Methods, devices, equipment and storage media for controlling the behavior of intelligent agents

By acquiring and analyzing the agent's environmental information, and using the target prediction module and behavior prediction module to predict the agent's selected target and behavior, the problem of irrationality and poor logic in agent behavior in the prior art is solved, the reliability and interpretability of agent behavior are improved, and the player's game experience is enhanced.

CN116726504BActive Publication Date: 2026-11-10SHENZHEN HONGXI TECHNOLOGY CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310587444.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-05-23
Publication Date
2026-11-10
Estimated Expiration
2043-05-23

AI Technical Summary

Technical Problem

Existing technologies suffer from irrationality and poor logic when controlling the behavior of intelligent agents, especially in the inability to adjust in time after failing to predict the attack position, resulting in a poor player experience, and support heroes tend to downplay the role of teammates in protecting them.

Method used

By acquiring environmental information of the target agent, determining target vector features and image features, analyzing them using the target prediction module and behavior prediction module, predicting the agent's target selection and behavior information, and executing the predicted behavior through the control module.

Benefits of technology

It improves the reliability and interpretability of agent behavior, ensuring that agents can rationally select targets and execute correct actions, thereby enhancing the player's gaming experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116726504B_ABST
    Figure CN116726504B_ABST
Patent Text Reader

Abstract

This invention provides a method, apparatus, device, and storage medium for controlling the behavior of an intelligent agent. The method is applied to a behavior control system, which includes a target prediction module and a behavior prediction module. The method includes: acquiring environmental information of the target intelligent agent and determining corresponding target vector features and target image features based on the environmental information; analyzing the target vector features and target image features through the target prediction module to obtain a first predicted target corresponding to the target intelligent agent; analyzing the target vector features, target image features, and the first predicted target through the behavior prediction module to obtain predicted behavior information corresponding to the target intelligent agent; and controlling the target intelligent agent to execute the behavior corresponding to the predicted behavior information based on the first predicted target. This application aims to predict the target selection and behavior of an intelligent agent, thereby improving the reliability and interpretability of the intelligent agent's behavior while achieving control over the agent's behavior.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of deep learning technology, and in particular to a method for controlling the behavior of an intelligent agent, a behavior control device, a computer device, and a computer-readable storage medium. Background Technology

[0002] With the development of computer technology, computer games and other human-computer interaction applications, such as Multiplayer Online Battle Arena (MOBA) games, have become increasingly popular forms of entertainment. In MOBA games, players are typically divided into multiple opposing factions, controlling virtual objects to compete and achieve game objectives to win. In human-computer battles or game automation, computers, such as game AI (Artificial Intelligence), automatically control computer players or corresponding managed virtual objects to engage in combat. Therefore, it is necessary to control the behavior of these intelligent agents to ensure a better gaming experience for players.

[0003] Existing technologies primarily predict the agent's next attack location and then its behavior at that location to control the agent based on that behavior. However, this method has certain drawbacks. For example, if the agent fails to attack at the predicted location and should move to another location, the next predicted attack location might still be the same. Furthermore, for characters with weaker attack capabilities, such as support heroes, this method can easily lead them to prioritize attacking targets while neglecting their role in protecting teammates. Therefore, this approach results in illogical and irrational agent behavior, leading to a poor player experience. Summary of the Invention

[0004] This application provides a behavior control method, behavior control device, computer equipment, and computer-readable storage medium for intelligent agents, aiming to predict the selection of targets and behaviors of intelligent agents, thereby improving the reliability and interpretability of intelligent agent behavior based on the realization of controlling the behavior of intelligent agents.

[0005] To achieve the above objectives, this application provides a behavior control method for an intelligent agent, applied to a behavior control system. The behavior control system includes a target prediction module and a behavior prediction module. The method includes:

[0006] Obtain the environmental information of the target intelligent agent, and determine the corresponding target vector features and target image features based on the environmental information;

[0007] The target prediction module analyzes the target vector features and the target image features to obtain the first predicted target corresponding to the target agent.

[0008] The behavior prediction module analyzes the target vector features, the target image features, and the first prediction target to obtain the prediction behavior information corresponding to the target agent.

[0009] The target agent is controlled to perform the behavior corresponding to the predicted behavior information based on the first predicted target.

[0010] To achieve the above objectives, this application also provides a behavior control device, the behavior control device comprising:

[0011] An acquisition module is used to acquire environmental information of the target intelligent agent and determine the corresponding target vector features and target image features based on the environmental information.

[0012] A target prediction module is used to analyze the target vector features and the target image features to obtain a first predicted target corresponding to the target agent;

[0013] A behavior prediction module is used to analyze the target vector features, the target image features, and the first prediction target to obtain the prediction behavior information corresponding to the target agent.

[0014] A control module is used to control the target agent to perform the behavior corresponding to the prediction behavior information based on the first prediction target.

[0015] Furthermore, to achieve the above objectives, this application also provides a computer device, the device including a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the steps of the behavior control method for an intelligent agent provided in any of the embodiments of this application.

[0016] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the steps of the behavior control method for an intelligent agent provided in any of the embodiments of this application.

[0017] This application discloses a method, device, computer equipment, and computer-readable storage medium for controlling the behavior of an intelligent agent. These can be applied to a behavior control system, which includes a target prediction module and a behavior prediction module. Specifically, it acquires environmental information about the target intelligent agent and determines corresponding target vector features and target image features based on this information. The target prediction module analyzes the target vector features and target image features to obtain a first predicted target for the target intelligent agent. Furthermore, the behavior prediction module analyzes the target vector features, target image features, and the first predicted target to obtain predicted behavior information for the target intelligent agent. Thus, it can control the target intelligent agent to execute the behavior corresponding to the predicted behavior information based on the first predicted target. This application can predict the target selection and behavior information of the intelligent agent, thereby improving the reliability and interpretability of the agent's behavior while achieving control over the agent's behavior. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating the steps of a behavior control method for an intelligent agent provided in an embodiment of this application;

[0020] Figure 2 This is a schematic diagram illustrating the steps for obtaining a first target prediction model according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of a scenario for obtaining a first predicted target, provided by an embodiment of this application;

[0022] Figure 4 This is a schematic diagram illustrating the steps for obtaining a target behavior prediction model according to an embodiment of this application;

[0023] Figure 5 This is a schematic diagram illustrating a step for obtaining predicted behavior information provided in an embodiment of this application;

[0024] Figure 6 This is a schematic diagram of a scenario for obtaining predicted behavior information provided in an embodiment of this application;

[0025] Figure 7 This is a schematic diagram of the steps of another intelligent agent behavior control method provided in the embodiments of this application;

[0026] Figure 8This is a schematic block diagram of a behavior control device provided in an embodiment of this application;

[0027] Figure 9 This is a schematic block diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation. Furthermore, although functional modules are divided in the device diagram, in some cases, a different module division may be used.

[0030] The term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.

[0031] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0032] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating the steps of a behavior control method for an intelligent agent provided in an embodiment of this application. The behavior control method for the intelligent agent can be applied to a behavior control system, which includes a target prediction module and a behavior prediction module.

[0033] like Figure 1 As shown, the behavior control method of the intelligent agent includes steps S11 to S14.

[0034] Step S11: Obtain the environmental information of the target agent and determine the corresponding target vector features and target image features based on the environmental information.

[0035] In this context, an intelligent agent refers to a game character capable of interacting with the game environment. For example, in a specific game environment, an intelligent agent can, based on its own perception of the game environment, follow existing instructions or learn autonomously, communicate, cooperate, or fight with other players to autonomously complete the set goals within its game environment.

[0036] It should be noted that this application does not limit the type of the target intelligent agent. For example, in multiplayer online tactical battle games, the target intelligent agent can be a jungler hero, a mage hero, a marksman hero, a support hero, etc.

[0037] Furthermore, environmental information refers to the virtual environment displayed when an application runs on a terminal. This environmental information can include a simulation of the real world, a semi-simulated / semi-fictional 3D environment, or a purely fictional 3D environment. Environmental information can include any of the following: a 2D game environment, a 2.5D game environment, or a 3D game environment.

[0038] Optionally, the environmental information is also used for game environment battles between at least two game characters, and the environmental information contains game resources available to at least two game characters. Furthermore, the map of the game environment is a square or rectangle, which includes a symmetrical lower left diagonal area and an upper right diagonal area; the victory conditions for battles in the environmental information include capturing or destroying the enemy faction's objective points, which can be all of the enemy faction's points or some of their points, such as the main base and defensive towers, etc., and this application does not limit this.

[0039] Based on the above environmental information, the corresponding target vector features and target image features can be determined.

[0040] Among them, the target vector feature is the vector feature of the target agent, which can be abstracted from units such as heroes, minions, monsters, and buildings in the environmental information corresponding to the target agent. The target image feature is the image feature of the target agent, which represents the hotspot distribution of each unit in the environmental information corresponding to the target agent.

[0041] In the embodiments of this application, the environmental information of the target intelligent agent can be obtained, and the target vector features and target image features corresponding to the target intelligent agent can be determined based on the environmental information, so as to perform target prediction on the target intelligent agent.

[0042] Step S12: Analyze the target vector features and target image features through the target prediction module to obtain the first predicted target corresponding to the target agent.

[0043] Here, the first predicted target is the target agent's chosen objective for its next action. Understandably, after obtaining the first predicted target, the target agent knows the object of its next action.

[0044] It should be noted that this application does not limit the type of the first predicted target. For example, the first predicted target can include units such as jungle monsters, enemy minions, and enemy buildings. The target agent can only select these units as targets when attacking them. The first predicted target can also be units such as friendly minions and friendly buildings. The target agent can select these units only when it appears near them and there are no other hero units nearby for a period of time. This allows the target agent to advance with friendly minions and defend friendly buildings. The first predicted target can also be units such as teammates or enemy heroes. The target agent can select these units only when it appears near them. When multiple targets appear simultaneously, the highest priority target can be selected as the final target according to a certain priority. Generally, enemy heroes have a higher priority than teammates.

[0045] In this embodiment, the target vector features and target image features can be analyzed by the target prediction module to obtain the first predicted target corresponding to the target agent. Thus, the target of the next action of the target agent can be determined based on the first predicted target. Compared to existing technologies that predict the agent's next attack position and thus predict the agent's action information, this embodiment solves the problem of incorrect positioning of intermediate processes caused by the late attack of the target agent in existing technologies. Furthermore, this application can also strengthen the awareness of support heroes to protect teammates, preventing support heroes from favoring attacking targets and neglecting their role in protecting teammates.

[0046] Step S13: Analyze the target vector features, target image features, and the first prediction target through the behavior prediction module to obtain the prediction behavior information corresponding to the target agent.

[0047] Among them, the predicted behavior information includes the next predicted behavior of the target intelligent agent.

[0048] It should be noted that this application does not limit the above analysis method. For example, the behavior prediction module may include a behavior prediction model, thereby analyzing the predicted behavior information of the target intelligent agent based on the behavior prediction model; in addition, the behavior prediction module may also include a preset trigger instruction, which can be triggered when the target vector features, target image features and the first prediction target of the target intelligent agent meet certain conditions, thereby obtaining the predicted behavior information.

[0049] In this embodiment of the application, the behavior prediction module can analyze the target vector features, target image features and the first prediction target of the target intelligent agent to obtain the predicted behavior information of the target intelligent agent, so as to control the behavior of the target intelligent agent.

[0050] Step S14: Control the target agent to execute the behavior corresponding to the predicted behavior information based on the first predicted target.

[0051] After obtaining the first predicted target and predicted behavior information for the target agent, the target agent can be controlled to execute the behavior corresponding to the predicted behavior information in response to the first predicted target. Because this method combines the prediction of the target agent's next action and the specific action itself, it improves the reliability and interpretability of the agent's behavior while achieving control over the agent's actions.

[0052] The behavior control method for an intelligent agent disclosed in this application can be applied to a behavior control system, wherein the behavior control system includes a target prediction module and a behavior prediction module. Specifically, it can acquire environmental information of the target intelligent agent, and then determine the corresponding target vector features and target image features based on the environmental information. The target prediction module analyzes the target vector features and target image features to obtain a first predicted target corresponding to the target intelligent agent. Further, the behavior prediction module analyzes the target vector features, target image features, and the first predicted target to obtain predicted behavior information corresponding to the target intelligent agent. Thus, it is possible to control the target intelligent agent to execute the behavior corresponding to the predicted behavior information based on the first predicted target. This application can predict the target selection and behavior information of the intelligent agent, thereby improving the reliability and interpretability of the intelligent agent's behavior while achieving control over the intelligent agent's behavior.

[0053] Please continue reading. Figure 2 as well as Figure 3 , Figure 2 This is a schematic diagram illustrating the steps for obtaining a first target prediction model according to an embodiment of this application; Figure 3 This is a schematic diagram of a scenario for obtaining a first predicted target, provided by an embodiment of this application. For example... Figure 2 As shown, the first target prediction model can be obtained through steps S121 to S124.

[0054] Step S121: Obtain several training data sets and determine the environmental information corresponding to several agents from the training data sets.

[0055] The training data may include game battle data. Thus, environmental information corresponding to several intelligent agents can be determined from several game battle data. For the limitations of environmental information, please refer to the above description, which will not be repeated here.

[0056] Furthermore, the intelligent agent can be a jungler hero, a mage hero, a marksman hero, a support hero, etc., and this application does not limit this.

[0057] Step S122: For each agent, determine the input features from the corresponding environmental information.

[0058] The input features include vector features and image features.

[0059] Furthermore, the vector features are the vector features of each agent, which can be abstracted from units such as heroes, minions, monsters, and buildings in the environment information corresponding to the agent. The image features are the image features of each agent, which represent the hotspot distribution of each unit in the environment information corresponding to the agent.

[0060] In this embodiment of the application, several training data can be obtained, and environmental information corresponding to several agents can be determined from the several training data. Then, vector features and image features corresponding to each agent can be determined based on the environmental information.

[0061] Step S123: Based on each agent, determine the target corresponding to the input feature as the label of the input feature.

[0062] Step S124: Input the input features and labels of each agent into the initial target prediction model for training to obtain the first target prediction model.

[0063] Specifically, the target corresponding to the input features of each agent can be used as the label of the input data. Then, each set of labeled input features is input into the initial target prediction model for supervised learning. When the training termination condition is met, such as when the number of training iterations reaches a threshold or the model's output accuracy reaches a threshold, the training ends, and the first target prediction model that has been trained is obtained.

[0064] Alternatively, please continue reading Figure 3 , Figure 3 This is a schematic diagram of a scenario for obtaining a first predicted target, provided by an embodiment of this application. For example... Figure 3 As shown, based on the above embodiments, the target prediction module analyzes the target vector features and target image features to obtain the first predicted target corresponding to the target agent, including: inputting the target vector features and target image features into the first target prediction model and outputting the first predicted target corresponding to the target agent.

[0065] Specifically, since the first target prediction model is trained based on existing agents and environmental information, the target vector features and target image features determined by the target agent through environmental information can be input into the first target prediction model to output the first predicted target for the next action of the target agent.

[0066] In this embodiment, input features and labels corresponding to several agents can be input into an initial target prediction model for supervised learning, thereby training a first target prediction model. Thus, target vector features and target image features can be input into the first target prediction model, outputting a first predicted target corresponding to the target agent, achieving the prediction of the first predicted target for the next action of the target agent.

[0067] Optionally, please refer to Figure 4 , Figure 4 This is a schematic diagram illustrating the steps for obtaining a target behavior prediction model according to an embodiment of this application. Figure 4 As shown, the behavior prediction module includes a target behavior prediction model, which can be obtained through steps S131 to S132.

[0068] Step S131: Obtain several training data sets and determine the sample data corresponding to several agents from the training data sets.

[0069] The training data consists of game battle data corresponding to several intelligent agents, which can be obtained from game replay files of historical game matches. These replay files may include, for example, built-in game files used to record player actions during gameplay; these files can be converted using the game's built-in engine and its decompression function to obtain the game battle data. This game battle data includes game state information corresponding to multiple player actions. Additionally, the game battle data may also include video data from the game process. This video data can be processed to extract the necessary information; the specific type of information depends on the input information required by the model and can be determined based on the input data required by different game intelligent agents.

[0070] Furthermore, the sample data includes behavioral information and input features for each agent. The input features consist of vector features, image features, and the prediction target for each agent; the specific methods for obtaining these features are described above and will not be repeated here.

[0071] Step S132: Input the sample data into the initial behavior prediction model for training to obtain the target behavior prediction model.

[0072] After obtaining the sample data, it can be input into the initial behavior prediction model for supervised learning, and then trained to obtain the target behavior prediction model.

[0073] Based on the above embodiments, the behavior prediction module analyzes the target vector features, target image features, and the first prediction target to obtain the prediction behavior information corresponding to the target agent. This includes: inputting the target vector features, target image features, and the first prediction target into the target behavior prediction model and outputting the prediction behavior information corresponding to the target agent.

[0074] The predicted behavioral information includes behaviors such as moving, releasing skills, and purchasing equipment, which are not limited in this application.

[0075] Specifically, after obtaining the target behavior prediction model, the target vector features, target image features, and the first prediction target corresponding to the target agent can be input into the target behavior prediction model to output the prediction behavior information corresponding to the target agent.

[0076] In this embodiment, sample data corresponding to several agents can be input into a supervised learning system to train a target behavior prediction model. Thus, target vector features, target image features, and a first prediction target can be input into the target behavior prediction model, outputting predicted behavior information corresponding to the target agent, thereby predicting the next action of the target agent.

[0077] Please continue reading. Figure 5 as well as Figure 6 , Figure 5 This is a schematic diagram illustrating a step for obtaining predicted behavior information provided in an embodiment of this application; Figure 6 This is a schematic diagram illustrating a scenario for obtaining predicted behavior information, provided in an embodiment of this application. For example... Figure 5 As shown, the predicted behavior information can be obtained through steps S1311 to S1313.

[0078] Step S1311: Encode the first prediction target to obtain the first prediction target after encoding.

[0079] It should be noted that this application does not limit the specific implementation method of the above encoding operation. For example, it can be one-hot encoding or encoder encoding. This application uses one-hot encoding as an example for explanation.

[0080] One-hot encoding, also known as single-hot coding, is a one-bit effective encoding method. It uses an N-bit state register to encode N states, with each state having its own independent register bit, and only one bit is active at any given time. One-hot encoding represents categorical variables as binary vectors. It requires mapping categorical values ​​to integer values, with each integer value represented as a binary vector. Except for the integer index, the value is zero, marked as 1.

[0081] In the embodiments of this application, the first prediction target can be encoded to obtain the first prediction target after encoding, thereby solving the problem that the classifier is not good at handling attribute data, and also playing a role in expanding features to a certain extent.

[0082] Step S1312: Concatenate the first predicted target after the encoding operation with the target vector features to obtain the concatenated vector features.

[0083] Step S1313: Input the spliced ​​vector features and target image features into the target behavior prediction model, and output the predicted behavior information corresponding to the target agent.

[0084] Specifically, the first predicted target after encoding can be concatenated with the target vector features to obtain concatenated vector features, thus achieving full feature fusion. Further, the concatenated vector features and target image features are input into the target behavior prediction model, outputting the predicted behavior information corresponding to the target agent. This enables the prediction of the target agent's behavior.

[0085] In this embodiment, the first predicted target can be encoded to obtain an encoded first predicted target. This solves the problem of classifiers' inability to handle attribute data and also expands the features to some extent. The encoded first predicted target is then concatenated with the target vector features to obtain concatenated vector features, achieving full feature fusion. Furthermore, the concatenated vector features and target image features are input into the target behavior prediction model to output the predicted behavior information corresponding to the target agent.

[0086] Please continue reading. Figure 7 , Figure 7 This is a schematic diagram illustrating the steps of another intelligent agent behavior control method provided in an embodiment of this application. For example... Figure 7 As shown, the behavior control system also includes an autonomous selection module. After the target prediction module analyzes the target vector features and target image features to obtain the first predicted target corresponding to the target agent, the behavior control of the agent can be realized through steps S21 to S24.

[0087] Step S21: Determine the second prediction target corresponding to the target agent through the autonomous selection module.

[0088] The autonomous selection module can be used to autonomously select a target. Therefore, a second predicted target corresponding to the target agent can be determined based on the autonomous selection module.

[0089] Understandably, neutral objectives like Baron Nashor will spawn during gameplay. Therefore, when the target AI is a jungler, it should proactively lead its teammates to attack Baron Nashor when its team has gained an advantage. However, the target AI's first predicted target might be minions, jungle monsters, or even an enemy hero. Therefore, the AI ​​can use a self-selection module to determine Baron Nashor as the target AI's second predicted target.

[0090] Step S22: Determine whether the first prediction target and the second prediction target are consistent.

[0091] Step S23: When the first prediction target and the second prediction target are inconsistent, the behavior prediction module analyzes the target vector features, target image features and the second prediction target to obtain the prediction behavior information corresponding to the target agent.

[0092] Step S24: Control the target agent to execute the behavior corresponding to the predicted behavior information based on the second predicted target.

[0093] Specifically, it can be determined whether the first predicted target and the second predicted target are consistent. If they are inconsistent, the second predicted target is identified as the target of the final action of the target agent. Thus, the action prediction module analyzes the target vector features, target image features, and the second predicted target to obtain the predicted action information corresponding to the target agent, and controls the target agent to execute the action corresponding to the predicted action information based on the second predicted target.

[0094] Optionally, the second predicted target determined by the autonomous selection module can be triggered by signals emitted by other players during the game. That is, during the game, when other players emit signals such as attacking enemy heroes, protecting friendly turrets, or attacking Baron Nashor, the autonomous selection module can determine the target corresponding to that signal as the second predicted target.

[0095] In this embodiment, a second predicted target corresponding to the target agent can be determined through an autonomous selection module. When the first and second predicted targets are inconsistent, the second predicted target is determined as the target for the final behavior of the target agent, thereby obtaining the predicted behavior information corresponding to the target agent. This allows the target agent to be controlled to execute the behavior corresponding to the predicted behavior information based on the second predicted target. Because of the introduction of the autonomous selection module, unreasonable predicted targets can be adjusted. Therefore, based on the control of agent behavior, the editability and interpretability of agent behavior control are achieved, thereby improving the player's gaming experience.

[0096] Please see Figure 8 , Figure 8 This is a schematic block diagram of a behavior control device provided in an embodiment of this application. The behavior control device can be configured in a server to execute the steps of the aforementioned behavior control method for an intelligent agent.

[0097] like Figure 8 As shown, the behavior control device 200 includes: an acquisition module 201, a target prediction module 202, a behavior prediction module 203, and a control module 204.

[0098] The acquisition module 201 is used to acquire environmental information of the target intelligent agent and determine the corresponding target vector features and target image features based on the environmental information.

[0099] The target prediction module 202 is used to analyze the target vector features and the target image features to obtain the first predicted target corresponding to the target agent;

[0100] The behavior prediction module 203 is used to analyze the target vector features, the target image features and the first prediction target to obtain the prediction behavior information corresponding to the target agent;

[0101] Control module 204 is used to control the target intelligent agent to perform the behavior corresponding to the prediction behavior information based on the first prediction target.

[0102] The acquisition module 201 acquires a plurality of training data and determines environmental information corresponding to a plurality of agents from the plurality of training data; based on each agent, it determines input features from the corresponding environmental information, wherein the input features include vector features and image features; based on each agent, it determines the target corresponding to the input feature as the label of the input feature; and it inputs the input feature corresponding to each agent and the label of the input feature into an initial target prediction model for training to obtain the first target prediction model.

[0103] The target prediction module 202 is further configured to input the target vector features and the target image features into the first target prediction model, and output the first predicted target corresponding to the target agent.

[0104] The acquisition module 201 is further configured to acquire a plurality of training data and determine sample data corresponding to a plurality of agents from the plurality of training data, wherein the sample data includes behavioral information and input features corresponding to each agent; and input the sample data into an initial behavior prediction model for training to obtain the target behavior prediction model.

[0105] The behavior prediction module 203 is further configured to input the target vector features, the target image features and the first prediction target into the target behavior prediction model, and output the predicted behavior information corresponding to the target agent.

[0106] The behavior prediction module 203 is further configured to encode the first prediction target to obtain the encoded first prediction target; concatenate the encoded first prediction target with the target vector features to obtain the concatenated vector features; input the concatenated vector features and the target image features into the target behavior prediction model, and output the predicted behavior information corresponding to the target agent.

[0107] The control module 204 is further configured to: determine the second prediction target corresponding to the target agent through the autonomous selection module; determine whether the first prediction target and the second prediction target are consistent; when the first prediction target and the second prediction target are inconsistent, analyze the target vector features, the target image features and the second prediction target through the behavior prediction module to obtain the prediction behavior information corresponding to the target agent; and control the target agent to execute the behavior corresponding to the prediction behavior information based on the second prediction target.

[0108] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and its modules and units can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0109] The methods and apparatus of this application can be used in a wide variety of general-purpose or special-purpose computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer terminal devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0110] For example, the above-described method and apparatus can be implemented as a computer program, which can be used in, for example... Figure 9 It runs on the computer device shown.

[0111] Please see Figure 9 , Figure 9 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device may be a server.

[0112] like Figure 9 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include volatile storage media, non-volatile storage media, and internal memory.

[0113] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any method of controlling the behavior of an intelligent agent.

[0114] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0115] Internal memory provides an environment for the execution of computer programs stored in non-volatile storage media. When these computer programs are executed by a processor, the processor can execute any method of controlling the behavior of an intelligent agent.

[0116] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that the structure of this device is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the device to which the present application is applied. Specific devices may include more or fewer components than shown in the diagram, or may combine certain components, or may have different component arrangements.

[0117] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0118] In some embodiments, the processor is used to run a computer program stored in a memory to perform the following steps: acquiring environmental information of the target agent and determining corresponding target vector features and target image features based on the environmental information; analyzing the target vector features and target image features through the target prediction module to obtain a first predicted target corresponding to the target agent; analyzing the target vector features, target image features, and the first predicted target through the behavior prediction module to obtain predicted behavior information corresponding to the target agent; and controlling the target agent to perform the behavior corresponding to the predicted behavior information based on the first predicted target.

[0119] In some embodiments, the processor is further configured to acquire a plurality of training data and determine environmental information corresponding to a plurality of agents from the plurality of training data; based on each agent, determine input features from the corresponding environmental information, wherein the input features include vector features and image features; based on each agent, determine the target corresponding to the input feature as the label of the input feature; input the input feature corresponding to each agent and the label of the input feature into an initial target prediction model for training to obtain the first target prediction model.

[0120] In some implementations, the processor is further configured to input the target vector features and the target image features into the first target prediction model, and output a first predicted target corresponding to the target agent.

[0121] In some embodiments, the processor is further configured to acquire a plurality of training data and determine sample data corresponding to a plurality of agents from the plurality of training data, wherein the sample data includes behavioral information and input features corresponding to each agent; and input the sample data into an initial behavior prediction model for training to obtain the target behavior prediction model.

[0122] In some implementations, the processor is further configured to input the target vector features, the target image features, and the first predicted target into the target behavior prediction model, and output the predicted behavior information corresponding to the target agent.

[0123] In some embodiments, the processor is further configured to encode the first predicted target to obtain an encoded first predicted target; concatenate the encoded first predicted target with the target vector features to obtain concatenated vector features; input the concatenated vector features and the target image features into the target behavior prediction model, and output the predicted behavior information corresponding to the target agent.

[0124] In some embodiments, the processor is further configured to: determine a second predicted target corresponding to the target agent through the autonomous selection module; determine whether the first predicted target and the second predicted target are consistent; when the first predicted target and the second predicted target are inconsistent, analyze the target vector features, the target image features and the second predicted target through the behavior prediction module to obtain predicted behavior information corresponding to the target agent; and control the target agent to execute the behavior corresponding to the predicted behavior information based on the second predicted target.

[0125] This application also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed, implement any of the intelligent agent behavior control methods provided in this application.

[0126] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0127] Furthermore, the computer-readable storage medium may primarily include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc.

[0128] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for controlling the behavior of an intelligent agent, characterized in that, Applied to a behavior control system, the behavior control system including a target prediction module and a behavior prediction module, the method includes: Obtain the environmental information of the target intelligent agent, and determine the corresponding target vector features and target image features based on the environmental information; The target prediction module analyzes the target vector features and the target image features to obtain the first predicted target corresponding to the target agent. The behavior prediction module analyzes the target vector features, the target image features, and the first prediction target to obtain the prediction behavior information corresponding to the target agent. The target intelligent agent is controlled to perform the behavior corresponding to the predicted behavior information based on the first predicted target. The behavior control system further includes an autonomous selection module. After the target prediction module analyzes the target vector features and the target image features to obtain the first predicted target corresponding to the target agent, the system further includes: The autonomous selection module determines the second prediction target corresponding to the target agent. Determine whether the first prediction target and the second prediction target are consistent; When the first prediction target and the second prediction target are inconsistent, the behavior prediction module analyzes the target vector features, the target image features and the second prediction target to obtain the prediction behavior information corresponding to the target agent. The target intelligent agent is controlled to perform the behavior corresponding to the predicted behavior information based on the second predicted target. The behavior prediction module includes a target behavior prediction model, and the method further includes: Acquire a number of training data and determine sample data corresponding to a number of intelligent agents from the number of training data, wherein the sample data includes behavioral information and input features corresponding to each intelligent agent; The sample data is input into the initial behavior prediction model for training to obtain the target behavior prediction model.

2. The method according to claim 1, characterized in that, The target prediction module includes a first target prediction model, and the method further includes: Acquire a number of training data, and determine the environmental information corresponding to a number of intelligent agents from the training data; Based on each of the aforementioned intelligent agents, input features are determined from the corresponding environmental information, wherein the input features include vector features and image features; Based on each of the aforementioned agents, the target corresponding to the input feature is determined as the label of the input feature; The input features and labels corresponding to each agent are input into the initial target prediction model for training to obtain the first target prediction model.

3. The method according to claim 2, characterized in that, The step of analyzing the target vector features and the target image features through the target prediction module to obtain the first predicted target corresponding to the target agent includes: The target vector features and the target image features are input into the first target prediction model, and the first predicted target corresponding to the target agent is output.

4. The method according to claim 1, characterized in that, The step of analyzing the target vector features, the target image features, and the first predicted target through the behavior prediction module to obtain the predicted behavior information corresponding to the target agent includes: The target vector features, the target image features, and the first predicted target are input into the target behavior prediction model, and the predicted behavior information corresponding to the target agent is output.

5. The method according to claim 4, characterized in that, The step of inputting the target vector features, the target image features, and the first predicted target into the target behavior prediction model, and outputting the predicted behavior information corresponding to the target agent, includes: The first prediction target is encoded to obtain the first prediction target after encoding. The first predicted target after the encoding operation is concatenated with the target vector feature to obtain the concatenated vector feature; The concatenated vector features and the target image features are input into the target behavior prediction model, and the predicted behavior information corresponding to the target agent is output.

6. A behavior control device, characterized in that, For implementing the behavior control method of an intelligent agent as described in any one of claims 1-5, the behavior control device comprises: An acquisition module is used to acquire environmental information of the target intelligent agent and determine the corresponding target vector features and target image features based on the environmental information. A target prediction module is used to analyze the target vector features and the target image features to obtain a first predicted target corresponding to the target agent; A behavior prediction module is used to analyze the target vector features, the target image features, and the first prediction target to obtain the prediction behavior information corresponding to the target agent. The control module is used to control the target agent to perform the behavior corresponding to the prediction behavior information based on the first prediction target; An autonomous selection module is configured to determine a second prediction target corresponding to the target agent; determine whether the first prediction target and the second prediction target are consistent; when the first prediction target and the second prediction target are inconsistent, analyze the target vector features, the target image features, and the second prediction target to obtain prediction behavior information corresponding to the target agent; and control the target agent to execute the behavior corresponding to the prediction behavior information based on the second prediction target.

7. A computer device, characterized in that, include: A memory and a processor; wherein the memory is connected to the processor and is used to store a program, and the processor is used to implement the steps of the behavior control method for an intelligent agent as described in any one of claims 1-5 by running the program stored in the memory.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the steps of the behavior control method for an intelligent agent as described in any one of claims 1-5.

Citation Information

Patent Citations

  • Artificial intelligence object control method and device, equipment and storage medium

    CN114272599A

  • Game AI training method and device, electronic equipment and storage medium

    CN115888119A