, screen coordinate system). Further, the object space setting unit 212 may arrange the instruction object in an object space (three-dimensional space, world coordinate system, virtual camera coordinate system, model coordinate system).

The movement/motion control unit 213 performs movement/motion calculations for objects in the object space. That is, based on the operation input received from the controller 30, the program (movement/motion algorithm), various data (motion data), etc., the object is moved within the object space, or the object is made to move (motion, animation). Perform the processing to do.

The movement/motion control unit 213 controls the behavior of the character object (character), which is the main object, in the object space.

Specifically, the movement/motion control unit 213 performs a process of controlling the motion of a player character to be operated by the player who operates the player's own machine, based on key data based on the operation input to the controller 30, Performs hit determination processing for attack techniques, damage processing when attacked by an opponent character, etc.

When the opponent is a real player, the movement/motion control unit 213 acquires the key data input on the opponent's terminal device 20, and controls the movement/motion of the attack technique performed by the opponent character based on the acquired key data. It performs actions, hit determination processing for the attack technique, and damage movement/action processing when attacked by the player character.

When the opponent is a computer (CPU), the movement/movement control unit 213 controls the movement/movement of attack techniques performed by a computer character (also referred to as an NPC (non-player character)) based on a predetermined program or algorithm. Performs hit determination processing for attack techniques, damage movement and movement processing when attacked by the player character, etc. That is, when the opponent is a computer, the movements and actions of the computer character are controlled without being operated by the player.

The movement/motion control unit 213 controls the computer character to
In the case of an AI character controlled by AI (abbreviation for Intelligence), the character moves and moves based on actions inferred using a learning model.

Furthermore, the movement/motion control unit 213 performs such processing every frame (for example, 1/60 seconds). Note that a frame is a unit of time for performing object movement/motion processing and image generation processing.

The operation input reception unit 214 receives operation instruction commands based on operation inputs input to the controller 30 by the player.

Specifically, the operation input reception unit 214 of this embodiment detects the operation instruction command input by the player based on the type of key, the direction of the key instruction, the number of times the key is pressed, the press time, the press timing, and a combination of these, input by the input unit 260.

The game control unit 215 executes processing related to the progress of the competitive game. For example, the process of starting the game when game start conditions are met (hereinafter also referred to as "game start process"), and controlling the player character in response to the player's operation input to control the progress of the competitive game. There are game processing, processing for ending the game when game ending conditions are met (hereinafter also referred to as "game ending processing"), and processing for advancing the ending when the final stage is cleared.

When a player plays against a computer, the game control unit 215 uses information on the player's operation input accepted by the operation input reception unit 214 of the terminal device 20 and information controlled by the computer (for example, determined by the behavior control unit 217). (including information on the actions taken), the game progresses.

When real players play against each other, the game control unit 215 receives information on the first player's operation input received by the operation input reception unit 214 of the terminal device 20A, and information on the operation input of the second player transmitted from the terminal device 20B. The game progresses in accordance with the information of the operation input by the second player on the terminal device 20B.

The game control unit 215 causes the first character OB1 and the second character OB2 to attack and defend against each other, thereby reducing the physical strength values of both characters. Then, the winner or loser is determined based on the physical strength values of both players. For example, during one round of the game, the player whose physical strength reaches a predetermined value (0) first becomes the loser and the other player becomes the winner. At the end of the game period of one round, if both players have a physical strength value of 1 or more, the one with the higher physical strength value is determined to be the winner, and the other player is determined to be the loser.

The virtual camera control unit 216 performs control processing of a virtual camera (viewpoint) to generate an image that can be seen from a given (arbitrary) viewpoint in the object space.

Specifically, when generating a three-dimensional image, the virtual camera control unit 216 controls the position (X, Y, Z) and rotation angle (for example, X, Y, Z axes) of the virtual camera in the world coordinate system. (rotation angle when rotating clockwise when viewed from the positive direction of each axis).

That is, the virtual camera control unit 216 performs processing to control at least one of the viewpoint position, line of sight direction, angle of view, moving direction, and moving speed of the virtual camera.

The behavior control unit 217 performs processing to control the behavior of the character. For example, the behavior control unit 217 controls the behavior of a character to be operated by the player (hereinafter also referred to as "first character") based on the player's operation input.

The behavior control unit 217 also uses learning data (for example, a learning model) learned (for example, machine learning) in the learning unit 218 to create a computer character (hereinafter also referred to as a “second character”) that is controlled by the computer. ) to control their behavior.

The behavior control unit 217 may control the behavior of the second character based on the related information searched by the search unit 221.

"Action" may be one action, or may include continuous action based on a plurality of actions that satisfy the continuation condition.

The behavior control unit 217 may predefine, for each type of behavior of the first character, a behavior of the second character that corresponds to that behavior.

The behavior control unit 217 may control the behavior of the second character using at least one of data defining the behavior of the second character and learning data. The number of predefined actions of the second character may be one or more.

For example, if the matching degree of the searched related information is not greater than or equal to a predetermined value, the behavior control unit 217 may change the predefined behavior corresponding to the detected unlearned behavior of the first character to a second one. You may also have the character do it.

In addition, when the degree of matching of the searched related information is not equal to or higher than a predetermined value, the behavior control unit 217 selects one of a plurality of predefined behaviors corresponding to the detected unlearned behavior of the first character. , one action may be selected under predetermined conditions, and the second character may be made to perform the selected action.

The learning unit 218 learns (for example, machine learning) the behavior of the first character and related information related to the behavior. For example, when the terminal device 20 receives related information, it generates a learning model that outputs a character's behavior (behavior information) by machine learning.

The setting unit 219 sets a predetermined period during the game period as a period in which learning (for example, machine learning) is performed frequently. For example, the setting unit 219 may set a predetermined period in the start period of the game period. Further, the setting unit 219 may be configured to be able to set the predetermined period based on the player's operation input.

The setting unit 219 may set a combination of a plurality of items that constitute the related information. Note that "item" may also be interpreted as an element or type.

Based on player information, the setting unit 219 determines, before learning (for example, before machine learning), a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information. , at least one of the following may be set.

Furthermore, the setting unit 219 determines the combination of the plurality of items forming the related information, the learning frequency, and the weight of each of the plurality of items forming the related information according to the game progress based on the player's information. At least one may be set.

The setting unit 219 may set weights for each of a plurality of items forming the related information according to the player's information.

The setting unit 219 may set the weight of each item forming the related information according to the information of the first character.

The setting unit 219 may set the weight of each item constituting the related information according to the game information.

The setting unit 219 may accept a selection of a high-weighted item from among multiple items constituting the related information based on the player's operation input. The setting unit 219 may set a priority order for each of the multiple items of the related information based on the player's operation input. The player information of the setting unit 219 may include information on the play history of the player. The player information may also include information on the player's past play tendencies. The player information may also include information related to other players that is determined based on predetermined information (e.g., level) of the player.

The determination unit 220 determines whether an unlearned action of the first character and related information of the action have been detected.

When the search unit 221 detects an unlearned behavior of the first character and related information of the behavior, the search unit 221 selects the detected unlearned behavior from among the related information related to the learned behavior of the first character. Search for related information that has a high degree of matching with related information related to the behavior of the first character.

That is, when the search unit 221 detects an unlearned behavior of the first character and related information of the behavior, the search unit 221 selects the detected unlearned behavior from among the related information related to the learned behavior of the first character. Search for related information that is the same as related information related to the behavior of the first character in learning. If there is no related information related to the behavior of the learned first character that is the same as the detected related information related to the behavior of the unlearned first character, the most similar related information Explore information.

When the search unit 221 detects a continuous action of an unlearned first character and related information related to the continuous action, the search unit 221 detects the detected continuous action from among the related information related to the continuous action of the learned first character. The related information that has a high degree of matching with the related information related to the continuous actions of the unlearned first character is searched.

The display control unit 222 performs processing to display a screen on the display unit 290. For example, the terminal device 20 displays the generated image on the display unit 290.

The display control unit 222 displays the screen display information received from the server device 10. Note that the display control unit 222 may display display information and screens using a web browser.

For example, the display control unit 222 displays on the game screen the character name, level (rank), and stamina value of the current first character OB1, and the character name, level (rank), and stamina value of the second character OB2.

The changing unit 223 changes the weighting period based on the player's information or the first character's information.

The image generation unit 230 performs drawing processing based on the results of various processes performed by the processing unit 200, thereby generating an image and outputting it to the display unit 290.

That is, the image generation unit 230 generates an image visible from the virtual camera in the object space.

For example, the image generation unit 230 receives object data (model data) including vertex data (vertex position coordinates, texture coordinates, color data, normal vector, α value, etc.) of each vertex of the object (model), Vertex processing (shading by a vertex shader) is performed based on vertex data included in the input object data. Note that when performing vertex processing, vertex generation processing (tessellation, curved surface division, polygon division) for redividing polygons may be performed as necessary.

In vertex processing, according to the vertex processing program (vertex shader program, first shader program), vertex movement processing, coordinate transformation, such as world coordinate transformation, field of view transformation (camera coordinate transformation), clipping processing, perspective transformation (projection transformation) ), geometry processing such as viewport conversion is performed, and based on the processing results, vertex data given for a group of vertices forming the object is changed (updated, adjusted).

Then, rasterization (scan conversion) is performed based on the vertex data after vertex processing, and the faces of polygons (primitives) and pixels are associated. Then, following rasterization, pixel processing (shading by a pixel shader, fragment processing) is performed to draw pixels that make up the image (fragments that make up the display screen).

In pixel processing, various processes such as reading textures (texture mapping), setting/changing color data, semi-transparent compositing, and anti-aliasing are performed according to a pixel processing program (pixel shader program, second shader program) to determine the final drawing color of the pixels that make up the image, and the drawing color of the perspective transformed object is output (drawn) to the image buffer 272 (a buffer that can store image information on a pixel-by-pixel basis; VRAM, rendering target, frame buffer).

That is, pixel processing involves per-pixel processing, where image information (color, normal, brightness, alpha value, etc.) is set or changed on a pixel-by-pixel basis. This generates an image that can be seen from a virtual camera (a given viewpoint) in object space.

Note that vertex processing and pixel processing are realized by hardware that allows drawing processing of polygons (primitives) to be programmed using a shader program written in a shading language, a so-called programmable shader (vertex shader or pixel shader).

Programmable shaders have a high degree of freedom in the content of drawing processing because they allow processing on a vertex-by-vertex and pixel-by-pixel basis, greatly improving expressiveness compared to fixed drawing processing using conventional hardware. I can do it.

The image generation unit 230 performs geometry processing, texture mapping, hidden surface removal processing, α-blending, etc. when drawing the object.

In the geometry processing, processing such as coordinate transformation, clipping processing, perspective projection transformation, or light source calculation is performed on the object. Then, object data (position coordinates of object vertices, texture coordinates, color data (luminance data), normal vector, or α value, etc.) after geometry processing (perspective projection transformation) is stored in the storage unit 270. .

Texture mapping is a process for mapping a texture (texel value) stored in the storage unit 270 to an object.

Specifically, the texture (surface properties such as color (RGB), alpha value, etc.) is read from the storage unit 270 using texture coordinates and the like set (assigned) to the vertices of the object. Then, the texture, which is a two-dimensional image, is mapped onto the object. In this case, a process is performed to associate pixels with texels, and bilinear interpolation is performed to interpolate the texels.

Hidden surface removal processing can be performed using a Z buffer method (depth comparison method, Z test) using a Z buffer (depth buffer) in which the Z value (depth information) of a drawn pixel is stored. .

That is, when drawing a drawing pixel corresponding to a primitive of an object, the Z value stored in the Z buffer is referred to. Then, the Z value of the referenced Z buffer is compared with the Z value of the drawing pixel of the primitive, and the Z value of the drawing pixel is the Z value that is on the near side as seen from the virtual camera (for example, a small Z value). If so, the drawing process for that drawing pixel is performed and the Z value of the Z buffer is updated to a new Z value.

α blending (α synthesis) is a translucent synthesis process (normal α blending, additive α blending, subtractive α blending, etc.) based on the α value (A value).

For example, in alpha blending, a linear blending process is performed based on the alpha value on the drawing color (overwrite color) C1 that will be drawn in the image buffer 272 and the drawing color (base color) C2 that has already been drawn in the image buffer 272 (rendering target). In other words, if the final drawing color is C, it can be calculated as C = C1 * alpha + C2 * (1 - alpha).

Note that the α value is information that can be stored in association with each pixel (texel, dot), and is, for example, plus alpha information other than color information. The α value can be used as mask information, translucency (equivalent to transparency or opacity), bump information, etc.

When playing a multiplayer online game in which data is sent and received over a network with another terminal (second terminal), the image generation unit 230 performs processing to generate an image seen from a virtual camera (a virtual camera controlled by the terminal (first terminal)) that follows the movement of an object that is the subject of operation of the terminal (first terminal). In other words, each terminal performs independent drawing processing.

The sound processing unit 240 performs sound processing based on the results of various processes performed by the processing unit 200, generates game sounds such as BGM, sound effects, or voices, and outputs them to the sound output unit 292. Note that the sound processing unit 240 may perform the same processing as the sound processing unit 140 of the server device 10 in whole or in part.

Note that the terminal device 20 of this embodiment may be controlled so that the game can be played in a single player mode in which only one player can play, or in a multiplayer mode in which a plurality of players can play. For example, when controlling in multi-player mode, game processing may be performed by transmitting and receiving data to and from other terminal devices 20 via a network, or one terminal device 20 may receive and receive data from multiple input units. The process may be performed based on the operation input.

When the terminal device 20 of this embodiment plays a game with a plurality of terminal devices 20 using P2P or the like, one terminal device 20 may play the role of a host and execute the process. In such a case, the terminal device 20 as a host may perform the processing of each processing section (a part or all of each processing section) of the server device 10.

Further, the terminal device 20 of this embodiment may perform the same processing as the server device 10 and display a screen on the display unit of the terminal device 20.

[4] Overview The terminal device (learning device) 20 of this embodiment is a device that can play fighting games. In the present embodiment, for convenience of explanation, an example will be described in which the player P1 plays a game on the terminal device 20.

FIG. 5 shows an example of a game screen displayed on the terminal device 20 of the player P1. In the game, the character OB1 (also referred to as the first character OB1) operated by the player P1 and the opponent character OB2 (also referred to as the second character OB2) attack and defend each other, and each other's physical strength It is a fighting game that determines victory or defeat based on physical strength.

In this embodiment, a practice mode game is provided in which player P1 can practice a fighting game by fighting against a non-existent virtual opponent. The virtual opponent is, for example, a computer character controlled by a computer (CPU). The computer character in this embodiment may also be called an AI character, since its behavior information is determined by AI.

As shown in FIG. 6, the terminal device 20 of the present embodiment includes information about the action of the first character OB1 (for example, right punch) and information related to the action (for example, when the first character OB1 executes the right punch). and the distance to the second character OB2) as learning data (for example, teacher data), and a learning model is generated by machine learning using the learning data.

In other words, the terminal device 20 performs machine learning of the "behavior of the first character OB1" and "related information related to the behavior of the first character OB1."

In this embodiment, the behavior of the second character OB2 is controlled using the learned learning model. In other words, the second character OB2 reproduces the operation mode of the player P1.

For example, when the first character OB1 performs a "right punch", the terminal device 20 performs machine learning on the action information of the "right punch" and the related information of the "right punch" as a set.

After machine learning, the terminal device 20 controls the second character OB2 as follows. For example, as shown in FIG. 6, the first character OB1 launches (starts) a "right punch." Then, relevant information when the first character OB1 delivers the "right punch" is input into the learning model to obtain output data. For example, when the terminal device 20 obtains a "footwork counter" action as output data, it begins to control the movement and motion of the "footwork counter" action for the second character OB2 based on the obtained "footwork counter" action.

However, if the actions of the first character OB1 and related information about said actions have already been learned in the learning model, there is no problem with the actions of the second character OB2. However, if the actions of the first character OB1 and related information about said actions have not yet been learned (lack of learning) in the learning model, there is a problem that the actions of the second character OB2 cannot be determined, or the actions become inappropriate.

Therefore, when the unlearned behavior of the first character OB1 and related information of the behavior are detected, the terminal device 20 performs the following process.

For example, as shown in FIG. 7, it is assumed that unlearned "jump kick" and "jump kick" related information of the first character OB1 are detected.

In the present embodiment, if the action of the first character OB1 (for example, "jump kick") and the related information of the "jump kick" have not been learned in the learning model, the "jump kick" of the first character OB1 is Based on the related information (for example, the distance between the first character OB1 and the second character OB2), information related to the behavior of the first character OB1 that has a high degree of matching and has already been learned is searched for. For example, as a search result, information related to "throwing technique" is searched. Then, the second character OB2 performs an action (for example, "backward defense") that corresponds to the action "throwing technique" learned in combination with the related information (related information of the action corresponding to "throwing technique"). let

In this way, in this embodiment, even if the second character OB2 encounters an unlearned action of the first character OB1 (e.g., a "jump kick"), it is possible to have the second character OB2 perform an action (e.g., "stepping back to defend") that corresponds to an action with a high degree of match in the related information (e.g., a "throw").

In addition, in this embodiment, since the behavior of the first character whose behavior is controlled based on the player's operation input and the related information of the behavior are machine learned, the player P1 can learn the behavior of the first character that controls the behavior based on the player's operation input, and therefore the player P1 can learn the behavior of the first character that controls the behavior based on the player's operation input. You can fight against character OB2.

Note that, of course, the game system of this embodiment allows a plurality of players who actually exist to play a competitive game, but for the sake of explanation, the character OB1 to be operated by the player P1 and the character OB2 controlled by the computer are An example of a competitive game will be explained.

[5] Description of behavior information [5.1] Types of behavior Next, the behavior of the character will be explained. ``Actions'' include moving movements such as moving forward, retreating, jumping, and crouching; attacking movements such as right punch, right kick, left kick, left punch, and throwing techniques; defensive movements such as retreating defense, guarding, and combos. Continuous actions, including neutrality of doing nothing.

Note that a "continuous action" is a plurality of actions that satisfy a predetermined continuation condition, and may also be referred to as a "combo" or a "continuous attack." Although there are multiple types (items) of actions in this embodiment, it is sufficient to have at least one type.

[5.2] Commands A command is information that defines the type of keys to be input by the player and the input order. Commands are defined in advance for each character in association with their actions.

The terminal device 20 of this embodiment determines an action based on the key history input from the player P1.

FIG. 8 shows an example of the action content of a character (for example, first character OB1) and commands associated with the action. Note that this is an example of a command when the character is facing right. Further, the action details and commands are stored in the game information storage unit 274 in association with the action ID.

For example, "forward" to move the character OB1 to be operated forward is performed by inputting a cross-shaped direction switch to the right. ``Backward'' to move the character OB1 to be operated backward is performed by inputting the cross-shaped direction switch to the left.

If the first character OB1 to be operated is to be punched to the right, the button 33a marked with a △ mark is input.

In order to have the first character OB1 to be operated perform Combo A, the player inputs the right punch command (for example, the △ mark button 33a) twice within 3 seconds, and then consecutively inputs the left punch command. A kick command (for example, the x mark button 33c) is input once.

[5.3] Action of the second character In the present embodiment, an action is determined using a machine-learned learning model, and the second character OB2 is caused to perform the determined action. For example, as shown in FIG. 9, the action (skill) of the first character OB1 (right punch) starts (activates) at timing T11. The terminal device 20 detects the right punch action of the first character OB1 at timing T11. For example, at timing T12, the terminal device 20 inputs the related information into the learning model, obtains the output result of the "foot technique counter" action, and starts the "foot skill counter" action on the second character OB2. There is a slight difference between timing T11 and timing T12 (a difference of one to several frames occurs). Furthermore, there may be a situation in which the second character OB2 cannot start a new action. For example, if the second character OB2 is in a rigid state (incapable of action) or is currently executing an action frame that has already been performed, a new action cannot be started. In such a case, the relevant information at the time when the second character OB2 is able to perform a new action is input into the learning model, the output result action is obtained, and the second character OB2 starts the obtained action. .

[5.4] Items of Behavior Information Behavior information is information for specifying behavior, and can be, for example, an action number (identifier). Note that the behavior information may be composed of a plurality of items (elements). For example, the action information items include an action number and an action frame.

The action number is a number assigned in response to the character's actions, such as forward movement, right punch attack movement, receiving attack when the right punch hits, throw and grab movement in throwing techniques, This is the number for the movement thrown in a throwing technique, the movement to throw through to avoid a throwing technique, etc. Note that the action frame indicates at which stage from the beginning to the end of the movement (behavior) corresponding to the action number. For example, if a right punch is performed in a period of 4 frames, the right punch technique is expressed in 4 action frames 1 to 4, and an action number is assigned to identify one action in all 4 action frames. be done.

[6] Related Information In the present embodiment, as shown in FIG. 9, related information at timing T11 when the action (for example, right punch) of the first character OB1 occurs is acquired, and the action (for example, right punch) of the first character OB1 is acquired. For example, a right punch) and the relevant information are stored in the learning storage unit 278 as a set.

The related information is information related to the action taken by the character, or in other words, the game situation. The terminal device 20 acquires information related to the action based on the timing of the action of the first character OB1. For example, as shown in FIG. 9, when the first character OB1 starts a right punch action at timing T11, related information is acquired at timing T11. Then, the information on the action of the right punch and the relevant information are stored as a set in the learning storage unit 278.

Further, when the character OB1 starts a left kick action at timing T14 after timing T11, related information is acquired at timing T14. Then, the information on the left kick action and the related information are stored as a set in the learning storage unit 278.

In this way, the terminal device 20 obtains information related to the timing at which the action of the character OB1 occurred.

The related information includes multiple items (multiple elements). Assign a variable name and its value to each item. FIG. 10 shows an example of related information related to the right punch based on the timing T11 when the first character OB1 performs the right punch action.

For example, the related information includes an item of distance between the first character OB1 and the second character OB2. The variable name of this item is Distance, which indicates that the value of Distance is "2" at timing T11, for example.

The related information also includes an item for the remaining time of one round (the period from when the first character OB1 takes action to the end of the game). The variable name of this item is Time, and for example, at timing T11, the value of Time is "47" (47 seconds remaining).

Further, the related information includes an item of the physical strength value of the first character OB1. The variable name of the item is MeHP, which indicates that the value of MeHP is "90" at timing T11, for example.

Further, the related information includes an item of physical strength value of the second character OB2. The variable name of the item is OpHP, which indicates that the value of OpHP is "80" at timing T11, for example.

It is assumed that a character with a remaining physical strength value is in a relaxed state of mind, and a character with a low physical strength value is in a desperate situation. Therefore, these physical strength values are used in machine learning.

Further, the related information includes an item of the height of the first character OB1 from the ground. The variable name of this item is MeOnFloor, which indicates that the value of MeOnFloor is "0" at timing T11, for example.

Further, the related information includes an item of the height of the first character OB1 from the ground. The variable name of the item is OpOnFloor, which indicates that the value of OpOnFloor is "0" at timing T11, for example.

Further, the related information includes an item of the defense period of the first character OB1 (the period from the start of defense to the action timing of the first character OB1). The variable name of this item is GuardTime, which indicates that the value of GuardTime is "0" at timing T11, for example.

Although not shown, the related information of this embodiment includes items such as the position coordinates (X, Y, Z) of the first character OB1 in the game space, and the position coordinates (X, Y, Z) of the second character OB2 in the game space. . Item of elapsed time (period from the start of the game of one round to the action timing of the first character OB1) Item of the character level (grade) of the first character OB1, Item of the first character OB1 OB1's orientation item, second character OB2's orientation item, first character OB1's height item, second character's height item, first character OB1's physical strength value and second character OB2's physical strength value It may also include other items such as an item of the difference in physical strength from the physical strength value.

Note that the related information may include an item of image information (video information). For example, the first character OB1 may include an item of image information of the timing at which the first character OB1 performed an action.

The related information may include items such as the brightness of the image, the display position (X, Y) of the character OB1 in the image, the display position (X, Y) of the character OB2 in the image, and the positional relationship between the first character OB1 and the second character OB2 on the image.

Fighting games often have characters facing each other with the center of the screen as the boundary, but if both characters are on the right or left side of the screen, the other player is considered to be in a corner, and this information can be used in machine learning. You can make use of it.

Further, the related information may include at least one item of the position, orientation, and angle of view of the virtual camera. For example, when the virtual camera is close to the first character OB1, the first character OB1 is displayed closer than when it is far away, and it can be considered that the game situation is heated. This situation can be utilized in machine learning.

The related information may include an item of information (flag) indicating whether the action information is continuous action information (combo action information). In this way, combo-specific predictions can be made during machine learning.

The related information may include items of key data (operation input). For example, it may include a key history that is a history of key data of the player P1. The key history stores operation inputs (key data) of the character operated by the player in association with frame numbers.

Further, the related information may include at least one item of the moving speed, acceleration, and moving direction of the first character OB1, and the moving speed, acceleration, and moving direction of the second character OB2.

[7] Description of Data to be Stored The terminal device 20 of the present embodiment stores information on the action of the character OB1 to be operated by the player P1 and related information related to the action as a set in the learning storage unit 278.

Note that the terminal device 20 may store this information in a storage area other than the learning storage unit 278 (for example, a storage area on the cloud).

In this embodiment, from the start of a given round of the game to the end of that round of the game (until the end of one round), information on the actions of the first character OB1 and information related to those actions are cumulatively stored as learning data in the learning memory unit 278.

Note that in this embodiment, learning data may be acquired and stored during a part of one round. Furthermore, the frequency of acquiring and storing learning data may be adjusted. Further, learning data may be acquired not only in one round but also in multiple rounds.

Further, in this embodiment, part or all of the learning data may be processed for machine learning of the learning model.

[8] Learning Model In this embodiment, a learning model is generated that performs machine learning on the behavior of the first character OB1 operated by the player P1 and related information related to the behavior. The learning model is a learning model that can reproduce the behavior of the first character OB1 operated by the player P1. The learning model can be rephrased as an equation or a function.

In this embodiment, an individual learning model is generated for each player so that the most suitable AI character for the player becomes the training partner. For example, the terminal device 20A of the player P1 generates a learning model corresponding to the player P1. Furthermore, the terminal device 20B of the player P2 generates a learning model corresponding to the player P2.

[9] Behavior control of second character in case of unlearning [9.1] Overview In this embodiment, the behavior of the first character OB1 determined based on the operation input of the player P1 (for example, "jump kick") ) and related information of the behavior have not been learned in the learning model, there is a problem that the behavior of the second character OB2 cannot be controlled or the behavior is inappropriate.

Therefore, in this embodiment, if the action of the first character OB1 (e.g., "jump kick") and related information about that action have not yet been learned in the learning model, the following processing is performed.

As shown in FIG. 7, the terminal device 20 associates related information with a high degree of matching with the related information of the action (for example, jump kick) of the first character OB1 to the learned action of the first character OB1. Search (search, extract) from related information. For example, the related information stored in the learning storage unit 278 is referred to.

For example, assume that "distance between first character OB1 and second character OB2", which is one item of information related to the action of first character OB1 (for example, jump kick information), is 1 meter. Then, related information in which the distance between the first character OB1 and the second character OB2 is 1 meter is searched from among the learned related information.

If there is no related information in which the distance between the first character OB1 and the second character OB2 is 1 meter from among the learned related information, search for related information whose distance is closest to 1 meter. do.

Then, when the related information with the highest degree of matching (for example, the distance between the first character OB1 and the second character OB2 is 1 meter) is searched from among the learned related information, the related information and the corresponding The behavior of the second character OB2 is controlled based on the learned behavior (eg, throwing technique) corresponding to the related information.

If it is decided in advance to perform "backward defense" as an action corresponding to the "throwing technique", the terminal device 20 controls the movement and movement of the "backward defense" action for the second character OB2. .

For example, when the first character OB1 performs a "throwing technique," which is an action learned in combination with the searched related information, the terminal device 20 determines the learned behavior when the second character OB2 responds. Relevant information may be input into the learning model and output data may be obtained. When the output data is the behavior of the "guard", the terminal device 20 controls the movement and movement of the "guard" on the second character OB2 based on the acquired behavior of the "guard".

Note that the timing at which the related information related to the action (for example, jump kick) of the first character OB1 is acquired is, for example, the timing at which the action (for example, the jump kick action) of the first character OB1 is started. , it may be a given timing of the action frame (during movement) of the action (for example, jump kick). The related information related to the action may be, for example, the distance between the first character OB1 and the second character OB2, a parameter of the first character OB1 (for example, physical strength value), or other items.

In this embodiment, related information with a high degree of match may be searched for by referring to items of high importance among multiple items of related information of the action of the first character OB1 (e.g., information on a jump kick). For example, "the distance between the first character OB1 and the second character OB2" and "the physical strength value of the first character OB1" which are of high importance are searched for from among the learned related information. In other words, in this embodiment, related information with a high degree of match for "the distance between the first character OB1 and the second character OB2" and for "the physical strength value of the first character OB1" may be searched for.

[9.2] Judgment when unlearned In the present embodiment, the action of the first character OB1 (for example, "jump kick") determined based on the operation input of the player P1 and the related information of the action are Various methods can be considered for determining whether the learning model has not yet been learned.

For example, the terminal device 20 registers (stores) in the storage unit 270 a pre-learned action list for the second character OB2, and the action of the first character OB1 (for example, "jump kick") is registered (stored) in the action list. If the information is not registered in , it may be determined that the information has not been learned.

Further, the terminal device 20 may refer to the learning storage unit 278 and determine that the first character OB1 has not yet been learned if the information on its behavior is not stored.

Alternatively, for example, the related information of the first character OB1 may be input into a learning model, and it may be determined whether the behavior is unlearned based on the output result of the learning model. For example, if the output value of the learning model is less than or equal to a predetermined value, it may be determined that the learning model has not been learned.

[9.3] Search method In this embodiment, if the action of the first character OB1 determined based on the operation input of the player P1 (for example, "jump kick") is unlearned in the learning model, it is determined that the action has been learned. Among the related information related to the behavior of the first character OB1, the association has a high degree of matching with the related information related to the detected unlearned behavior of the first character (for example, "jump kick") Explore information.

In such a case, for example, the terminal device 20 may determine the detected unlearned behavior of the first character (for example, " Search for related information that has a high degree of matching with related information related to "Jump Kick").

There may be multiple pieces of information related to the search results. For example, as a search result, "throwing technique"
It is possible that information related to "High Kick" and information related to "High Kick" may be detected. In such a case, the action of the second character OB2 is determined using any one of the actions resulting from the search.

[9.4] Specific example of search (A) Example of referring to physical strength value of related information In this embodiment, if the "jump kick" action of the first character OB1 is not learned in the learning model, " From the physical strength value, which is one item of the related information of the first character OB1 of "Jump Kick" (that is, the physical strength value of the first character OB1 at the time of jump kick), to the learned related information of the first character OB1. You may search for related information with the highest degree of matching.

For example, the physical strength value of the first character OB1 when performing a jump kick is "10", and the physical strength value of the first character OB1 when the learned first character OB1 performs a "left kick" is "10". , the physical strength values are the same, and it is determined that the degree of matching is the highest. Then, the action of the second character OB2 may be controlled based on the information related to the "left kick" action of the first character OB1. For example, when the action corresponding to "left kick" is "guard", the action of the second character OB2 is determined to be "guard", and the second character OB2 is caused to perform "guard". That is, when the first character OB1 performs a "jump kick", the terminal device 20 causes the second character OB2 to perform a "guard" action.

(B) Example of referencing remaining game time In this embodiment, when the "jump kick" action of the first character OB1 has not been learned in the learning model, the remaining game time (e.g., the remaining game time at the time of the jump kick), which is one item of the related information of the first character OB1 for the "jump kick," may be used to search for related information of the first character OB1 that has already been learned that has the highest degree of match.

For example, the remaining game time of the first character OB1 when performing a jump kick is "15", and the remaining game time of the first character OB1 when the learned first character OB1 performs a "left punch" is "15". , and it is determined that the degree of coincidence is highest since the remaining game time is the same. Then, the action of the second character OB2 is determined based on the related information of the "left punch" action of the first character OB1. For example, when the action corresponding to "left punch" is "retreat", the action of the second character OB2 is determined to be "retreat", and the second character OB2 is caused to perform "retreat". In other words, when the first character OB1 performs a "jump kick", the terminal device 20 causes the second character OB2 to perform a "retreat" action.

[9.5] Effects According to this embodiment, even if the player encounters the first character OB1 that has not been machine-learned, the action of the second character OB2 is controlled from the related information of the action of the learned first character OB1, so that the second character OB2 can be made to take an appropriate action. In addition, according to this embodiment, an optimal AI character (e.g., the second character OB2) that reflects the player's habits can be provided as a practice partner. For example, if the first character OB1 of the player P1 has many punch attacks, the AI character will also be a character that has many punch attacks. The player P1 can practice correcting his/her habits or making use of his/her habits by practicing against a computer that plays similarly to the player P1. In addition, the player P1 can perform an unlearned action (e.g., a jump kick) and the AI character can take an appropriate action.

[9.6] Learning of second character's behavior and related information of the behavior In this embodiment, the behavior of the second character OB2 and related information related to the behavior may be learned by machine.

For example, in the present embodiment, when an unlearned action of the first character OB1 (e.g., "jump kick") and related information of the action are encountered, the action taken by the second character OB2 (e.g., "backwards") is encountered. Machine learning may be performed on a set of information related to the behavior (for example, when the behavior of "backward defense" is started) and the relevant information at the time of the behavior (for example, when the behavior of "backward defense" is started).

[10] Learning mode and actual battle mode In this embodiment, a learning mode (learning period) in which learning is performed using a learning model, and the learning model is used to predict (infer) the behavior of the second character OB2 and advance the game. A practical mode (practical period) is conducted at the same time. However, in the present embodiment, the period of the learning mode for generating the learning model and the period of the practical mode may be set so that some or all of them are different periods.

For example, learning data may be collected in a match between a first character and a computer character, and a learning model may be generated based on the collected learning data when a certain amount of learning data has been accumulated.

Furthermore, in this embodiment, the learning model may be generated in a device other than the terminal device 20 of the player P1 (for example, a terminal device 20 of another player, a server device, or a learning device). Then, the generated learning model may be received from the device, stored in the learning model storage section 279, and used. Further, the learning data may be received from the device and stored in the learning storage section 278.

[11] Period where learning is performed frequently As shown in FIG. A predetermined period may be provided as the high period. The length of the predetermined period is shorter than the game period (for example, 60 seconds), and is, for example, 20 seconds.

"Learning is performed frequently" means, for example, that the number of times learning data is acquired and machine learning is performed on a learning model in a predetermined period is greater than in periods other than the predetermined period.

For example, as shown in FIG. 11, the number of times machine learning is performed during the predetermined period (period TS to T1) is at least 10 times, and the number of times machine learning is performed outside the predetermined period (period T1 to TE) is at least 10. No more than 5 times.

Note that the terminal device 20 sets a predetermined period of time before learning. Note that the terminal device 20 may set or change the predetermined period even during learning. For example, the terminal device 20 may set a predetermined period or change a predetermined period depending on the progress of the game.

[11.1] Example of setting a predetermined period The terminal device 20 of this embodiment may set a predetermined period in the start period of the game period.

For example, as shown in FIG. 11, the start time of the game period and the start time of the predetermined period are set to the same timing TS. The end time T1 of the predetermined period is the time 10 seconds after the start time TS.

In addition, the types of actions to be acquired as learning data during the predetermined period (period from TS to T1) are all types of actions, such as simple actions, techniques, and continuous actions, and the types of actions to be acquired as learning data during the predetermined period (period from T1 to T1) are all types of actions, and outside the predetermined period (period from T1 to TE), the learning data is The types of actions to be acquired may be limited to techniques and continuous actions.

The starting period of a game period is an important moment that influences the flow of the game and the direction of the match. For example, important factors in the game include which character makes the first attack, and which character makes the move first. During the start period of the game period, players tend to develop their habits.

Therefore, in this embodiment, in order to learn efficiently, a predetermined period is set at the start of the game period to learn many actions. Then, after the predetermined period has passed, the player learns actions such as techniques and consecutive actions (combos).

[11.2] Example of Setting Predetermined Period Based on Operation Input by Player The terminal device 20 of this embodiment may set a predetermined period based on operation input by the player P1 who controls the first character OB1.

FIG. 12 shows an example of a setting screen. For example, as shown in FIG. 12, the player P1 inputs, on the setting screen, a game start time point of 0, a game end time point of 60, and a start timing A1 and an end timing A2 of a predetermined period.

When the player P1 receives the start timing A1 (e.g., 50) and the end timing A2 (e.g., 60), the terminal device 20 sets a predetermined period based on the start timing A1 (e.g., 50) and the end timing A2 (e.g., 60).

In this way, the player P1 can play the game keeping in mind the predetermined period. For example, if you practice a technique or a series of actions for a predetermined period of time, you can learn the corresponding counterattack behavior. As a result, an optimal practice environment can be provided to the player P1.

Furthermore, if player P1 is aware that his own habits tend to emerge at the end, by setting a predetermined period for the final period (10 seconds remaining), an optimal practice environment for player P1 can be created. .

[11.3] Setting Learning Frequency The terminal device 20 may divide the game period of one round (for example, 60 seconds) into a plurality of periods in chronological order, and may set the learning frequency for each period. For example, until 10 seconds have elapsed from the game start timing, learning is performed 10 or more times, and from 10 seconds after the game start timing until 20 seconds have elapsed, learning is performed 5 or more times less than 10 times, and at the game start timing After 20 seconds have passed, the learning frequency is set so that learning is performed less than 5 times.

Further, the terminal device 20 may divide the game period of one round (for example, 60 seconds) into a plurality of periods in chronological order, and set the type of behavior to be learned in each period.

First, the types of actions are divided into three types: simple actions, techniques, and combos. For example, all types of simple actions, techniques, and combos are learned until 10 seconds have passed from the game start timing, and all types of simple actions, techniques, and combos have been learned from 10 seconds after the game start timing until 20 seconds have passed. Among actions, techniques and combos are learned, and after 20 seconds from the game start timing, the learning frequency is set so that the types of combos among simple actions, techniques, and combos are learned.

The terminal device 20 of this embodiment may set the learning frequency based on the operation input of the player P1. In other words, the terminal device 20 may set the learning frequency in each period in which the game period is divided into a plurality of periods and the type of behavior to be learned in each period based on the operation input from the player P1.

[12] Description regarding continuous action [12.1] Description of continuous action The actions of this embodiment include continuous actions (combo) that are a plurality of actions that satisfy a predetermined continuation condition. Continuous actions are, in other words, continuous attacks. Continuous attacks are a useful attack method that can prevent the opponent character from attacking, and can also cause great damage to the opponent character. When making a character perform continuous actions, multiple command inputs are required. The number of consecutive actions may be two or three or more actions.

A consecutive action is, for example, a sequential combination of multiple actions (a consecutive action of right punch, right punch, left kick). As shown in FIG. 8, in this embodiment, a combo command for performing consecutive actions is defined in advance.

For example, when activating Combo A, as shown in Figure 13, commands for three actions, right punch, right punch, and left kick, must be input consecutively within a continuous action period (for example, within 3 seconds). is required.

An example of a successful combo A is as follows. As shown in FIG. 13, for example, at timing T21, the first character OB1 inputs a right punch action command based on an operation input from the player P1, and the right punch action is started. Based on the action of the right punch of the character OB1, the first character OB1 hits the second character OB2 at timing T22, and the right punch is successful. Note that when the second character OB2 receives a hit from the first character OB1, the second character OB2 remains in a rigid state for several frames and is defenseless, with no key data being reflected. Then, at timing T23, the first character OB1 inputs a right punch action command based on the operation input from the player P1, and the right punch action is started. Based on the right punch action input by the character OB1 at timing T23, the first character OB1 hits the second character OB2 at timing T24, and the right punch is successful. Then, at timing T25, the first character OB1 inputs a left kick action command based on the operation input from the player P1, and the left kick action is started. Based on the left kick action performed by the character OB1 at timing T25, the first character OB1 hits the second character OB2 at timing T26, and the left kick is successful. In this way, Combo A is successful.

Various "continuation conditions" can be considered. For example, the "continuation condition" is to perform a predetermined continuous action (in the case of Combo A, a continuous action of right punch, right punch, and left kick) within the continuous action period. In other words, the "continuation condition" is to perform the action multiple times in succession without being attacked by the opponent character.

Continuous actions cause more damage to the character than single actions (decreases the attack target's physical strength more than single actions). For example, when a right punch at timing T24 hits, stronger damage may be caused than a single right punch. Furthermore, when the left kick at timing T26 hits, it may cause more damage than a single left kick.

[12.2] Learning continuous actions and information related to continuous actions Continuous actions and hit judgments associated with continuous actions occur in a given continuous action period (period T21 to T27), as shown in Figure 13. It is something. In this embodiment, information occurring during this continuous action period (period T21 to T27) is acquired as related information related to the continuous action.

For example, the continuous action related information includes an item of the average value (average distance) of the distance between character OB1 and character OB2 acquired in each frame (every 1/60 second) of the continuous action period.

Further, the continuous action related information may include an item of the decrease value of the physical strength value of the character OB2 during the continuous action period. For example, if character OB2's physical strength value decreases from 100 to 10 during the continuous action period (period T21 to T27), the difference, 90, becomes the decreased value.

In addition, the related information of the continuous action may include an item of the total value (total value, total distance) of the distance between character OB1 and character OB2 obtained in each frame (every 1/60th of a second) of the continuous action period.

Further, the continuous action related information may include an item of a rate of change in damage inflicted by character OB1 on character OB2 during the continuous action period.

Moreover, the related information of a continuous action may include an item of a combination of a plurality of actions that constitute a continuous action. For example, combo A is a combination of two right punches and one left kick.

Further, the related information of a continuous action may include an item of the order (attack order) of a plurality of actions that constitute the continuous action. For example, in the case of combo A, the information is the order of right punch, right punch, and left kick.

Further, the related information of the continuous action may include an item of the number of actions (for example, the number of attacks) of a plurality of actions that constitute the continuous action. For example, in the case of Combo A, the number of attacks is three because the attacks are a right punch, a right punch, and a left kick.

[12.3] Learning Data The terminal device 20 uses information on a continuous action and information acquired as information related to the continuous action as learning data.

For example, the terminal device 20 includes information on the continuous action of the first character OB1 (combo A information) and information acquired as related information of the continuous action (combo A) (for example, average distance between characters, etc.) The information is stored in the learning storage unit 278 as a set.

Then, the terminal device 20 performs machine learning on the information on the continuous action and the related information on the continuous action.

Note that during the continuous action period, the continuous actions of the first character OB1 and related information are acquired, but in addition, the single actions of the first character OB1 (for example, right punch, right punch, left kick) are acquired. Each action) and each related information may be acquired and machine learning may be performed.

[12.4] Searching for consecutive actions In this embodiment, when the consecutive actions of the first character OB1 and related information for the consecutive actions have not been learned in the learning model, related information for learned actions (single actions or consecutive actions) is searched for based on related information related to the consecutive actions.

When the action of "combo A" of the first character OB1 has not been learned in the learning model, the terminal device 20 searches for related information that has a high degree of match with the related information of combo A from among the learned related information.

For example, when performing a search based on the average distance between the first character OB1 and the second character OB2 among related information, the following processing is performed.

An example of a case where a search is performed with emphasis on the item "combo attack order", which is one item of related information, will be explained. For example, when combo A performs three attacks in the order of right punch, right punch, and left kick, the terminal device 20 searches for a learned combo that has the most similar attack order. For example, combo B is searched as a learned combo that is most similar to combo A, including a right punch, right punch, and right kick. Then, the terminal device 20 causes the second character OB2 to perform an action corresponding to combo B.

For example, as shown in FIG. 13, when the action corresponding to combo B is "get up kick", the terminal device 20 causes the second character OB2 to perform the "get up kick" action. That is, the terminal device 20 uses the learning model to determine that the second character OB2 is ready to act after the continuous action period of "Combo A" of the first character OB1 ends (for example, after timing T27). At timing T28, the second character OB2 is controlled to move and perform a "stand-up kick."

Next, an example of a case where a search is performed with emphasis on the "average distance between character OB1 and character OB2", which is one item of related information, will be described. For example, if the average distance during the continuous action period of combo A (average distance between the first character OB1 and the second character OB2) is 0.8 meters, the terminal device 20 searches for learned actions (continuous actions or single actions) that match the average distance. For example, the average distance (average distance between the first character OB1 and the second character OB2) of learned combo C is 0.8 meters, and combo C is searched for as the learned combo that matches combo A most closely. Then, the terminal device 20 causes the second character OB2 to perform the action corresponding to combo C.

For example, when the action corresponding to combo C is "backward passive", the terminal device 20 causes the second character OB2 to perform the "backward passive" action. In other words, the terminal device 20 uses the second character OB2 in the learning model at timing T28 after the end of the continuous action period of "Combo A" of the first character OB1 and when the second character OB2 becomes ready for action. ``Backward passive'' movement and movement control is performed for the character OB2.

In this way, according to the present embodiment, even if combo A has not been learned, the degree to which the sequence of combo A matches is searched among the learned combos, and the second character OB2 Conduct appropriate actions.

Note that when searching for continuous actions, the items of related information to be emphasized may be changed. For example, in addition to the average distance between characters in a combo and the order of attacks, other related information items include, for example, if emphasis is placed on the number of attacks, Combo A is a total of two right punches and one left kick. Since the action is based on the number of attacks of three times, a continuous action that has been learned and has the number of attacks of "3" is searched for. For example, when combo C has three attacks and is searched as a learned combo that has the highest degree of matching with combo A, the terminal device 20 transmits the action corresponding to combo C to second character OB2. have it done.

[13] Setting of Related Information Items The terminal device 20 of the present embodiment may set items of related information to be used in machine learning in advance. In addition, the terminal device 20 of the present embodiment may set multiple items of related information based on input information of the player P1.

The terminal device 20 may determine a combination of at least two or more items to be used in machine learning among the plurality of items of related information based on the operation input of the player P1.

A combination of items of related information means, for example, a set of at least two or more items among a plurality of items of related information.

[14] Weighting The terminal device 20 determines the weighting of each item of related information. Weighting involves evaluating (scoring) the importance of each item. That is, in this embodiment, weighting is determined by quantifying the degree of importance for each item of related information to be learned.

For example, the terminal device 20 may determine the weighting according to at least one of player information, first character information, and game information.

The terminal device 20 may use the weighted numerical value (weight value) for machine learning calculations.

The terminal device 20 performs weighting before learning (before machine learning). In this way, machine learning can be performed efficiently at an early stage. That is, when weighting is performed in real time, it takes a certain amount of time, but in this embodiment, by performing weighting early, learning can be achieved efficiently.

The terminal device 20 may perform weighting depending on the progress of the game (for example, during learning). In this way, you can learn efficiently in real time. Further, it is possible to perform weighting appropriate to the progress of the game.

For example, in the initial state, the weight value of each item of related information is set to "1", and the value of an item with high importance is set to a value larger than "1". There may be one or more items with high importance.

Specifically, based on the operation input from the player P1, the terminal device 20 selects the item "distance between the first character OB1 and the second character OB2" among the items of related information to be learned; If the item "Remaining time of 1 round" is selected as an item to be prioritized, set the weight of these two items to ``3'', and maintain the weight of other items as ``1''. .

[14.1] Setting Weighting Based on Player Information The terminal device 20 may determine the weighting of each item of related information to be learned, depending on the player information. Based on player information set in advance, weighting can be set for items with high priority during learning, and learning can be performed to achieve actions that are highly satisfying to the user.

The player information includes the player's level, rank, information on whether the player is a beginner, intermediate, or advanced player, game play time of the fighting game, winning rate, number of battles, battle record, and the like.

For example, beginners tend to be concerned about their physical strength, while intermediate and advanced players tend not to care about their physical strength but rather the distance to their opponent. Therefore, a beginner sets the weight value of the physical strength value item to "3" and maintains the weight value of other items as "1".

Intermediate and advanced players should set the weight value for the stamina item to "3" and keep the weights for all other items at "1".

Player information may determine the weighting of each item to be learned based on information about player P1 (age and gender).

[14.2] Setting Weighting Based on Player Character Information The terminal device 20 may determine the weighting of each item of related information to be studied according to the player character information.

The player character information includes, for example, the number of characters operated by the player P1, the type of characters, and the like. Alternatively, the learning level of character skills (including combos) for each character managed by the player may be used. The degree of skill acquisition indicates the ratio of the number of skills that the character has learned out of all the number of skills of the character. For example, a technique that the player P1 succeeds in in the practice mode is determined to have been learned.

For example, in the case of a character with a high learning level (for example, a character with a learning level of 50% or more), the weight of the distance item among the related information items is set to "3", which is higher than the other items.

For example, for a character with a low level of proficiency (e.g., a character with a level of proficiency of less than 50%), the weight of the stamina value among the related information items is set to "3", making it higher than the other items.

[14.3] Setting Weighting Based on Game Information Then, the terminal device 20 may determine the weighting of each item of related information of the learning target according to the game information.

The game information includes, for example, the level and level difference between the player P1 and the opponent player, the level and level difference between the character OB1 to be operated by the player P1 and the opponent character OB2, and the like.

For example, if the level difference is 10 or more, the weight of the distance item among the related information items is set to "3", so that it is higher than the weight of non-distance items.Also, if the level difference is 9 or less, Among the items of related information, the weight of the physical strength value is set to "3", making it higher than items other than the physical strength value.

Further, the game information may be the type of mode such as casual mode or serious mode. The casual mode is a mode for playing casually, and the serious mode is a mode for playing serious fighting games such as events, tournaments, and E-sports held by the management company.

In addition, when the game mode is serious mode, the weighting of the distance and stamina value among the related information items is set to "3", which is higher than the weighting of items other than stamina value.

[15] Settings based on player's operation input [15.1] Specific example of weight setting screen In this embodiment, based on the player P1's operation input, among the plurality of items of related information of the first character OB1, It may also be possible to accept selections of items with high weights.

Fig. 14 shows an example of a weight setting screen. For example, as shown in Fig. 14, if the player P1 is concerned about the vitality value of the first character OB1 and wants to place importance on the vitality value of the first character OB1, the player P1 selects the item of vitality value as an item to be weighted higher.

Further, although not shown, if the player P1 wants to place importance on the distance between the first character OB1 and the second character OB2, he selects the distance item as an item to have a higher weight.

In the weight setting screen of this embodiment, as shown in FIG. 15, based on the operation input of the player P1, a weight distribution rate is determined in percentage for each of a plurality of items of related information of the first character OB1. Good too. Then, the weight value of each item of related information is determined according to the distribution ratio. The weights are set such that the higher the distribution rate, the higher the weight value. Conversely, the weights are set such that the lower the allocation rate, the lower the weight value.

Further, in the weight setting screen of this embodiment, the priority order of each of the plurality of items of related information may be set based on the operational input of the player P1.

For example, as shown in FIG. 16, priorities may be determined for a plurality of items of information related to the first character OB1. The terminal device 20 determines the weight of each of the plurality of items of related information of the first character OB1 based on the priority order. In other words, the weights are determined so that the higher the priority, the higher the weight. In this way, the player can easily set the weights. Furthermore, priorities may be set for some items (for example, the top five items) among all items of related information.

[16] Setting weights, etc. based on player information The terminal device 20 of the present embodiment configures the related information of the first character OB1 before learning (for example, before machine learning) based on the player information. At least one of a combination of a plurality of items, a learning frequency, and a weight of each of a plurality of items constituting the related information may be set.

Furthermore, the terminal device 20 determines the combination of a plurality of items constituting the related information of the first character OB1, the learning frequency, and the plurality of items constituting the related information according to the game progress based on the player's information. At least one weight may be set for each item.

In other words, the terminal device 20 configures the related information of the first character OB1 during the game, during learning, after learning, the progress of the game, or according to the progress of the game, based on the player's information. At least one of a combination of a plurality of items, a learning frequency, and a weight of each of a plurality of items making up the related information may be set.

[16.1] Explanation of Player Information The player information will be explained in detail. Player information includes fighting game playing time, experience with the game, fighting game history, number of matches, match results, winning percentage, information on opponents played so far, and the level of the opponent (dan, rank). .

The player information also includes information on the character used by the player in the fighting game, techniques and continuous actions (combos) learned by the character, and techniques and continuous actions used by the character.

It also includes information about the character that the player uses most often among the characters used in the fighting game.

For example, the game playing time of a fighting game is the total period from when the player starts the fighting game (login, power on) until it ends (sleep, logout, power off, etc.).

If the game play time is less than one hour, the player is considered to be a beginner, and the weight is determined so that weight is given to the physical strength value. For example, the weight is set so that the item of the physical strength value of the character OB1 in the related information has a higher weight than other items.

If there are more short-range attacks than long-range attacks among the actions performed by the player, there is a high possibility that the player is concerned about the distance between the first character OB1 and the second character OB2. Therefore, the weight is set so that the item of the distance between the first character OB1 and the second character OB2 of the related information has a higher weight than other items.

[16.2] Play History The player information includes information regarding the player's play history. For example, the game play history of the player P1 is a history of behavior information and related information of the first character OB1 to be operated by the player P1. For example, the attack pattern is determined from the game play history of the player P1. In other words, when the first character OB1 often corners the second character OB2 against a wall and performs techniques, the positional information of the first character OB1 is considered important. Therefore, the weight is set so that the item of position information of the first character OB1 in the related information has a higher weight than other items.

[16.3] Past Play Tendency The player information may include information regarding the player's past play tendency.

For example, player P1's past play tendencies include information indicating whether he plays games in casual mode or serious mode, and which player he plays most often among advanced players, intermediate players, or beginners. , and information indicating what kind of play style the player P1 tends to play.

In terms of play style, what kind of character do you use and what kind of flow does the match have? (For example, players tend to perform a big move within the first 10 seconds of the game and reduce the opponent's character's health to 0. (a tendency for an onslaught of attacks to unfold within seconds).

For example, if the player's past playing tendency is to play in serious mode more often than casual mode, the weighting of the stamina value item of the first character OB1 in the related information is set higher than the weighting of other items.

Furthermore, if the player's past play tendency is that there are many beginners, the weight is set so that the item of the physical strength value of the first character OB1 in the related information is weighted higher than other items.

In addition, if the player's past play tendency is to perform a big move and reduce the opponent character's physical strength to 0 within 10 seconds of the start of the game, the item of physical strength of the first character OB1 in the related information and the The weight is set so that the weight with the physical strength value item of the character OB2 of No. 2 is higher than the other items.

[16.4] Information related to other players The player information may include information related to other players determined based on the player information.

For example, the information related to other players determined based on the information of player P1 is other players whose level is similar to player P1.

For example, when player P1 is at an expert level, an item to be given a higher weight is determined for the same item as an item with a higher weight among related information of another player P2 at an expert level.

If the item with a high weight among the related information of another player P2 at an advanced level is the distance from the wall, the weight of the distance from the wall item of the related information of the player P1 will be higher than other items. Set the weight as follows.

For example, the server device 10 may collect and centrally manage information related to each player managed by the server device 10. Then, the terminal device 20 of the player P1 receives, from the server device 10, information related to another player P2 determined based on the information of the player P1, for example, "the weight of the plurality of items of related information of the player P2". Get "high item". The "item with the highest weight among the plurality of items of related information about the player P2" may be the single item with the highest weight, or may be the item with the highest weight (for example, the top five).

[16.5] Supplementary explanation of the example of setting weighting according to the game progress The terminal device 20 sets the related information of the first character OB1 in real time according to the game progress based on the player's information. Weights may be set for each of the plurality of constituent items.

For example, when the terminal device 20 detects a physical strength reduction situation (an example of a game progress situation) in which the physical strength value of the first character OB1 is decreased, the terminal device 20 changes the weight of each item constituting the related information of the first character OB1. The value is set as the weight value of each item corresponding to the physical strength reduction situation.

Further, for example, when the terminal device 20 detects a successful attack situation (an example of a game progress situation) in which the attack of the first character OB1 is successful, the terminal device 20 determines the weight of each item constituting the related information of the first character OB1. The value of is set as the weight value of each item corresponding to the attack success situation.

[16.6] Setting a combination of a plurality of items Before learning, the terminal device 20 may set a combination of a plurality of items constituting the related information based on the player's information before learning. For example, if the player is a beginner, a combination of a plurality of items constituting the related information for beginners is set. If the player is an advanced player, a combination of a plurality of items constituting related information for advanced players is set.

Further, after the game starts, the terminal device 20 may set a combination of a plurality of items that constitute the related information of the first character OB1 in real time according to the progress of the game. For example, when a state of decrease in physical strength is detected, a combination of a plurality of items corresponding to the state of decrease in physical strength is set. For example, when a successful attack situation is detected, a combination of a plurality of items corresponding to the successful attack situation is set.

[16.7] Setting Learning Frequency Before learning, the terminal device 20 may set the learning frequency based on the player's information before learning. For example, if the player is a beginner, a learning frequency for beginners is set. If the player is an advanced player, a learning frequency for advanced players is set.

Further, the terminal device 20 may set the learning frequency in real time according to the progress of the game after starting the game. For example, when a state of decrease in physical strength is detected, a learning frequency corresponding to the state of decrease in physical strength is set. For example, when a successful attack situation is detected, a learning frequency corresponding to the successful attack situation is set.

17A and 17B are flowcharts showing an example of behavior control according to the present embodiment. Using these flowcharts, a flow of behavior control of a second character OB2, which is an AI character, will be described when a character OB1 operated by a player P1 competes against a second character OB2 on the terminal device 20 of the player P1.

First, as shown in FIG. 17A, weights are set (step S1). For example, a weight is set for each of a plurality of items of related information related to an action. The weights may be set based on the operational input of the player P1.

Then, when the game starts (step S2), the behavior of the first character OB1 is controlled based on the operation input from the player P1 (step S3).

Determine whether or not the unlearned behavior of the first character OB1 and related information related to that behavior have been encountered (step S4).

If the action of the unlearned first character OB1 and related information related to the action are not encountered (N in step S4), the action of the second character OB2 is determined using the learning model based on the related information. (Step S5). Then, learning is performed (step S6). For example, the behavior of the first character OB1 and related information related to the behavior are learned. Then, the process advances to step S14.

Note that if the action of the unlearned first character OB1 and related information related to the action are encountered (Y in step S4), the process advances to step S11 in FIG. 17B. Then, as shown in FIG. 17B, based on the related information related to the behavior of the first character OB1, information related to the learned behavior of the first character OB1 is searched (step S11).

Then, based on the searched related information, the action of the second character OB2 is determined (step S12).

Then, learning is performed (step S13). For example, the behavior of the first character OB1 and related information related to the behavior are learned. Furthermore, the behavior of the second character OB2 and related information related to the behavior are learned. Then, the process advances to step S14.

It is determined whether the game has ended (step S14), and if the game has ended (Y in step S14), the process ends. On the other hand, if the game has not ended (N in step S14), the process returns to step S3. This completes the process.

[18] Detailed explanation of machine learning In this embodiment, weighting is optimized by machine learning. The terminal device 20 optimizes the weights related to learning by performing deep learning processing based on the behavior (behavior information) and related information of the behavior stored in the learning storage unit 278. In the subsequent competitive game between the first character and the second character, the terminal device 20 reflects the weight of the optimized learning result and controls the movement of the second character OB2 without being based on the input from the input unit 260. Control.

The learning unit 218 performs deep learning calculation processing on each relevant information using a recurrent neural network (recursive neural network) consisting of multiple layers, calculates behavioral information, and optimizes the weight of the learning result. do.

For example, the learning unit 218 sends behavioral information of the second character OB2 to the movement and movement control unit 213, and the movement and movement control unit 213 receives behavioral information reflecting the weight of the learning result from the learning unit 218 and controls the behavior (movement and movement) of the second character OB2.

The learning storage unit 278 also includes a history data storage unit that stores a history of related information related to an action, and a stratified weight storage unit that stores calculation results of a plurality of stratified weights as deep learning processing results. May include.

In addition, the terminal device 20 writes the collected related information to the learning storage unit 278, and uses each related information stored in the history data storage unit as input to the first layer, and weights each input in the plurality of layers. The output of each layer is calculated by multiplying by , and the output is calculated as the input of the next layer. The weights for each layer may be optimized using the difference and stored in the weight storage section for each layer.

The learning unit 218 also includes an inference processing unit for deriving an action that the character would have taken by multiplying the current related information by the latest weight at that time, and a It may also include a learning processing unit for reflecting the behavior of the character OB1 based on the actual operation input in the weight of deep learning.

The inference processing unit writes the collected related information to the historical data storage unit, uses each related information stored in the historical data storage unit as input to the first layer, multiplies the weights in each layer, and creates a layer. By calculating another output and sequentially calculating that output as input to the next layer, behavioral information is obtained as the output of the final layer.

The learning processing unit optimizes deep learning weights based on the behavior information calculated by the inference processing unit and the behavior information determined based on the player's operation input corresponding to the time of the current related information. .

The learning unit 218 executes, as deep learning processing, calculation processing using at least one all-to-fully connected layer and calculation processing using a plurality of gated recurrent unit layers. In the calculation process using the all-to-all connected layer, a process is performed in which a weight matrix is multiplied by an input vector. In the calculation process by each gated recurrent unit layer, in each of the multiple gated recurrent unit layers, the amount to be forgotten is calculated using the weight, the reflected amount is calculated using the weight, and the output candidate is calculated. may be performed sequentially.

As a calculation process performed by each gated recurrent unit layer, the learning unit 218 may selectively perform dropout processing and output after calculating the output candidates.

The terminal device 20 may store operation data corresponding to the player's operation of the input unit 260 in the game information storage unit 274 in a first-in, first-out manner.

In the terminal device 20, the learning storage unit 278 may store stratified weights related to learning.

The learning storage unit 278 (weight storage unit) may include weight storage units for multiple layers that store the weights of each layer after deep learning.

The learning unit 218 stores actions (action information) during game play and information related to the actions in the game information storage unit 274 in a first-in, first-out manner, and also stores the weight data of each layer after deep learning as the weight of the corresponding layer. It may also be stored in a data storage unit.

Next, generation of a learning model for obtaining behavior information of the second character OB2 as an output result when related information is input will be described.

In this embodiment, the neural network is configured with multiple layers (e.g., six layers). The neural network is configured, for example, as a recurrent neural network.

In the inference process, when the related information in the storage unit 270 is stored, each item of the related information is input in parallel to the first layer of the neural network. Find the output by multiplying the respective weights for the inputs.

Next, in each layer of the neural network, the output of the previous layer is weighted and calculated in the next layer, and becomes the input to the next layer. In this way, the output result in the output layer becomes the inferred behavior information of the second character OB2. The inferred behavior information of the second character OB2 is used as input for the learning process. Note that inference and learning do not need to be performed at the same time, and may be performed separately.

Next, learning will be explained. In the learning process, first, the error of the difference between the inferred behavior information and the correct behavior information is determined. The weight data optimized for each layer is written in an updated manner. The weights optimized for each layer are updated and stored for each layer.

Next, an example of processing (inference processing) for calculating the output of each layer will be described. The inference process includes, for example, a first input layer, hidden layers (second to fifth layers), and a sixth output layer. Then, in the calculation of the nodes of each layer, the output is obtained by calculating the sum of products of the input value and the weight. Note that for node calculation, a bias may be added to the sum of products of input values and weights of each layer.

For example, the input layer is the number of items of related information, and the output is the number of elements of the behavior information of the second character OB2.

The output of the output layer (final layer) is the action information of the second character OB2 that is estimated to correspond to the related information. In other words, the output layer is provided with nodes equal to the number of action information items (for example, if there are 64 types of actions, then 64). For example, when the values of each node are obtained, if the item with the highest value is, for example, the "footwork counter" item, the action information of the second character OB2 is determined to be footwork counter.

Next, a process (inference process) for calculating an output using weight data of each layer will be explained. In this embodiment, weights are adjusted in inference processing. If it is necessary to modify the weights, it must be done from the layer closest to the output side. The weights are, for example, a 64x64 matrix. You may prepare multiple types of weights. These weights may be set in advance based on the player's operational input or predetermined information in the initial state, but as the learning process is repeated, they are gradually optimized and finally optimized.

In this embodiment, each time one round of competition ends, weight data for six layers of each round is cumulatively stored in the learning storage unit 278 as each weight of the neural network.

The weight data may be read out again the next time the game is played for further learning, or may be used to reproduce learned behavioral information. The weight data may be transmitted and received between multiple terminal devices 20 via a network.

According to the present embodiment, using machine learning, it is possible to create an optimal opponent character OB2 that can reproduce the operations of the player P1's character OB1. Furthermore, by using the results of pre-learning, more human-like behavioral information can be realized even when playing against a computer opponent.

[19] Other supplementary explanations [19.1] Server device In the present embodiment, an example has been described in which the terminal device 20 has a learning function and an inference function when providing a game based on machine learning. However, it may have the learning function and inference function performed by the terminal device 20 described above.

[19.2] Learning Data In this embodiment, a game has been described in which the player P1 learns while playing the game and predicts the actions of the second character OB2. A period in which the first character OB1 to be operated by P1 and the second character OB2, which is a computer character, compete is provided, and learning data (information on the behavior of the first character OB1 and information related to the behavior) is provided during the period. ) may be collected. Furthermore, a trained learning model may be generated based on the collected learning data.

In this embodiment, a period during which player P1 plays against real player P2 (a period during which a first character OB1 operated by player P1 plays against a second character OB2 operated by player P1) may be set in advance, and learning data (information on the actions of first character OB1 and related information about said actions and/or information on the actions of second character OB2 and related information about said actions) may be collected during that period. In addition, a learned learning model may be generated based on the collected learning data.

[19.3] Supplementary explanation of weighting period and level In this embodiment, the weighting period may be changed based on the player's information or the first character's information. The first character information may be a concept including learning data of the first character (information on the first character's behavior and information related to the behavior). Note that the "weighting period" is a period in which a given item (for example, an item selected based on the player's operation input) is given a higher weight among the multiple items that make up the related information. be.

For example, the terminal device 20 determines whether the player is a beginner, intermediate, or advanced based on the player's information. If the player's level is, for example, less than 10, the player is determined to be a beginner. If the player's level is 10 or higher but lower than 80, the player is determined to be an intermediate player. If the player's level is 80 or higher, the player is determined to be an advanced player (veteran).

Then, the terminal device 20 determines a period for weighting according to the level of the player. For example, in the case of a novice player, a period of 30 seconds from the game start time TS during the game period is determined as the period for weighting. For intermediate players, weighting is applied to the 10 second period from the game start time TS and the 10 second period from 30 seconds after the game start time TS to 40 seconds after the game start point TS. Decide on the period in which it will be carried out. In the case of an expert player, a period of 20 seconds from the game start time TS and the last 5 seconds including the game end time TE are determined as weighting periods during the game period. In this way, learning can be weighted in various periods depending on the level of the player.

Note that, similarly to the level of the player, it may be determined whether the first character OB1 to be operated by the player falls under the category of beginner, intermediate, or advanced based on the level of the first character OB1 to be operated by the player. Then, the period during which weighting is performed may be changed based on the information (for example, level) of the first character OB1.

[19.4] Supplementary explanation of second character's behavior control In this embodiment, for each type of behavior of the first character, the behavior of the second character corresponding to the behavior may be defined in advance.

For example, "guard" is defined as the action of the second character that corresponds to the action "jump kick" of the first character OB1. Furthermore, "throw through" is defined as the second character's action corresponding to the "throwing move" action of the first character OB1. Further, "crouch" is defined as the second character's action corresponding to the "kick" action of the first character OB1. In this way, the terminal device 20 defines in advance the actions of the second character OB2 in association with each action (action ID) of the first character OB1.

The terminal device 20 stores at least one of data (hereinafter also referred to as "fixed data") that defines the behavior of a second character corresponding to the behavior of the first character for each type of behavior of the first character, and learning data. may be used to control the actions of the second character.

For example, if the learning data is insufficient and learning of the behavior of the first character OB1 and information related to the behavior is insufficient, the terminal device 20 uses at least one of the fixed data and the learning data to learn the second character OB1's behavior and information related to the behavior. Control your character's actions. In such a case, whether to use fixed data or learning data may be selected based on a given lottery process. For example, if the training data is insufficient (for example, the training dataset is 100 or less), the winning probability of the fixed data is set to 80%, so that the fixed data has a higher probability of being selected by lottery. do. In this way, unnatural behavior toward the second character can be reduced as much as possible. Note that the terminal device 20 may reduce the probability of winning as the number of learning data increases.

[19.5] Supplementary explanation on the degree of similarity The terminal device 20 compares the related information related to the behavior of the unlearned first character OB1 with the searched related information, and quantifies the degree of similarity K between the related information related to the behavior of the unlearned first character OB1 and the searched related information as a percentage.

The degree of matching K is calculated based on a given algorithm or formula. In addition, when searching for learned related information that has a high degree of match with the related information related to the detected unlearned first character's action (for example, "jump kick"), the search process or the search Afterwards, the matching degree K of the searched related information is calculated.

The terminal device 20 may calculate the matching degree K of the searched related information based on the weight of each item.

The terminal device 20 also calculates the degree of match L for each item that constitutes the searched related information. The degree of match L for an item may also be referred to as the degree of item match.

For example, an example will be described in which the degree of coincidence L of "physical strength value", which is one item of related information, is calculated. If the physical strength value of the first character OB1 at the time of the jump kick of the first character OB1 is "100", and the learned "physical strength value" that has been searched is "100", the "physical strength value" It is determined that the degree of coincidence L is 100%. If the learned "physical strength value" found is "80", it is determined that the degree of matching L is 80%. If the learned "physical strength value" that has been searched for is "30", it is determined that the matching degree L of the "physical strength value" is 30%.

The terminal device 20 obtains K of the related information based on the matching degree L of each item of the searched related information. For example, the average value of the matching degree L of each item of the related information searched may be calculated as once K of the related information, or for all the number of items (for example, "100" if there are 100 items), The ratio (for example, 20%) of the number of items with a matching degree L of 80% or more (for example, if there are 20 items with a matching degree of 80% or more, "20") may be set as the degree K of related information.

[19.6] Supplementary explanation of the search method When the terminal device 20 performs a search focusing on weighted items, the terminal device 20 performs the search based on the weighting value of the item. For example, the terminal device 20 may search for related information with a high degree of match L on an item-by-item basis, starting from the item with the highest weighting value.

Specifically, among multiple items of related information, the weight of "physical strength" is "3", the weight of "remaining game time" is "2", and the weight of other items is "1". If there is, related information with the highest matching degree L is searched for the item "physical strength value" which has a high weight value. When multiple pieces of related information with the highest matching degree are searched for "physical strength value" (for example, when multiple pieces of related information are searched for where the matching degree L of physical strength value is 100%), among them, "remaining Regarding the item "game time", related information with the highest matching degree L is searched.

When multiple pieces of related information with the highest matching degree are searched for each item of "physical strength value" and "remaining game time" (for example, if the matching degree is 100% for both "physical strength value" and "remaining game time") If a plurality of pieces of related information are searched), the related information with the highest matching degree L for other items is searched.

[19.7] Behavior control of second character based on matching degree In this embodiment, when the matching degree K of the searched related information is not equal to or higher than a predetermined value, the detected unlearned first character The second character may be made to perform a predefined action corresponding to the action. In this way, unnatural behavior toward the second character can be reduced as much as possible.

For example, assume that related information of "jump kick" and "jump kick" is detected as the unlearned actions of the first character. The terminal device 20 executes a predefined action corresponding to the "jump kick" (for example, "guard") when the matching degree K of the related information of "jump kick" is not equal to or higher than a predetermined value (for example, 70% or higher). ) to the second character.

In addition, in this embodiment, when the matching degree K of the searched related information is not equal to or higher than a predetermined value, a plurality of predefined actions corresponding to the detected action of the unlearned first character are selected. , one action may be selected under predetermined conditions, and the second character may be made to perform the selected action.

For example, it is assumed that three predefined actions corresponding to the "jump kick" of the first character OB1 are "guard", "backward defense", and "crouch". The terminal device 20 selects one action from among the three actions of "guard", "backward defense", and "crouch" under predetermined conditions, and causes the second character to perform the selected action. The predetermined conditions include selection in a lottery process, a case where the game situation (game information) is a predetermined situation, and the like.

For example, if "guard" is selected in the lottery process from among the three actions "guard", "backward defense", and "crouch", the terminal device 20 selects "guard" and selects the second character. have them perform a “guard”.

Furthermore, when the game situation is a predetermined situation, the terminal device 20 performs control such as ensuring that "guard" is selected instead of the lottery process. The predetermined situation is, for example, when the remaining game time is within 10 seconds, when the physical strength value of the second character is 10 or less, and the like.

Furthermore, in this embodiment, similar processing is performed for continuous actions (combos). That is, when the continuous action of the unlearned first character is detected, the terminal device 20 detects the unlearned first character when the matching degree K of the related information searched from the learned related information is not higher than the predetermined value. The second character is caused to perform a predefined action corresponding to the unlearned continuous action of the first character.

For example, if the first character OB1 performs a consecutive action of kicking twice in succession, one of the actions "low guard," "kick," or "moving away" is selected by lottery. The second character OB2 is then made to perform the selected action.

Further, when the first character OB1 performs a continuous action of kicking twice in a row, the physical strength value of the second character OB2 is less than the first predetermined value (for example, less than 50% of the upper limit). When the physical strength value is equal to or higher than the second predetermined value (for example, 30% or more of the upper limit value), the second character OB2 performs "lower guard", and the physical strength value of the second character OB2 is set to the second predetermined value. If the value is less than the upper limit value (for example, less than 30% of the upper limit value), the second character OB2 may be controlled to "move away".

[20] Application Example This embodiment is applicable not only to competitive games, but also to racing games, shooting games, music games, RPGs (role playing games), action games, sports games, training simulation games, and the like.

It is also applicable to a service where a user's character (avatar) is placed in a three-dimensional virtual space to enjoy the virtual space.

[21] Others The present invention is not limited to the above-described embodiment, and various modifications are possible. For example, a term cited in the description of the specification or drawings as a term with a broad meaning or synonymous meaning can be replaced with a term with a broad meaning or synonymous meaning in other descriptions of the specification or drawings.

The present invention includes configurations that are substantially the same as the configurations described in the embodiments (for example, configurations with the same functions, methods, and results, or configurations with the same purpose and effect). The present invention also includes configurations that replace non-essential parts of the configurations described in the embodiments. The present invention also includes configurations that achieve the same effects as the configurations described in the embodiments, or that can achieve the same purpose. The present invention also includes configurations in which publicly known technology is added to the configurations described in the embodiments.

As mentioned above, the embodiments of the present invention have been described in detail, but those skilled in the art will easily understand that many modifications can be made without substantially departing from the novelty and effects of the present invention. . Therefore, all such modifications are included within the scope of the present invention.

10 server device, 20, 20A, 20B, 20C terminal device,
100 processing unit, 111 game processing unit, 112 display control unit,
119 reception unit, 120 communication control unit, 121 web processing unit,
122 Management Department, 123 Notification Department,
160 input section, 170 storage section, 171 main storage section,
174 Game information storage unit, 176 Player information storage unit,
180 Information storage medium, 196 Communication Department,
200 processing unit, 211 communication control unit, 212 object space setting unit,
213 movement/motion control unit, 214 operation input reception unit, 215 game control unit,
216 virtual camera control unit, 217 behavior control unit, 218 learning unit, 219 setting unit,
220 judgment unit, 221 search unit, 222 display control unit,
223 changing unit, 230 image generating unit,
240 sound processing unit, 260 input unit, 262 detection unit, 270 storage unit,
271 main storage unit, 272 image buffer, 274 game information storage unit,
276 player information storage unit, 278 learning storage unit, 279 learning model storage unit,
280 information storage medium, 290 display section,
292 Sound output section, 296 Communication section

Claims (13)

A game system for controlling the actions of a second character in a competitive game in which a first character whose actions are controlled based on a player's operation input competes with a second character controlled by a computer, the game system comprising:
a learning section that learns the actions of the first character and related information related to the actions;
a behavior control unit that controls the behavior of the second character using the learning data learned in the learning unit;
A setting for setting at least one of a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information, based on the information of the player, before learning. A game system comprising: A game system for controlling the actions of a second character in a competitive game in which a first character whose actions are controlled based on a player's operation input competes with a second character controlled by a computer, the game system comprising:
a learning section that learns the actions of the first character and related information related to the actions;
a behavior control unit that controls the behavior of the second character using the learning data learned in the learning unit;
Based on the player's information, at least one of a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information is determined according to the game progress. A game system comprising: a setting section for making settings. In claim 1 or 2,
The setting section includes:
A game system characterized in that, based on an operation input by the player, selection of an item having a higher weight among a plurality of items forming the related information is accepted. In claim 1 or 2,
The setting unit is
A game system characterized in that a priority order is set for each of a plurality of items of said related information based on operational input by said player. In claim 1 or 2,
A game system, wherein the player information includes information regarding the player's play history. In claim 1 or 2,
The game system is characterized in that the player's information includes information regarding the player's past play tendencies. In claim 1 or 2,
A game system characterized in that the player's information includes information related to other players determined based on predetermined information of the player. A program that causes a computer to control the actions of a second character in a competitive game in which a first character whose actions are controlled based on a player's operation input competes with a second character controlled by a computer. hand,
a learning section that learns the actions of the first character and related information related to the actions;
a behavior control unit that controls the behavior of the second character using the learning data learned in the learning unit;
A setting for setting at least one of a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information, based on the information of the player, before learning. A program that, as a part, causes a computer to function. In a battle game in which a first character whose behavior is controlled based on a player's operation input competes against a second character controlled by a computer, a program for causing a computer to control the behavior of the second character, comprising:
a learning unit that learns an action of a first character and related information related to the action;
a behavior control unit that controls a behavior of a second character using the learning data learned by the learning unit;
A program that causes a computer to function as a setting unit that sets at least one of the combination of multiple items that make up the related information, the learning frequency, and the weighting of each of the multiple items that make up the related information, based on the player's information and in accordance with the game progress. A learning device for controlling the behavior of a second character in a competitive game in which a first character whose behavior is controlled based on a player's operation input competes with a second character controlled by a computer, the learning device comprising:
a learning section that learns the actions of the first character and related information related to the actions;
a behavior control unit that controls the behavior of the second character using the learning data learned in the learning unit;
A setting for setting at least one of a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information, based on the information of the player, before learning. A learning device comprising: A learning device for controlling the behavior of a second character in a competitive game in which a first character whose behavior is controlled based on a player's operation input competes with a second character controlled by a computer, the learning device comprising:
a learning section that learns the actions of the first character and related information related to the actions;
a behavior control unit that controls the behavior of the second character using the learning data learned in the learning unit;
Based on the player's information, at least one of a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information is determined according to the game progress. A learning device comprising: a setting section for setting. In a competitive game in which a first character whose actions are controlled based on the player's operation input competes with a second character whose actions are controlled by a computer, the player's terminal is sent from a server device that controls the actions of the second character. A game providing method for providing information regarding a game to a device, the method comprising:
transmitting information regarding the player's operation input accepted by the terminal device to the server device;
transmitting information regarding the game from the server device to the terminal device;
The server device includes:
a learning section that learns the actions of the first character and related information related to the actions;
a behavior control unit that controls the behavior of the second character using the learning data learned in the learning unit;
A setting for setting at least one of a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information, based on the information of the player, before learning. A game providing method characterized by comprising: In a competitive game in which a first character whose actions are controlled based on the player's operation input competes with a second character whose actions are controlled by a computer, the player's terminal is sent from a server device that controls the actions of the second character. A game providing method for providing information regarding a game to a device, the method comprising:
transmitting information regarding the player's operation input accepted by the terminal device to the server device;
transmitting information regarding the game from the server device to the terminal device;
The server device includes:
a learning section that learns the actions of the first character and related information related to the actions;
a behavior control unit that controls the behavior of the second character using the learning data learned in the learning unit;
Based on the player's information, at least one of a combination of a plurality of items constituting the related information, a learning frequency, and a weight of each of the plurality of items constituting the related information is determined according to the game progress. A game providing method, comprising: a setting section for making settings.

Images