Intelligent agent behavior control method, electronic device, and storage medium

Abstract

The application relates to the technical field of agent control, and discloses an agent behavior control method, an electronic device and a storage medium. The method comprises the following steps: in response to an agent behavior request received, generating a corresponding behavior request object; acquiring a policy rule set matched with action type information and resource type information; acquiring environment boundary parameters related to execution terminal information and target resource information, and generating an environment boundary snapshot according to the environment boundary parameters; performing joint constraint judgment on the behavior request object and the environment boundary snapshot according to the policy rule set, action risk information and parameter type information, to obtain a constraint judgment result; and controlling a corresponding execution terminal to perform a corresponding action according to the constraint judgment result. The embodiment of the application can improve the agent behavior control precision in a complex physical environment.

Description

Technical Field

[0001] This application relates to the field of intelligent agent control technology, and in particular to an intelligent agent behavior control method, electronic device and storage medium. Background Technology

[0002] Existing intelligent agent behavior control solutions typically focus on role permissions, interface authentication, action whitelists, call frequency control, or manual approval. While these solutions can limit the scope of tools an intelligent agent can invoke to some extent, they still have significant shortcomings when facing physically-dependent scenarios such as robots / IoT. For the same control command, such as "open the valve," "start the robot," "read the camera," or "call the remote device interface," it may have completely different security implications in different geographical areas, with different device states, different network boundaries, and different on-site risk conditions. Relying solely on static policy rules makes it difficult to accurately determine whether the current action should be allowed, thus affecting the accuracy of intelligent agent behavior control. Summary of the Invention

[0003] The purpose of this application is to provide an intelligent agent behavior control method, electronic device, and storage medium that can improve the accuracy of intelligent agent behavior control in complex physical environments.

[0004] This application provides an intelligent agent behavior control method, including: In response to a received action request from the intelligent agent, a corresponding action request object is generated; the action request object includes action type information, action risk information, resource type information, parameter type information, execution terminal information, and target resource information. Obtain a set of policy rules that match the action type information and the resource type information; Obtain environmental boundary parameters related to the execution terminal information and the target resource information, and generate an environmental boundary snapshot based on the environmental boundary parameters; Based on the set of policy rules, the action risk information, and the parameter type information, a joint constraint determination is performed on the behavior request object and the environmental boundary snapshot to obtain the constraint determination result. Based on the constraint determination result, the corresponding execution terminal is controlled to perform the corresponding action.

[0005] In some embodiments, obtaining the set of policy rules that match the action type information and the resource type information includes: Based on the action type information, select a first strategy rule that covers the action type information; Based on the resource type information, select a second strategy rule from the first strategy rule whose applicable scope covers the resource type information; The second strategy rules are sorted according to their priority, and the strategy rule set is generated based on the sorting result.

[0006] In some embodiments, obtaining environmental boundary parameters related to the execution terminal information and the target resource information includes: Based on the execution terminal information, determine the device address and collectable parameter types of the execution terminal; Based on the target resource information, determine the resource location information and associated sensors of the target resource; Based on the device address, the type of collectable parameter, the resource location information, and the associated sensor, obtain environmental boundary parameters related to the execution terminal and the target resource; The environmental boundary parameters are normalized and timestamps are added.

[0007] In some embodiments, the step of jointly constraining the behavior request object and the environmental boundary snapshot based on the policy rule set, the action risk information, and the parameter type information includes: Determine whether the behavior request object matches the rejection condition in the policy rule set; If the condition is met, a constraint rejection result is generated. If the target is not hit, weights are configured based on the action risk information and the parameter type information to obtain multiple risk dimension weights. Based on the risk dimension weights, multi-dimensional boundary determination is performed on the environmental boundary snapshot to obtain the constraint determination result.

[0008] In some embodiments, the step of performing multi-dimensional boundary determination on the environmental boundary snapshot based on the determination dimension weight includes: The geographic boundary parameters, equipment boundary parameters, network boundary parameters, time boundary parameters, and risk boundary parameters in the environmental boundary snapshot are normalized and mapped to obtain geographic risk, equipment risk, network risk, time window risk, and on-site hazard signal risk. Based on the audit level of the strategy rules in the aforementioned strategy rule set, determine the strategy risk; Based on the weights of the judgment dimensions, the risk scores are calculated by weighted summation of the strategy risk, geographical risk, equipment risk, network risk, time window risk, and on-site danger signal risk. The constraint determination result is obtained based on the comprehensive risk score.

[0009] In some embodiments, controlling the corresponding execution terminal to perform the corresponding action based on the constraint determination result includes: When the constraint determination result is to allow passage, the execution terminal corresponding to the execution terminal information is controlled to execute the action corresponding to the action type information; When the constraint determination result is substitution, a substitution action is generated based on the boundary failure reason and the action type information, and the execution terminal corresponding to the substitution action is controlled to execute the substitution action; If the constraint determination result is a rejection, the execution of the agent behavior request is terminated.

[0010] In some embodiments, generating an alternative action based on the boundary failure reason and the action type information, and controlling the execution terminal corresponding to the alternative action to execute the alternative action, includes: Based on the action type information, determine the task objective; Based on the reasons for the boundary failure, determine the boundary conditions that are not met; Based on the task objective and the unmet boundary conditions, select the alternative action corresponding to the action type from the preset alternative action mapping relationship; Based on the execution terminal and alternative resources corresponding to the alternative action, obtain the environmental boundary parameters corresponding to the alternative action and generate an environmental boundary snapshot of the alternative action; Obtain the set of policy rules that match the alternative action; Based on the set of policy rules that match the alternative action, a joint constraint determination is performed on the alternative action and the snapshot of the alternative action's environment boundary to obtain a secondary determination result; When the secondary determination result is to allow passage, the execution terminal corresponding to the alternative action is controlled to execute the alternative action.

[0011] In some embodiments, the behavior request object further includes information on the expected effect of the action, and the agent behavior control method further includes: Obtain the execution feedback information after the execution terminal performs the corresponding action; The execution feedback information is compared with the expected action effect information to obtain verification information; The behavior request object, the policy rules triggered during the joint constraint determination process, the environment boundary snapshot, the constraint determination result, the execution feedback information, and the verification information are associated and stored to generate a behavior control record.

[0012] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described intelligent agent behavior control method.

[0013] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described intelligent agent behavior control method.

[0014] The beneficial effects of this application are as follows: Upon receiving a behavior request from an intelligent agent, a behavior request object is generated, containing action type information, action risk information, resource type information, parameter type information, execution terminal information, and target resource information, providing a comprehensive behavior request context. Then, a set of policy rules matching the action type and resource type information is obtained, along with environmental boundary parameters related to the execution terminal and target resource information. An environmental boundary snapshot is generated based on these parameters, dynamically perceiving the real-time physical environment. Subsequently, based on the policy rule set, action risk information, and parameter type information, a joint constraint judgment is performed on the behavior request object and the environmental boundary snapshot to obtain the constraint judgment result. Based on the constraint judgment result, the corresponding execution terminal is controlled to perform the corresponding action, realizing a multi-dimensional and dynamic decision-making mechanism. Therefore, after obtaining a comprehensive behavior request context from the intelligent agent's behavior request, the policy rules and dynamic environmental boundary parameters are integrated through a joint constraint judgment mechanism to achieve multi-dimensional real-time evaluation of the intelligent agent's behavior. This enables dynamic adjustment of behavior control judgments based on real-time environmental boundary parameters, improving the accuracy of intelligent agent behavior control in complex physical environments. Attached Figure Description

[0015] Figure 1 This is a flowchart of the intelligent agent behavior control method provided in the embodiments of this application.

[0016] Figure 2 This is a flowchart of a method for obtaining a set of policy rules that match action type information and resource type information, as provided in an embodiment of this application.

[0017] Figure 3 This is a flowchart of a method for obtaining and executing environmental boundary parameters related to terminal information and target resource information, provided in an embodiment of this application.

[0018] Figure 4 This is a flowchart of a method for jointly determining the constraints of a behavior request object and an environmental boundary snapshot, provided in an embodiment of this application.

[0019] Figure 5 This is a flowchart of a method for controlling a corresponding execution terminal to perform corresponding actions, as provided in an embodiment of this application.

[0020] Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and drawings are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit the application. Furthermore, the information, data, and signals involved in the embodiments of this application are all authorized by the relevant parties or have been fully authorized by the parties, and the collection, use, and processing of related data comply with relevant laws, regulations, and standards.

[0024] Existing intelligent agent behavior control technologies primarily rely on static rules such as role-based access control, interface authentication mechanisms, action whitelists, call frequency control strategies, or manual approval processes for constraint. When applied to physically-dependent scenarios such as robots or IoT devices, these solutions inherently suffer from an inability to dynamically respond to real-time impacts from geographical differences, device status fluctuations, network boundary changes, and on-site risk conditions. Specifically, for the same control command, such as "open the valve" or "start the robot," its safety attributes differ fundamentally across different environmental contexts. Static policy rules lack the ability to integrate multi-dimensional environmental parameters in real time, leading to decreased accuracy in behavior control decisions and consequently affecting the overall security and reliability of the system.

[0025] If the aforementioned problems are not addressed, the behavioral control of intelligent agents will remain constrained by a lack of adaptability to dynamic environmental changes. In critical infrastructure sectors, this deficiency will lead to a disconnect between control decisions and the actual physical environment, increasing the probability of erroneous operations and triggering a chain reaction of system failures. Furthermore, the lack of joint constraint mechanisms on geographical boundaries, device boundaries, network boundaries, and risk boundaries will prevent the behavioral control process from effectively avoiding potential security risks, ultimately weakening the system's robustness and reliability in complex physical environments.

[0026] Based on this, embodiments of this application provide an intelligent agent behavior control method, an electronic device, and a storage medium. By integrating policy rules and dynamic environmental boundary parameters through a joint constraint judgment mechanism, it achieves multi-dimensional real-time evaluation of intelligent agent behavior. It can dynamically adjust behavior control judgments according to real-time environmental boundary parameters, thereby improving the accuracy of intelligent agent behavior control in complex physical environments.

[0027] See Figure 1 In one embodiment, an agent behavior control method is provided, which includes, but is not limited to, steps S101 to S105.

[0028] Step S101: In response to the received intelligent agent behavior request, generate a corresponding behavior request object.

[0029] The behavior request object includes action type information, action risk information, resource type information, parameter type information, execution terminal information, and target resource information.

[0030] A behavior request object refers to a data entity that has been structured and encapsulated from a received intelligent agent behavior request. Specifically, action risk information refers to potential risk assessment data associated with a specific action type; resource type information refers to data in the behavior request object describing the type of resource the action affects, such as valves, robots, cameras, and remote device interfaces; parameter type information refers to data in the behavior request object describing additional parameters required for action execution, such as the degree of valve opening, robot speed, and camera data acquisition frequency; execution terminal information refers to data in the behavior request object identifying the physical or logical device that will perform the action; and target resource information refers to data in the behavior request object identifying the specific resource instance affected by the action.

[0031] Upon receiving a behavior request from an agent, it is parsed, and a structured behavior request object is generated. This object is designed to contain a series of key information to comprehensively describe the intent and context of the request. Specifically, the object may contain action type information, indicating the operation the agent expects to perform, such as "open," "close," or "read." It also contains action risk information to preliminarily assess the potential risk level of the action. Furthermore, resource type information identifies the resource category affected by the action, such as "valve," "sensor," or "robot." Parameter type information carries the specific parameters required for the action, such as "opening angle" or "reading frequency." Execution terminal information identifies the device that will perform the action, while target resource information specifies the concrete resource instance affected by the action. For example, an agent requesting to "open the valve" can generate a behavior request object where the action type is "open," the resource type is "valve," the execution terminal is "robot arm," the target resource is "valve 100," and it may also include pre-defined action risk and parameter type information. The generation of this behavior request object provides a unified data structure for subsequent policy matching and constraint determination.

[0032] Step S102: Obtain a set of policy rules that match the action type information and resource type information.

[0033] A policy rule set refers to a predefined set of rules used to guide the decision-making process of an agent's behavior. These rules can specify the conditions under which an action is permitted, denied, or substituted, based on factors such as action type, resource type, and environmental conditions.

[0034] After generating the behavior request object, it is necessary to determine the applicable policy rules for that request. To do this, based on the action type and resource type information contained in the behavior request object, the corresponding set of policy rules is retrieved from a pre-defined policy library. For example, if the action type information is "open" and the resource type information is "valve," all policy rules related to "open valve" are searched. These policy rules can be manually configured or filtered using simple keyword matching. This approach initially narrows down the range of policies to be considered, ensuring that subsequent decision-making processes are based on relevant rules. For example, for a "open valve" request, a series of rules regarding "valve operation permissions" and "valve opening conditions" might be obtained.

[0035] Step S103: Obtain environmental boundary parameters related to the execution terminal information and target resource information, and generate an environmental boundary snapshot based on the environmental boundary parameters.

[0036] Environmental boundary parameters refer to real-time or near-real-time environmental data related to the execution terminal and target resources. These parameters reflect the external conditions during action execution, such as geographical location, equipment status, network conditions, time window, and on-site risks.

[0037] An environmental boundary snapshot is a data set that is collected and archived at a specific point in time regarding environmental boundary parameters. This snapshot provides a momentary view of the environmental state at the time an action is performed.

[0038] To perform dynamic, environment-based determinations, real-time data related to the action execution environment needs to be acquired. Based on the execution terminal information and target resource information in the behavior request object, the data types and sources of the environmental data to be collected are identified. For example, if the execution terminal is a "robot arm" and the target resource is "valve 100," it might be configured to acquire data from the robot arm's sensors, valve 100's sensors, and surrounding environment sensors. These environmental boundary parameters can include geographical location, equipment status, network connectivity, time information, and on-site risk indicators. After acquiring these raw environmental boundary parameters, they are processed, such as through data format conversion or simple aggregation, to generate an environmental boundary snapshot. This environmental boundary snapshot is a fixed representation of the environmental state at a specific point in time, providing a stable environmental context for subsequent joint constraint determinations.

[0039] Step S104: Based on the set of policy rules, action risk information, and parameter type information, perform joint constraint determination on the behavior request object and the environmental boundary snapshot to obtain the constraint determination result.

[0040] Joint constraint determination refers to the process of comprehensively considering information such as the requesting object, environmental boundary snapshot, and policy rule set to conduct a multi-dimensional and dynamic security and compliance assessment of the agent's behavior request.

[0041] The constraint determination result refers to the output of the joint constraint determination process, indicating whether the agent's behavior request is allowed to be executed, rejected, or requires an alternative action.

[0042] After acquiring the policy rule set, the action request object, and the environmental boundary snapshot, the core joint constraint determination process is executed. This process comprehensively utilizes the action risk information and parameter type information from the policy rule set and the action request object, as well as the real-time environmental state reflected in the environmental boundary snapshot. The determination process may include a step-by-step comparison of the policy rules, for example, checking whether the action request violates any policies that would prohibit execution. Simultaneously, based on the action risk information and parameter type information, various parameters in the environmental boundary snapshot are weighted and evaluated to determine whether the current environment meets the safety and compliance requirements for action execution. For example, if the action risk information indicates that the action is high-risk, certain parameters in the environmental boundary snapshot (such as on-site personnel density) may be given higher weight. Through this multi-dimensional comprehensive evaluation, a clear constraint determination result can be obtained, indicating whether the agent's action request is allowed to execute.

[0043] Step S105: Based on the constraint determination result, control the corresponding execution terminal to perform the corresponding action.

[0044] An execution terminal refers to a physical device or software entity that actually performs the action specified in the intelligent agent's behavior request. Examples include robots, smart valves, sensors, and remote servers.

[0045] After the joint constraint determination is completed, the agent's behavior is guided by the constraint determination result. If the constraint determination result indicates that the action can be permitted, an instruction is sent to the execution terminal specified in the action request object to control it to execute the corresponding action. For example, if the determination result is "permitted," an instruction to "open valve 100" will be sent to the "robotic arm." If the determination result is "rejected," the agent's action request will be terminated, and no action will be taken. If the determination result is of another type (e.g., "alternative"), further processes may be triggered to generate and execute alternative actions. In this way, the agent's behavior is dynamically controlled based on the real-time environment and policy rules, ensuring the safety and compliance of the operation.

[0046] The intelligent agent behavior control method provided in this application, upon receiving an intelligent agent behavior request, generates a behavior request object containing action type information, action risk information, resource type information, parameter type information, execution terminal information, and target resource information, providing a comprehensive behavior request context. Then, it acquires a set of policy rules matching the action type information and resource type information, as well as environmental boundary parameters related to the execution terminal information and target resource information. Based on these environmental boundary parameters, it generates an environmental boundary snapshot, dynamically perceiving the real-time physical environment. Subsequently, based on the policy rule set, action risk information, and parameter type information, it performs joint constraint judgment on the behavior request object and the environmental boundary snapshot, obtaining a constraint judgment result. Based on the constraint judgment result, it controls the corresponding execution terminal to perform the corresponding action, realizing a multi-dimensional and dynamic decision-making mechanism. Therefore, after acquiring a comprehensive behavior request context from the intelligent agent behavior request, it integrates policy rules and dynamic environmental boundary parameters through a joint constraint judgment mechanism, achieving multi-dimensional real-time evaluation of intelligent agent behavior. This enables dynamic adjustment of behavior control judgments based on real-time environmental boundary parameters, improving the accuracy of intelligent agent behavior control in complex physical environments.

[0047] See Figure 2 In one embodiment, the method for obtaining a set of policy rules that match the action type information and resource type information includes, but is not limited to, steps S201 to S203.

[0048] Step S201: Select the first strategy rule that covers the action type information based on the action type information.

[0049] Step S202: Based on the resource type information, select a second strategy rule from the first strategy rule whose applicable scope covers the resource type information.

[0050] Step S203: Sort the second strategy rules according to their priority, and generate a set of strategy rules based on the sorting results.

[0051] Based on action type information, a first policy rule covering that action type is selected. This allows the agent to initially filter rules related to the action type from all available policy rules based on the action type information contained in the agent's action request. For example, a policy rule base can be maintained, where each policy rule explicitly identifies the action type it applies to. When specific action type information is received, all policy rules covering that action type are identified and extracted through a query or indexing mechanism. Alternatively, policy rules can contain conditional expressions that, during evaluation, check whether the action type information meets their requirements, thus selecting a first policy rule that meets the conditions.

[0052] Based on resource type information, a second strategy rule whose scope of application covers the resource type information is selected from the first strategy rule. Building upon the initial selection of the first strategy rules, they are further refined using the resource type information in the request object. For example, it is checked whether the first strategy rule contains limiting conditions for a specific resource type. If the scope of application of a strategy rule clearly covers the resource type in the request, it is selected as the second strategy rule. Alternatively, a strategy rule can define a set of resource types; if the requested resource type is a subset or member of this set, the strategy rule is selected.

[0053] The second policy rules are sorted according to their priority, and a policy rule set is generated based on the sorting results. This ensures that rules are evaluated according to predetermined importance or execution order in subsequent joint constraint determination. For example, each policy rule can be assigned a preset priority value, and the second policy rules can be sorted in descending or ascending order based on these values. Alternatively, the priority of the policy rules can be dynamically calculated based on specific attributes (such as rule creation time, rule granularity, rule enforcement level, etc.), and sorted accordingly. After sorting, the final set of policy rules used for joint constraint determination is obtained.

[0054] This application's solution addresses the efficiency and accuracy issues in obtaining a set of matching policy rules through a phased, selective, and ranking mechanism. First, an initial screening based on action type information quickly eliminates a large number of policy rules irrelevant to the current action, forming a preliminary subset of relevant rules—the first policy rules. Building upon the first policy rules, a second screening is performed using resource type information. This ensures that the selected second policy rules are not only relevant to the action type but also highly matched to the resource type of the specific operation, thus guaranteeing the contextual relevance and accuracy of the policy rules. Finally, these highly relevant second policy rules are prioritized, allowing rules to be evaluated according to predetermined importance or conflict resolution principles in subsequent joint constraint determination, avoiding the risk of rule conflicts or improper execution. Through this progressive screening and ranking, the final set of policy rules is both comprehensive and accurate, providing a solid foundation for joint constraint determination of agent behavior and significantly improving the intelligence and reliability of behavior control.

[0055] See Figure 3 In one embodiment, the method for obtaining environmental boundary parameters related to execution terminal information and target resource information includes, but is not limited to, steps S301 to S304.

[0056] Step S301: Determine the device address and collectable parameter types of the execution terminal based on the execution terminal information.

[0057] Step S302: Based on the target resource information, determine the resource location information and associated sensors of the target resource.

[0058] Step S303: Based on the device address, the type of collectable parameters, resource location information, and associated sensors, obtain the environmental boundary parameters related to the execution terminal and the target resource.

[0059] Step S304: Normalize the environmental boundary parameters and add a timestamp.

[0060] Determining the device address and collectable parameter types of the execution terminal based on the execution terminal information can be achieved by maintaining an execution terminal registry, which contains the unique device address (e.g., IP address, MAC address, or device ID) of each execution terminal and its supported sensor types or a list of reportable parameters. When execution terminal information is received, this registry is queried to retrieve the corresponding data. Alternatively, the execution terminal reports its device address and collectable parameter types to the control system upon initial access or periodically. The control system receives and stores this information and retrieves it when needed.

[0061] Based on the target resource information, the resource location information and associated sensors of the target resource can be determined. This can be achieved by establishing a resource management database, which records the unique identifier of each target resource, its physical location coordinates (e.g., latitude and longitude or region ID), and a list of sensors deployed nearby or functionally related to it (e.g., temperature sensors, humidity sensors, or smoke sensors). The database can then be queried using the target resource information. Alternatively, the target resource itself may possess self-describing capabilities, providing its own resource location information and a list of associated sensors through its information interface. The control system obtains this information by communicating with the target resource.

[0062] Based on the device address, the type of collectable parameters, resource location information, and associated sensors, environmental boundary parameters related to the execution terminal and the target resource are acquired. This can be achieved by the control system establishing a communication connection with the execution terminal via the device address (e.g., using protocols such as MQTT, HTTP, or Modbus) and requesting the terminal to report real-time data based on the type of collectable parameters. Simultaneously, environmental data surrounding the target resource is obtained from sensor networks, environmental monitoring platforms, or Geographic Information Systems (GIS) based on resource location information and associated sensors. Alternatively, a unified data acquisition platform can exist, integrating data from the execution terminal, various sensors, and external environmental data sources. Query requests can be sent to this platform, specifying the device address, parameter type, resource location, and associated sensors; the platform will then aggregate and return the corresponding environmental boundary parameters.

[0063] Normalizing environmental boundary parameters and adding timestamps can be achieved by pre-setting corresponding normalization algorithms (e.g., linear normalization or Z-score normalization) for different types of environmental parameters (e.g., temperature, humidity, pressure, or distance). After acquiring the parameters, the corresponding algorithm is applied to transform them to a uniform numerical range (e.g., 0-1 or -1 to 1) according to their type. Simultaneously, the current UTC time or local time is recorded as a timestamp upon completion of data acquisition or processing. Alternatively, a data conversion service can be used, which can identify the original format and unit of the input environmental parameters and convert them to a standard format and uniform unit according to predefined rules. Timestamp addition can be automatically completed by the data acquisition module before data is entered into the database, or uniformly added by the data processing module after normalization.

[0064] The proposed solution, when acquiring environmental boundary parameters, firstly, by parsing the execution terminal information, identifies the unique identifier (device address) and data acquisition capabilities (collectible parameter types) of the execution terminal, enabling precise location of the specific execution terminal and understanding of the environmental data it can provide. Simultaneously, by parsing the target resource information, it identifies the geographical location of the target resource (resource location information) and its associated sensing devices (associated sensors), ensuring comprehensive acquisition of various data related to the environment in which the target resource is located. Based on this, using the determined device address, collectible parameter types, resource location information, and associated sensors, it actively or passively acquires environmental boundary parameters related to the current action request from the execution terminal itself, sensor networks, geographic information systems, or other data sources. To ensure that these heterogeneous, multi-source environmental boundary parameters can be effectively utilized by the subsequent joint constraint determination module, this solution further normalizes the acquired environmental boundary parameters to eliminate differences in units and numerical ranges between different parameters. Furthermore, a timestamp is added to each normalized parameter, not only recording the timeliness of the data but also providing a basis for subsequent possible data tracing, anomaly analysis, or time-series determination.

[0065] See Figure 4 In one embodiment, the method for jointly determining the constraints of the behavior request object and the environmental boundary snapshot includes, but is not limited to, steps S401 to S403.

[0066] Step S401: Determine whether the behavior request object meets the rejection condition in the policy rule set.

[0067] If a match is found, proceed to step S402; if a match is not found, proceed to step S403.

[0068] Step S402: Generate the constraint decision result for characterization rejection.

[0069] Step S403: Configure weights based on action risk information and parameter type information to obtain multiple risk dimension weights, and perform multi-dimensional boundary determination on the environmental boundary snapshot based on the risk dimension weights to obtain constraint determination results.

[0070] The rejection conditions are pre-defined rules with the highest priority. Once met, the action request will be immediately rejected without further analysis. Determining whether an action request matches a rejection condition in the policy rule set can be achieved through a pre-defined blacklist mechanism. For example, if the action type information or target resource information matches an entry in the blacklist, it is considered a rejection condition. Alternatively, a series of strict policy rules can be defined, explicitly specifying actions that are not allowed under specific conditions. Upon receiving an action request, these rules are checked first.

[0071] If a match is found, a constraint rejection result is generated to prevent non-compliant or high-risk behaviors. This ensures an immediate response to critical security and compliance requirements. This result can be achieved by returning a predefined rejection code or status identifier, which explicitly indicates that the behavior request has been rejected and may include the reason for the rejection. Alternatively, it can be achieved by triggering a rejection event or message to notify the relevant execution endpoint or management system that the request has been rejected and to log the rejection.

[0072] If a match is not found, weights are configured based on action risk information and parameter type information to dynamically adjust the importance of different risk dimensions according to the characteristics of the behavior request (action risk information and parameter type information), ensuring the accuracy and relevance of the assessment. For example, for high-risk actions, the safety dimension will have a higher weight. This weight configuration can be achieved through a preset weight configuration table, which looks up the corresponding set of risk dimension weights based on action risk information (such as high risk, medium risk, low risk) and parameter type information (such as involving sensitive data, involving physical operations). Alternatively, weights can be dynamically generated through a machine learning model. The model predicts the degree of influence of different risk dimensions on the final decision based on historical data and the characteristics of the current behavior request, and outputs the corresponding weights.

[0073] Multi-dimensional boundary determination can be performed on environmental boundary snapshots based on risk dimension weights. This can be achieved by defining a series of rules and thresholds for each determination dimension. Based on the specific parameters in the environmental boundary snapshot, judgments are made against these rules and thresholds, and combined with the risk dimension weights to calculate the risk score for each dimension, ultimately resulting in a comprehensive constraint determination result. Alternatively, a decision tree or expert system can be constructed, where each node represents a determination dimension. Based on the input from the environmental boundary snapshot and the risk dimension weights, the system evaluates along a decision path, eventually reaching a leaf node, which corresponds to a constraint determination result. Specifically, policy authorization determination checks whether the behavior request conforms to the preset authorization policy; geographical boundary determination checks whether the agent's current location is within the permitted geographical area; device boundary determination assesses whether the health status, battery level, and critical component status of the execution terminal meet the execution requirements; network boundary determination assesses the security and reliability of the current network environment; time boundary determination checks whether the behavior request is executed within the permitted time window; and risk boundary determination assesses the potential hazards of the on-site environment.

[0074] In some embodiments, multi-dimensional boundary determination includes at least two of policy authorization determination, geographical boundary determination, device boundary determination, network boundary determination, time boundary determination, and risk boundary determination.

[0075] In the agent behavior control method of this application, after receiving an agent behavior request and generating a behavior request object, the matching policy rule set and environmental boundary snapshot are first obtained. When performing joint constraint determination, the scheme of this application first performs preliminary screening of the behavior request object to determine whether it hits the preset rejection conditions in the policy rule set. This pre-judgment mechanism can quickly identify and prevent obviously non-compliant or high-risk behaviors, thereby avoiding unnecessary subsequent complex evaluations and improving decision-making efficiency and security. If the behavior request does not hit the rejection conditions, multiple risk dimension weights are configured based on the action risk information and parameter type information contained in the behavior request object. Subsequently, multi-dimensional boundary determination is performed on the environmental boundary snapshot based on the dynamically configured risk dimension weights. By combining the preliminary rejection condition judgment with dynamically weighted multi-dimensional boundary determination, fine-grained control of agent behavior can be achieved. This not only enables rapid response to high-risk situations but also allows for comprehensive and accurate risk assessment in complex environments, effectively improving the security, compliance, and decision-making efficiency of agent behavior.

[0076] In some embodiments, multi-dimensional boundary determination is performed on the environmental boundary snapshot based on the determination dimension weight, including: Normalize and map the geographic boundary parameters, equipment boundary parameters, network boundary parameters, time boundary parameters, and risk boundary parameters in the environmental boundary snapshot to obtain geographic risk, equipment risk, network risk, time window risk, and on-site hazard signal risk; Determine the strategy risk based on the audit level of the strategy rules in the strategy rule set; Based on the weights of the judgment dimensions, the risk scores are calculated by weighted summation of strategic risk, geographical risk, equipment risk, network risk, time window risk, and on-site danger signal risk. The constraint determination result is obtained based on the comprehensive risk score.

[0077] Normalizing and mapping geographic boundary parameters, device boundary parameters, network boundary parameters, temporal boundary parameters, and risk boundary parameters in environmental boundary snapshots allows geographic boundary parameters (such as distance) to be mapped to geographic risk values ​​between 0 and 1 using linear or nonlinear functions; the closer the distance is to the restricted area, the higher the risk value. Device boundary parameters (such as battery percentage) can be mapped to device risk; the lower the battery level, the higher the risk value. Network boundary parameters (such as packet loss rate) can be mapped to network risk; the higher the packet loss rate, the higher the risk value. Temporal boundary parameters (such as action-allowed time windows) can be mapped to time window risk; for example, the risk value may increase when the action request time is close to the edge of the allowed time window. Risk boundary parameters (such as on-site personnel density levels) can be mapped to on-site hazard signal risk; the higher the personnel density, the higher the risk value. This mapping process can be implemented using pre-defined lookup tables, piecewise functions, or machine learning models.

[0078] In some embodiments, the geographic boundary parameters include at least one of geographic area identifier, target area restriction level, and distance from the current location to the permitted execution area; the device boundary parameters include at least one of device health score, power ratio, and actuator status; the network boundary parameters include at least one of network trust level, network latency, and packet loss rate; the time boundary parameters include the action allowable time window; and the risk boundary parameters include at least one of on-site personnel density level and danger signal level.

[0079] Strategic risk can be determined based on the audit rating of strategic rules within a set of strategic rules. The audit rating reflects the significance, sensitivity, or stringency of a particular strategic rule. For example, a strategic rule involving high-value assets or high-risk operations might be assigned a "high" audit rating, thus resulting in higher strategic risk in risk assessment. Determining strategic risk can involve directly mapping audit ratings to numerical risk values, or calculating them in conjunction with other factors.

[0080] Based on the comprehensive risk score, a constraint determination result can be obtained, and a low-risk threshold, a medium-risk threshold, and a high-risk threshold can be set. If the comprehensive risk score is lower than the low-risk threshold, the determination result is "allow"; if it is between the low-risk and medium-risk thresholds, the determination result may be "substitute"; if it is higher than the high-risk threshold, the determination result is "reject". This determination method based on quantitative scores makes the decision-making process more objective and traceable.

[0081] This application's solution maps multi-dimensional environmental boundary parameters and policy rule audit levels into standardized risks, and calculates a comprehensive risk score by weighting and summing them according to preset judgment dimension weights. This comprehensive risk score can quantitatively reflect the overall risk level faced by the agent's behavioral request. Based on this comprehensive risk score, more refined and objective constraint judgments can be made, such as allowing, substituting, or rejecting. This method overcomes the limitations of qualitative judgments and the difficulty in unified evaluation that may exist in multi-dimensional judgments, providing a unified risk assessment standard and decision-making basis for agent behavior control.

[0082] See Figure 5 In one embodiment, the method for controlling the corresponding execution terminal to perform the corresponding action includes, but is not limited to, steps S501 to S503.

[0083] Step S501: When the constraint determination result is to allow passage, control the execution terminal corresponding to the execution terminal information to perform the action corresponding to the action type information.

[0084] In step S502, when the constraint determination result is substitution, a substitution action is generated based on the boundary failure reason and action type information, and the execution terminal corresponding to the substitution action is controlled to execute the substitution action.

[0085] Step S503: If the constraint determination result is rejection, terminate the execution of the agent behavior request.

[0086] When the joint constraint determination indicates that the agent's behavior request fully complies with all policy rules and environmental boundary conditions, the execution of the request will be permitted. This means that the action type information, resource type information, action risk information, parameter type information, and current environment (such as geographical location, device status, network status, time window, etc.) in the agent's behavior request are all within acceptable ranges. The execution terminal corresponding to the control execution terminal information will execute the action corresponding to the action type information. This can be achieved by sending direct instruction signals to the execution terminal, such as through API calls, message queue notifications, or remote procedure calls (RPC), triggering the execution of a preset action. Another implementation method is to encapsulate the execution instructions into an executable script or program package and send it to the execution terminal for parsing and execution.

[0087] When the joint constraint decision result is neither completely acceptable nor completely rejectable, the specific boundary reasons causing partial constraint failure are identified, and an alternative action is generated by combining this with the original action type information. This generation of alternative actions can be achieved through a pre-defined rule engine or decision tree, selecting a suitable alternative (e.g., downgrading to a low-precision operation) from a predefined library of alternative actions based on specific boundary failure reasons (e.g., high network latency preventing real-time performance) and the original action type information (e.g., high-precision operation). Another approach is to utilize a machine learning model to dynamically recommend or generate alternative actions by analyzing the correlation between boundary failure reasons and successful alternative actions in historical data.

[0088] Alternative actions are flexible and feasible alternatives provided when the original action cannot be performed under specific constraints. In some embodiments, alternative actions include degraded actions, delayed actions, read-only actions, manual transfer actions, and / or edge-local execution actions. Degraded actions refer to reducing the execution standard or functional scope of the original action when it cannot be performed at the highest standard or with full functionality; for example, high-precision inspection is downgraded to low-precision inspection. Delayed actions refer to postponing the execution time of the original action until specific conditions are met (such as improved network conditions or resource availability); for example, delaying large file transfer tasks to off-peak hours when the network is congested. Read-only actions refer to restricting operations that originally included write or modify permissions to read-only operations; for example, when in maintenance mode or with limited resources, configuration modification is not allowed, only status viewing is permitted. Manual transfer actions refer to transferring tasks that cannot be completed automatically by the intelligent agent or are too risky to a human operator; for example, transferring control to the driver when encountering complex road conditions that cannot be recognized. Edge local execution refers to transferring tasks that would normally need to be executed in the cloud or on a central server to edge devices closer to the execution terminal, in order to reduce latency or dependence on central resources. For example, real-time data analysis tasks can be transferred from the cloud to a local gateway device for initial processing.

[0089] When the joint constraint determination result indicates that an agent's behavior request seriously violates policy rules or environmental boundary conditions, and cannot be remedied by alternative actions, the execution of the request will be explicitly rejected. Termination of an agent's behavior request can be achieved by returning a rejection status code or error message to the request initiator, explicitly informing them that the request has been rejected. Alternatively, the behavior request can be discarded directly without any further processing, and a rejection log can be logged for auditing purposes.

[0090] The solution in this application, through the aforementioned agent behavior control method, after completing the joint constraint determination of the behavior request object and the environmental boundary snapshot, further refines the control of the execution terminal's actions based on the type of determination result. When the determination result is "allow," it indicates that the agent behavior request fully conforms to the preset policy rule set and the current environmental boundary snapshot. In this case, the execution terminal will be directly instructed to execute the action corresponding to the original action type information, ensuring the smooth progress of the task. However, when the determination result is not fully conforming, but not completely rejecting, the solution in this application demonstrates its flexibility. Based on the specific boundary reasons that caused the constraint failure, and combined with the original action type information, an alternative action is generated. When the determination result is "reject," it indicates that the agent behavior request has a serious non-compliance with policy rules or environmental boundary conditions, and cannot be compensated for by an alternative action. In this case, the execution of the agent behavior request will be terminated to ensure the security and stability of the system. Through this hierarchical response mechanism, the solution in this application not only ensures the compliance and security of agent behavior, but also provides more flexible and intelligent decision-making in the face of complex and ever-changing environments and constraints, effectively improving the adaptability and task execution efficiency of the agent system.

[0091] In some embodiments, an alternative action is generated based on the boundary failure cause and action type information, and the execution terminal corresponding to the alternative action is controlled to execute the alternative action, including: Determine the task objective based on the action type information; Based on the reasons for boundary failure, determine the boundary conditions that are not met; Based on the task objective and the unmet boundary conditions, select the alternative action corresponding to the action type from the preset alternative action mapping relationship; Based on the execution terminal and alternative resources corresponding to the alternative action, obtain the environmental boundary parameters corresponding to the alternative action and generate an environmental boundary snapshot of the alternative action; Obtain the set of policy rules that match the alternative actions; Based on the set of policy rules that match the alternative action, a joint constraint determination is performed on the alternative action and the snapshot of the boundary of the alternative action environment to obtain a secondary determination result; When the secondary judgment result is to allow passage, the execution terminal corresponding to the alternative action is controlled to execute the alternative action.

[0092] A task objective refers to the ultimate goal or effect that an intelligent agent expects to achieve by requesting a behavior. For example, for the action type information of "spraying pesticides," the task objective might be "to complete pesticide coverage of a designated area." The task objective can be determined by parsing the semantic content of the action type information or by querying a predefined action-objective mapping table. Alternatively, the task objective can be further clarified by combining user input or system context information.

[0093] Unmet boundary conditions are technical descriptions of the reasons for boundary failures, such as "current location outside the allowed area," "device health score below the threshold," or "network packet loss rate above the upper limit." Unmet boundary conditions can be identified by analyzing the failure dimensions and specific parameter values ​​indicated in the joint constraint determination results. For example, if the geographical boundary determination fails, the unmet boundary condition might be "current wind speed exceeds the safety threshold."

[0094] An alternative action mapping is a data structure that stores the association between an original action type and one or more alternative actions. This mapping can be a lookup table, a rule engine, or a machine learning-based model. When the original action fails due to specific boundary conditions, one or more of the most suitable alternative actions can be intelligently selected from this mapping based on the original action type, task objective, and the unmet boundary conditions. For example, if "high-altitude spraying" fails due to excessive wind speed, the mapping might suggest "low-altitude spraying" or "delayed spraying."

[0095] After determining the alternative action, its execution environment needs to be reassessed. This includes identifying the execution terminal required to perform the alternative action (which may be the same terminal as the original action or a different terminal) and alternative resources (e.g., different sensors, different network paths, or different target areas). Subsequently, based on this information, environmental boundary parameters related to the alternative action are re-acquired or calculated. For example, if the alternative action is "low-altitude spraying," information such as wind speed and obstacles in the low-altitude environment needs to be obtained. These parameters are integrated to form a snapshot of the alternative action's environmental boundary that reflects the execution environment of the alternative action.

[0096] Similar to the initial action request, alternative actions also need to comply with corresponding policy rules. Based on the action type information and the resource types it may involve, all policy rules related to the alternative action are filtered from the policy rule base. Then, the alternative action (including its action type information, action risk information, parameter type information, etc.) is combined with a newly generated snapshot of the alternative action's environment boundary, and the set of policy rules obtained for the alternative action, and a joint constraint judgment is performed again. This judgment process is similar to that of the initial action request, considering multiple dimensions such as policy authorization judgment, geographical boundary judgment, device boundary judgment, network boundary judgment, time boundary judgment, and risk boundary judgment. Finally, a secondary judgment result is obtained, indicating whether the alternative action can be allowed, whether further substitution is needed, or whether it should be rejected. Only when the alternative action passes the secondary joint constraint judgment, and the secondary judgment result explicitly indicates "allowed," will an instruction be issued to the execution terminal corresponding to the alternative action to execute the alternative action.

[0097] The proposed solution, after the agent behavior control method performs joint constraint determination on the initial behavior request, indicates that if the determination result is an alternative, it means that the original request has unmet boundary conditions in the current environment, but is not completely unexecutable; rather, there are adjustable alternative solutions. To ensure the effectiveness and security of the alternative solutions, the proposed solution does not directly execute the alternative action, but introduces a mechanism for secondary verification of the alternative action. First, based on the original action type information, its task objective is clarified, and combined with the specific boundary failure reasons that caused the original request to fail, the unmet boundary conditions are accurately identified. Based on this information, one or more alternative actions that match the original action type and are expected to resolve the unmet boundary conditions can be intelligently selected from a pre-set alternative action mapping relationship. After determining the alternative action, the execution environment context of the alternative action is reconstructed. This includes identifying the execution terminals and alternative resources that the alternative action may involve, and obtaining the environmental boundary parameters related to the alternative action, thereby generating a new snapshot of the alternative action's environmental boundary. Simultaneously, based on the characteristics of the alternative action, a set of matching policy rules is obtained. Subsequently, this solution performs a comprehensive joint constraint assessment again on the alternative action itself, the newly generated snapshot of the alternative action's environment boundary, and the matching set of policy rules. This secondary assessment process maintains the same logic as the assessment of the initial behavior request, ensuring that the alternative action meets requirements across all key dimensions, including policy authorization, geography, device, network, time, and risk. Only when the secondary assessment result is clearly "allowed" will the corresponding execution terminal be controlled to execute the alternative action.

[0098] In some embodiments, the behavior request object further includes information on the expected effect of the action, and the method further includes: Obtain execution feedback information after the execution terminal performs the corresponding action; The execution feedback information is compared with the expected effect information of the action to obtain verification information; The behavior request object, the policy rules triggered during the joint constraint determination process, the environment boundary snapshot, the constraint determination result, the execution feedback information and the verification information are associated and stored to generate a behavior control record.

[0099] Action expectation effect information refers to the state or result that is pre-defined in the agent's behavior request and is expected to be achieved after the action is executed. This information can be a description of the target resource's state, such as "valve opened to 50%", or a marker of a specific task's completion, such as "data transmission completed". Its purpose is to provide a benchmark for verifying the results of subsequent action execution. This information can be represented in the form of structured data (such as JSON, XML), including expected values, allowable error ranges, success conditions, etc., or it can be represented by predefined enumerated values ​​to indicate specific expected results.

[0100] Execution feedback information refers to the actual data received by the execution terminal after performing a corresponding action, regarding the status and result of the action execution. This information is an objective reflection of the action execution and is used to evaluate whether the action was completed as expected.

[0101] In some embodiments, the execution feedback information includes at least one of execution receipts, device status changes, sensor feedback, or API return values. An execution receipt is a confirmation message returned by the execution terminal, indicating that the action command has been received and processed; a device status change is, for example, the actual change in device parameters (such as temperature, pressure, or position) monitored by sensors; sensor feedback is environmental or device data directly collected by sensors associated with the target resource; and an API return value is the action execution result or status code obtained by calling the application programming interface (API) provided by the execution terminal.

[0102] This application's solution, based on the intelligent agent behavior control method, further improves the monitoring, verification, and recording mechanism for intelligent agent behavior. Specifically, upon receiving an intelligent agent behavior request and generating a behavior request object, this object includes not only action type information, action risk information, resource type information, parameter type information, execution terminal information, and target resource information, but also additional action expectation effect information. After jointly constraining the behavior request object and the environmental boundary snapshot according to the policy rule set, action risk information, and parameter type information, and controlling the corresponding execution terminal to execute the corresponding action based on the constraint determination result, this solution proactively acquires the execution feedback information after the execution terminal executes the corresponding action. This feedback information is an objective reflection of the actual execution of the action and may include at least one of the following: execution receipt, device state change, sensor feedback, or API return value. Subsequently, the acquired feedback information is compared with the preset action expectation effect information in the behavior request object to obtain verification information, thereby determining whether the action was successfully executed as expected. Ultimately, to ensure the traceability and auditability of the entire agent behavior control process, the behavior request object, the policy rules triggered during the joint constraint determination process, the environmental boundary snapshot, the constraint determination result, the execution feedback information, and the verification information are stored together to generate a complete behavior control record. In this way, this solution not only achieves effective control of agent behavior but also provides the ability to verify the action execution results and record the entire process, forming a closed-loop agent behavior management system.

[0103] This application also provides an electronic device. Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application. For example... Figure 6As shown, the electronic device in this embodiment mainly includes a processor 601 and a memory 602. The memory 602 can be configured to store a program for executing the intelligent agent behavior control method of the above-described method embodiments. The processor 601 can be configured to execute the program in the memory 602, which includes, but is not limited to, a program for executing the intelligent agent behavior control method of the above-described method embodiments. For ease of explanation, only the parts related to the embodiments of this application are shown. For specific technical details not disclosed, please refer to the method section of the embodiments of this application.

[0104] In some embodiments, the electronic device may include multiple processors 601 and multiple memories 602. The program executing the agent behavior control method of the above-described method embodiments can be divided into multiple subroutines, each of which can be loaded and run by a processor 601 to perform different steps of the agent behavior control method of the above-described method embodiments. Specifically, each subroutine can be stored in a different memory 602, and each processor 601 can be configured to execute programs in one or more memories 602 to jointly implement the agent behavior control method of the above-described method embodiments; that is, each processor 601 executes different steps of the agent behavior control method of the above-described method embodiments to jointly implement the agent behavior control method of the above-described method embodiments.

[0105] The aforementioned multiple processors 601 can be processors deployed on the same device. For example, the aforementioned electronic device can be a high-performance device composed of multiple processors, and the aforementioned multiple processors 601 can be processors configured on that high-performance device. Alternatively, the aforementioned multiple processors 601 can also be processors deployed on different devices. For example, the aforementioned electronic device can be a server cluster, and the aforementioned multiple processors 601 can be processors on different servers within the server cluster.

[0106] This application also provides a computer-readable storage medium. In one embodiment of the computer-readable storage medium according to this application, the computer-readable storage medium can be configured to store a program that performs the intelligent agent behavior control method of the above-described method embodiments. This program can be loaded and run by a processor to implement the above-described intelligent agent behavior control method. For ease of explanation, only the parts related to the embodiments of this application are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of this application. The computer-readable storage medium can be a memory formed by various electronic devices. Optionally, in the embodiments of this application, the computer-readable storage medium is a non-transitory computer-readable storage medium.

[0107] The intelligent agent behavior control method, electronic device, and storage medium provided in this application, upon receiving an intelligent agent behavior request, generate a behavior request object containing action type information, action risk information, resource type information, parameter type information, execution terminal information, and target resource information, providing a comprehensive behavior request context. Then, it acquires a set of policy rules matching the action type information and resource type information, as well as environmental boundary parameters related to the execution terminal information and target resource information. Based on these environmental boundary parameters, it generates an environmental boundary snapshot, dynamically perceiving the real-time physical environment. Subsequently, based on the policy rule set, action risk information, and parameter type information, it performs joint constraint judgment on the behavior request object and the environmental boundary snapshot, obtaining a constraint judgment result. Based on the constraint judgment result, it controls the corresponding execution terminal to perform the corresponding action, realizing a multi-dimensional and dynamic decision-making mechanism. Therefore, after acquiring a comprehensive behavior request context from the intelligent agent behavior request, it integrates policy rules and dynamic environmental boundary parameters through a joint constraint judgment mechanism, achieving multi-dimensional real-time evaluation of intelligent agent behavior. It can dynamically adjust behavior control judgments based on real-time environmental boundary parameters, improving the accuracy of intelligent agent behavior control in complex physical environments.

[0108] Exemplary embodiments of this disclosure have been specifically shown and described above. It should be understood that this disclosure is not limited to the detailed structures, arrangements, or implementations described herein; rather, this disclosure is intended to cover various modifications and equivalent arrangements contained within the spirit and scope of the appended claims.

Claims

1. A method for controlling the behavior of an intelligent agent, characterized in that, include: In response to the received action request from the agent, a corresponding action request object is generated; The behavior request object includes action type information, action risk information, resource type information, parameter type information, execution terminal information, and target resource information; Obtain a set of policy rules that match the action type information and the resource type information; Obtain environmental boundary parameters related to the execution terminal information and the target resource information, and generate an environmental boundary snapshot based on the environmental boundary parameters; Based on the set of policy rules, the action risk information, and the parameter type information, a joint constraint determination is performed on the behavior request object and the environmental boundary snapshot to obtain the constraint determination result. Based on the constraint determination result, the corresponding execution terminal is controlled to perform the corresponding action.

2. The intelligent agent behavior control method according to claim 1, characterized in that, The process of obtaining the set of strategy rules that match the action type information and the resource type information includes: Based on the action type information, select a first strategy rule that covers the action type information; Based on the resource type information, select a second strategy rule from the first strategy rule whose applicable scope covers the resource type information; The second strategy rules are sorted according to their priority, and the strategy rule set is generated based on the sorting result.

3. The intelligent agent behavior control method according to claim 1, characterized in that, The acquisition of environmental boundary parameters related to the execution terminal information and the target resource information includes: Based on the execution terminal information, determine the device address and collectable parameter types of the execution terminal; Based on the target resource information, determine the resource location information and associated sensors of the target resource; Based on the device address, the type of collectable parameter, the resource location information, and the associated sensor, obtain environmental boundary parameters related to the execution terminal and the target resource; The environmental boundary parameters are normalized and timestamps are added.

4. The intelligent agent behavior control method according to claim 1, characterized in that, The step of jointly constraining the behavior request object and the environmental boundary snapshot based on the policy rule set, the action risk information, and the parameter type information includes: Determine whether the behavior request object matches the rejection condition in the policy rule set; If the condition is met, a constraint rejection result is generated. If the target is not hit, weights are configured based on the action risk information and the parameter type information to obtain multiple risk dimension weights. Based on the risk dimension weights, multi-dimensional boundary determination is performed on the environmental boundary snapshot to obtain the constraint determination result.

5. The intelligent agent behavior control method according to claim 4, characterized in that, The step of performing multi-dimensional boundary determination on the environmental boundary snapshot based on the determination dimension weight includes: The geographic boundary parameters, equipment boundary parameters, network boundary parameters, time boundary parameters, and risk boundary parameters in the environmental boundary snapshot are normalized and mapped to obtain geographic risk, equipment risk, network risk, time window risk, and on-site hazard signal risk. Based on the audit level of the strategy rules in the aforementioned strategy rule set, determine the strategy risk; Based on the weights of the judgment dimensions, the risk scores are calculated by weighted summation of the strategy risk, geographical risk, equipment risk, network risk, time window risk, and on-site danger signal risk. The constraint determination result is obtained based on the comprehensive risk score.

6. The intelligent agent behavior control method according to claim 1, characterized in that, The step of controlling the corresponding execution terminal to perform corresponding actions based on the constraint determination result includes: When the constraint determination result is to allow passage, the execution terminal corresponding to the execution terminal information is controlled to execute the action corresponding to the action type information; When the constraint determination result is substitution, a substitution action is generated based on the boundary failure reason and the action type information, and the execution terminal corresponding to the substitution action is controlled to execute the substitution action; If the constraint determination result is a rejection, the execution of the agent behavior request is terminated.

7. The intelligent agent behavior control method according to claim 6, characterized in that, The step of generating an alternative action based on the boundary failure reason and the action type information, and controlling the execution terminal corresponding to the alternative action to execute the alternative action, includes: Based on the action type information, determine the task objective; Based on the reasons for the boundary failure, determine the boundary conditions that are not met; Based on the task objective and the unmet boundary conditions, select the alternative action corresponding to the action type from the preset alternative action mapping relationship; Based on the execution terminal and alternative resources corresponding to the alternative action, obtain the environmental boundary parameters corresponding to the alternative action and generate an environmental boundary snapshot of the alternative action; Obtain the set of policy rules that match the alternative action; Based on the set of policy rules that match the alternative action, a joint constraint determination is performed on the alternative action and the snapshot of the alternative action's environment boundary to obtain a secondary determination result; When the secondary determination result is to allow passage, the execution terminal corresponding to the alternative action is controlled to execute the alternative action.

8. The intelligent agent behavior control method according to claim 1, characterized in that, The behavior request object also includes information on the expected effect of the action, and the intelligent agent behavior control method further includes: Obtain the execution feedback information after the execution terminal performs the corresponding action; The execution feedback information is compared with the expected action effect information to obtain verification information; The behavior request object, the policy rules triggered during the joint constraint determination process, the environment boundary snapshot, the constraint determination result, the execution feedback information, and the verification information are associated and stored to generate a behavior control record.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the intelligent agent behavior control method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent agent behavior control method according to any one of claims 1 to 8.

Images