WO2023024762A1 - 人工智能对象控制方法、装置、设备及存储介质 - Google Patents

人工智能对象控制方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2023024762A1
WO2023024762A1 PCT/CN2022/106330 CN2022106330W WO2023024762A1 WO 2023024762 A1 WO2023024762 A1 WO 2023024762A1 CN 2022106330 W CN2022106330 W CN 2022106330W WO 2023024762 A1 WO2023024762 A1 WO 2023024762A1
Authority
WO
WIPO (PCT)
Prior art keywords
artificial intelligence
virtual
instruction
control
target
Prior art date
Application number
PCT/CN2022/106330
Other languages
English (en)
French (fr)
Inventor
张树宝
Original Assignee
腾讯科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Publication of WO2023024762A1 publication Critical patent/WO2023024762A1/zh
Priority to US18/135,914 priority Critical patent/US20230293995A1/en

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/55—Controlling game characters or game objects based on the game progress
    • A63F13/56—Computing the motion of game characters with respect to other game characters, game objects or elements of the game scene, e.g. for simulating the behaviour of a group of virtual soldiers or for path finding
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/30—Interconnection arrangements between game servers and game devices; Interconnection arrangements between game devices; Interconnection arrangements between game servers
    • A63F13/35—Details of game servers
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/40—Processing input control signals of video game devices, e.g. signals generated by the player or derived from the environment
    • A63F13/42—Processing input control signals of video game devices, e.g. signals generated by the player or derived from the environment by mapping the input signals into game commands, e.g. mapping the displacement of a stylus on a touch screen to the steering angle of a virtual vehicle
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/45—Controlling the progress of the video game
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/60—Generating or modifying game content before or while executing the game program, e.g. authoring tools specially adapted for game development or game-integrated level editor
    • A63F13/67—Generating or modifying game content before or while executing the game program, e.g. authoring tools specially adapted for game development or game-integrated level editor adaptively or by learning from player actions, e.g. skill level adjustment or by storing successful combat sequences for re-use
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/80—Special adaptations for executing a specific game genre or game mode
    • A63F13/822—Strategy games; Role-playing games
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
    • G06F3/0481—Interaction techniques based on graphical user interfaces [GUI] based on specific properties of the displayed interaction object or a metaphor-based environment, e.g. interaction with desktop elements like windows or icons, or assisted by a cursor's changing behaviour or appearance
    • G06F3/04815—Interaction with a metaphor-based environment or interaction object displayed as three-dimensional, e.g. changing the user viewpoint with respect to the environment or object
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
    • G06F3/0481—Interaction techniques based on graphical user interfaces [GUI] based on specific properties of the displayed interaction object or a metaphor-based environment, e.g. interaction with desktop elements like windows or icons, or assisted by a cursor's changing behaviour or appearance
    • G06F3/04817—Interaction techniques based on graphical user interfaces [GUI] based on specific properties of the displayed interaction object or a metaphor-based environment, e.g. interaction with desktop elements like windows or icons, or assisted by a cursor's changing behaviour or appearance using icons
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
    • G06F3/0484—Interaction techniques based on graphical user interfaces [GUI] for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range
    • G06F3/04842—Selection of displayed objects or displayed text elements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
    • G06F3/0484—Interaction techniques based on graphical user interfaces [GUI] for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range
    • G06F3/0486—Drag-and-drop
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
    • G06F3/0487—Interaction techniques based on graphical user interfaces [GUI] using specific features provided by the input device, e.g. functions controlled by the rotation of a mouse with dual sensing arrangements, or of the nature of the input device, e.g. tap gestures based on pressure sensed by a digitiser
    • G06F3/0488—Interaction techniques based on graphical user interfaces [GUI] using specific features provided by the input device, e.g. functions controlled by the rotation of a mouse with dual sensing arrangements, or of the nature of the input device, e.g. tap gestures based on pressure sensed by a digitiser using a touch-screen or digitiser, e.g. input of commands through traced gestures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/451—Execution arrangements for user interfaces
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects

Definitions

  • the embodiments of the present application relate to the technical field of artificial intelligence, and in particular to an artificial intelligence object control method, device, device, and storage medium.
  • An artificial intelligence object is an object controlled based on a neural network model, rather than an object controlled by a user based on a terminal.
  • Embodiments of the present application provide an artificial intelligence object control method, device, device, and storage medium, which expand the functions of the artificial intelligence object and provide an artificial intelligence object that is controllable in operation. Described technical scheme is as follows:
  • a method for controlling an artificial intelligence object comprising:
  • the terminal displays a virtual scene interface, and the virtual scene interface displays the controlled virtual object of the terminal;
  • the terminal detects a control instruction through the virtual scene interface, and the control instruction is used to instruct the artificial intelligence object in the own camp of the controlled virtual object to perform a target operation;
  • the terminal controls the artificial intelligence object to perform the target operation based on the control instruction.
  • a method for controlling an artificial intelligence object comprising:
  • the server receives the control instruction sent by the terminal, and the control instruction is used to instruct the artificial intelligence object to perform the target operation;
  • the server controls the artificial intelligence object to perform the target operation based on the control instruction
  • the artificial intelligence object belongs to the own camp of the controlled virtual object of the terminal.
  • an artificial intelligence object control device comprising:
  • a display module configured to display a virtual scene interface, and the virtual scene interface displays the controlled virtual object at the local end;
  • a detection module configured to detect a control instruction through the virtual scene interface, and the control instruction is used to instruct the artificial intelligence object in the own camp of the controlled virtual object to perform a target operation;
  • a first control module configured to control the artificial intelligence object to perform the target operation based on the control instruction.
  • an artificial intelligence object control device comprising:
  • a receiving module configured to receive a control instruction sent by the terminal, and the control instruction is used to instruct the artificial intelligence object to perform a target operation;
  • control module configured to control the artificial intelligence object to perform the target operation based on the control instruction
  • the artificial intelligence object belongs to the own camp of the controlled virtual object of the terminal.
  • a terminal in another aspect, includes a processor and a memory, at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the above aspects The operations performed by the artificial intelligence object control method described above.
  • a server in another aspect, includes a processor and a memory, at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the above aspects The operations performed by the artificial intelligence object control method described above.
  • a computer-readable storage medium wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to realize the artificial A smart object controls what a method does.
  • a computer program product or computer program includes computer program code, the computer program code is stored in a computer-readable storage medium, and a processor of a computer device reads from the computer Reading the storage medium reads the computer program code, and the processor executes the computer program code, so that the computer device implements the operations performed by the artificial intelligence object control method as described in the above aspects.
  • the embodiment of the present application expands the functions of the artificial intelligence object and provides a controllable artificial intelligence object.
  • the user only needs to issue a control command to control the artificial intelligence object belonging to the same camp as the controlled virtual object to perform the target operation.
  • the control of the artificial intelligence object is realized, thereby realizing the synergy between the human and the artificial intelligence object.
  • FIG. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present application.
  • Fig. 2 is a flowchart of a method for controlling an artificial intelligence object provided by an embodiment of the present application.
  • Fig. 3 is a schematic diagram of a virtual scene interface provided by an embodiment of the present application.
  • Fig. 4 is a schematic diagram of another virtual scene interface provided by the embodiment of the present application.
  • Fig. 5 is a schematic diagram of another virtual scene interface provided by the embodiment of the present application.
  • Fig. 6 is a schematic diagram of another virtual scene interface provided by the embodiment of the present application.
  • Fig. 7 is a schematic diagram of another virtual scene interface provided by the embodiment of the present application.
  • Fig. 8 is a schematic diagram of another virtual scene interface provided by the embodiment of the present application.
  • Fig. 9 is a schematic diagram of another virtual scene interface provided by the embodiment of the present application.
  • Fig. 10 is a schematic diagram of another virtual scene interface provided by the embodiment of the present application.
  • Fig. 11 is a flow chart of a method for controlling an artificial intelligence object provided by an embodiment of the present application.
  • Fig. 12 is a flow chart of another artificial intelligence object control method provided by the embodiment of the present application.
  • FIG. 13 is a schematic diagram of a deep hidden variable neural network model provided by an embodiment of the present application.
  • FIG. 14 is a schematic diagram of interaction between a terminal and a server provided by an embodiment of the present application.
  • Fig. 15 is a flow chart of a method for verifying a control instruction provided by an embodiment of the present application.
  • FIG. 16 is a schematic diagram of a gridded map provided by an embodiment of the present application.
  • FIG. 17 is a schematic diagram of a grid at multiple times provided by an embodiment of the present application.
  • Fig. 18 is a schematic diagram of a prediction model provided by an embodiment of the present application.
  • Fig. 19 is a schematic diagram of another deep hidden variable neural network model provided by the embodiment of the present application.
  • Fig. 20 is a schematic diagram of an intention prediction model provided by an embodiment of the present application.
  • Fig. 21 is a schematic diagram of a neural network model provided by an embodiment of the present application.
  • FIG. 22 is a schematic diagram of a shared coding network provided by an embodiment of the present application.
  • Fig. 23 is a schematic structural diagram of an artificial intelligence object control device provided by an embodiment of the present application.
  • Fig. 24 is a schematic structural diagram of another artificial intelligence object control device provided by an embodiment of the present application.
  • FIG. 25 is a schematic structural diagram of a terminal provided by an embodiment of the present application.
  • FIG. 26 is a schematic structural diagram of a server provided by an embodiment of the present application.
  • the virtual scene involved in this application is used to simulate a three-dimensional virtual space
  • the three-dimensional virtual space is an open space
  • the virtual scene is used to simulate the real environment in reality
  • the virtual scene includes sky, land, ocean, etc.
  • the land includes environmental elements such as deserts and cities.
  • virtual items may also be included in the virtual scene, for example, buildings, vehicles, and props such as weapons required by virtual objects in the virtual scene to arm themselves or fight with other virtual objects.
  • the virtual scene can also be used to simulate a real environment under different weathers, for example, weather such as sunny days, rainy days, foggy days or dark nights.
  • the user controls a virtual object to move in the virtual scene.
  • the virtual object is a virtual avatar representing the user in the virtual scene.
  • the avatar is in any form, such as a person or an animal. This application There is no limit to this. Taking shooting games as an example, the user controls the virtual object to freely fall, glide, or open a parachute to fall in the sky of the virtual scene, run, jump, crawl, bend forward, etc. on land, and can also control the virtual object. The object swims, floats, or dives in the ocean. Of course, the user can also control the virtual object to move in the virtual scene on a vehicle.
  • the user can also control the virtual object to enter and exit the building in the virtual scene, find and pick up the virtual items (for example, props such as weapons) in the virtual scene, so as to fight with other virtual objects through the picked up virtual items
  • the virtual Items can be clothing, helmets, body armor, medical supplies, cold weapons or hot weapons, etc., and can also be virtual items left after other virtual objects are eliminated.
  • the above-mentioned scenario is used as an example for illustration, and this embodiment of the present application does not specifically limit it.
  • the embodiment of the present application takes an electronic game scene as an example.
  • the user operates on the terminal in advance. After the terminal detects the user's operation, it downloads the game configuration file of the electronic game.
  • the game configuration file includes the application program and interface of the electronic game. Display data or virtual scene data, etc., so that when the user logs in the electronic game on the terminal, the game configuration file is invoked to render and display the electronic game interface.
  • the user performs a touch operation on the terminal. After the terminal detects the touch operation, it determines the game data corresponding to the touch operation, and renders and displays the game data.
  • the game data includes virtual scene data, Behavioral data of virtual objects, etc.
  • the terminal When the terminal renders and displays the virtual scene, it displays the virtual scene in full screen.
  • the terminal can also independently display the global map in the first preset area of the current display interface while displaying the virtual scene on the current display interface.
  • the terminal The global map may also be displayed only when a click operation on a preset button is detected.
  • the global map is used to display a thumbnail of the virtual scene, and the thumbnail is used to describe geographical features such as terrain, landform, and geographical location corresponding to the virtual scene.
  • the terminal can also display thumbnails of virtual scenes within a certain distance around the current virtual object on the current display interface. Thumbnails of the scene so that users can view not only the virtual scene around them, but also the virtual scene as a whole.
  • the terminal When the terminal detects a zoom operation on the full thumbnail, it can also zoom and display the full thumbnail.
  • the specific display positions and shapes of the first preset area and the second preset area can be set according to user's operating habits. For example, in order not to cause too much occlusion to the virtual scene, the first preset area can be a rectangular area in the upper right corner, lower right corner, upper left corner or lower left corner of the current display interface, etc., and the second preset area can be the current display interface. The square area on the right or left side of the interface. Of course, the first preset area and the second preset area can also be circular areas or areas of other shapes.
  • the specific display position and shape of the preset area in the embodiment of the present application Not limited.
  • MOBA Multiplayer Online Battle Arena
  • It is a game that provides several strongholds in a virtual scene, and users in different camps control virtual objects to fight in the virtual scene, occupy strongholds or destroy enemy camp strongholds.
  • users are divided into at least two hostile camps, and different virtual teams belonging to at least two hostile camps occupy their respective map areas and compete with a certain victory condition as the goal.
  • the victory conditions include but are not limited to: Occupy strongholds or destroy enemy faction strongholds, kill the virtual objects of the enemy faction, ensure their own survival in the specified scene and time, rob a certain resource, and exceed the opponent's interaction score within the specified time at least one of the
  • mobile MOBA games can divide users into two hostile camps, disperse virtual objects controlled by users in virtual scenes and compete with each other, and destroy or occupy all the enemy's strongholds as victory conditions.
  • the skill types of this skill can include attack skills, defense skills, healing skills, auxiliary skills, killing skills, etc.
  • Each virtual object has Each may have one or more fixed skills, but different virtual objects generally have different skills, and different skills may produce different effects. For example, if the virtual object releases the attack skill and hits the hostile virtual object, it will cause certain damage to the hostile virtual object, which usually means deducting a part of the virtual life value of the hostile virtual object.
  • the virtual object releases the healing skill and hits friendly virtual objects, then it will produce certain healing for the friendly virtual objects, usually in the form of recovering part of the virtual health of the friendly virtual objects, and other various skills can produce corresponding effects, so I won’t enumerate them one by one here .
  • Human-AI collaboration Humans and AI collaborate (cooperate to complete a certain task) and interact (transfer signals to each other, such as text, etc.) in the same environment. This type of AI is usually called collaborative AI. Compared with the automation attributes of traditional AI agents, collaborative AI also emphasizes the collaboration and interactivity of AI agents.
  • Action-controllable AI agents AI agents whose action strategies can be changed by control commands. Some AI agents usually only have the attribute of automation, that is, the agent can make autonomous decisions and act independently; but its actions cannot be controlled, that is, the control commands initiated by the outside world cannot change the actions of the AI agent. In the scenario of human-AI collaboration, in addition to automation attributes, AI agents must also be cooperative and interactive, that is, agents must be able to cooperate and interact with other agents. Action-controllable AI agents can meet the needs of this kind of collaborative interaction, and its actions can be changed by external control instructions, so that collaborative interaction can be completed.
  • FIG. 1 is a schematic structural diagram of an implementation environment provided by an embodiment of the present application.
  • the implementation environment includes a terminal 101 and a server 102 .
  • the terminal 101 is a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart TV, a smart watch, etc., but is not limited thereto.
  • the server 102 is an independent physical server, or, the server 102 is a server cluster or a distributed system composed of multiple physical servers, or, the server 102 provides cloud services, cloud databases, cloud computing, cloud function Cloud servers for basic cloud computing services such as cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network, content distribution network), and big data and artificial intelligence platforms.
  • the terminal 101 and the server 102 are directly or indirectly connected through wired or wireless communication, which is not limited in this application.
  • the server 102 provides a virtual scene for the terminal 101. Through the virtual scene provided by the server 102, the terminal 101 can display virtual objects, virtual props, etc., and the terminal 101 provides an operating environment for the user to detect operations performed by the user. The server 102 can perform background processing for the operations detected by the terminal, and provide background support for the terminal 101 .
  • the terminal 101 installs a game application provided by the server 102, through which the terminal 101 and the server 102 can interact.
  • the terminal 101 runs the game application, provides the user with an operating environment for the game application, can detect the user's operation on the game application, and sends an operation instruction to the server 102, and the server 102 responds according to the operation instruction, and returns the response result to the
  • the terminal 101 is displayed by the terminal 101, so as to realize human-computer interaction.
  • the terminal runs a competitive game, and the game is divided into an own camp and an enemy camp, wherein the own camp includes the controlled virtual object and the artificial intelligence object of the terminal.
  • the terminal can detect the control command, and based on the control command, control the artificial intelligence object in our camp to perform the target operation corresponding to the control command.
  • the control instruction is an assembly instruction, and the terminal will control the artificial intelligence object of our camp to move closer to the controlled virtual object of the terminal until the assembly of all members is completed, and the control of the artificial intelligence object is realized.
  • FIG. 2 is a flow chart of a method for controlling an artificial intelligence object provided in an embodiment of the present application. The method is executed by a terminal. As shown in FIG. 2 , the method includes:
  • the terminal displays a virtual scene interface.
  • the virtual scene in the embodiment of the present application includes virtual objects of at least two hostile camps, wherein each camp may include virtual objects controlled by a terminal, artificial intelligence objects, and objects such as defense towers.
  • the artificial intelligence object refers to an object controlled based on a neural network model, and the artificial intelligence object may also be called an AI agent.
  • the terminal displays a virtual scene interface, which displays the controlled virtual object of the terminal, and can also display other objects other than the controlled virtual object, such as objects in the controlled virtual object's own camp , or an object in the non-own faction of the controlled dummy.
  • the objects in the non-own camp of the charged virtual object include: objects in the enemy camp of the charged virtual object, and neutral objects, such as wild monsters.
  • the virtual scene interface displays the virtual scene in the first-person perspective of the controlled virtual object of the terminal, and then displays a part of the virtual scene within the viewing angle range of the controlled virtual object on the virtual scene interface, and the other objects in the virtual scene.
  • the virtual scene interface displays the virtual scene in a third-person perspective
  • the virtual scene interface displays a part of the virtual scene within the perspective range of the third-person perspective, and the controlled virtual scene located in the part of the virtual scene. objects and other objects.
  • the terminal detects the control instruction through the virtual scene interface.
  • the embodiment of the present application provides a scheme for interaction between humans and artificial intelligence objects.
  • the user triggers a control instruction in the virtual scene interface, and the control instruction is used to instruct the artificial intelligence object in the controlled virtual object's own camp to execute the target.
  • the artificial intelligence object can be controlled to perform the target operation, thereby simulating the effect of the artificial intelligence object performing the operation according to the user's instructions, that is, realizing the interaction between the user and the artificial intelligence object.
  • the synergy also creates an effect of synergy between the controlled virtual object and the artificial intelligence object in the virtual scene.
  • control instruction can be triggered by at least one operation of the user in the virtual scene interface, such as clicking a certain control, dragging a certain control, sliding or selecting a certain object.
  • control instruction is an attack instruction, or a retreat instruction, or an assembly instruction, or other control instructions.
  • control instruction is an attack instruction
  • the terminal detects the attack instruction through the virtual scene interface, and the attack instruction is used to instruct the artificial intelligence object to attack the first target virtual object that is not its own camp.
  • the detection of attack instructions through the virtual scene interface includes the following methods:
  • the first is to detect the trigger operation of the attack control through the virtual scene interface, and generate an attack instruction, which is used to instruct the artificial intelligence object to attack the virtual object closest to the controlled virtual object of the terminal in the non-own camp.
  • the virtual scene interface displays an attack control, and the user triggers the attack control to initiate the attack command.
  • the default first target virtual object to be attacked this time is: the accused virtual object in the non-own camp and the terminal. The virtual object closest to the object.
  • attack instruction is also used to instruct other virtual objects in the own camp of the controlled virtual object to also attack the virtual object closest to the controlled virtual object in the non-own camp.
  • the triggering operation on the attack control may be a click operation on the attack control, a long press operation on the attack control, an operation of dragging the attack control, or other operations on the attack control.
  • the second is to detect the operation of dragging the attack control to the object identifier of the first target virtual object through the virtual scene interface, and generate an attack instruction, which is used to instruct the artificial intelligence object to attack the virtual object corresponding to the object identifier.
  • the virtual scene interface displays an attack control and an object identifier of at least one object in the virtual scene, wherein the object identifier is used to determine a unique object, and may be an icon or a name of a virtual object.
  • the user drags the attack control to the object ID of the first target virtual object to initiate the attack command, and the object to be attacked this time is the first target virtual object corresponding to the object ID.
  • the first target virtual object can be any object in a camp other than your own.
  • the first target virtual object is a virtual object in the enemy camp, or an artificial intelligence object in the enemy camp, or an The defensive towers in the camp, or wild monsters, or other virtual objects.
  • attack instruction is also used to instruct other virtual objects in the own camp of the controlled virtual object to also attack the first target virtual object corresponding to the object identifier.
  • the virtual scene interface displays a thumbnail map of the virtual scene
  • the map displays icons for each object.
  • the object identifier 302 is an icon of a certain hero in the enemy camp.
  • the virtual scene interface displays a thumbnail map of the virtual scene
  • the map displays icons for each object.
  • the object identifier 402 is an icon of a wild monster.
  • the third method is to detect the selection operation of the object identifier of the first target virtual object through the virtual scene interface, and generate an attack instruction, and the attack instruction is used to instruct the artificial intelligence object to attack the virtual object corresponding to the object identifier.
  • the virtual scene interface displays an object identifier of at least one object in the virtual scene, wherein the object identifier is used to determine a unique object, and may be an icon or a name of a virtual object.
  • the user triggers the selection operation of the object identifier of the first target virtual object to initiate the attack command, and the object to be attacked this time is the first target virtual object corresponding to the object identifier.
  • the first target virtual object can be any object in a camp other than your own.
  • the first target virtual object is a virtual object in the enemy camp, or an artificial intelligence object in the enemy camp, or an The defensive towers in the camp, or wild monsters, or other virtual objects.
  • attack instruction is also used to instruct other virtual objects in the camp of the controlled virtual object to also attack the virtual object corresponding to the object identifier.
  • the virtual scene interface displays a map of the virtual scene and icons of various objects, wherein the object identifier 501 is an icon of a certain hero in the enemy camp.
  • the user triggers the selection operation of the object identifier 501 to generate an attack instruction, and the attack instruction is used to instruct the artificial intelligence object to attack the hero corresponding to the object identifier 501 .
  • the virtual scene interface displays a map of the virtual scene and icons of various objects, wherein the object identifier 601 is an icon of a wild monster.
  • the user triggers the selection operation of the object identifier 601 to generate an attack instruction, and the attack instruction is used to instruct the artificial intelligence object to attack the wild monster corresponding to the object identifier 601 .
  • the virtual scene interface displays a map of the virtual scene and icons of various objects, wherein the object identifier 701 is an icon of a soldier in the enemy camp.
  • the user triggers the selection operation of the object identifier 701 to generate an attack instruction, and the attack instruction is used to instruct the artificial intelligence object to attack the soldier corresponding to the object identifier 701 .
  • the user only needs to trigger the attack command to instruct the artificial intelligence object in the own camp to attack, realizing the interaction between the user and the artificial intelligence object, thereby improving the operation efficiency of the own camp,
  • the team combat capability of the own camp is enhanced, which helps the camp to better complete team tasks.
  • control instruction is an evacuation instruction
  • the terminal detects the evacuation instruction through the virtual scene interface, and the evacuation instruction is used to instruct the artificial intelligence object to evacuate to a safe location.
  • the withdrawal instruction is detected through the virtual scene interface, including the following methods:
  • the first one is to detect the trigger operation of the retreat control through the virtual scene interface to generate a retreat instruction, which is used to instruct the artificial intelligence object to retreat to a safe position.
  • the virtual scene interface displays a evacuation control, and the user triggers the evacuation control to initiate the evacuation command, instructing the artificial intelligence object to evacuate to a safe location.
  • the default safe location is the safe location closest to each artificial intelligence object, or the location of the defense tower closest to each artificial intelligence object in the own camp, or other locations.
  • the evacuation instruction is also used to instruct other virtual objects in the own camp of the controlled virtual object to also evacuate to a safe location.
  • the triggering operation on the retreat control may be a click operation on the retreat control, a long press operation on the retreat control, an operation of dragging the retreat control, or other operations on the retreat control.
  • the second is to detect the operation of dragging the retreat control to the object identification of the third target virtual object of the own camp through the virtual scene interface, and generate a retreat command, which is used to instruct the artificial intelligence object to retreat to the third target virtual object s position.
  • the virtual scene interface displays a withdrawal control and an object identifier of at least one object in the virtual scene, wherein the object identifier is used to determine a unique object, and may be an icon or a name of a virtual object.
  • the user drags the retreat control to the object identifier of the third target virtual object to initiate the retreat command, and this time the artificial intelligence object needs to retreat to the position where the third target virtual object is located.
  • the first target virtual object can be any object in the own camp, for example, the third target virtual object is a virtual object in the own camp, or an artificial intelligence object in the own camp, or an object of the own camp. Defensive towers in the faction.
  • the user can drag the retreat control to different object identifiers to instruct the artificial intelligence object to retreat to the positions of different virtual objects.
  • the retreat instruction is also used to instruct other virtual objects in the own camp of the controlled virtual object to also retreat to the position where the third target virtual object is located.
  • the virtual scene interface displays a thumbnail map of the virtual scene
  • the user clicks on the thumbnail map to enlarge the thumbnail map, that is, the map of the virtual scene and the retreat control 801 are displayed in the virtual scene interface
  • the map displays icons for each object.
  • the object identifier 802 is an icon of a defense tower in the own camp.
  • the user drags the attack control 801 to the object identifier 802 to generate a retreat instruction, and the retreat instruction is used to instruct the artificial intelligence object to retreat to the position of the defense tower corresponding to the object identifier 802 .
  • the third way is to detect the selection operation of the object identifier of the third target virtual object through the virtual scene interface, and generate a retreat instruction, and the retreat instruction is used to instruct the artificial intelligence object to retreat to the position of the third target virtual object.
  • the virtual scene interface displays an object identifier of at least one object in the virtual scene, wherein the object identifier is used to determine a unique object, and may be an icon or a name of a virtual object.
  • the user triggers the selection operation of the object identifier of the third target virtual object to initiate the retreat command, and the artificial intelligence object needs to retreat to the position where the third target virtual object is located this time.
  • the first target virtual object can be any object in the own camp, for example, the third target virtual object is a virtual object in the own camp, or an artificial intelligence object in the own camp, or an object of the own camp. Defensive towers in the faction. By selecting different object identifiers, the user can instruct the artificial intelligence object to retreat to the location of different virtual objects.
  • the retreat instruction is also used to instruct other virtual objects in the own camp of the controlled virtual object to also retreat to the position where the third target virtual object is located.
  • the virtual scene interface displays a map of the virtual scene and icons of various objects, wherein the object identifier 901 is an icon of a defense tower in the own camp.
  • the user triggers the selection operation of the object identifier 901 to generate a retreat instruction, and the retreat instruction is used to instruct the artificial intelligence object to retreat to the position where the defense tower corresponding to the object identifier 901 is located.
  • control instruction is a collection instruction
  • the terminal detects the collection instruction through the virtual scene interface, and the collection instruction is used to instruct the artificial intelligence object to move to the position of the controlled virtual object or the second target virtual object of the own camp.
  • the user only needs to trigger the retreat command to instruct the artificial intelligence objects in the camp to retreat to a safe location, realizing the interaction between the user and the artificial intelligence objects, thereby improving the operation of the camp Efficiency, which enhances the team combat capability of the own camp, thus helping to protect the artificial intelligence objects in the own camp.
  • the collection instruction is detected through the virtual scene interface, including the following methods:
  • the first one is to detect the trigger operation on the collection control through the virtual scene interface, and generate a collection instruction, which is used to instruct the artificial intelligence object to move to the position of the controlled virtual object.
  • the virtual scene interface displays a collection control, and the user triggers the collection control to initiate the collection command.
  • the default artificial intelligence object needs to reach the location of the controlled virtual object, so as to gather with the virtual objects in the camp. That is to say, the artificial intelligence object needs to move to the position where the controlled virtual object is located.
  • the set instruction is also used to instruct other virtual objects in the controlled virtual object's own camp to also move to the position where the controlled virtual object is located.
  • the triggering operation on the collection control may be a click operation on the collection control, a long press operation on the collection control, an operation of dragging the collection control, or other operations on the collection control.
  • the second is to detect the operation of dragging the collection control to the object identification of the second target virtual object of the own camp through the virtual scene interface, and generate a collection instruction, which is used to instruct the artificial intelligence object to move towards the second target virtual object.
  • the location moves.
  • the virtual scene interface displays a collection control and an object identifier of at least one object in the virtual scene, wherein the object identifier is used to determine a unique object, and may be an icon or a name of a virtual object.
  • the user drags the collection control to the object identifier of the second target virtual object to initiate the collection instruction, instructing the artificial intelligence object to go to the location of the second target virtual object.
  • the first target virtual object can be any object in the own camp, for example, the second target virtual object is a virtual object in the own camp, or an artificial intelligence object in the own camp, or is Defensive towers in the faction.
  • the user drags the collection control to the object identifiers of different virtual objects to instruct the artificial intelligence object to retreat to the position of the different virtual objects.
  • the set instruction is also used to instruct other virtual objects in the own camp of the controlled virtual object to also move to the position where the second target virtual object corresponding to the object identifier is located.
  • the virtual scene interface displays a thumbnail map of the virtual scene
  • the user clicks on the thumbnail map to enlarge the thumbnail map, that is, the map of the virtual scene and the collection control 1001 are displayed in the virtual scene interface
  • the map displays icons for each object.
  • the object identifier 1002 is an icon of a certain hero in the own camp.
  • the user drags the collection control 1001 to the object identifier 1002 to generate a collection instruction, which is used to instruct the artificial intelligence object to move to the position of the hero corresponding to the object identifier 1002 .
  • the third way is to detect the selection operation of the object identifier of the third target virtual object through the virtual scene interface, and generate a set instruction, which is used to instruct the artificial intelligence object to move to the position of the second target virtual object.
  • the virtual scene interface displays an object identifier of at least one object in the virtual scene, wherein the object identifier is used to determine a unique object, and may be an icon or a name of a virtual object.
  • the set instruction can be initiated, instructing the artificial intelligence object to move to the position where the second target virtual object is located.
  • the second target virtual object can be any object in the own camp, for example, the second target virtual object is a virtual object in the own camp, or an artificial intelligence object in the own camp, or is an object of the own camp. Defensive towers in the faction.
  • the set instruction is also used to instruct other virtual objects in the own camp of the controlled virtual object to also move to the position where the second target virtual object corresponding to the object identifier is located.
  • the user only needs to trigger the collection command to indicate the collection of artificial intelligence objects in his own camp, which realizes the interaction between the user and the artificial intelligence objects, thereby improving the operation efficiency of the own camp and enhancing Improve the team combat capability of the own camp, thus helping the camp to better complete the team task.
  • the terminal controls the artificial intelligence object to perform the target operation based on the control instruction.
  • the terminal After the terminal detects the control instruction, it can control the artificial intelligence object to perform the target operation based on the control instruction. For example, if the control instruction is an attack instruction, then based on the control instruction, the artificial intelligence object is controlled to attack the first target virtual object of the non-own camp. Alternatively, if the control instruction is an evacuation instruction, the artificial intelligence object is controlled to evacuate to a safe location based on the evacuation instruction. Alternatively, the control instruction is a collection instruction, and the collection of artificial intelligence objects is controlled based on the collection instruction.
  • control instruction carries the object identifier of the target virtual object
  • target operation is an operation of moving to the target virtual object.
  • the terminal can control the artificial intelligence object to perform the operation of moving to the target virtual object.
  • the above-mentioned attack command, retreat command or assembly command can be regarded as an operation instructing the artificial intelligence object to move towards the target virtual object, so based on these instructions, the movement of the artificial intelligence object can be controlled.
  • controlling the artificial intelligence object to perform the target operation based on the control instruction refers to controlling the artificial intelligence object to start executing the target operation, and does not limit whether the artificial intelligence object must complete the target operation .
  • controlling the movement of an artificial intelligence object if the artificial intelligence object encounters other virtual objects that are not in its own camp, the artificial intelligence object will conduct battle operations with the other virtual objects, and will not continue to follow the Move as directed by the control command. After the battle is completed, the artificial intelligence object can continue to move according to the instruction of the control instruction, or no longer respond to the control instruction.
  • this embodiment of the present application is only described by taking the execution subject as the terminal as an example.
  • the terminal is connected to the server, and the artificial intelligence object on the terminal is actually controlled by the server based on the neural network model. If To realize the control of artificial intelligence objects, interaction between the terminal and the server is required.
  • the terminal after the terminal detects the control instruction, it sends the control instruction to the server, and the server is used to call the neural network model to determine at least one sub-operation for completing the target operation based on the control instruction, and control the artificial intelligence object to perform at least one sub-operation .
  • a neural network model is set in the server, and the neural network model is used to predict the operations that the artificial intelligence object needs to perform.
  • the server obtains the state data at the current moment, calls the neural network to process the state data, thereby predicting the operation that the artificial intelligence object needs to perform, and sends the operation to the terminal corresponding to the artificial intelligence object, so that the The terminal displays the scene where the artificial intelligence object performs the operation, thereby realizing the effect of controlling the artificial intelligence object to perform the operation.
  • the artificial intelligence object in order to respond to the control instruction, the artificial intelligence object is required to perform the target operation. Therefore, at each moment, the server will call the neural network model to determine the sub-operation used to complete the target operation, thereby controlling the artificial intelligence object. Execute the sub-action.
  • the interval between any two adjacent moments can be a preset target duration, then the server controls the artificial intelligence object to perform a sub-operation every time the target duration, then after at least one control, the artificial intelligence object will Execute at least one sub-action to satisfy the requirements of this control instruction. Therefore, sub-operations can be seen as decomposition operations of the target operation.
  • the embodiment of the present application expands the functions of the artificial intelligence object and provides a controllable artificial intelligence object.
  • the user only needs to issue a control command to control the artificial intelligence object belonging to the same camp as the controlled virtual object to perform the target operation.
  • the control of the artificial intelligence object is realized, thereby realizing the synergy between the human and the artificial intelligence object.
  • control instructions include various instructions such as attack instructions, retreat instructions, and assembly instructions, which enhances the diversity of control instructions and realizes different types of control over artificial intelligence objects.
  • multiple triggering modes of each control command are provided, providing users with multiple options, and facilitating the user's operation.
  • the artificial intelligence object can be controlled by the server calling the neural network model.
  • the process of controlling the artificial intelligence object will be described in detail below with reference to FIG. 11 .
  • Fig. 11 is a flow chart of a method for controlling an artificial intelligence object provided by an embodiment of the present application. The method is executed by a server. As shown in Fig. 11 , the method includes:
  • the server receives the control instruction sent by the terminal.
  • the server is connected to the terminal, and the terminal has a corresponding controlled virtual object, while the server creates a virtual scene and an artificial intelligence object, and sets a neural network model to control the artificial intelligence object to perform operations based on the neural network model.
  • the controlled virtual object of the terminal and the artificial intelligence object belong to the same camp and can cooperate to perform operations.
  • the user triggers a control instruction on the terminal, and the terminal sends the control instruction to the server, and the control instruction is used to instruct the artificial intelligence object to perform a target operation.
  • the process for the terminal to obtain the control instruction is detailed in the embodiment shown in FIG. 2 above, and will not be repeated here.
  • the server controls the artificial intelligence object to perform the target operation based on the control instruction.
  • the artificial intelligence object belongs to the own camp of the controlled virtual object of the terminal.
  • control instruction is an attack instruction
  • the artificial intelligence object is controlled to attack the first target virtual object of the non-own camp.
  • the control instruction is an evacuation instruction
  • the artificial intelligence object is controlled to evacuate to a safe location based on the evacuation instruction.
  • control instruction is a collection instruction, and the collection of artificial intelligence objects is controlled based on the collection instruction.
  • control instruction carries the object identifier of the target virtual object
  • target operation is an operation of moving to the target virtual object.
  • the artificial intelligence object can be controlled to perform the operation of moving to the target virtual object.
  • attack instruction, retreat instruction or assembly instruction can all be regarded as an operation instructing the artificial intelligence object to move towards the target virtual object, so the movement of the artificial intelligence object can be controlled based on these instructions.
  • the embodiment of the present application expands the functions of the artificial intelligence object and provides a controllable artificial intelligence object.
  • the user only needs to issue a control command to control the artificial intelligence object belonging to the same camp as the controlled virtual object to perform the target operation.
  • the control of the artificial intelligence object is realized, thereby realizing the synergy between the human and the artificial intelligence object.
  • Fig. 12 is a flow chart of a method for controlling an artificial intelligence object provided by an embodiment of the present application, the method is executed by a server, and the method describes in detail the process of the server controlling the artificial intelligence object to perform a certain sub-operation. As shown in Figure 12, the method includes:
  • the server receives the control instruction sent by the terminal.
  • control instruction is used to instruct the artificial intelligence object to perform the target operation.
  • This step 1201 is the same as the above-mentioned embodiments shown in FIG. 2 and FIG. 11 , and will not be repeated here.
  • the server After receiving the control instruction, the server needs to invoke the neural network model to control the artificial intelligence object to perform the target operation corresponding to the control instruction.
  • the server considering the complex and changeable conditions of each object in the virtual scene, directly controlling the artificial intelligence object to perform the target operation is likely to be inconsistent with the current state of the virtual scene. Therefore, in order to ensure the authenticity of the virtual scene and ensure the artificial intelligence The continuity of the operations performed by the object, the server will call the neural network model based on the control instruction, determine at least one sub-operation for completing the target operation, and control the artificial intelligence object to execute at least one sub-operation so as to meet the requirements of the control instruction.
  • the server invokes the neural network model to determine a sub-operation, sends the sub-operation to the terminal, and the terminal can control the artificial intelligence object to perform the sub-operation, and display the sub-operation on the display interface.
  • the AI object performs the sub-operation.
  • the server invokes the neural network model to determine the next sub-operation, sends the next sub-operation to the terminal, and the terminal can control the artificial intelligence object to execute the next sub-operation.
  • the artificial intelligence object can be controlled to perform multiple sub-operations according to the instruction of the control instruction. Wherein, if the interval between any two adjacent moments is a preset target duration, the server controls the artificial intelligence object to perform a sub-operation every the target duration.
  • the embodiment of the present application is described by taking the neural network model including the encoding network, the fusion network and the operation prediction network as an example, and the process of calling the neural network model for processing is described in the following steps 1202-1204.
  • the server invokes an encoding network to encode the state data of multiple virtual objects to obtain encoded features of the multiple virtual objects.
  • the state data of the virtual object is used to represent the current state of the virtual object, for example, the state data of the virtual object includes the name, type, blood volume, skill type, etc. of the virtual object.
  • the encoding network is used to encode the state data of the virtual object to obtain encoded features. Then the coded feature can also reflect the current state of the virtual object. Subsequent processing of the coded feature can consider the current state of the virtual object to determine the operations that the artificial intelligence object can perform.
  • the server invokes the fusion network to perform weighted fusion on the coded features of multiple virtual objects based on weight features to obtain fusion features.
  • the weight features include weight parameters of multiple virtual objects in the virtual scene, and the weight features are determined based on control instructions.
  • the fusion network includes weight features, and the weight features include weight parameters of multiple virtual objects in the virtual scene, which are used to represent the degree of influence of the multiple virtual objects when predicting the operations to be performed by artificial intelligence objects.
  • the weighted fusion of the encoded features of each virtual object can take into account the different degrees of influence of different virtual objects, so as to ensure that the operation determined for the artificial intelligence object matches the current state of the multiple virtual objects in the virtual scene. Operation is accurate.
  • the operation performed corresponds to the control instruction.
  • the server Before predicting the operation, the server first determines the weight characteristics in the fusion network based on the control instruction.
  • the control instruction includes the first object identifier of the controlled virtual object and the second object identifier of the target virtual object, and the target operation is an operation of moving toward the target virtual object, that is, the control instruction is used to instruct manual The smart object moves towards the target dummy.
  • determining the weight feature in the fusion network includes: forming a coding matrix from the coding features of multiple virtual objects, and determining the first weight feature based on the matrix obtained by multiplying the coding matrix and the transposed matrix of the coding matrix , and set the weight parameter of the non-associated virtual object in the first weight feature to negative infinity to obtain the second weight feature.
  • the distance between the non-associated virtual object and the controlled virtual object, and the distance between the non-associated virtual object and the target virtual object are not less than a distance threshold.
  • the following provides an operation process for determining weight features, including the following steps:
  • Input F [f 1 ; f 2 ; . . . ; f n ];
  • w_logit represents the weight feature
  • F represents the coding matrix
  • f i represents the coding feature of the i-th virtual object
  • i and n are positive integers
  • n represents the number of virtual objects
  • i is less than or equal to n
  • [ ] 1 represents the first row vector of the matrix.
  • the process of inputting the weight feature includes: extracting the first row vector from the matrix obtained by multiplying the encoding matrix and the transposed matrix of the encoding matrix, and determining the product of this vector and the scaling factor as the weight feature.
  • the distance between the coding feature of the i-th virtual object and the coding feature of the controlled virtual object of the terminal is less than the distance threshold, it means that the i-th virtual object is a non-associated virtual object of the controlled virtual object, and the i-th virtual object is Virtual objects are added to the selective attention set S.
  • the distance between the coded feature of the i-th virtual object and the coded feature of the target virtual object is less than the distance threshold, it means that the i-th virtual object is a non-associated virtual object of the target virtual object, then add the i-th virtual object to into the selective attention set S. Thus, it can be determined which objects are included in the selective attention set S.
  • Using the above method to determine the weight feature is equivalent to using the attention mechanism to process the coding features of multiple virtual objects, which can take into account the influence of important virtual objects on the operations that the target virtual object needs to perform, and weaken the influence of irrelevant virtual objects on the target virtual object. The impact of the operation to be performed, so that the determined operation is more accurate and more in line with the current state of the virtual scene.
  • weighted fusion can be performed on the coded features of the plurality of virtual objects based on the weight feature to obtain a fusion feature.
  • the coded features of the multiple virtual objects are weighted and summed to obtain the fusion feature.
  • the coding features of the plurality of virtual objects are weighted and averaged to obtain the fusion feature.
  • the server invokes the operation prediction network, performs operation prediction on the fusion features, and obtains operation information.
  • the operation information includes at least the sub-operations that the artificial intelligence object needs to perform, that is, it realizes the prediction of the sub-operations that the artificial intelligence object will perform.
  • the operation information may also include the expected duration of the artificial intelligence object to perform the sub-operation, or include prompt information, which is used to display on the terminal to prompt that the artificial intelligence object is performing the sub-operation, or may also include other information.
  • the fusion feature can take into account the instruction of the control instruction, as well as the current state and degree of influence of multiple virtual objects, the operation information predicted according to the fusion feature is more accurate.
  • the server delivers the operation information to the terminal.
  • the terminal receives the operation information, and controls the artificial intelligence object to execute the sub-operation.
  • the embodiment of the present application only takes the process of controlling the artificial intelligence object once as an example. In fact, after receiving the control instruction, the steps in the embodiment of the present application can be executed multiple times, thereby controlling the artificial intelligence object to execute the determined sub-operations.
  • Figure 13 provides a schematic diagram of a deep hidden variable neural network model
  • Figure 14 provides a schematic diagram of the interaction between a terminal and a server .
  • the player sends a control command to his teammates
  • the terminal where the player is located sends the control command back to the server
  • the server invokes the deep hidden variable neural network model to predict the player's AI agent based on the control command Teammate's action strategy (that is, the above-mentioned operation information).
  • the server sends the action strategy back to the terminal, and on the terminal, the AI agent teammates can perform corresponding operations according to the action strategy.
  • the input of the deep hidden variable neural network model is the current game state data, including the game image displayed by the current terminal, the state data of each virtual object, the global information used to represent the global state of the virtual scene, and the controlled state of the terminal. available skills of virtual objects, etc.
  • each virtual object may include the main hero of the terminal, other heroes, soldiers line, defense tower, wild monster and so on.
  • the deep latent variable neural network model includes a convolutional network, an encoding network, a fusion network and an operation prediction network.
  • the convolution network performs convolution operation on the image to obtain image features
  • the encoding network encodes the state data of each virtual object, and obtains the encoding features of multiple virtual objects
  • the fusion network can be regarded as an implicit alignment module.
  • the implicit alignment module performs weighted fusion on the encoded features of multiple virtual objects
  • the features obtained by the weighted fusion are fused with the image features to obtain the fusion features, for example, using the Concat operator (an operator that combines multiple character strings)
  • the features obtained by weighted fusion are fused with the image features to obtain fused features.
  • the operation prediction network includes at least one fully connected layer and at least one fusion layer, so that the input features can be fully connected or fused.
  • the hidden alignment module has a weight feature, which is a hidden variable of the hidden alignment module, and the weight feature includes weight parameters of multiple virtual objects, and is used to weight the coding features of the multiple virtual objects.
  • the implicit alignment module is a linear weighted sum operator, and the weight vector is restricted in a special latent space.
  • the mathematical expression of the implicit alignment operator is as follows:
  • f fusion ⁇ R d is the fused feature vector
  • the weight parameter of the i-th virtual object is a probability simplex
  • 0 is the l 0 norm of the vector, that is, the number of non-zero elements.
  • the weight feature w ⁇ R n is a hidden random variable, and the calculation formula is as follows:
  • i and n are positive integers
  • n represents the number of virtual objects
  • i is less than or equal to n
  • [ ] 1 represents the first row vector of the matrix.
  • the weight feature is determined based on the control command issued by the player, and the specific determination method is similar to the above-mentioned operation process for determining the weight feature, which will not be repeated here.
  • the weight feature w ⁇ R n is a sparse vector belonging to the probability simplex ⁇ n-1 .
  • the sparsity of weight features can be understood as the feature selection of virtual objects, that is, only the encoding features of virtual objects whose weight parameters are not zero are reserved to participate in the process of predicting action strategies, which is actually aligning the encoding of virtual objects
  • the feature is associated with the predicted action strategy, so the weight feature w ⁇ R n is also called a hidden alignment variable.
  • This implicit feature selection mechanism is in line with the intuition of human players playing games: the decisions made by players are only determined by a few virtual objects that are focused on, which is actually a selective attention mechanism. If the predicted action strategy needs to be changed, we only need to sample a weight feature from the probability simplex ⁇ n -1 according to the received control instruction to predict the corresponding action strategy.
  • the output of the operation prediction network is a game button.
  • the game button includes a movement key and a skill key.
  • the movement key is used to instruct the AI agent teammate to move
  • the skill key is used to instruct the AI agent teammate to release a certain skill. Such as summoning skills, positional skills, directional skills, target skills, etc.
  • the embodiment of this application proposes a deep hidden variable neural network to solve human-AI cooperation in MOBA games. Its principle is based on the method of implicit alignment to control AI agents with controllable actions. Specifically, the predicted action strategy is changed by changing the latent variables (weight features) in the neural network. In a game where a human player is paired with an AI agent teammate whose actions can be controlled, the human player can actively cooperate with the player's intention by sending a purposeful control command to the AI agent teammate, and letting the AI agent teammate execute the instruction , thus enhancing the fun of the game and the player's experience.
  • FIG. 12 illustrates the process of calling the neural network model, but in order to ensure the accuracy of the neural network model, it is necessary to train the neural network model first.
  • the training process of the neural network model includes: obtaining sample data, the sample data includes sample state data and sample operation information at any sample time, the sample state data includes at the sample time, more than The state data of a virtual object, the sample operation information includes the operation that the artificial intelligence object should perform at the sample moment, then the neural network model is trained based on the sample data, that is, the sample state is based on the neural network model
  • the data is processed to obtain predicted operation information.
  • the predicted operation information includes the operation that the artificial intelligence object should perform at the sample moment predicted by the neural network model.
  • the neural network The network model is trained to improve the accuracy of the neural network model. After one or more training sessions, a neural network model with satisfactory accuracy can be obtained, and the trained neural network model can be called to predict the operation that the artificial intelligence object should perform.
  • the specific content of the sample status data is similar to the status data in step 1202 above, and the specific content of the sample operation information and the predicted operation information is similar to the operation information in step 1204 above.
  • the specific process of obtaining the predicted operation information is similar to the above steps 1202-1204, and will not be repeated here.
  • the neural network model can also be trained in a reinforcement learning manner. During the training process, if the artificial intelligence object responds to the control command successfully, it will be rewarded. After one or more training sessions, That is, the neural network model can learn the ability to respond to control instructions, thereby obtaining a more accurate neural network model.
  • the server is provided with a behavior tree model, and the operation rules of the artificial intelligence object are defined in the behavior tree model, then when the server receives the control instruction, it controls the artificial intelligence object based on the instruction of the behavior tree model Execute the target action.
  • the control instruction is a collection instruction
  • the rule in the behavior tree model is: always move to the collection target until it reaches the vicinity of the collection target.
  • the artificial intelligence model can be controlled to move towards the set target until it reaches the vicinity of the set target.
  • the embodiment of the present application also provides a method for verifying the control instruction, which can first verify the control instruction and judge whether it is allowed to respond to the control instruction. Instructions, in the case of allowing to respond to the control instructions, and then control the artificial intelligence object based on the control instructions. As shown in Figure 15, the method includes:
  • the server receives the control instruction sent by the terminal, and parses the control instruction into a format supported by the server.
  • the server verifies the control instruction.
  • control instruction After the control instruction is analyzed, the analyzed control instruction is verified to determine whether the control instruction needs to be responded to at present, so as to screen out reasonable control instructions and unreasonable control instructions.
  • the server predetermines the target rule, and the target rule may be manually determined. If the control instruction satisfies the target rule, then the verification passes, and the subsequent step 1503 can be performed; but if the control instruction does not meet the target rule, the verification fails, and the subsequent step 1504 is not performed.
  • the intent prediction model can also be invoked for verification, to filter out control instructions that require a response, and filter out control instructions that do not require a response.
  • the intention prediction model is used to predict the first position that the artificial intelligence object will reach at the next moment in the virtual scene based on the current state data, so that according to the first position and the second position indicated by the control instruction ( Indicate the location where the artificial intelligence object needs to reach), determine whether to respond to the control instruction.
  • the map of the virtual scene is divided into multiple grids, for example, the map is divided into 12*12 grids.
  • determine the predicted position information of the artificial intelligence object determines the predicted position information of the artificial intelligence object, the state data includes the current position of the artificial intelligence object, the predicted position information includes the probability corresponding to each grid in the map, and the probability indicates that the artificial intelligence object will The probability that an AI object will reach this mesh.
  • the first K grids with the highest probability in the predicted position information constitute a set, and the set can represent the K positions that the artificial intelligence object is most likely to reach at the next moment, and K is a positive integer.
  • control instruction is verified, indicating that the control instruction is allowed to respond. However, if the second location does not belong to the set, the control instruction fails to pass the verification, indicating that the control instruction is not allowed to be responded to.
  • each grid in the map is numbered in sequence
  • the artificial intelligence object is currently in grid 121
  • the target position that the control instruction requires the artificial intelligence object to reach is in grid 89
  • the intention is to predict
  • the set P ⁇ 26, 27, 89, 116, 117 ⁇ predicted by the model, the set includes the grid 89
  • control instruction 1503 If the control instruction is verified and passed, predict an action strategy of the artificial intelligence object based on the control instruction.
  • the server sends back the action policy to the terminal, so that the terminal receives the action policy returned by the server, and controls the artificial intelligence object according to the action policy.
  • the intent prediction model includes a coding network, a fusion network and a position prediction network
  • the process of calling the intent prediction model to determine the predicted position information of the artificial intelligence object includes: the server calls the coding network to encode the state data of multiple virtual objects , to obtain the encoding features of multiple virtual objects, call the fusion network, and based on the weight feature, carry out weighted fusion on the encoding features of multiple virtual objects to obtain the fusion feature, the weight feature includes the weight parameters of multiple virtual objects in the virtual scene, the server Call the position prediction network to predict the position of the fusion feature and obtain the predicted position information.
  • the process of invoking the encoding network, the fusion network and the position prediction network is similar to the process of the above steps 1202-1204, the difference is that in the above step 1203, the weight characteristics of the fusion network in the neural network model are based on the received control instructions If it is determined, the weight feature may change depending on the received control instruction. However, in this embodiment, after the training of the intention prediction model is completed, the weight characteristics of the fusion network in the intention prediction model remain unchanged.
  • the process of training the intention prediction model includes: obtaining sample data, the sample data includes state data and corresponding position labels of multiple virtual objects at the sample time, the state data includes at least the image frame at the sample time, the state data is the same as the above
  • the status data in the embodiments are similar, and the location label is the number of the grid where target events such as attack, assembly, and retreat occur at the next moment of the sample moment.
  • the intent prediction model can be trained based on the sample data, so as to obtain the trained intent prediction model.
  • the above-mentioned events are defined as the following types: 1. An attack behavior occurred; 2. A collection behavior occurred; 3.
  • the sample data may include state data and corresponding location labels at multiple sample moments, so as to train the intention prediction model according to the state data and location labels at multiple consecutive moments.
  • the intention prediction model can be trained based on the data of human players, that is, the sample data obtained during the training of the intention prediction model belongs to human players, that is, the sample data includes the controlled virtual objects of human players.
  • the status data and corresponding position labels of the game can ensure that the intention prediction model can learn the events of human players, and then when the intention prediction model is called to predict the intention of the control command, it can ensure that the intention is in line with the behavior of human players.
  • the embodiment of the present application also provides a way to train the neural network model and the intention prediction model at the same time, that is, as shown in Figure 18, the server pre-creates a prediction model , the prediction model includes an encoding network 1801, a fusion network 1802, and a prediction network 1803, and the prediction network 1803 includes an operation prediction network 1831 and a position prediction network 1832.
  • the operation prediction network 1831 is used to predict the operation information of artificial intelligence objects to determine artificial intelligence
  • the location prediction network 1832 is used to predict the predicted location information of the artificial intelligence object, so as to determine the location where the artificial intelligence object will arrive at the next moment.
  • sample data is obtained, the sample data includes sample state data and sample operation information at any sample time, and location labels, the sample state data includes state data of multiple virtual objects at the sample time , the sample operation information includes the operation that the artificial intelligence object should perform at the sample time, and the location label is the number of the grid where attack, assembly, retreat and other events occur at the next time of the sample time.
  • the prediction model is trained based on the sample data, that is, the sample state data is processed based on the prediction model to obtain the prediction operation information and the prediction location label, and the prediction operation information includes The operation that the artificial intelligence object should perform, the predicted location information includes the location where the artificial intelligence object will arrive at the next moment, then according to the sample operation information and the predicted operation information, as well as the location label and the predicted location information, the prediction The model is trained to improve the accuracy of the predictive model. After one or more times of training, a prediction model whose accuracy meets the requirements can be obtained, and the trained prediction model can be called for prediction.
  • This comprehensive training method can simultaneously train a prediction model with operation prediction and position prediction functions, which improves the training speed and saves training time.
  • an intention prediction model as shown in Figure 20 and a neural network as shown in Figure 21 are respectively constructed based on the trained network Model.
  • the intention prediction model includes an encoding network 2001 , a fusion network 2002 and a position prediction network 2003
  • the neural network model includes an encoding network 2101 , a fusion network 2102 and an operation prediction network 2103 .
  • the encoding network 2001 and the encoding network 2101 are trained by the encoding network 1801 in FIG. 18, the fusion network 2002 and the fusion network 2102 are trained by the fusion network 1802 in FIG.
  • the position prediction network 1832 in FIG. 18 is trained, and the operation prediction network 2103 is trained by the operation prediction network 1831 in FIG. 18 .
  • the server after the server receives the control instruction, it first invokes the intention prediction model to determine the predicted position information of the artificial intelligence object based on the state data of multiple virtual objects. If the control command indicates that the location to be reached by the artificial intelligence object belongs to the set, it means that the control command has passed the verification. Then the server determines the weight characteristics in the fusion network 2102 based on the control instruction, and then calls the neural network model to determine the operation information of the artificial intelligence object.
  • the server can also determine Encoded features of multiple virtual objects to predict operational information for artificial intelligence objects.
  • the server can modify the state data of the plurality of virtual objects.
  • the control instruction includes the second object identifier of the target virtual object, and the target operation is an operation of moving to the target virtual object, that is, the control instruction indicates that the artificial intelligence object arrives The location of the target dummy.
  • the neural network model will not control the movement of the artificial intelligence object to the location of the target virtual object, so in order to enable the neural network model to control the movement of the artificial intelligence object
  • the location information of the attack object of the artificial intelligence object is changed to the location information near the target virtual object, for example, the attack object of the artificial intelligence object , so that the distance between the changed coordinates and the coordinates of the target virtual object is less than the preset distance, the artificial intelligence object will mistakenly think that the attack object is near the target virtual object, so it will move towards the attack object, Thus, the effect of moving to the target virtual object is realized.
  • the intention prediction model and the neural network model can share the same encoding network, as shown in Figure 22, the encoding network 2001 and the encoding network 2101 are the same encoding network, and the server calls the encoding
  • the network 2001 determines the coding features of multiple virtual objects based on the state data of multiple virtual objects, and then calls the fusion network 2002 and the position prediction network 2003 to determine the predicted position information of the artificial intelligence object.
  • the predicted position information includes the next moment of the artificial intelligence object A set of possible reachable locations. If the control instruction indicates that the location to be reached by the artificial intelligence object belongs to the set, it means that the control instruction has passed the verification.
  • the server determines the weight features in the fusion network 2102, and then calls the fusion network 2102 and the operation prediction network 2103 to predict the coding features obtained by the coding network 2001, thereby determining the operation information of the artificial intelligence object.
  • Fig. 23 is a schematic structural diagram of an artificial intelligence object control device provided by an embodiment of the present application.
  • the device includes:
  • the display module 2301 is used to display the virtual scene interface, and the virtual scene interface displays the controlled virtual object at the end;
  • the detection module 2302 is configured to detect a control instruction through the virtual scene interface, and the control instruction is used to instruct the artificial intelligence object in the own camp of the controlled virtual object to perform the target operation;
  • the control module 2303 is configured to control the artificial intelligence object to perform the target operation based on the control instruction.
  • the artificial intelligence object control device realizeds the interaction between the human and the artificial intelligence object.
  • the operation of the artificial intelligence object is controllable.
  • the user only needs to issue a control command to control and control the virtual object.
  • the artificial intelligence objects belonging to the same camp perform target operations and realize the control of the artificial intelligence objects, thereby realizing the synergy between humans and artificial intelligence objects, and expanding the functions of the artificial intelligence objects.
  • the detection module 2302 includes:
  • the first detection unit is configured to detect an attack instruction through the virtual scene interface, and the attack instruction is used to instruct the artificial intelligence object to attack the first target virtual object that is not its own camp; or,
  • the second detection unit is configured to detect a retreat instruction through the virtual scene interface, and the retreat instruction is used to instruct the artificial intelligence object to retreat to a safe position; or,
  • the third detection unit is configured to detect a collection instruction through the virtual scene interface, and the collection instruction is used to instruct the artificial intelligence object to move to the position of the controlled virtual object or the second target virtual object of the own camp.
  • the first detection unit is used for:
  • the first target virtual object is the virtual object closest to the accused virtual object in the non-own camp;
  • the selection operation of the object identifier of the first target virtual object is detected through the virtual scene interface, and an attack instruction is generated.
  • the second detection unit is used for:
  • the retreat instruction is used to instruct the artificial intelligence object to retreat to a safe position;
  • the retreat instruction is used to instruct the artificial intelligence object to retreat to the position of the third target virtual object;
  • the selection operation of the object identifier of the third target virtual object is detected through the virtual scene interface, and a retreat instruction is generated, and the retreat instruction is used to instruct the artificial intelligence object to retreat to the position of the third target virtual object.
  • the third detection unit is used for:
  • the collection instruction is used to instruct the artificial intelligence object to move to the position of the second target virtual object;
  • the selection operation of the object identifier of the third target virtual object is detected through the virtual scene interface, and a set instruction is generated, and the set instruction is used to instruct the artificial intelligence object to move to the position of the second target virtual object.
  • control instruction carries the object identifier of the target virtual object, and the target operation is an operation of moving to the target virtual object;
  • Control module 2303 including:
  • the first control unit is configured to control the artificial intelligence object to perform an operation of moving to the target virtual object based on the control instruction.
  • control module 2303 includes:
  • the second control unit is configured to send a control instruction to the server, and the server is configured to call the neural network model to determine at least one sub-operation for completing the target operation based on the control instruction, and control the artificial intelligence object to execute at least one sub-operation.
  • Fig. 24 is a schematic structural diagram of an artificial intelligence object control device provided by an embodiment of the present application. Referring to Figure 24, the device includes:
  • the receiving module 2401 is configured to receive a control instruction sent by the terminal, and the control instruction is used to instruct the artificial intelligence object to perform a target operation;
  • a control module 2402 configured to control the artificial intelligence object to perform the target operation based on the control instruction
  • the artificial intelligence object belongs to the own camp of the controlled virtual object of the terminal.
  • the artificial intelligence object control device realizes the interaction between the terminal and the server.
  • the server receives the control instruction, it can control the artificial intelligence object belonging to the same camp as the terminal to perform the corresponding target operation, realizing It realizes the control of artificial intelligence objects, thereby realizing the cooperation between humans and artificial intelligence objects, and expanding the functions of artificial intelligence objects.
  • control module 2402 includes:
  • the first control unit is configured to call the neural network model to determine at least one sub-operation for completing the target operation based on the control instruction, and control the artificial intelligence object to execute at least one sub-operation.
  • the neural network model includes an encoding network, a fusion network and an operation prediction network
  • the first control unit includes:
  • the encoding subunit is used to call the encoding network to encode the state data of multiple virtual objects to obtain the encoding features of multiple virtual objects;
  • the fusion subunit is used to call the fusion network, and based on the weight feature, perform weighted fusion on the coding features of multiple virtual objects to obtain the fusion feature.
  • the weight feature includes the weight parameters of multiple virtual objects in the virtual scene, and the weight feature is based on the control order confirmation;
  • the prediction sub-unit is used to call the operation prediction network, perform operation prediction on the fusion features, and obtain operation information, and the operation information includes at least the sub-operations required by the artificial intelligence object.
  • control instruction includes a first object identifier of the controlled virtual object and a second object identifier of the target virtual object, and the target operation is an operation of moving to the target virtual object;
  • the first control unit further comprising:
  • control module 2402 includes:
  • a verification unit is used to call the intention prediction model to verify the control instruction
  • the second control unit is configured to control the artificial intelligence object to perform the target operation based on the control instruction when the control instruction is verified to pass.
  • the artificial intelligence object control device controls the artificial intelligence object to perform the target operation, it only uses the division of the above-mentioned functional modules as an example. In practical applications, the above-mentioned functions can be allocated according to needs. Completion of different functional modules means that the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.
  • the artificial intelligence object control device and the artificial intelligence object control method embodiment provided by the above embodiment belong to the same concept, and the specific implementation process thereof is detailed in the method embodiment, and will not be repeated here.
  • Fig. 25 shows a schematic structural diagram of a terminal 2500 provided by an exemplary embodiment of the present application.
  • the terminal 2500 includes: a processor 2501 and a memory 2502 .
  • the processor 2501 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like.
  • Processor 2501 can be realized by at least one hardware form in DSP (Digital Signal Processing, digital signal processing), FPGA (Field Programmable Gate Array, field programmable gate array), PLA (Programmable Logic Array, programmable logic array) .
  • the processor 2501 may include an AI (Artificial Intelligence, artificial intelligence) processor, and the AI processor is used to process computing operations related to machine learning.
  • AI Artificial Intelligence, artificial intelligence
  • Memory 2502 may include one or more computer-readable storage media, which may be non-transitory.
  • the memory 2502 may also include high-speed random access memory and non-volatile memory, such as one or more magnetic disk storage devices and flash memory storage devices.
  • the non-transitory computer-readable storage medium in the memory 2502 is used to store at least one computer program, and the at least one computer program is used to be possessed by the processor 2501 to implement the methods provided by the method embodiments in this application. Artificial intelligence object control method.
  • the terminal 2500 may optionally further include: a peripheral device interface 2503 and at least one peripheral device.
  • the processor 2501, the memory 2502, and the peripheral device interface 2503 may be connected through buses or signal lines.
  • Each peripheral device can be connected to the peripheral device interface 2503 through a bus, a signal line or a circuit board.
  • the peripheral device includes: at least one of a radio frequency circuit 2504 or a display screen 2505 . Additional components may also be included.
  • the peripheral device interface 2503 may be used to connect at least one peripheral device related to I/O (Input/Output, input/output) to the processor 2501 and the memory 2502 .
  • the processor 2501, memory 2502 and peripheral device interface 2503 are integrated on the same chip or circuit board; in some other embodiments, any one of the processor 2501, memory 2502 and peripheral device interface 2503 or The two can be implemented on a separate chip or circuit board, which is not limited in this embodiment.
  • the radio frequency circuit 2504 is used to receive and transmit RF (Radio Frequency, radio frequency) signals, also called electromagnetic signals.
  • the radio frequency circuit 2504 communicates with the communication network and other communication devices through electromagnetic signals.
  • the radio frequency circuit 2504 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals.
  • the radio frequency circuit 2504 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a subscriber identity module card, and the like.
  • the radio frequency circuit 2504 can communicate with other devices through at least one wireless communication protocol.
  • the wireless communication protocol includes but is not limited to: a metropolitan area network, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and/or a WiFi (Wireless Fidelity, wireless fidelity) network.
  • the radio frequency circuit 2504 may also include circuits related to NFC (Near Field Communication, short-range wireless communication), which is not limited in this application.
  • the display screen 2505 is used to display a UI (User Interface, user interface).
  • the UI can include graphics, text, icons, video, and any combination thereof.
  • the display screen 2505 also has the ability to collect touch signals on or above the surface of the display screen 2505 .
  • the touch signal can be input to the processor 2501 as a control signal for processing.
  • the display screen 2505 can also be used to provide virtual buttons and/or virtual keyboards, also called soft buttons and/or soft keyboards.
  • FIG. 25 does not constitute a limitation to the terminal 2500, and may include more or less components than shown in the figure, or combine some components, or adopt a different component arrangement.
  • FIG. 26 is a schematic structural diagram of a server provided by an embodiment of the present application.
  • the server 2600 may have relatively large differences due to different configurations or performances, and may include one or more than one processor (Central Processing Units, CPU) 2601 and one Or more than one memory 2602, wherein at least one computer program is stored in the memory 2602, and the at least one computer program is loaded and executed by the processor 2601 to implement the methods provided by the above method embodiments.
  • the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the server may also include other components for realizing device functions, which will not be repeated here.
  • the embodiment of the present application also provides a computer-readable storage medium, at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to realize the artificial intelligence object of the above-mentioned embodiment Controls what the method does.
  • the embodiment of the present application also provides a computer program product or computer program, the computer program product or computer program includes computer program code, the computer program code is stored in a computer-readable storage medium, and the processor of the computer device reads from the computer-readable storage medium The computer program code is read, and the processor executes the computer program code, so that the computer device implements the operations performed by the method for controlling an artificial intelligence object in the above-mentioned embodiments.
  • the computer programs involved in the embodiments of the present application can be deployed and executed on one computer device, or executed on multiple computer devices at one location, or distributed in multiple locations and communicated Executed on multiple computer devices interconnected by the network, multiple computer devices distributed in multiple locations and interconnected through a communication network can form a blockchain system.
  • the program can be stored in a computer-readable storage medium.
  • the above-mentioned The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, and the like.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • General Physics & Mathematics (AREA)
  • Human Computer Interaction (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • User Interface Of Digital Computer (AREA)

Abstract

一种人工智能对象控制方法、装置、设备及存储介质，属于人工智能技术领域。该方法包括：终端显示虚拟场景界面(201)，虚拟场景界面显示有本终端的被控虚拟对象；终端通过虚拟场景界面检测控制指令(202)，控制指令用于指示被控虚拟对象的本方阵营中的人工智能对象执行目标操作；终端基于控制指令，控制人工智能对象执行目标操作(203)。

Description

人工智能对象控制方法、装置、设备及存储介质
本申请要求于2021年08月23日提交、申请号为202110969023.3、发明名称为“人工智能对象控制方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及人工智能技术领域，特别涉及一种人工智能对象控制方法、装置、设备及存储介质。
背景技术
随着AI(Artificial Intelligence，人工智能)技术的发展，在游戏、医疗等领域中逐渐出现了人工智能对象。人工智能对象是基于神经网络模型操控的对象，而不是由用户基于终端操控的对象。
发明内容
本申请实施例提供了一种人工智能对象控制方法、装置、设备及存储介质，扩展了人工智能对象的功能，提供了一种操作可控的人工智能对象。所述技术方案如下：
一方面，提供了一种人工智能对象控制方法，所述方法包括：
终端显示虚拟场景界面，所述虚拟场景界面显示有本端的被控虚拟对象；
所述终端通过所述虚拟场景界面检测控制指令，所述控制指令用于指示所述被控虚拟对象的本方阵营中的人工智能对象执行目标操作；
所述终端基于所述控制指令，控制所述人工智能对象执行所述目标操作。
另一方面，提供了一种人工智能对象控制方法，所述方法包括：
服务器接收终端发送的控制指令，所述控制指令用于指示人工智能对象执行目标操作；
所述服务器基于所述控制指令，控制所述人工智能对象执行所述目标操作；
其中，所述人工智能对象属于所述终端的被控虚拟对象的本方阵营。
另一方面，提供了一种人工智能对象控制装置，所述装置包括：
显示模块，用于显示虚拟场景界面，所述虚拟场景界面显示有本端的被控虚拟对象；
检测模块，用于通过所述虚拟场景界面检测控制指令，所述控制指令用于指示所述被控虚拟对象的本方阵营中的人工智能对象执行目标操作；
第一控制模块，用于基于所述控制指令，控制所述人工智能对象执行所述目标操作。
另一方面，提供了一种人工智能对象控制装置，所述装置包括：
接收模块，用于接收终端发送的控制指令，所述控制指令用于指示人工智能对象执行目标操作；
控制模块，用于基于所述控制指令，控制所述人工智能对象执行所述目标操作；
其中，所述人工智能对象属于所述终端的被控虚拟对象的本方阵营。
另一方面，提供了一种终端，所述终端包括处理器和存储器，所述存储器中存储有至少一条计算机程序，所述至少一条计算机程序由所述处理器加载并执行以实现如上述方面所述的人工智能对象控制方法所执行的操作。
另一方面，提供了一种服务器，所述服务器包括处理器和存储器，所述存储器中存储有至少一条计算机程序，所述至少一条计算机程序由所述处理器加载并执行以实现如上述方面所述的人工智能对象控制方法所执行的操作。
另一方面，提供了一种计算机可读存储介质，所述计算机可读存储介质中存储有至少一条计算机程序，所述至少一条计算机程序由处理器加载并执行以实现如上述方面所述的人工智能对象控制方法所执行的操作。
另一方面，提供了一种计算机程序产品或计算机程序，所述计算机程序产品或计算机程序包括计算机程序代码，所述计算机程序代码存储在计算机可读存储介质中，计算机设备的处理器从计算机可读存储介质读取所述计算机程序代码，处理器执行所述计算机程序代码，使得所述计算机设备实现如上述方面所述的人工智能对象控制方法所执行的操作。
本申请实施例扩展了人工智能对象的功能，提供了一种操作可控的人工智能对象，用户只需发出控制指令，即可控制与被控虚拟对象属于同一阵营的人工智能对象执行目标操作，实现了对人工智能对象的控制，从而实现了人与人工智能对象之间的协同配合。
附图说明
图1是本申请实施例提供的一种实施环境的示意图。
图2是本申请实施例提供的一种人工智能对象控制方法的流程图。
图3是本申请实施例提供的一种虚拟场景界面的示意图。
图4是本申请实施例提供的另一种虚拟场景界面的示意图。
图5是本申请实施例提供的另一种虚拟场景界面的示意图。
图6是本申请实施例提供的另一种虚拟场景界面的示意图。
图7是本申请实施例提供的另一种虚拟场景界面的示意图。
图8是本申请实施例提供的另一种虚拟场景界面的示意图。
图9是本申请实施例提供的另一种虚拟场景界面的示意图。
图10是本申请实施例提供的另一种虚拟场景界面的示意图。
图11是本申请实施例提供的一种人工智能对象控制方法的流程图。
图12是本申请实施例提供的另一种人工智能对象控制方法的流程图。
图13是本申请实施例提供的一种深度隐变量神经网络模型的示意图。
图14是本申请实施例提供的一种终端与服务器的交互示意图。
图15是本申请实施例提供的一种验证控制指令的方法的流程图。
图16是本申请实施例提供的一种地图网格化示意图。
图17是本申请实施例提供的一种多个时刻的网格的示意图。
图18是本申请实施例提供的一种预测模型的示意图。
图19是本申请实施例提供的另一种深度隐变量神经网络模型的示意图。
图20是本申请实施例提供的一种意图预测模型的示意图。
图21是本申请实施例提供的一种神经网络模型的示意图。
图22是本申请实施例提供的一种共享编码网络的示意图。
图23是本申请实施例提供的一种人工智能对象控制装置的结构示意图。
图24是本申请实施例提供的另一种人工智能对象控制装置的结构示意图。
图25是本申请实施例提供的一种终端的结构示意图。
图26是本申请实施例提供的一种服务器的结构示意图。
具体实施方式
本申请涉及到的虚拟场景用于模拟一个三维虚拟空间，该三维虚拟空间是一个开放空间，该虚拟场景用于模拟现实中的真实环境，例如，该虚拟场景中包括天空、陆地、海洋等，该陆地包括沙漠、城市等环境元素。当然，在该虚拟场景中还可以包括虚拟物品，例如，建筑物、载具、虚拟场景中的虚拟对象用于武装自己或与其他虚拟对象进行战斗所需的兵器等道具。该虚拟场景还可以用于模拟不同天气下的真实环境，例如，晴天、雨天、雾天或黑夜等天气。
用户控制虚拟对象在该虚拟场景中进行移动，该虚拟对象是该虚拟场景中的一个虚拟的用于代表用户的虚拟形象，该虚拟形象是任一种形态，例如，人、动物等，本申请对此不限定。以射击类游戏为例，用户控制虚拟对象在该虚拟场景的天空中自由下落、滑翔或者打开降落伞进行下落等，在陆地上中跑动、跳动、爬行、弯腰前行等，也可以控制虚拟对象在海洋中游泳、漂浮或者下潜等，当然，用户也可以控制虚拟对象乘坐载具在该虚拟场景中进行移动。用户还可以控制虚拟对象在该虚拟场景中进出建筑物，发现并拾取该虚拟场景中的虚拟物品(例如，兵器等道具)，从而通过拾取的虚拟物品与其他虚拟对象进行战斗，例如，该虚拟物品可以是衣物、头盔、防弹衣、医疗品、冷兵器或热兵器等，也可以是其他虚拟对象被淘汰后遗留的虚拟物品。在此仅以上述场景进行举例说明，本申请实施例对此不作具体限定。
本申请实施例以电子游戏场景为例，用户提前在该终端上进行操作，该终端检测到用户的操作后，下载电子游戏的游戏配置文件，该游戏配置文件包括该电子游戏的应用程序、界面显示数据或虚拟场景数据等，以使得该用户在该终端上登录电子游戏时调用该游戏配置文件，对电子游戏界面进行渲染显示。用户在终端上进行触控操作，该终端检测到触控操作后，确定该触控操作所对应的游戏数据，并对该游戏数据进行渲染显示，该游戏数据包括虚拟场景数据、该虚拟场景中虚拟对象的行为数据等。
终端在对虚拟场景进行渲染显示时，全屏显示该虚拟场景，终端还可以在当前显示界面显示虚拟场景的同时，在该当前显示界面的第一预设区域独立显示全局地图，实际应用中，终端也可以在检测到对预设按钮的点击操作时，才对该全局地图进行显示。其中，该全局地图用于显示该虚拟场景的缩略图，该缩略图用于描述该虚拟场景对应的地形、地貌、地理位置等地理特征。当然，终端还可以在当前显示界面显示当前虚拟对象周边一定距离内的虚拟场景的缩略图，在检测到对该全局地图的点击操作时，在终端当前显示界面的第二预设区域显示整体虚拟场景的缩略图，以便于用户不仅可以查看其周围的虚拟场景，也可以查看整体虚拟场景。终端在检测到对该完整缩略图的缩放操作时，也可以对完整缩略图进行缩放显示。该第一预设区域和第二预设区域的具体显示位置和形状可以根据用户操作习惯来设定。例如，为了不对虚拟场景造成过多的遮挡，该第一预设区域可以为该当前显示界面右上角、右下角、左上角或左下角的矩形区域等，该第二预设区域可以为当前显示界面的右边或者左边的正方形区域，当然，该第一预设区域和第二预设区域也可以是圆形区域或其他形状的区域，本申请实施例对该预设区域的具体显示位置和形状不作限定。
为了便于理解本申请实施例，先对本申请实施例涉及到的关键词进行解释：
MOBA(Multiplayer Online Battle Arena，多人在线战术竞技)游戏：是一种在虚拟场景中提供若干个据点，处于不同阵营的用户控制虚拟对象在虚拟场景中对战，占领据点或摧毁敌对阵营据点的游戏。例如，MOBA游戏将用户分成至少两个敌对阵营，分属至少两个敌对阵营的不同虚拟队伍分别占据各自的地图区域，以某一种胜利条件作为目标进行竞技。该胜利条件包括但不限于：占领据点或摧毁敌对阵营据点、击杀敌对阵营的虚拟对象、在指定场景和时间内保证自身的存活、抢夺到某种资源、在指定时间内互动比分超过对方中的至少一种。例如，手机MOBA游戏可将用户分成两个敌对阵营，将用户控制的虚拟对象分散在虚拟场景中互相竞争，以摧毁或占领敌方的全部据点作为胜利条件。
在MOBA游戏中，用户控制虚拟对象释放技能从而与其他虚拟对象进行战斗，例如，该技能的技能类型可以包括攻击技能、防御技能、治疗技能、辅助技能、斩杀技能等，每个虚拟对象都可以具有各自固定的一个或多个技能，而不同的虚拟对象通常具有不同的技能，不同的技能可以产生不同的作用效果。比如，若虚拟对象释放攻击技能击中了敌对虚拟对象，那么会对敌对虚拟对象造成一定的伤害，通常表现为扣除敌对虚拟对象的一部分虚拟生命值，又比如，若虚拟对象释放治疗技能命中了友方虚拟对象，那么会对友方虚拟对象产生一定的治疗，通常表现为回复友方虚拟对象的一部分虚拟生命值，其他各类技能均可以产生相应的 作用效果，这里不再一一枚举。
人AI协作：人类和AI在同一个环境中进行协作(共同合作完成某个任务)和交互(互相传递信号比如文字等)。这类AI通常称作协作AI，相对于传统AI智能体的自动化属性，协作AI也强调AI智能体的协作性和交互性。
动作可控制AI智能体：动作策略能够被控制指令改变的AI智能体。有些AI智能体通常只具备自动化属性，即智能体能够自主决策自主行动；但是它的动作不能被控制，即外界发起的控制指令无法改变AI智能体的动作。在人AI协作的场景中，AI智能体除了自动化属性外，也要具备协作性和交互性，即智能体要能够跟其它智能体进行协作交互。动作可控制AI智能体就能够满足这种协作交互的需求，它的动作可以被外部控制指令改变，从而可以完成协作交互。
图1是本申请实施例提供的一种实施环境的结构示意图，如图1所示，该实施环境包括终端101和服务器102。可选地，该终端101是智能手机、平板电脑、笔记本电脑、台式计算机、智能音箱、智能电视、智能手表等，但并不局限于此。可选地，该服务器102是独立的物理服务器，或者，该服务器102是多个物理服务器构成的服务器集群或者分布式系统，或者，该服务器102是提供云服务、云数据库、云计算、云函数、云存储、网络服务、云通信、中间件服务、域名服务、安全服务、CDN(Content Delivery Network，内容分发网络)、以及大数据和人工智能平台等基础云计算服务的云服务器。终端101以及服务器102通过有线或无线通信方式进行直接或间接地连接，本申请在此不做限制。
服务器102为终端101提供虚拟场景，终端101通过服务器102提供的虚拟场景，能够显示虚拟对象、虚拟道具等，并且终端101为用户提供操作环境，来检测用户执行的操作。服务器102能够针对终端检测到的操作，进行后台处理，为终端101提供后台支持。
可选地，终端101安装由服务器102提供服务的游戏应用，通过该游戏应用，终端101与服务器102能够进行交互。终端101运行该游戏应用，为用户提供针对该游戏应用的操作环境，能够检测用户对游戏应用的操作，并向服务器102发送操作指令，由服务器102根据操作指令进行响应，并将响应结果返回给终端101，由终端101进行展示，从而实现人机交互。
本申请实施例提供的方法，可用于多种场景。
例如，竞技游戏场景下：
终端运行竞技游戏，该游戏分为我方阵营和敌方阵营，其中，我方阵营包括终端的被控虚拟对象和人工智能对象。在玩家对我方阵营中的人工智能对象发出控制指令的情况下，终端能够检测到该控制指令，并基于该控制指令控制我方阵营中的人工智能对象执行该控制指令对应的目标操作。例如，该控制指令为集合指令，终端将控制我方阵营的人工智能对象向该终端的被控虚拟对象靠拢，直到完成全员集合，实现了对人工智能对象的控制。
又如，在实战模拟场景下：
玩家在购买新英雄之后，需要进行多局游戏才能熟悉该英雄，而实战模拟很好地满足了玩家训练新英雄的需求。在实战模拟中，将2个人类玩家的英雄搭配3个人工智能对象，组成一队，将5个人工智能对象组成敌方队，由这两个队伍进行对战。则采用本申请实施例提供的方法，玩家不仅可以控制自己的英雄，而且还可以控制同队中的人工智能对象。
图2是本申请实施例提供的一种人工智能对象控制方法的流程图，该方法由终端执行，如图2所示，该方法包括：
201、终端显示虚拟场景界面。
本申请实施例的虚拟场景包括至少两个敌对阵营的虚拟对象，其中每个阵营中既可以包括由终端操控的虚拟对象，也可以包括人工智能对象，另外还可以包括防御塔等对象。其中，人工智能对象是指基于神经网络模型操控的对象，人工智能对象也可称为AI智能体。
而终端显示虚拟场景界面，该虚拟场景界面显示有该终端的被控虚拟对象，另外还可以 显示除该被控虚拟对象之外的其他对象，例如该被控虚拟对象的本方阵营中的对象，或者该被控虚拟对象的非本方阵营中的对象。其中，该被控虚拟对象的非本方阵营中的对象包括：该被控虚拟对象的敌方阵营中的对象，以及中立对象，如野怪等。
在一种可能实现方式中，虚拟场景界面以终端的被控虚拟对象的第一人称视角显示虚拟场景，则在虚拟场景界面显示该被控虚拟对象的视角范围内的部分虚拟场景，及位于该部分虚拟场景中的其他对象。
在另一种可能实现方式中，虚拟场景界面以第三人称视角显示虚拟场景，则在虚拟场景界面显示第三人称视角的视角范围内的部分虚拟场景，及位于该部分虚拟场景中的该被控虚拟对象和其他对象。
202、终端通过虚拟场景界面检测控制指令。
本申请实施例提供了一种人与人工智能对象之间交互的方案，用户在虚拟场景界面中触发控制指令，该控制指令用于指示被控虚拟对象的本方阵营中的人工智能对象执行目标操作，后续基于该控制指令，即可控制该人工智能对象执行该目标操作，从而模拟了该人工智能对象按照用户的指示来执行操作的效果，也即是实现了用户与人工智能对象之间的协同配合，也在虚拟场景中营造出了一种被控虚拟对象与人工智能对象之间协同配合的效果。
其中，该控制指令可以通过用户在虚拟场景界面中的至少一种操作触发，如点击某个控件的操作、拖动某个控件的操作、滑动操作或选中某个对象的操作等。
可选地，该控制指令为攻击指令，或者为撤退指令，或者为集合指令，或者是其他控制指令。
以下将对该控制指令进行举例说明：
可选地，该控制指令为攻击指令，该终端通过虚拟场景界面检测攻击指令，该攻击指令用于指示人工智能对象攻击非本方阵营的第一目标虚拟对象。
在一些实施例中，通过虚拟场景界面，检测攻击指令，包括以下方式：
第一种，通过虚拟场景界面检测对攻击控件的触发操作，生成攻击指令，该攻击指令用于指示人工智能对象攻击非本方阵营中与终端的被控虚拟对象距离最近的虚拟对象。
其中，该虚拟场景界面显示有攻击控件，用户触发该攻击控件，即可发起该攻击指令，此时默认本次需攻击的第一目标虚拟对象为：非本方阵营中与终端的被控虚拟对象距离最近的虚拟对象。
另外，该攻击指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也攻击非本方阵营中与该被控虚拟对象距离最近的虚拟对象。
另外，对攻击控件的触发操作可以为对该攻击控件的点击操作、对该攻击控件的长按操作，或者拖动该攻击控件的操作，或者为对该攻击控件的其他操作。
第二种，通过虚拟场景界面检测将攻击控件拖动至第一目标虚拟对象的对象标识的操作，生成攻击指令，该攻击指令用于指示人工智能对象攻击该对象标识对应的虚拟对象。
其中，该虚拟场景界面显示有攻击控件和虚拟场景中的至少一个对象的对象标识，其中，该对象标识用于确定唯一的对象，可以为虚拟对象的图标或者名称等。
用户将该攻击控件拖动至第一目标虚拟对象的对象标识，即可发起该攻击指令，则本次需攻击的对象即为该对象标识所对应的第一目标虚拟对象。其中，该第一目标虚拟对象可以为非本方阵营中的任一对象，如，第一目标虚拟对象为敌方阵营中的虚拟对象，或者为敌方阵营中的人工智能对象，或者为敌方阵营中的防御塔，或者为野怪，或者为其他虚拟对象。用户将该攻击控件拖动至不同的对象标识，即可指示攻击不同的虚拟对象。
另外，该攻击指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也攻击该对象标识所对应的第一目标虚拟对象。
如图3所示，该虚拟场景界面显示有虚拟场景的缩略地图，用户点击该缩略地图，将该缩略地图放大，即在该虚拟场景界面中显示虚拟场景的地图以及攻击控件301，并且该地图 中显示有各个对象的图标。其中，对象标识302为敌方阵营中的某个英雄的图标。用户将该攻击控件301拖动至对象标识302处，即可生成攻击指令，该攻击指令用于指示本方阵营中的人工智能对象攻击对象标识302对应的英雄。
如图4所示，该虚拟场景界面显示有虚拟场景的缩略地图，用户点击该缩略地图，将该缩略地图放大，即在该虚拟场景界面中显示虚拟场景的地图以及攻击控件401，并且该地图中显示有各个对象的图标。其中，对象标识402为某只野怪的图标。用户将该攻击控件401拖动至对象标识402处，即可生成攻击指令，该攻击指令用于指示本方阵营中的人工智能对象攻击对象标识402对应的野怪。
第三种，通过虚拟场景界面检测对第一目标虚拟对象的对象标识的选中操作，生成攻击指令，该攻击指令用于指示人工智能对象攻击该对象标识对应的虚拟对象。
其中，该虚拟场景界面显示有虚拟场景中的至少一个对象的对象标识，其中，该对象标识用于确定唯一的对象，可以为虚拟对象的图标或者名称等。
用户触发对第一目标虚拟对象的对象标识的选中操作，即可发起该攻击指令，则本次需攻击的对象即为该对象标识所对应的第一目标虚拟对象。其中，该第一目标虚拟对象可以为非本方阵营中的任一对象，如，第一目标虚拟对象为敌方阵营中的虚拟对象，或者为敌方阵营中的人工智能对象，或者为敌方阵营中的防御塔，或者为野怪，或者为其他虚拟对象。用户选中不同的对象标识，即可指示攻击不同的虚拟对象。
另外，该攻击指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也攻击该对象标识所对应的虚拟对象。
如图5所示，该虚拟场景界面显示有虚拟场景的地图及各个对象的图标，其中对象标识501为敌方阵营中的某个英雄的图标。用户触发对该对象标识501的选中操作，即可生成攻击指令，该攻击指令用于指示人工智能对象攻击该对象标识501对应的英雄。
如图6所示，该虚拟场景界面显示有虚拟场景的地图及各个对象的图标，其中对象标识601为某只野怪的图标。用户触发对该对象标识601的选中操作，即可生成攻击指令，该攻击指令用于指示人工智能对象攻击该对象标识601对应的野怪。
如图7所示，该虚拟场景界面显示有虚拟场景的地图及各个对象的图标，其中对象标识701为敌方阵营中的小兵的图标。用户触发对该对象标识701的选中操作，即可生成攻击指令，该攻击指令用于指示人工智能对象攻击该对象标识701对应的小兵。
在本申请实施例中，用户只需触发攻击指令，即可指示本方阵营中的人工智能对象进行攻击，实现了用户与人工智能对象之间的交互，从而提高了本方阵营的操作效率，增强了本方阵营的团队作战能力，从而有助于本方阵营更好地完成团队任务。
可选地，该控制指令为撤退指令，该终端通过虚拟场景界面检测撤退指令，该撤退指令用于指示人工智能对象撤退至安全位置。
在一些实施例中，通过虚拟场景界面，检测撤退指令，包括以下方式：
第一种，通过虚拟场景界面检测对撤退控件的触发操作，生成撤退指令，该撤退指令用于指示人工智能对象撤退至安全位置。
其中，该虚拟场景界面显示有撤退控件，用户触发该撤退控件，即可发起该撤退指令，指示人工智能对象撤退至安全位置。此时默认该安全位置为距离各个人工智能对象最近的安全位置，或者为本方阵营中距离各个人工智能对象最近的防御塔所在的位置，或者为其他位置等。
另外，该撤退指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也撤退至安全位置。
另外，对撤退控件的触发操作可以为对该撤退控件的点击操作、对该撤退控件的长按操作，或者拖动该撤退控件的操作，或者为对该撤退控件的其他操作。
第二种，通过虚拟场景界面检测将撤退控件拖动至本方阵营的第三目标虚拟对象的对象 标识的操作，生成撤退指令，该撤退指令用于指示人工智能对象撤退至第三目标虚拟对象的位置。
其中，该虚拟场景界面显示有撤退控件和虚拟场景中的至少一个对象的对象标识，其中，该对象标识用于确定唯一的对象，可以为虚拟对象的图标或者名称等。
用户将该撤退控件拖动至第三目标虚拟对象的对象标识，即可发起该撤退指令，则本次人工智能对象需撤退至第三目标虚拟对象所在的位置。其中，该第一目标虚拟对象可以为本方阵营中的任一对象，如，第三目标虚拟对象为本方阵营中的虚拟对象，或者为本方阵营中的人工智能对象，或者为本方阵营中的防御塔。用户将该撤退控件拖动至不同的对象标识，即可指示人工智能对象撤退至不同虚拟对象所在的位置。
另外，该撤退指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也撤退至该第三目标虚拟对象所在的位置。
如图8所示，该虚拟场景界面显示有虚拟场景的缩略地图，用户点击该缩略地图，将该缩略地图放大，即在该虚拟场景界面中显示虚拟场景的地图以及撤退控件801，并且该地图中显示有各个对象的图标。其中，对象标识802为本方阵营中的某个防御塔的图标。用户将该攻击控件801拖动至对象标识802处，即可生成撤退指令，该撤退指令用于指示人工智能对象撤退至对象标识802所对应的防御塔所在的位置。
第三种，通过虚拟场景界面检测对第三目标虚拟对象的对象标识的选中操作，生成撤退指令，该撤退指令用于指示人工智能对象撤退至第三目标虚拟对象的位置。
其中，该虚拟场景界面显示有虚拟场景中的至少一个对象的对象标识，其中，该对象标识用于确定唯一的对象，可以为虚拟对象的图标或者名称等。
用户触发对第三目标虚拟对象的对象标识的选中操作，即可发起该撤退指令，则本次人工智能对象需撤退至第三目标虚拟对象所在的位置。其中，该第一目标虚拟对象可以为本方阵营中的任一对象，如，第三目标虚拟对象为本方阵营中的虚拟对象，或者为本方阵营中的人工智能对象，或者为本方阵营中的防御塔。用户选中不同的对象标识，即可指示人工智能对象撤退至不同虚拟对象所在的位置。
另外，该撤退指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也撤退至该第三目标虚拟对象所在的位置。
如图9所示，该虚拟场景界面显示有虚拟场景的地图及各个对象的图标，其中对象标识901为本方阵营中的某个防御塔的图标。用户触发对该对象标识901的选中操作，即可生成撤退指令，该撤退指令用于指示人工智能对象撤退至对象标识901对应的防御塔所在的位置。
可选地，该控制指令为集合指令，该终端通过虚拟场景界面检测集合指令，该集合指令用于指示人工智能对象向被控虚拟对象或本方阵营的第二目标虚拟对象的位置移动。
在本申请实施例中，用户只需触发撤退指令，即可指示本方阵营中的人工智能对象撤退至安全位置，实现了用户与人工智能对象之间的交互，从而提高了本方阵营的操作效率，增强了本方阵营的团队作战能力，从而有助于保护本方阵营中的人工智能对象。
在一些实施例中，通过虚拟场景界面，检测集合指令，包括以下方式：
第一种，通过虚拟场景界面检测对集合控件的触发操作，生成集合指令，该集合指令用于指示人工智能对象向被控虚拟对象的位置移动。
其中，该虚拟场景界面显示有集合控件，用户触发该集合控件，即可发起该集合指令，此时默认人工智能对象需到达被控虚拟对象所在位置，从而与本方阵营中的虚拟对象集合，也即是人工智能对象需向该被控虚拟对象所在位置移动。
另外，该集合指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也向该被控虚拟对象所在位置移动。
另外，对集合控件的触发操作可以为对该集合控件的点击操作、对该集合控件的长按操作，或者拖动该集合控件的操作，或者为对该集合控件的其他操作。
第二种，通过虚拟场景界面检测将集合控件拖动至本方阵营的第二目标虚拟对象的对象标识的操作，生成集合指令，该集合指令用于指示人工智能对象向第二目标虚拟对象的位置移动。
其中，该虚拟场景界面显示有集合控件和虚拟场景中的至少一个对象的对象标识，其中，该对象标识用于确定唯一的对象，可以为虚拟对象的图标或者名称等。
用户将该集合控件拖动至第二目标虚拟对象的对象标识，即可发起该集合指令，指示人工智能对象需前往第二目标虚拟对象所在的位置。其中，该第一目标虚拟对象可以为本方阵营中的任一对象，如，第二目标虚拟对象为本方阵营中的虚拟对象，或者为本方阵营中的人工智能对象，或者为本方阵营中的防御塔。用户将该集合控件拖动至不同虚拟对象的对象标识，即可指示人工智能对象撤退至不同虚拟对象所在的位置。
另外，该集合指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也向该对象标识对应的第二目标虚拟对象所在的位置移动。
如图10所示，该虚拟场景界面显示有虚拟场景的缩略地图，用户点击该缩略地图，将该缩略地图放大，即在该虚拟场景界面中显示虚拟场景的地图以及集合控件1001，并且该地图中显示有各个对象的图标。其中，对象标识1002为本方阵营中的某个英雄的图标。用户将该集合控件1001拖动至对象标识1002处，即可生成集合指令，该集合指令用于指示人工智能对象向对象标识1002对应的英雄所在位置移动。
第三种，通过虚拟场景界面检测对第三目标虚拟对象的对象标识的选中操作，生成集合指令，该集合指令用于指示人工智能对象向第二目标虚拟对象的位置移动。
其中，该虚拟场景界面显示有虚拟场景中的至少一个对象的对象标识，其中，该对象标识用于确定唯一的对象，可以为虚拟对象的图标或者名称等。
用户触发对第二目标虚拟对象的对象标识的选中操作，即可发起该集合指令，指示人工智能对象需向该第二目标虚拟对象所在的位置移动。其中，该第二目标虚拟对象可以为本方阵营中的任一对象，如，第二目标虚拟对象为本方阵营中的虚拟对象，或者为本方阵营中的人工智能对象，或者为本方阵营中的防御塔。
另外，该集合指令还用于指示该被控虚拟对象的本方阵营中的其他虚拟对象，也向该对象标识对应的第二目标虚拟对象所在的位置移动。
在本申请实施例中，用户只需触发集合指令，即可指示本方阵营中的人工智能对象集合，实现了用户与人工智能对象之间的交互，从而提高了本方阵营的操作效率，增强了本方阵营的团队作战能力，从而有助于本方阵营更好地完成团队任务。
203、终端基于控制指令，控制人工智能对象执行目标操作。
终端检测到控制指令之后，即可基于该控制指令，控制人工智能对象执行目标操作。例如，该控制指令为攻击指令，则基于该控制指令，控制人工智能对象攻击非本方阵营的第一目标虚拟对象。或者，该控制指令为撤退指令，则基于该撤退指令，控制人工智能对象撤退至安全位置。或者，该控制指令为集合指令，则基于该集合指令，控制人工智能对象集合。
可选地，该控制指令携带目标虚拟对象的对象标识，该目标操作为向目标虚拟对象移动的操作，则基于该控制指令，终端即可控制人工智能对象执行向目标虚拟对象移动的操作。例如，上述攻击指令、撤退指令或者集合指令，均可看做是指示人工智能对象向目标虚拟对象移动的操作，因此基于这些指令，即可控制人工智能对象移动。
需要说明的是，在本申请实施例中，基于控制指令控制人工智能对象执行目标操作，是指控制人工智能对象开始执行该目标操作，而不限定人工智能对象是否一定要将该目标操作执行完成。例如，在控制人工智能对象移动的过程中，如果该人工智能对象与非本方阵营中的其他虚拟对象相遇，则该人工智能对象与该其他虚拟对象进行对战操作，此时将不再继续按照控制指令的指示进行移动。待对战完成后，该人工智能对象可以继续按照控制指令的指示进行移动，也可以不再响应该控制指令。
另外，本申请实施例仅是以执行主体为终端为例进行说明，而在另一实施例中，终端与服务器连接，终端上的人工智能对象其实是由服务器基于神经网络模型进行控制，则若要实现对人工智能对象的控制，需要终端与服务器之间进行交互。
可选地，终端检测到控制指令之后，向服务器发送该控制指令，服务器用于基于控制指令，调用神经网络模型确定用于完成目标操作的至少一个子操作，控制人工智能对象执行至少一个子操作。
服务器中设置有神经网络模型，该神经网络模型用于对人工智能对象所需执行的操作进行预测。在每个时刻，服务器获取当前时刻的状态数据，调用该神经网络对该状态数据进行处理，从而预测出人工智能对象所需执行的操作，向人工智能对象对应的终端发送该操作，以使该终端显示该人工智能对象执行该操作的场景，从而实现了控制该人工智能对象执行该操作的效果。
而在本申请实施例中，为了响应控制指令，需要人工智能对象执行目标操作，因此，在每个时刻，服务器会调用神经网络模型确定用于完成该目标操作的子操作，从而控制人工智能对象执行该子操作。其中，任两个相邻的时刻之间的间隔可以为预先设置的目标时长，则服务器每隔该目标时长控制该人工智能对象执行一次子操作，则在至少一次控制之后，该人工智能对象会执行至少一个子操作，以满足该控制指令的要求。因此，子操作可以看做是目标操作的分解操作。
需要说明的是，服务器响应控制指令的详细过程可参见下述图11所示的实施例，在此暂不赘述。
本申请实施例扩展了人工智能对象的功能，提供了一种操作可控的人工智能对象，用户只需发出控制指令，即可控制与被控虚拟对象属于同一阵营的人工智能对象执行目标操作，实现了对人工智能对象的控制，从而实现了人与人工智能对象之间的协同配合。
并且，控制指令包括攻击指令、撤退指令、集合指令等多种指令，增强了控制指令的多样性，实现了对人工智能对象的不同类型的控制。
并且，提供了各个控制指令的多种触发方式，为用户提供了多种选择，便于用户的操作。
在上述实施例的基础上，该人工智能对象可以由服务器调用神经网络模型进行控制。以下将结合图11对控制该人工智能对象的过程进行详细说明。
图11是本申请实施例提供的一种人工智能对象控制方法的流程图，该方法由服务器执行，如图11所示，该方法包括：
1101、服务器接收终端发送的控制指令。
在本申请实施例中，服务器与终端连接，终端具有对应的被控虚拟对象，而服务器创建有虚拟场景以及人工智能对象，并且会设置神经网络模型，基于神经网络模型控制人工智能对象执行操作。在虚拟场景中，终端的被控虚拟对象与人工智能对象属于同一阵营，可以配合执行操作。
为了便于与人工智能对象协同配合，用户在终端上触发控制指令，终端向服务器发送该控制指令，该控制指令用于指示人工智能对象执行目标操作。其中，终端获取该控制指令的过程详见上述图2所示的实施例，在此不再赘述。
1102、服务器基于控制指令，控制人工智能对象执行目标操作。
其中，该人工智能对象属于终端的被控虚拟对象的本方阵营。
例如，该控制指令为攻击指令，则基于该控制指令，控制人工智能对象攻击非本方阵营的第一目标虚拟对象。或者，该控制指令为撤退指令，则基于该撤退指令，控制人工智能对象撤退至安全位置。或者，该控制指令为集合指令，则基于该集合指令，控制人工智能对象集合。
可选地，该控制指令携带目标虚拟对象的对象标识，该目标操作为向目标虚拟对象移动的操作，则基于该控制指令，即可控制人工智能对象执行向目标虚拟对象移动的操作。例如， 上述攻击指令、撤退指令或者集合指令，均可看作是指示人工智能对象向目标虚拟对象移动的操作，因此基于这些指令，即可控制人工智能对象移动。
本申请实施例扩展了人工智能对象的功能，提供了一种操作可控的人工智能对象，用户只需发出控制指令，即可控制与被控虚拟对象属于同一阵营的人工智能对象执行目标操作，实现了对人工智能对象的控制，从而实现了人与人工智能对象之间的协同配合。
图12是本申请实施例提供的一种人工智能对象控制方法的流程图，该方法由服务器执行，该方法对服务器控制人工智能对象执行某一子操作的过程进行详细说明。如图12所示，该方法包括：
1201、服务器接收终端发送的控制指令。
其中，控制指令用于指示人工智能对象执行目标操作。
该步骤1201与上述图2和图11所示实施例同理，在此不再赘述。
服务器在接收到控制指令后，需要调用神经网络模型，控制人工智能对象执行与该控制指令对应的目标操作。但是考虑到虚拟场景中各个对象的情况复杂多变，直接控制该人工智能对象执行目标操作，很可能会与当前的虚拟场景的状态不符，因此，为了保证虚拟场景的真实性，以及保证人工智能对象所执行操作的连续性，该服务器会基于控制指令，调用神经网络模型，确定用于完成目标操作的至少一个子操作，控制人工智能对象执行至少一个子操作，以便满足该控制指令的要求。
例如，在每个时刻，服务器基于该控制指令，调用神经网络模型确定一个子操作，向终端下发该子操作，终端即可控制该人工智能对象执行该子操作，并在显示界面中显示该人工智能对象执行该子操作。之后在下一个时刻，服务器基于该控制指令，调用神经网络模型确定下一个子操作，向终端下发下一个子操作，终端即可控制该人工智能对象执行该下一个子操作。以此类推，即可按照该控制指令的指示，控制人工智能对象执行多个子操作。其中，任两个相邻时刻之间的间隔为预先设置的目标时长，则服务器每隔该目标时长控制该人工智能对象执行一次子操作。
并且，本申请实施例以神经网络模型包括编码网络、融合网络和操作预测网络为例进行说明，调用神经网络模型进行处理的过程详见下述步骤1202-1204。
1202、服务器调用编码网络，对多个虚拟对象的状态数据进行编码，得到多个虚拟对象的编码特征。
其中，该虚拟对象的状态数据用于表示该虚拟对象当前的状态，例如，该虚拟对象的状态数据包括虚拟对象的名称、类型、血量、技能类型等。该编码网络用于对虚拟对象的状态数据进行编码，得到编码特征。则该编码特征也能够反映出该虚拟对象当前的状态，后续对该编码特征进行处理，即可考虑该虚拟对象当前的状态，来确定人工智能对象可以执行的操作。
1203、服务器调用融合网络，基于权重特征，对多个虚拟对象的编码特征进行加权融合，得到融合特征，权重特征包括虚拟场景中的多个虚拟对象的权重参数，且权重特征基于控制指令确定。
该融合网络中包括权重特征，权重特征包括虚拟场景中的多个虚拟对象的权重参数，用于表示在预测人工智能对象需执行的操作时该多个虚拟对象的影响程度，基于权重特征对多个虚拟对象的编码特征进行加权融合，能够考虑到不同虚拟对象的不同影响程度，以保证为人工智能对象确定的操作是与虚拟场景中该多个虚拟对象的当前状态是匹配的，所确定的操作是准确的。而为了控制人工智能对象所执行的操作是与控制指令对应的操作，服务器在预测操作之前，首先基于控制指令来确定融合网络中的权重特征。
在一种可能实现方式中，控制指令包括被控虚拟对象的第一对象标识和目标虚拟对象的第二对象标识，目标操作为向目标虚拟对象移动的操作，也即是控制指令用于指示人工智能对象向目标虚拟对象移动。相应地，基于控制指令，确定融合网络中的权重特征，包括：将 多个虚拟对象的编码特征构成编码矩阵，基于编码矩阵与编码矩阵的转置矩阵相乘得到的矩阵，确定第一权重特征，将第一权重特征中的非关联虚拟对象的权重参数设置为负无穷，得到第二权重特征。对第二权重特征进行归一化处理，将归一化处理后的权重特征确定为融合网络中的权重特征。其中，非关联虚拟对象与被控虚拟对象之间的距离，以及非关联虚拟对象与目标虚拟对象之间的距离均不小于距离阈值。
可选地，以下提供一种确定权重特征的操作流程，包括如下步骤：
1、输入
Figure PCTCN2022106330-appb-000001
F＝[f 1；f 2；...；f n]；
其中，w_logit表示权重特征，F表示编码矩阵，f i表示第i个虚拟对象的编码特征，i和n为正整数，n表示虚拟对象的数量，i小于或等于n，
Figure PCTCN2022106330-appb-000002
为缩放因子，[·] 1表示矩阵的第一行向量。
由上述公式可知，输入的权重特征的过程包括：从编码矩阵与编码矩阵的转置矩阵相乘得到的矩阵中提取第一行向量，将此向量与缩放因子的乘积确定为该权重特征。
2、获取选择性注意力集合S，将终端的被控虚拟对象和目标虚拟对象加入到选择性注意力集合S中。
3、如果第i个虚拟对象的编码特征与终端的被控虚拟对象的编码特征之间的距离小于距离阈值，代表第i个虚拟对象为被控虚拟对象的非关联虚拟对象，则将第i个虚拟对象加入到选择性注意力集合S中。
4、如果第i个虚拟对象的编码特征与目标虚拟对象的编码特征之间的距离小于距离阈值，代表第i个虚拟对象为目标虚拟对象的非关联虚拟对象，则将第i个虚拟对象加入到选择性注意力集合S中。由此，可以确定选择性注意力集合S中都包括哪些对象。
5、针对选择性注意力集合S中不包含的对象，将这些对象的权重参数设置为负无穷，保留选择性注意力集合S中包含的对象的权重参数，从而得到更新后的权重特征w_logit’。之后对该权重特征进行归一化处理，即可得到归一化处理后的权重特征softmax(w_logit’)。
采用上述方式确定权重特征，相当于对多个虚拟对象的编码特征采用注意力机制进行处理，能够考虑到重要虚拟对象对目标虚拟对象需执行的操作的影响，而弱化无关虚拟对象对目标虚拟对象需执行的操作的影响，这样确定的操作才更为准确，更符合虚拟场景的当前状态。
在确定该权重特征之后，即可基于该权重特征，对该多个虚拟对象的编码特征进行加权融合，得到融合特征。可选地，基于该权重特征中的每个虚拟对象的权重参数，对该多个虚拟对象的编码特征进行加权求和，得到融合特征。或者，基于该权重特征中的每个虚拟对象的权重参数，对该多个虚拟对象的编码特征进行加权平均，得到融合特征。
1204、服务器调用操作预测网络，对融合特征进行操作预测，得到操作信息。
其中，该操作信息至少包括人工智能对象所需执行的子操作，也即是实现了为人工智能对象预测其要执行的子操作。另外该操作信息还可以包括人工智能对象预计执行该子操作的持续时长，或者还包括提示信息，该提示信息用于展示在终端上，以提示人工智能对象正在执行该子操作，或者还可以包括其他信息。
由于该融合特征能够考虑到该控制指令的指示，也能考虑到多个虚拟对象当前的状态以及影响程度，因此根据该融合特征预测出的操作信息更为准确。
1205、服务器向终端下发该操作信息。
1206、终端接收该操作信息，控制人工智能对象执行该子操作。
本申请实施例仅是以控制一次人工智能对象的过程为例，实际上，在接收到该控制指令之后，可以多次执行本申请实施例中的步骤，从而多次控制人工智能对象执行所确定的子操作。
以下将结合图13和图14，对游戏过程中控制AI智能体的过程进行举例说明，图13提供了一种深度隐变量神经网络模型的示意图，图14提供了一种终端与服务器的交互示意图。
如图13和图14所示，首先，玩家向队友发送控制指令，则玩家所在的终端将控制指令传回服务器，服务器调用深度隐变量神经网络模型，基于控制指令，预测此玩家的AI智能体队友的动作策略(即上述操作信息)。之后服务器将动作策略回传给终端，在终端上AI智能体队友即可根据动作策略执行相应的操作。
并且，该深度隐变量神经网络模型的输入为当前的游戏状态数据，包括当前终端所显示的游戏图像，各个虚拟对象的状态数据，用于表示虚拟场景全局状态的全局信息，以及终端的被控虚拟对象的可用技能等。其中，各个虚拟对象可以包括终端的主英雄、其他英雄、兵线、防御塔、野怪等。
该深度隐变量神经网络模型包括卷积网络、编码网络、融合网络和操作预测网络。其中，卷积网络对图像进行卷积运算，得到图像特征，编码网络对各个虚拟对象的状态数据进行编码，得到多个虚拟对象的编码特征，而融合网络可以看作是一个隐对齐模块，该隐对齐模块对多个虚拟对象的编码特征进行加权融合之后，再将加权融合得到的特征与该图像特征进行融合，得到融合特征，例如采用Concat算子(联结合并多个字符串的算子)将加权融合得到的特征与该图像特征进行融合，得到融合特征。而操作预测网络包括至少一个全连接层和至少一个融合层，从而能够对输入的特征进行全连接或者融合。
该隐对齐模块具有权重特征，该权重特征即为隐对齐模块的隐变量，该权重特征包括多个虚拟对象的权重参数，用来对该多个虚拟对象的编码特征进行加权。本质上，隐对齐模块是一个线性加权求和的算子，且权重向量被限制在一个特殊的隐空间内。隐对齐算子的数学表达式如下所示：
Figure PCTCN2022106330-appb-000003
Figure PCTCN2022106330-appb-000004
在以上算子中，
Figure PCTCN2022106330-appb-000005
是虚拟对象的编码特征，f fusion∈R d是融合后的特征向量，
Figure PCTCN2022106330-appb-000006
是第i个虚拟对象的权重参数，
Figure PCTCN2022106330-appb-000007
是概率单纯形，||v|| 0是向量的l 0范数即非零元素个数。权重特征w∈R n是一个隐随机变量，计算公式如下：
Figure PCTCN2022106330-appb-000008
其中，i和n为正整数，n表示虚拟对象的数量，i小于或等于n，
Figure PCTCN2022106330-appb-000009
为缩放因子，[·] 1表示矩阵的第一行向量。
而该权重特征基于玩家发出的控制指令确定，具体确定方式与上述确定权重特征的操作流程类似，在此不再赘述。
隐对齐成功的关键在于：权重特征w∈R n是一个隶属于概率单纯形Δ n-1内的稀疏向量。权重特征的稀疏性可以理解为是在做虚拟对象的特征选择，即只保留权重参数不为零的虚拟对象的编码特征参与到预测动作策略的过程中，这实际上是在对齐虚拟对象的编码特征与预测的动作策略，所以权重特征w∈R n也被称作隐对齐变量。这种隐含的特征选择机制是与人类玩家打游戏的直觉相符的：玩家做出的决策只是由少数被重点关注的虚拟对象所决定，这其实是一种选择性注意力机制。如果需要改变预测的动作策略，我们只需要根据接收到的控制指令从概率单纯形Δ n-1内采样一个权重特征用于预测相应的动作策略即可。
该操作预测网络的输出为游戏按键，该游戏按键包括移动键和技能键，该移动键用于指示控制AI智能体队友进行移动，该技能键用于指示控制AI智能体队友释放某种技能，如召 唤技能、位置型技能、方向型技能、目标型技能等。
本申请实施例提出了一种解决MOBA游戏中人AI协作的深度隐变量神经网络，其原理是基于隐对齐的方法，控制动作可控制的AI智能体。具体地，通过改变神经网络中的隐变量(权重特征)来改变预测的动作策略。在人类玩家与动作可控制的AI智能体队友搭配的游戏中，人类玩家通过发送具有某种目的性的控制指令给AI智能体队友，并让AI智能体队友执行指令从而能够主动配合玩家的意图，进而提升游戏的趣味和玩家的体验。
需要说明的是，上述图12所示的实施例对调用神经网络模型的过程进行了说明，而为了保证神经网络模型的准确性，需要先训练该神经网络模型。
在一种可能实现方式中，该神经网络模型的训练过程包括：获取样本数据，该样本数据包括在任一样本时刻的样本状态数据和样本操作信息，该样本状态数据包括在该样本时刻下，多个虚拟对象的状态数据，该样本操作信息包括在该样本时刻下，该人工智能对象应当执行的操作，则基于该样本数据训练该神经网络模型，也即是基于该神经网络模型对该样本状态数据进行处理，得到预测操作信息，该预测操作信息包括由神经网络模型预测的、在该样本时刻下该人工智能对象应执行的操作，则根据该样本操作信息与该预测操作信息，对该神经网络模型进行训练，以提高该神经网络模型的准确度。在经过一次或多次训练之后，即可得到准确度满足要求的神经网络模型，从而调用训练完成的神经网络模型来预测人工智能对象应执行的操作。
其中，该样本状态数据的具体内容与上述步骤1202中的状态数据类似，该样本操作信息和该预测操作信息的具体内容与上述步骤1204中的操作信息类似。息，而得到该预测操作信息的具体处理过程与上述步骤1202-1204类似，在此不再赘述。
在另一种可能实现方式中，还可以采用强化学习的方式训练该神经网络模型，在训练过程中，如果人工智能对象响应控制指令成功，则会获得奖励，在经过一次或多次训练后，即可使神经网络模型学习到响应控制指令的能力，从而得到更为准确的神经网络模型。
需要说明的另一点是，除该神经网络模型之外，还可以采用其他方式控制人工智能对象。在一个实施例中，该服务器设置有行为树模型，该行为树模型中定义了人工智能对象的操作规则，则该服务器接收到控制指令时，基于该行为树模型的指示，控制该人工智能对象执行目标操作。例如，该控制指令为集合指令，行为树模型中的规则是：一直向集合目标移动，直至到达集合目标附近为止。则基于该行为树模型的指示，即可控制该人工智能模型一直向集合目标移动，直至到达集合目标附近为止。
在上述实施例的基础上，为了便于判断是否控制人工智能对象对该控制指令进行响应，本申请实施例还提供了一种验证控制指令的方法，能够先验证控制指令，判断是否允许响应该控制指令，在允许响应该控制指令的情况下，再基于该控制指令控制人工智能对象。如图15所示，该方法包括：
1501、服务器接收终端发送的控制指令，并将该控制指令解析为服务器支持的格式。
1502、服务器对该控制指令进行验证。
解析该控制指令之后，对解析的控制指令进行验证，判断当前是否需要响应该控制指令，从而筛选出合理的控制指令，以及不合理的控制指令。
可选地，服务器预先确定目标规则，该目标规则可以是人工确定的。该控制指令满足该目标规则，则验证通过，即可执行后续步骤1503，而该控制指令不满足该目标规则，则验证不通过，不再执行后续步骤1504。
可选地，还可以调用意图预测模型来进行验证，筛选出需要响应的控制指令，过滤掉不需要响应的控制指令。其中，该意图预测模型用于基于当前的状态数据，在虚拟场景中，预测人工智能对象在下一时刻将要到达的第一位置，从而根据该第一位置以及该控制指令所指示的第二位置(指示人工智能对象需要到达的位置)，确定是否要响应该控制指令。
可选地，将虚拟场景的地图划分为多个网格，如将地图划分为12*12的网格。根据当前 的状态数据，确定该人工智能对象的预测位置信息，该状态数据包含该人工智能对象当前的位置，该预测位置信息包括地图中每个网格对应的概率，该概率表示在下一时刻该人工智能对象到达该网格的概率。则该预测位置信息中网格的概率越大，在下一时刻该人工智能对象越有可能到达该网格。那么，将该预测位置信息中概率最大的前K个网格构成集合，该集合即可表示下一时刻该人工智能对象最有可能到达的K个位置，K为正整数。
之后将该集合与该控制指令所指示的第二位置进行对比，在该第二位置属于该集合的情况下，该控制指令验证通过，表示允许响应该控制指令。而在该第二位置不属于该集合的情况下，该控制指令验证未通过，表示不允许响应该控制指令。
举例来说，如图16所示，为地图中的每个网格按照顺序依次编号，人工智能对象当前在网格121，控制指令要求人工智能对象到达的目标位置在网格89，如果意图预测模型预测的集合P＝{26，27，89，116，117}，该集合中包括网格89，那么人工智能对象可以执行当前的控制指令，从而前往网格89。如果意图预测模型预测的集合P＝{26，27，80，116，117}，该集合中不包括网格89，那么需要拒绝执行当前的控制指令。
1503、在控制指令验证通过的情况下，基于该控制指令预测人工智能对象的动作策略。
1504、服务器将动作策略回传给终端，以便终端接收服务器回传的动作策略，按照该动作策略控制人工智能对象。
可选地，意图预测模型包括编码网络、融合网络和位置预测网络，调用意图预测模型确定人工智能对象的预测位置信息的过程，包括：服务器调用编码网络，对多个虚拟对象的状态数据进行编码，得到多个虚拟对象的编码特征，调用融合网络，基于权重特征，对多个虚拟对象的编码特征进行加权融合，得到融合特征，权重特征包括虚拟场景中的多个虚拟对象的权重参数，服务器调用位置预测网络，对融合特征进行位置预测，得到预测位置信息。
其中，调用编码网络、融合网络和位置预测网络的过程与上述步骤1202-1204的过程类似，区别在于，上述步骤1203中，神经网络模型中的融合网络的权重特征是基于所接收到的控制指令确定的，接收到的控制指令不同，该权重特征可能会发生变化。而在本实施例中，意图预测模型训练完成之后，意图预测模型中的融合网络的权重特征是不变的。
而在调用该意图预测模型，预测控制指令的意图之前，还需要先训练上述意图预测模型。训练意图预测模型的过程包括：获取样本数据，该样本数据包括在样本时刻多个虚拟对象的状态数据以及对应的位置标签，该状态数据至少包括在该样本时刻的图像帧，该状态数据与上述实施例中的状态数据类似，而该位置标签为在该样本时刻的下一个时刻发生进攻、集合、撤退等目标事件的网格的编号。之后即可基于该样本数据对意图预测模型进行训练，从而得到训练完成的意图预测模型。其中，上述事件定义为以下几种类型：1.发生了攻击行为；2.发生了集合行为；3.发生了撤退行为(从危险之处撤退到安全位置)，当然还可以包括其他的事件类型。并且，为了保证训练的准确性，该样本数据可以包括多个样本时刻的状态数据以及对应的位置标签，从而根据连续的多个时刻的状态数据和位置标签训练该意图预测模型。
可选地，可以基于人类玩家的数据训练意图预测模型，也即是，在训练意图预测模型的过程中获取到的样本数据是属于人类玩家的，即样本数据中包括人类玩家的被控虚拟对象的状态数据以及对应的位置标签，这样能够保证意图预测模型能够学习到人类玩家的事件，进而在调用意图预测模型预测控制指令的意图时，能够保证该意图是符合人类玩家的行为的。
如图17所示，某一英雄从t_(s-1)时刻出发，之后在t_s时刻发生了攻击行为，所以将[t_(s-1),t_s)时间段内的图像帧的位置标签标定为t_s时刻的网格编号；同理，该英雄在t_(s+1)时刻发生了撤退行为，则将[t_s,t_(s+1))时间段内的图像帧的位置标签标定为t_(s+1)时刻的网格编号，从而得到了各个时刻的样本数据。
而综合上述神经网络模型和上述意图预测模型，本申请实施例还提供了一种同时训练神经网络模型和意图预测模型的方式，也即是，如图18所示，服务器预先创建一种预测模型，该预测模型包括编码网络1801、融合网络1802和预测网络1803，而预测网络1803包括操作 预测网络1831和位置预测网络1832，该操作预测网络1831用于预测人工智能对象的操作信息，以确定人工智能对象要执行的操作，而该位置预测网络1832用于预测人工智能对象的预测位置信息，以确定人工智能对象下一时刻要到达的位置。
在训练该预测模型时，获取样本数据，该样本数据包括在任一样本时刻的样本状态数据和样本操作信息，以及位置标签，该样本状态数据包括在该样本时刻下，多个虚拟对象的状态数据，该样本操作信息包括在该样本时刻下，该人工智能对象应当执行的操作，而该位置标签为在该样本时刻的下一个时刻发生进攻、集合、撤退等事件的网格的编号。则基于该样本数据训练该预测模型，也即是基于该预测模型对该样本状态数据进行处理，得到预测操作信息和预测位置标签，该预测操作信息包括由预测模型预测的、在该样本时刻下该人工智能对象应执行的操作，该预测位置信息包括人工智能对象在下一时刻要到达的位置，则根据该样本操作信息与该预测操作信息，以及该位置标签和该预测位置信息，对该预测模型进行训练，以提高该预测模型的准确度。在经过一次或多次训练之后，即可得到准确度满足要求的预测模型，从而调用训练完成的预测模型来进行预测。
举例来说，从图13和图19可以看出，在图13所示的深度隐变量神经网络模型中增添另一个全连接层，用于预测人工智能对象下一时刻要到达的位置，从而得到图19所示的深度隐变量神经网络模型，训练图19所示的深度隐变量神经网络模型即可。
这种综合训练的方式，能够同时训练出具备操作预测和位置预测功能的预测模型，提高了训练速度，节省了训练时间。
在一种可能实现方式中，在训练完成之后，为了分别实现操作预测功能和位置预测功能，基于已训练好的网络分别构建如图20所示的意图预测模型和如图21所示的神经网络模型。意图预测模型包括编码网络2001、融合网络2002和位置预测网络2003，神经网络模型包括编码网络2101、融合网络2102和操作预测网络2103。其中，编码网络2001和编码网络2101是由图18中的编码网络1801训练得到的，融合网络2002和融合网络2102是由图18中的融合网络1802训练得到的，位置预测网络2003是由图18中的位置预测网络1832训练得到的，操作预测网络2103是由图18中的操作预测网络1831训练得到的。
相应的，在服务器接收到控制指令之后，先调用意图预测模型，基于多个虚拟对象的状态数据确定人工智能对象的预测位置信息，该预测位置信息包括人工智能对象下一时刻可能到达的位置组成的集合，控制指令指示人工智能对象需到达的位置属于该集合的情况下，表示控制指令验证通过。则服务器基于该控制指令，确定融合网络2102中的权重特征，之后调用神经网络模型确定人工智能对象的操作信息。
并且，虽然在调用意图预测模型时，已经基于多个虚拟对象的状态数据确定多个虚拟对象的编码特征，但是后续在调用神经网络模型时，服务器还可以重新基于多个虚拟对象的状态数据确定多个虚拟对象的编码特征，以预测人工智能对象的操作信息。而且服务器可以对该多个虚拟对象的状态数据进行更改，例如，控制指令包括目标虚拟对象的第二对象标识，目标操作为向目标虚拟对象移动的操作，也即是控制指令指示人工智能对象到达该目标虚拟对象所在的位置。但是如果该目标虚拟对象不是该人工智能对象的攻击对象，神经网络模型是不会控制该人工智能对象移动至该目标虚拟对象所在的位置的，因此为了使神经网络模型能够控制该人工智能对象移动至该目标虚拟对象所在的位置，首先将该多个虚拟对象的状态数据中，人工智能对象的攻击对象的位置信息更改为该目标虚拟对象附近的位置信息，例如将该人工智能对象的攻击对象的坐标更改，以使更改后的坐标与该目标虚拟对象的坐标之间的距离小于预设距离，则人工智能对象会误以为攻击对象在该目标虚拟对象附近，因此会向该攻击对象移动，从而实现了向该目标虚拟对象移动的效果。
在另一种可能实现方式中，为了节省处理流程，意图预测模型和神经网络模型可以共享相同的编码网络，如图22所示，编码网络2001和编码网络2101是同一个编码网络，服务器调用编码网络2001，基于多个虚拟对象的状态数据确定多个虚拟对象的编码特征，之后调用 融合网络2002和位置预测网络2003确定人工智能对象的预测位置信息，该预测位置信息包括人工智能对象下一时刻可能到达的位置组成的集合，控制指令指示人工智能对象需到达的位置属于该集合的情况下，表示控制指令验证通过。则服务器基于该控制指令，确定融合网络2102中的权重特征，之后调用融合网络2102和操作预测网络2103，对编码网络2001得到的编码特征进行预测，从而确定人工智能对象的操作信息。
图23是本申请实施例提供的一种人工智能对象控制装置的结构示意图。参见图23，该装置包括：
显示模块2301，用于显示虚拟场景界面，虚拟场景界面显示有本端的被控虚拟对象；
检测模块2302，用于通过虚拟场景界面检测控制指令，控制指令用于指示被控虚拟对象的本方阵营中的人工智能对象执行目标操作；
控制模块2303，用于基于控制指令，控制人工智能对象执行目标操作。
本申请实施例提供的人工智能对象控制装置，实现了人与人工智能对象之间的交互，该人工智能对象的操作是可控的，用户只需发出控制指令，即可控制与被控虚拟对象属于同一阵营的人工智能对象执行目标操作，实现了对人工智能对象的控制，从而实现了人与人工智能对象之间的协同配合，扩展了人工智能对象的功能。
可选地，检测模块2302，包括：
第一检测单元，用于通过虚拟场景界面，检测攻击指令，攻击指令用于指示人工智能对象攻击非本方阵营的第一目标虚拟对象；或者，
第二检测单元，用于通过虚拟场景界面，检测撤退指令，撤退指令用于指示人工智能对象撤退至安全位置；或者，
第三检测单元，用于通过虚拟场景界面，检测集合指令，集合指令用于指示人工智能对象向被控虚拟对象或本方阵营的第二目标虚拟对象的位置移动。
可选地，第一检测单元，用于：
通过虚拟场景界面检测对攻击控件的触发操作，生成攻击指令，第一目标虚拟对象为非本方阵营中与被控虚拟对象距离最近的虚拟对象；或者，
通过虚拟场景界面检测将攻击控件拖动至第一目标虚拟对象的对象标识的操作，生成攻击指令；或者，
通过虚拟场景界面检测对第一目标虚拟对象的对象标识的选中操作，生成攻击指令。
可选地，第二检测单元，用于：
通过虚拟场景界面检测对撤退控件的触发操作，生成撤退指令，撤退指令用于指示人工智能对象撤退至安全位置；或者，
通过虚拟场景界面检测将撤退控件拖动至本方阵营的第三目标虚拟对象的对象标识的操作，生成撤退指令，撤退指令用于指示人工智能对象撤退至第三目标虚拟对象的位置；或者，
通过虚拟场景界面检测对第三目标虚拟对象的对象标识的选中操作，生成撤退指令，撤退指令用于指示人工智能对象撤退至第三目标虚拟对象的位置。
可选地，第三检测单元，用于：
通过虚拟场景界面检测对集合控件的触发操作，生成集合指令，集合指令用于指示人工智能对象向被控虚拟对象的位置移动；或者，
通过虚拟场景界面检测将集合控件拖动至本方阵营的第二目标虚拟对象的对象标识的操作，生成集合指令，集合指令用于指示人工智能对象向第二目标虚拟对象的位置移动；或者，
通过虚拟场景界面检测对第三目标虚拟对象的对象标识的选中操作，生成集合指令，集合指令用于指示人工智能对象向第二目标虚拟对象的位置移动。
可选地，控制指令携带目标虚拟对象的对象标识，目标操作为向目标虚拟对象移动的操作；
控制模块2303，包括：
第一控制单元，用于基于控制指令，控制人工智能对象执行向目标虚拟对象移动的操作。
可选地，控制模块2303，包括：
第二控制单元，用于向服务器发送控制指令，服务器用于基于控制指令，调用神经网络模型确定用于完成目标操作的至少一个子操作，控制人工智能对象执行至少一个子操作。
上述所有可选技术方案，可以采用任意结合形成本申请的可选实施例，在此不再一一赘述。
图24是本申请实施例提供的一种人工智能对象控制装置的结构示意图。参见图24，该装置包括：
接收模块2401，用于接收终端发送的控制指令，该控制指令用于指示人工智能对象执行目标操作；
控制模块2402，用于基于控制指令，控制人工智能对象执行目标操作；
其中，人工智能对象属于终端的被控虚拟对象的本方阵营。
本申请实施例提供的人工智能对象控制装置，实现了终端与服务器之间的交互，服务器在接收到控制指令的情况下，能够控制与终端属于同一阵营的人工智能对象执行相应的目标操作，实现了对人工智能对象的控制，从而实现了人与人工智能对象之间的协同配合，扩展了人工智能对象的功能。
可选地，控制模块2402，包括：
第一控制单元，用于基于控制指令，调用神经网络模型确定用于完成目标操作的至少一个子操作，控制人工智能对象执行至少一个子操作。
可选地，神经网络模型包括编码网络、融合网络和操作预测网络，第一控制单元，包括：
编码子单元，用于调用编码网络，对多个虚拟对象的状态数据进行编码，得到多个虚拟对象的编码特征；
融合子单元，用于调用融合网络，基于权重特征，对多个虚拟对象的编码特征进行加权融合，得到融合特征，权重特征包括虚拟场景中的多个虚拟对象的权重参数，且权重特征基于控制指令确定；
预测子单元，用于调用操作预测网络，对融合特征进行操作预测，得到操作信息，操作信息至少包括人工智能对象所需执行的子操作。
可选地，控制指令包括被控虚拟对象的第一对象标识和目标虚拟对象的第二对象标识，目标操作为向目标虚拟对象移动的操作；
第一控制单元，还包括：
确定子单元，用于：
将多个虚拟对象的编码特征构成编码矩阵；
基于编码矩阵与编码矩阵的转置矩阵相乘得到的矩阵，确定第一权重特征；
将第一权重特征中的非关联虚拟对象的权重参数设置为负无穷，得到第二权重特征，非关联虚拟对象与被控虚拟对象之间的距离，以及非关联虚拟对象与目标虚拟对象之间的距离均不小于距离阈值；
对第二权重特征进行归一化处理，将归一化处理后的权重特征确定为融合网络中的权重特征。
可选地，控制模块2402，包括：
验证单元，用于调用意图预测模型，对控制指令进行验证；
第二控制单元，用于在控制指令验证通过的情况下，基于控制指令，控制人工智能对象执行目标操作。
上述所有可选技术方案，可以采用任意结合形成本申请的可选实施例，在此不再一一赘述。
需要说明的是：上述实施例提供的人工智能对象控制装置在控制人工智能对象执行目标 操作时，仅以上述各功能模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能模块完成，即将设备的内部结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。另外，上述实施例提供的人工智能对象控制装置与人工智能对象控制方法实施例属于同一构思，其具体实现过程详见方法实施例，这里不再赘述。
图25示出了本申请一个示例性实施例提供的终端2500的结构示意图。终端2500包括有：处理器2501和存储器2502。
处理器2501可以包括一个或多个处理核心，比如4核心处理器、8核心处理器等。处理器2501可以采用DSP(Digital Signal Processing，数字信号处理)、FPGA(Field Programmable Gate Array，现场可编程门阵列)、PLA(Programmable Logic Array，可编程逻辑阵列)中的至少一种硬件形式来实现。一些实施例中，处理器2501可以包括AI(Artificial Intelligence，人工智能)处理器，该AI处理器用于处理有关机器学习的计算操作。
存储器2502可以包括一个或多个计算机可读存储介质，该计算机可读存储介质可以是非暂态的。存储器2502还可包括高速随机存取存储器，以及非易失性存储器，比如一个或多个磁盘存储设备、闪存存储设备。在一些实施例中，存储器2502中的非暂态的计算机可读存储介质用于存储至少一条计算机程序，该至少一条计算机程序用于被处理器2501所具有以实现本申请中方法实施例提供的人工智能对象控制方法。
在一些实施例中，终端2500还可选包括有：外围设备接口2503和至少一个外围设备。处理器2501、存储器2502和外围设备接口2503之间可以通过总线或信号线相连。各个外围设备可以通过总线、信号线或电路板与外围设备接口2503相连。可选地，外围设备包括：射频电路2504或显示屏2505中的至少一种。另外还可以包括其他组件。
外围设备接口2503可被用于将I/O(Input/Output，输入/输出)相关的至少一个外围设备连接到处理器2501和存储器2502。在一些实施例中，处理器2501、存储器2502和外围设备接口2503被集成在同一芯片或电路板上；在一些其他实施例中，处理器2501、存储器2502和外围设备接口2503中的任意一个或两个可以在单独的芯片或电路板上实现，本实施例对此不加以限定。
射频电路2504用于接收和发射RF(Radio Frequency，射频)信号，也称电磁信号。射频电路2504通过电磁信号与通信网络以及其他通信设备进行通信。射频电路2504将电信号转换为电磁信号进行发送，或者，将接收到的电磁信号转换为电信号。可选地，射频电路2504包括：天线系统、RF收发器、一个或多个放大器、调谐器、振荡器、数字信号处理器、编解码芯片组、用户身份模块卡等等。射频电路2504可以通过至少一种无线通信协议来与其它设备进行通信。该无线通信协议包括但不限于：城域网、各代移动通信网络(2G、3G、4G及5G)、无线局域网和/或WiFi(Wireless Fidelity，无线保真)网络。在一些实施例中，射频电路2504还可以包括NFC(Near Field Communication，近距离无线通信)有关的电路，本申请对此不加以限定。
显示屏2505用于显示UI(User Interface，用户界面)。该UI可以包括图形、文本、图标、视频及其它们的任意组合。当显示屏2505是触摸显示屏时，显示屏2505还具有采集在显示屏2505的表面或表面上方的触摸信号的能力。该触摸信号可以作为控制信号输入至处理器2501进行处理。此时，显示屏2505还可以用于提供虚拟按钮和/或虚拟键盘，也称软按钮和/或软键盘。
本领域技术人员可以理解，图25中示出的结构并不构成对终端2500的限定，可以包括比图示更多或更少的组件，或者组合某些组件，或者采用不同的组件布置。
图26是本申请实施例提供的一种服务器的结构示意图，该服务器2600可因配置或性能不同而产生比较大的差异，可以包括一个或一个以上处理器(Central Processing Units，CPU)2601和一个或一个以上的存储器2602，其中，所述存储器2602中存储有至少一条计算机程序，所述至少一条计算机程序由所述处理器2601加载并执行以实现上述各个方法实施例提供 的方法。当然，该服务器还可以具有有线或无线网络接口、键盘以及输入输出接口等部件，以便进行输入输出，该服务器还可以包括其他用于实现设备功能的部件，在此不做赘述。
本申请实施例还提供了一种计算机可读存储介质，该计算机可读存储介质中存储有至少一条计算机程序，该至少一条计算机程序由处理器加载并执行，以实现上述实施例的人工智能对象控制方法所执行的操作。
本申请实施例还提供了一种计算机程序产品或计算机程序，计算机程序产品或计算机程序包括计算机程序代码，计算机程序代码存储在计算机可读存储介质中，计算机设备的处理器从计算机可读存储介质读取计算机程序代码，处理器执行计算机程序代码，使得计算机设备实现如上述实施例的人工智能对象控制方法所执行的操作。在一些实施例中，本申请实施例所涉及的计算机程序可被部署在一个计算机设备上执行，或者在位于一个地点的多个计算机设备上执行，又或者，在分布在多个地点且通过通信网络互连的多个计算机设备上执行，分布在多个地点且通过通信网络互连的多个计算机设备可以组成区块链系统。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成，也可以通过程序来指令相关的硬件完成，所述的程序可以存储于一种计算机可读存储介质中，上述提到的存储介质可以是只读存储器，磁盘或光盘等。
以上所述仅为本申请实施例的可选实施例，并不用以限制本申请实施例，凡在本申请实施例的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本申请的保护范围之内。

Claims (16)

  1. 一种人工智能对象控制方法，所述方法包括：
    终端显示虚拟场景界面，所述虚拟场景界面显示有所述终端的被控虚拟对象；
    所述终端通过所述虚拟场景界面检测控制指令，所述控制指令用于指示所述被控虚拟对象的本方阵营中的人工智能对象执行目标操作；
    所述终端基于所述控制指令，控制所述人工智能对象执行所述目标操作。
  2. 根据权利要求1所述的方法，其中，所述终端通过所述虚拟场景界面检测控制指令，包括：
    所述终端通过所述虚拟场景界面，检测攻击指令，所述攻击指令用于指示所述人工智能对象攻击非本方阵营的第一目标虚拟对象；或者，
    所述终端通过所述虚拟场景界面，检测撤退指令，所述撤退指令用于指示所述人工智能对象撤退至安全位置；或者，
    所述终端通过所述虚拟场景界面，检测集合指令，所述集合指令用于指示所述人工智能对象向所述被控虚拟对象或本方阵营的第二目标虚拟对象的位置移动。
  3. 根据权利要求2所述的方法，其中，所述终端通过所述虚拟场景界面，检测攻击指令，包括：
    所述终端通过所述虚拟场景界面检测对攻击控件的触发操作，生成所述攻击指令，所述第一目标虚拟对象为非本方阵营中与所述被控虚拟对象距离最近的虚拟对象；或者，
    所述终端通过所述虚拟场景界面检测将所述攻击控件拖动至所述第一目标虚拟对象的对象标识的操作，生成所述攻击指令；或者，
    所述终端通过所述虚拟场景界面检测对所述第一目标虚拟对象的对象标识的选中操作，生成所述攻击指令。
  4. 根据权利要求2所述的方法，其中，所述终端通过所述虚拟场景界面，检测撤退指令，包括：
    所述终端通过所述虚拟场景界面检测对撤退控件的触发操作，生成所述撤退指令，所述撤退指令用于指示所述人工智能对象撤退至安全位置；或者，
    所述终端通过所述虚拟场景界面检测将所述撤退控件拖动至本方阵营的第三目标虚拟对象的对象标识的操作，生成所述撤退指令，所述撤退指令用于指示所述人工智能对象撤退至所述第三目标虚拟对象的位置；或者，
    所述终端通过所述虚拟场景界面检测对所述第三目标虚拟对象的对象标识的选中操作，生成所述撤退指令，所述撤退指令用于指示所述人工智能对象撤退至所述第三目标虚拟对象的位置。
  5. 根据权利要求2所述的方法，其中，所述终端通过所述虚拟场景界面，检测集合指令，包括：
    所述终端通过所述虚拟场景界面检测对集合控件的触发操作，生成所述集合指令，所述集合指令用于指示所述人工智能对象向所述被控虚拟对象的位置移动；或者，
    所述终端通过所述虚拟场景界面检测将所述集合控件拖动至本方阵营的第二目标虚拟对象的对象标识的操作，生成所述集合指令，所述集合指令用于指示所述人工智能对象向所述第二目标虚拟对象的位置移动；或者，
    所述终端通过所述虚拟场景界面检测对所述第三目标虚拟对象的对象标识的选中操作， 生成所述集合指令，所述集合指令用于指示所述人工智能对象向所述第二目标虚拟对象的位置移动。
  6. 根据权利要求1-5任一项所述的方法，其中，所述控制指令携带目标虚拟对象的对象标识，所述目标操作为向所述目标虚拟对象移动的操作；
    所述终端基于所述控制指令，控制所述人工智能对象执行所述目标操作，包括：
    所述终端基于所述控制指令，控制所述人工智能对象执行向所述目标虚拟对象移动的操作。
  7. 根据权利要求1-5任一项所述方法，其中，所述终端基于所述控制指令，控制所述人工智能对象执行所述目标操作，包括：
    所述终端向服务器发送所述控制指令，所述服务器用于基于所述控制指令，调用神经网络模型确定用于完成所述目标操作的至少一个子操作，控制所述人工智能对象执行所述至少一个子操作。
  8. 一种人工智能对象控制方法，所述方法包括：
    服务器接收终端发送的控制指令，所述控制指令用于指示人工智能对象执行目标操作；
    所述服务器基于所述控制指令，控制所述人工智能对象执行所述目标操作；
    其中，所述人工智能对象属于所述终端的被控虚拟对象的本方阵营。
  9. 根据权利要求8所述的方法，其中，所述服务器基于所述控制指令，控制所述人工智能对象执行所述目标操作，包括：
    所述服务器基于所述控制指令，调用神经网络模型确定用于完成所述目标操作的至少一个子操作，控制所述人工智能对象执行所述至少一个子操作。
  10. 根据权利要求9所述的方法，其中，所述神经网络模型包括编码网络、融合网络和操作预测网络，所述服务器基于所述控制指令，调用神经网络模型确定用于完成所述目标操作的至少一个子操作，包括：
    所述服务器调用所述编码网络，对所述多个虚拟对象的状态数据进行编码，得到所述多个虚拟对象的编码特征；
    所述服务器调用所述融合网络，基于权重特征，对所述多个虚拟对象的编码特征进行加权融合，得到融合特征，所述权重特征包括虚拟场景中的多个虚拟对象的权重参数，且所述权重特征基于所述控制指令确定；
    所述服务器调用所述操作预测网络，对所述融合特征进行操作预测，得到操作信息，所述操作信息至少包括所述人工智能对象所需执行的子操作。
  11. 一种人工智能对象控制装置，所述装置包括：
    显示模块，用于显示虚拟场景界面，所述虚拟场景界面显示有本端的被控虚拟对象；
    检测模块，用于通过所述虚拟场景界面检测控制指令，所述控制指令用于指示所述被控虚拟对象的本方阵营中的人工智能对象执行目标操作；
    控制模块，用于基于所述控制指令，控制所述人工智能对象执行所述目标操作。
  12. 一种人工智能对象控制装置，所述装置包括：
    接收模块，用于接收终端发送的控制指令，所述控制指令用于指示人工智能对象执行目标操作；
    控制模块，用于基于所述控制指令，控制所述人工智能对象执行所述目标操作；
    其中，所述人工智能对象属于所述终端的被控虚拟对象的本方阵营。
  13. 一种终端，所述终端包括处理器和存储器，所述存储器中存储有至少一条计算机程序，所述至少一条计算机程序由所述处理器加载并执行，以实现如权利要求1至7任一项所述的人工智能对象控制方法所执行的操作。
  14. 一种服务器，所述服务器包括处理器和存储器，所述存储器中存储有至少一条计算机程序，所述至少一条计算机程序由所述处理器加载并执行，以实现如权利要求8至10任一项所述的人工智能对象控制方法所执行的操作。
  15. 一种计算机可读存储介质，所述计算机可读存储介质中存储有至少一条计算机程序，所述至少一条计算机程序由处理器加载并执行，以实现如权利要求1至7任一项所述的人工智能对象控制方法所执行的操作，或者以实现如权利要求8至10任一项所述的人工智能对象控制方法所执行的操作。
  16. 一种计算机程序产品，所述计算机程序产品包括计算机程序代码，所述计算机程序代码存储在计算机可读存储介质中，计算机设备的处理器从所述计算机可读存储介质读取所述计算机程序代码，所述处理器执行所述计算机程序代码，使得所述计算机设备实现如权利要求1至7任一项所述的人工智能对象控制方法所执行的操作，或者以实现如权利要求8至10任一项所述的人工智能对象控制方法所执行的操作。
PCT/CN2022/106330 2021-08-23 2022-07-18 人工智能对象控制方法、装置、设备及存储介质 WO2023024762A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/135,914 US20230293995A1 (en) 2021-08-23 2023-04-18 Artificial intelligence object control method and apparatus, device, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110969023.3 2021-08-23
CN202110969023.3A CN114272599A (zh) 2021-08-23 2021-08-23 人工智能对象控制方法、装置、设备及存储介质

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/135,914 Continuation US20230293995A1 (en) 2021-08-23 2023-04-18 Artificial intelligence object control method and apparatus, device, and storage medium

Publications (1)

Publication Number Publication Date
WO2023024762A1 true WO2023024762A1 (zh) 2023-03-02

Family

ID=80868440

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/106330 WO2023024762A1 (zh) 2021-08-23 2022-07-18 人工智能对象控制方法、装置、设备及存储介质

Country Status (3)

Country Link
US (1) US20230293995A1 (zh)
CN (1) CN114272599A (zh)
WO (1) WO2023024762A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110276408B (zh) * 2019-06-27 2022-11-22 腾讯科技（深圳）有限公司 3d图像的分类方法、装置、设备及存储介质
CN114272599A (zh) * 2021-08-23 2022-04-05 腾讯科技（深圳）有限公司 人工智能对象控制方法、装置、设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111282279A (zh) * 2020-02-05 2020-06-16 腾讯科技（深圳）有限公司 模型训练的方法、基于交互式应用的对象控制方法及装置
CN111450533A (zh) * 2020-03-31 2020-07-28 腾讯科技（深圳）有限公司 虚拟场景中的虚拟对象控制方法、装置、终端及存储介质
CN111514585A (zh) * 2020-03-17 2020-08-11 清华大学 智能体的控制方法及系统、计算机装置以及存储介质
WO2020252400A1 (en) * 2019-06-14 2020-12-17 Fighter Base Publishing, Inc. Methods and system for artificial intelligence powered user interface
CN112973117A (zh) * 2021-04-15 2021-06-18 腾讯科技（深圳）有限公司 虚拟对象的交互方法、奖励发放方法、装置、设备及介质
CN114272599A (zh) * 2021-08-23 2022-04-05 腾讯科技（深圳）有限公司 人工智能对象控制方法、装置、设备及存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020252400A1 (en) * 2019-06-14 2020-12-17 Fighter Base Publishing, Inc. Methods and system for artificial intelligence powered user interface
CN111282279A (zh) * 2020-02-05 2020-06-16 腾讯科技（深圳）有限公司 模型训练的方法、基于交互式应用的对象控制方法及装置
CN111514585A (zh) * 2020-03-17 2020-08-11 清华大学 智能体的控制方法及系统、计算机装置以及存储介质
CN111450533A (zh) * 2020-03-31 2020-07-28 腾讯科技（深圳）有限公司 虚拟场景中的虚拟对象控制方法、装置、终端及存储介质
CN112973117A (zh) * 2021-04-15 2021-06-18 腾讯科技（深圳）有限公司 虚拟对象的交互方法、奖励发放方法、装置、设备及介质
CN114272599A (zh) * 2021-08-23 2022-04-05 腾讯科技（深圳）有限公司 人工智能对象控制方法、装置、设备及存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
"Tencent's "Juewu" paper discloses technical details", DEEP LEARNING AND ADVANCED INTELLIGENT DECISION MAKING, AIJISHU, CN, CN, pages 1 - 5, XP009543997, Retrieved from the Internet <URL:https://aijishu.com/a/1060000000100723> *

Also Published As

Publication number Publication date
CN114272599A (zh) 2022-04-05
US20230293995A1 (en) 2023-09-21

Similar Documents

Publication Publication Date Title
CN112691377B (zh) 虚拟角色的控制方法、装置、电子设备及存储介质
WO2023024762A1 (zh) 人工智能对象控制方法、装置、设备及存储介质
JP7390400B2 (ja) 仮想オブジェクトの制御方法並びにその、装置、端末及びコンピュータプログラム
JP7492611B2 (ja) バーチャルシーンにおけるデータ処理方法、装置、コンピュータデバイス、及びコンピュータプログラム
US11969654B2 (en) Method and apparatus for determining target virtual object, terminal, and storage medium
US20220379214A1 (en) Method and apparatus for a control interface in a virtual environment
CN111249735A (zh) 控制对象的路径规划方法、装置、处理器及电子装置
TW202222395A (zh) 預購道具的顯示方法、裝置、設備、媒體及產品
TWI821779B (zh) 虛擬對象的控制方法、裝置、計算機設備及儲存媒體
CN115888119A (zh) 一种游戏ai训练方法、装置、电子设备及存储介质
CN113018862B (zh) 虚拟对象的控制方法、装置、电子设备及存储介质
JP2023164687A (ja) 仮想オブジェクトの制御方法及び装置並びにコンピュータ装置及びプログラム
CN114225372B (zh) 虚拟对象的控制方法、装置、终端、存储介质及程序产品
CN115645923A (zh) 游戏交互方法、装置、终端设备及计算机可读存储介质
CN114307150B (zh) 虚拟对象之间的互动方法、装置、设备、介质及程序产品
CN113769379A (zh) 虚拟对象的锁定方法、装置、设备、存储介质及程序产品
WO2024027292A1 (zh) 虚拟场景中的交互方法、装置、电子设备、计算机可读存储介质及计算机程序产品
WO2023231557A9 (zh) 虚拟对象的互动方法、装置、设备、存储介质及程序产品
CN118036694A (zh) 一种训练智能体的方法、装置和设备及计算机存储介质
CN116549974A (zh) 虚拟对战中的信息通讯方法、装置及产品
CN116549972A (zh) 虚拟资源处理方法、装置、计算机设备及存储介质
CN117643723A (zh) 游戏交互方法、装置、计算机设备及计算机可读存储介质
CN117462949A (zh) 游戏技能的处理方法、装置、计算机设备及存储介质
CN116510319A (zh) 虚拟资源获取方法、装置、计算机设备及存储介质
JP2024522484A (ja) 仮想シーンに基づくグラフィック表示方法、装置、機器及びコンピュータプログラム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22860108

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE