Method, device, equipment, medium and program product for controlling embodied agent

Abstract

The present application relates to the technical field of robot autonomous control, and discloses a control method, device, equipment, medium and program product of embodied intelligent agent, wherein the method comprises: obtaining task execution parameters, the task execution parameters being used to represent a task execution environment; comparing the task execution parameters with a first control rule set to determine a first execution action, the first control rule set comprising a plurality of parallel independent root nodes, the root nodes comprising a plurality of parallel dimension branches, and the dimension branches comprising a plurality of serial first sub-rules; controlling the embodied intelligent agent to execute the first execution action and obtaining a first execution result; updating rule parameters of the first control rule set based on an evaluation result of the first execution result; and when the first execution action is triggered again, controlling the embodied intelligent agent to execute the first execution action based on the first control rule set with updated rule parameters. Through implementation of the technical solution of the present application, the control capability of the embodied intelligent agent is improved.

Description

Technical Field

[0001] This invention relates to the field of robot autonomous control technology, specifically to control methods, devices, equipment, media, and program products for embodied intelligent agents. Background Technology

[0002] With the rapid development of embodied intelligence and autonomous mobile robot technologies, various intelligent entities have been widely applied in fields such as inspection, outdoor operations, and high-risk scenarios. However, among these technologies, end-to-end black-box control methods lack interpretability and are difficult to adapt to environmental changes, while traditional fixed-rule methods rely on manually preset rules and cannot evolve autonomously based on real feedback, resulting in poor control accuracy of embodied intelligent entities in practical applications. Summary of the Invention

[0003] This invention provides a control method, device, equipment, medium, and program product for embodied intelligent agents to solve the problem of poor control accuracy of embodied intelligent agents.

[0004] In a first aspect, the present invention provides a control method for an embodied intelligent agent, comprising: Obtain task execution parameters, which are used to characterize the task execution environment; The task execution parameters are compared with the first control rule set to determine the first execution action. The first control rule set includes multiple parallel independent root nodes, each root node includes multiple parallel dimensional branches, and each dimensional branch includes multiple serial first sub-rules. The first sub-rules are used to characterize the mapping relationship between the task execution parameters and the first execution action. Control the embodied intelligent agent to execute the first action and obtain the first execution result; Based on the evaluation results of the first execution result, update the rule parameters of the first control rule set; When the first execution action is triggered again, the embodied agent is controlled to execute the first execution action based on the updated first control rule set with updated rule parameters.

[0005] The embodied agent control method provided in this invention obtains task execution parameters and compares them with a first control rule set to determine a first execution action. The first control rule set includes multiple parallel and independent root nodes, each containing multiple parallel dimensional branches. Each dimensional branch contains multiple serial first sub-rules, which decouples and structures complex control logic, facilitating the understanding of priority relationships between different control rules. The embodied agent is then controlled to execute the first execution action, and the rule parameters of the first control rule set are updated based on the obtained first execution result. This allows the control rules to adaptively adjust according to task execution feedback in the real physical environment, avoiding control rule mismatch caused by dynamic environmental changes and the lag in manual iterative adjustments. When the first execution action is triggered again, the embodied agent is controlled to execute the first execution action based on the updated first control rule set. This allows for the continuous accumulation of historical execution experience and its application to subsequent decisions, forming a closed-loop evolution mechanism and improving the control accuracy of the embodied agent.

[0006] In one optional implementation, the task execution parameters are compared with a first set of control rules to determine a first execution action, including: The task execution parameters are compared with multiple first sub-rules corresponding to each parallel dimension branch to determine at least one first objective sub-rule that the task execution parameters satisfy under each parallel dimension branch. Obtain the target execution action corresponding to each first target sub-rule; Based on the priority of the root node corresponding to each parallel dimension branch, the first execution action is determined from the execution actions of each target.

[0007] The embodied agent control method provided in this invention compares the task execution parameters with multiple first sub-rules corresponding to each parallel dimension branch to determine at least one first target sub-rule satisfied by the task execution parameters under each parallel dimension branch. This fully utilizes the multi-branch parallel structure of the first control rule set to determine the current environmental state, avoiding decision blind spots caused by single-dimensional judgment. Then, the target execution action corresponding to each first target sub-rule is obtained, and based on the root node priority corresponding to each parallel dimension branch, the first execution action is determined from each target execution action. This allows for arbitration decisions on target execution actions in case of conflict, ensuring the uniqueness of the embodied agent's decision output.

[0008] In one optional implementation, based on the evaluation result of the first execution result, the rule parameters of the first control rule set are updated, including: If the evaluation result indicates that the first execution result failed, then the rule path of the first execution result is traced back to locate the second sub-rule, which is the first sub-rule that caused the first execution result to fail. Based on the action threshold corresponding to the second sub-rule, determine the threshold range of the second sub-rule; The rule parameters of the first control rule set are updated based on the threshold range.

[0009] The control method for embodied intelligent agents provided in this invention, when the evaluation result indicates that the first execution result has failed, traces back the rule path of the first execution result to locate the second sub-rule. The second sub-rule is the first sub-rule that caused the first execution result to fail. This allows for precise identification of the first sub-rule corresponding to the failure from the multi-level structure of the first control rule set, shortening the response time from rule failure to triggering correction. Based on the action threshold corresponding to the second sub-rule, the threshold range of the second sub-rule is determined, and the rule parameters of the first control rule set are updated based on this threshold range. This constrains the range of rule parameters within a controllable range, avoiding the problem of other rule functions degrading or triggering hard security risks due to blind modifications, and ensuring the stability of rule updates.

[0010] In one optional implementation, the threshold range of the second sub-rule is determined based on the action threshold corresponding to the second sub-rule, including: The action threshold is sampled first and second according to a preset step size to obtain multiple sampled thresholds; the first sampling is in the direction of decreasing action threshold; the second sampling is in the direction of increasing action threshold. The first execution action is simulated for each sampling threshold to obtain the second execution result of the first execution action under each sampling threshold; The threshold range is determined based on the execution result type corresponding to the second execution result.

[0011] The control method for embodied intelligent agents provided in this invention obtains multiple sampling thresholds by performing first and second sampling on action thresholds according to a preset step size. This allows for a systematic exploration of the variation range of action thresholds near the current value, avoiding threshold omissions caused by unilateral sampling. By simulating the first execution action for each sampling threshold, the second execution result of the first execution action under each sampling threshold is obtained. This enables rapid verification and evaluation of each sampling threshold, ensuring evaluation efficiency while avoiding security risks caused by threshold testing in a real physical environment. By determining the threshold range based on the execution result type corresponding to the second execution result, the threshold space can be divided into sub-intervals corresponding to different types, accurately selecting feasible threshold ranges suitable for task execution.

[0012] In one optional implementation, updating the rule parameters of the first control rule set based on a threshold range includes: Within the threshold range, obtain the first sub-range representing the first execution result, which indicates that all execution actions were successfully executed, the second sub-range representing the partial execution success, and the third sub-range representing the complete failure of execution. Update the parameter range corresponding to the rule parameter to the first sub-range; or, Update the parameter range corresponding to the rule parameter to the first sub-range, the second sub-range, and the third sub-range.

[0013] The control method for embodied intelligent agents provided in this invention, by acquiring a first sub-interval representing the complete success of the first execution action, a second sub-interval representing partial success, and a third sub-interval representing complete failure within a threshold interval, can further refine the distinction between continuous threshold intervals based on actual execution results. By updating the parameter interval corresponding to the rule parameters to the first sub-interval, the rule parameters can always remain within a safe and efficient value range, ensuring the reliability of task execution. By updating the parameter interval corresponding to the rule parameters to the first, second, and third sub-intervals, the parameter intervals corresponding to different execution results can be retained, and a conservative and reliable parameter interval can be selected according to actual needs, improving the flexibility of parameter interval adjustment.

[0014] In an optional implementation, the method further includes: If the second sub-rule cannot be located from the rule path of the first execution result, the first execution result is simulated based on the task execution parameters to obtain the simulated execution result corresponding to the first execution result; Based on the evaluation results of the simulation execution, the target parameters that caused the first execution result to fail are determined. The task execution parameters include the target parameters. Based on the causal relationship between the target parameters and the first execution result, new sub-rules are determined; Update the newly added sub-rule to the first control rule set.

[0015] The control method for embodied intelligent agents provided in this invention, if the second sub-rule cannot be located from the rule path of the first execution result, simulates the first execution result based on the task execution parameters to obtain the simulated execution result corresponding to the first execution result. This method can proactively explore the correlation between task execution failure and each task execution parameter, avoiding abandoning rule evolution due to path backtracking failure. Based on the evaluation results of the simulated execution result, the target parameters that caused the first execution result to fail are determined. The task execution parameters include the target parameters. Based on the causal relationship between the target parameters and the first execution result, new sub-rules are determined. This method can accurately identify the influencing factors of task execution failure at the task execution parameter level and transform the causal relationship between the influencing factors and the execution result into explicit control rules. Then, the new sub-rules are updated to the first control rule set, which can effectively fill the decision blind spots of the existing first control rule set in specific scenarios.

[0016] In one alternative implementation, the first execution result is evaluated, including: Obtain the execution metrics corresponding to the first execution result, and evaluate the execution metrics from multiple dimensions to obtain the first evaluation result; Obtain the historical execution results of the first execution action under the first control rule set, compare the historical execution results with the first execution result, and obtain the second evaluation result; Based on the preset simulation model and the first set of control rules, the first execution action is simulated to obtain the third evaluation result; In response to the evaluation operation on the first execution result, obtain the fourth evaluation result; The evaluation result of the first execution result is determined based on the fusion of the first evaluation result, the second evaluation result, the third evaluation result, and the fourth evaluation result.

[0017] The control method for embodied intelligent agents provided in this invention obtains the execution indicators corresponding to the first execution result and performs multi-dimensional evaluation of the execution indicators to obtain a first evaluation result. This allows for quantitative evaluation of the execution indicators corresponding to the first execution result from multiple dimensions, avoiding the one-sidedness caused by a single indicator evaluation. Next, the historical execution results of the first execution action under the first control rule set are obtained, and the historical execution results are compared with the first execution result to obtain a second evaluation result. This verifies whether the current rule has degraded compared to historical performance, avoiding the problem of losing historical capabilities due to over-adaptation to current failed use cases during rule evolution. Then, based on a preset simulation model and the first control rule set, the first execution action is simulated to obtain a third evaluation result. This allows for advance deduction of the rules in the first control rule set, achieving risk screening before rule updates. Following the evaluation operation on the first execution result, a fourth evaluation result is obtained, enabling automatic evaluation and human experience judgment to complement each other. Finally, based on the fusion result of the first, second, third, and fourth evaluation results, the evaluation result of the first execution result is determined, avoiding the influence of the limitations or biases of a single evaluation dimension on the evaluation conclusion and improving the reliability of the evaluation result.

[0018] In a second aspect, the present invention provides a control device for an embodied intelligent agent, comprising: The parameter acquisition module is used to acquire task execution parameters, which are used to characterize the task execution environment. The action determination module is used to compare the task execution parameters with the first control rule set to determine the first execution action. The first control rule set includes multiple parallel independent root nodes, each root node includes multiple parallel dimensional branches, and each dimensional branch includes multiple serial first sub-rules. The first sub-rules are used to characterize the mapping relationship between the task execution parameters and the first execution action. The execution result determination module is used to control the embodied intelligent agent to perform the first execution action and obtain the first execution result; The parameter update module is used to update the rule parameters of the first control rule set based on the evaluation result of the first execution result; The action execution module is used to control the embodied intelligent agent to execute the first execution action based on the updated first control rule set of rule parameters when the first execution action is triggered again.

[0019] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the control method of the embodied intelligent agent described in the first aspect or any corresponding embodiment thereof.

[0020] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the control method of the embodied intelligent agent described in the first aspect or any corresponding embodiment thereof.

[0021] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the control method of the embodied intelligent agent described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a schematic flowchart of a first method for controlling an embodied intelligent agent according to an embodiment of the present invention; Figure 2 This is a second flowchart illustrating the control method for an embodied intelligent agent according to an embodiment of the present invention; Figure 3 This is a third flowchart illustrating the control method for an embodied intelligent agent according to an embodiment of the present invention; Figure 4 This is a structural block diagram of a control device for an embodied intelligent agent according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the first control rule set in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0026] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0027] With the rapid development of embodied intelligence and autonomous mobile robot technologies, various intelligent entities have been widely applied in fields such as inspection, outdoor operations, and high-risk scenarios. Related technologies mainly include end-to-end black-box control methods and traditional fixed-rule logic control methods. However, they still face many limitations in practical applications. Specifically: (1) End-to-end black box control methods usually require large-scale labeled data and high-precision simulation environment for pre-training to achieve strategy iteration. However, the simulation environment is difficult to fully reproduce factors such as friction changes, light fluctuations, terrain disturbances and sensor temperature drift in the real physical environment. After the model is deployed, it is easy to make decision-making errors in unfamiliar environments.

[0028] (2) Traditional fixed-rule logic control methods rely on fixed preset thresholds and logical judgments. Once the parameters are set, they will not be automatically adjusted during operation. However, the working environment of embodied intelligent agents often changes dynamically, such as the ground changing from dry to slippery, or changes in light intensity causing sensor readings to deviate. At this time, the preset fixed logic is often no longer applicable, which can easily lead to inaccurate control of the embodied intelligent agent or task failure.

[0029] According to an embodiment of the present invention, a control method embodiment for an embodied intelligent agent is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] This embodiment provides a control method for an embodied intelligent agent, which can be used in electronic devices, such as a concrete intelligent agent. This concrete intelligent agent is equipped with an intelligent agent control system based on unmanned scenarios, and this system is used to control the embodied intelligent agent. Figure 1 This is a flowchart of a control method for an embodied intelligent agent according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain task execution parameters.

[0031] Task execution parameters refer to the parameters collected by the embodied intelligent agent during the execution of a task, which are used to characterize the task execution environment.

[0032] Specifically, the embodied intelligent agent reads raw signals such as lidar point clouds, attitude data, and joint encoder feedback values ​​through built-in sensors and external sensing devices. After preprocessing the raw signals through filtering, coordinate transformation, and semantic understanding, the task execution parameters are obtained.

[0033] Step S102: Compare the task execution parameters with the first control rule set to determine the first execution action.

[0034] The first control rule set refers to a set of rules organized according to a preset structure, consisting of multiple explicit control rules. The first control rule set includes multiple parallel and independent root nodes, each root node containing multiple parallel dimensional branches. Each dimensional branch contains multiple sequential first sub-rules, which characterize the mapping relationship between task execution parameters and the first execution action. For example... Figure 6 As shown.

[0035] In this context, the root node refers to the top-level node of the first control rule set. Multiple root nodes operate independently and in parallel, with each root node corresponding to a different task type. Dimensional branches are branches extending downwards from the root nodes. Multiple dimensional branches under the same root node operate in parallel, with each dimensional branch corresponding to a type of task execution parameter dimension.

[0036] The first execution action is the instruction given to the control embodied intelligent agent after comparing the task execution parameters with the first set of control rules. For example, the first execution action can be a specific action instruction, a set of action parameter constraints, or a safety alarm event.

[0037] Specifically, the task execution parameters are input data into the first set of control rules. Then, each root node is traversed, and under each root node, all parallel dimension branches are traversed. In each dimension branch, the first sub-rule in the dimension branch is matched sequentially. When the task execution parameters meet the condition of a certain first sub-rule, the first execution action is determined through the mapping relationship corresponding to the first sub-rule.

[0038] Step S103: Control the embodied intelligent agent to execute the first execution action and obtain the first execution result.

[0039] The first execution result refers to the actual task output result obtained in the real environment after the embodied intelligent agent performs the first execution action.

[0040] Specifically, after acquiring the first execution action corresponding to the first sub-rule through its built-in motion controller, the embodied intelligent agent decomposes the first execution action into low-level hardware driver instructions that the embodied intelligent agent can recognize and execute. These hardware driver instructions are then sent to the execution mechanism of the embodied intelligent agent to control it to execute the first execution action. After the action is completed, the embodied intelligent agent collects the results generated after executing the first execution action through its built-in sensors, thus obtaining the first execution result.

[0041] Hardware driving instructions refer to control signals or data frames that the embodied intelligent agent's hardware layer can directly recognize and execute. Actuators refer to hardware components within the embodied intelligent agent that directly generate physical actions and interact physically with the real environment. For example, hardware driving instructions include motor speed, joint rotation angle, output torque, and runtime. Actuators include servo motors for each joint of the robotic arm, drive hub motors and steering servos for the wheeled chassis, and electric actuators for the lifting platform.

[0042] Step S104: Based on the evaluation result of the first execution result, update the rule parameters of the first control rule set.

[0043] The evaluation result refers to the judgment result obtained after a comprehensive quantitative evaluation of the first execution result, which is used to characterize the completion quality and success or failure of the first execution action.

[0044] Rule parameters refer to the adjustable threshold parameters used for conditional judgments in each of the first sub-rules within the first control rule set. For example, rule parameters include safety distance thresholds, abnormal temperature thresholds, etc.

[0045] Specifically, a preset evaluation method (such as evaluation based on historical execution results) is invoked to evaluate the first execution result, and an evaluation result is obtained. Then, based on the process of obtaining the first execution result, a reverse lookup is performed starting from the first execution action to obtain the first sub-rule corresponding to the first execution action. Then, the rule parameters corresponding to the first sub-rule are read, and the rule parameters are updated based on the comprehensive evaluation results.

[0046] Step S105: When the first execution action is triggered again, the embodied intelligent agent is controlled to execute the first execution action based on the first control rule set updated with the rule parameters.

[0047] Specifically, during the subsequent execution of tasks, the embodied intelligent agent continuously collects task execution parameters from the current real environment and continuously compares these parameters with the first control rule set updated based on the rule parameters. When the collected task execution parameters trigger the execution of the first execution action again, the first execution action is decomposed into hardware driver instructions, and then the embodied intelligent agent is controlled to execute the first execution action.

[0048] The embodied agent control method provided in this invention obtains task execution parameters and compares them with a first control rule set to determine a first execution action. The first control rule set includes multiple parallel and independent root nodes, each containing multiple parallel dimensional branches. Each dimensional branch contains multiple serial first sub-rules, which decouples and structures complex control logic, facilitating the understanding of priority relationships between different control rules. The embodied agent is then controlled to execute the first execution action, and the rule parameters of the first control rule set are updated based on the obtained first execution result. This allows the control rules to adaptively adjust according to task execution feedback in the real physical environment, avoiding control rule mismatch caused by dynamic environmental changes and the lag in manual iterative adjustments. When the first execution action is triggered again, the embodied agent is controlled to execute the first execution action based on the updated first control rule set. This allows for the continuous accumulation of historical execution experience and its application to subsequent decisions, forming a closed-loop evolution mechanism and improving the control accuracy of the embodied agent.

[0049] This embodiment provides a control method for an embodied intelligent agent, which can be used in electronic devices, such as a concrete intelligent agent. This concrete intelligent agent is equipped with an intelligent agent control system based on unmanned scenarios, and this system is used to control the embodied intelligent agent. Figure 2 This is a flowchart of a control method for an embodied intelligent agent according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain task execution parameters. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0050] Step S202: Compare the task execution parameters with the first control rule set to determine the first execution action.

[0051] Specifically, step S202 includes: Step S2021: Compare the task execution parameters with the multiple first sub-rules corresponding to each parallel dimension branch to determine at least one first target sub-rule that the task execution parameters satisfy under each parallel dimension branch.

[0052] The first target sub-rule refers to the first sub-rule in a specific parallel dimension branch where the current task execution parameters satisfy its triggering conditions.

[0053] Specifically, after obtaining the task execution parameters, these parameters are input into the first control rule set. Then, each root node is traversed, and under each root node, all parallel dimension branches are traversed. For each parallel dimension branch, along the multiple first sub-rules arranged serially within that branch, the task execution parameters are compared one by one to see if they meet the judgment condition corresponding to that first sub-rule. If the task execution parameters meet the judgment condition corresponding to a certain first sub-rule, then that first sub-rule is marked as the first target sub-rule.

[0054] Step S2022: Obtain the target execution action corresponding to each first target sub-rule.

[0055] Specifically, once the first sub-rule is marked as the first target sub-rule, the target execution action corresponding to the first target sub-rule is determined through the mapping relationship between the first sub-rule and the execution action.

[0056] Step S2023: Based on the priority of the root node corresponding to each parallel dimension branch, determine the first execution action from the execution actions of each target.

[0057] Root node priority refers to the pre-defined execution order among the root nodes in the first control rule set. For example, the security protection root node has the highest priority, ensuring that security-related control commands are executed before other control commands under any circumstances.

[0058] Specifically, the priority flags of the root nodes corresponding to the dimension branches of each first target sub-rule are read. Then, all target execution actions are sorted according to the priority of their root nodes, and the target execution action corresponding to the root node with the highest priority is selected as the first execution action.

[0059] Step S203: Control the embodied intelligent agent to execute the first execution action and obtain the first execution result. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0060] Step S204: Based on the evaluation result of the first execution result, update the rule parameters of the first control rule set. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0061] Step S205: When the first execution action is triggered again, the embodied agent is controlled to execute the first execution action based on the updated first control rule set. For details, please refer to [link to details]. Figure 1 Step S105 of the illustrated embodiment will not be described again here.

[0062] The embodied agent control method provided in this invention compares the task execution parameters with multiple first sub-rules corresponding to each parallel dimension branch to determine at least one first target sub-rule satisfied by the task execution parameters under each parallel dimension branch. This fully utilizes the multi-branch parallel structure of the first control rule set to determine the current environmental state, avoiding decision blind spots caused by single-dimensional judgment. Then, the target execution action corresponding to each first target sub-rule is obtained, and based on the root node priority corresponding to each parallel dimension branch, the first execution action is determined from each target execution action. This allows for arbitration decisions on target execution actions in case of conflict, ensuring the uniqueness of the embodied agent's decision output.

[0063] This embodiment provides a control method for an embodied intelligent agent, which can be used in electronic devices, such as a concrete intelligent agent. This concrete intelligent agent is equipped with an intelligent agent control system based on unmanned scenarios, and this system is used to control the embodied intelligent agent. Figure 3 This is a flowchart of a control method for an embodied intelligent agent according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: Obtain task execution parameters. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0064] Step S302: Compare the task execution parameters with the first control rule set to determine the first execution action. For details, please refer to [link to relevant documentation]. Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0065] Step S303: Control the embodied intelligent agent to execute the first execution action and obtain the first execution result. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0066] Step S304: Based on the evaluation result of the first execution result, update the rule parameters of the first control rule set.

[0067] Specifically, step S304 includes: Step S3041: If the evaluation result indicates that the first execution result has failed, then backtrack the rule path of the first execution result and locate the second sub-rule. The second sub-rule is the first sub-rule that caused the first execution result to fail.

[0068] The rule path refers to the complete matching link that the task execution parameters traverse when mapped to the first execution action in the first control rule set, including the first sub-rule triggered during this task execution, the first execution result, and the evaluation result corresponding to the first execution result.

[0069] Specifically, the evaluation result obtained after evaluating the first execution result is acquired. If the evaluation result indicates that the first execution result failed, the first execution action corresponding to the first execution result is used as the starting point to perform a reverse search of the rule path. For each first sub-rule on the rule path, the first sub-rule with the strongest correlation is obtained through causal attribution analysis or a confusion distribution matrix of success or failure of historical triggers, and is determined as the second sub-rule.

[0070] Step S3042: Determine the threshold range of the second sub-rule based on the action threshold corresponding to the second sub-rule.

[0071] Action threshold refers to the adjustable threshold parameter used for conditional judgment in the second sub-rule. For example, action threshold can be represented as an integer value, a percentage value, etc.

[0072] The threshold range refers to an adjustable numerical range determined with the action threshold as the central reference point, and is usually defined by a lower limit and an upper limit.

[0073] Specifically, sampling is performed using the action threshold corresponding to the second sub-rule as the central reference point, according to sampling parameters (such as sampling step size, sliding window, etc.), to obtain multiple sampling points. Then, the multiple sampling points are evaluated, and the threshold range of the second sub-rule is obtained based on the evaluation results.

[0074] Here, the sampling point refers to the candidate value within the threshold interval.

[0075] In some optional implementations, step S3042 above includes: Step a1: Perform first sampling and second sampling on the action threshold according to the preset step size to obtain multiple sampling thresholds.

[0076] Step a2: Simulate the first execution action for each sampling threshold to obtain the second execution result of the first execution action under each sampling threshold.

[0077] Step a3: Determine the threshold range based on the execution result type corresponding to the second execution result.

[0078] The preset step size refers to the pre-set sampling interval value.

[0079] The first sampling refers to sampling in the direction of decreasing action threshold according to a preset step size, and the resulting sampling threshold is less than the action threshold; the second sampling refers to sampling in the direction of increasing action threshold according to a preset step size, and the resulting sampling threshold is greater than the action threshold.

[0080] The execution result type refers to the category label obtained after classifying the second execution result. For example, the execution result type can include execution success, execution failure, partial execution success, risk warning, etc.

[0081] Specifically, using the action threshold corresponding to the second sub-rule as the central reference point, sampling is performed in both the decreasing and increasing directions of the action threshold according to a preset step size, resulting in multiple sampling thresholds. Then, through a preset simulation model in the intelligent agent control system, all sampling thresholds are iterated over, and the action threshold is replaced with the sampling threshold one by one, and the first action is executed, obtaining the second execution result of the first execution action under each sampling threshold. Finally, the second execution result is evaluated, and based on the evaluation result, the execution result type corresponding to the second execution result is obtained. Based on this execution result type, the threshold range is determined.

[0082] Among them, the preset simulation model refers to the virtual environment used to simulate the execution of tasks by an embodied intelligent agent.

[0083] In the above embodiments, by performing first and second sampling on the action threshold according to a preset step size, multiple sampling thresholds are obtained. This allows for a systematic exploration of the variation range of the action threshold near the current value, avoiding the threshold omission problem caused by unilateral sampling. By simulating the first execution action for each sampling threshold, the second execution result of the first execution action under each sampling threshold is obtained. This enables rapid verification and evaluation of each sampling threshold, ensuring evaluation efficiency while avoiding the security risks caused by threshold testing in a real physical environment. By determining the threshold interval based on the execution result type corresponding to the second execution result, the threshold space can be divided into sub-intervals corresponding to different types, and feasible threshold ranges suitable for task execution can be accurately selected.

[0084] Step S3043: Update the rule parameters of the first control rule set based on the threshold range.

[0085] Specifically, the threshold range corresponding to the second sub-rule is obtained, and then a sub-range within this threshold range is selected that can prevent the first execution result from failing without disrupting the successful execution case. Here, the successful execution case refers to the task case whose evaluation result indicates that the first execution result was successfully executed.

[0086] Then, the rule parameters of the second sub-rule in the first control rule set are updated to that sub-interval.

[0087] It should be noted that, except for the second sub-rule, the rule parameters of the other first sub-rules in the first control rule set are not updated.

[0088] In some optional implementations, step S3043 above includes: Step b1: Within the threshold interval, obtain the first sub-interval representing the first execution result, indicating that all first execution actions were successfully executed, the second sub-interval representing the partial execution success, and the third sub-interval representing the complete failure of execution.

[0089] Step b2: Update the parameter range corresponding to the rule parameter to the first sub-range; The first subinterval, the second subinterval, and the third subinterval are all subintervals of the threshold interval.

[0090] Partial execution success means that the first execution result corresponding to the first execution action in the second sub-interval fluctuates, and the action is not completely executed but the performance is unstable.

[0091] Specifically, based on the obtained execution result types, the numerical range corresponding to all successful executions is identified and marked as the first sub-interval; the numerical range corresponding to partially successful executions is identified and marked as the second sub-interval; and the numerical range corresponding to all failed executions is identified and marked as the third sub-interval.

[0092] Then, the first sub-interval is selected as the new range of rule parameter values, and the parameter range corresponding to the rule parameter of the second sub-rule is updated to the first sub-interval.

[0093] In some optional implementations, step S3043 above further includes: updating the parameter interval corresponding to the rule parameter to a first sub-interval, a second sub-interval, and a third sub-interval.

[0094] After determining the first, second, and third sub-intervals, the first, second, and third sub-intervals can be selected together as a new range of rule parameter values. That is, the parameter range corresponding to the rule parameter of the second sub-rule is divided into three sub-intervals to retain the complete range of values.

[0095] In the above embodiments, by obtaining a first sub-interval representing the complete success of the first execution action, a second sub-interval representing partial success, and a third sub-interval representing complete failure within the threshold interval, continuous threshold intervals can be further refined based on actual execution results. By updating the parameter interval corresponding to the rule parameters to the first sub-interval, the rule parameters can always remain within a safe and efficient value range, ensuring the reliability of task execution. By updating the parameter interval corresponding to the rule parameters to the first, second, and third sub-intervals, the parameter intervals corresponding to different execution results can be retained, and a conservative and reliable parameter interval can be selected according to actual needs, improving the flexibility of parameter interval adjustment.

[0096] Step S3044: If the second sub-rule is not located from the rule path of the first execution result, the first execution result is simulated based on the task execution parameters to obtain the simulated execution result corresponding to the first execution result.

[0097] Specifically, the evaluation result obtained after evaluating the first execution result is acquired. If the evaluation result indicates that the first execution result failed, the first execution action corresponding to the first execution result is used as the starting point to perform a reverse search of the rule path. For each first sub-rule on the rule path, after analysis through causal attribution analysis or a confusion distribution matrix of success or failure of historical triggers, no second sub-rule is determined. Then, the task execution parameters are used as input data and input into a preset simulation model to simulate the first execution result, obtaining the simulated execution result.

[0098] Step S3045: Based on the evaluation results of the simulation execution results, determine the target parameters that caused the first execution result to fail. The task execution parameters include the target parameters.

[0099] The target parameter refers to the parameter variable in the task execution parameters that is related to the failure of the first execution result.

[0100] Specifically, using a univariate comparison method, task execution parameters are input into a preset simulation model in the intelligent agent control system to obtain simulated execution results. These results are then evaluated using a preset evaluation method to obtain evaluation results. Based on these evaluation results, the target parameters that caused the first execution result to fail are determined.

[0101] The univariate control method refers to simulation in which only one parameter of the task execution parameters is changed, while the other parameters remain unchanged.

[0102] If the evaluation result changes from indicating that the first execution result failed to indicating that the first execution result succeeded, then the parameter is marked as a highly relevant parameter that caused the first execution result to fail. If the evaluation result still indicates that the first execution result failed, then the parameter is excluded.

[0103] Step S3046: Based on the causal relationship between the target parameters and the first execution result, determine the new sub-rule.

[0104] Causal correlation refers to the causal relationship between the target parameter and the success or failure of the first execution result. For example, if the target parameter is within a certain range of values, the first execution result will fail; if the target parameter is within another range of values, the first execution result will succeed.

[0105] A new sub-rule refers to a completely new judgment rule that does not belong to the original first sub-rule in the first control rule set.

[0106] Specifically, the first execution results are analyzed when the target parameter is within different value ranges to determine the causal relationship between the target parameter and the first execution result. Then, based on this causal relationship, the target parameter is mapped to an explicit rule, resulting in a new sub-rule. For example, when the ambient wind speed is >12m / s, the drone's hovering maneuver fails; when the ambient wind speed is <12m / s, the hovering maneuver succeeds. The new sub-rule obtained through the causal relationship mapping is: when the ambient wind speed is >12m / s, execute a descent and return to base.

[0107] Step S3047: Update the newly added sub-rule to the first control rule set.

[0108] Specifically, based on the dimension involved in the triggering condition of the newly added sub-rule, the root node and parallel dimension branch to which it belongs in the first control rule set are determined, and then the newly added sub-rule is serially merged into the dimension branch as a new rule under that dimension branch.

[0109] In one alternative implementation, evaluating the first execution result may include: Step c1: Obtain the execution metrics corresponding to the first execution result, and evaluate the execution metrics from multiple dimensions to obtain the first evaluation result.

[0110] Execution metrics refer to quantitative data used to measure the quality of the initial execution result. For example, execution metrics include task completion rate, operation time, operating energy consumption, number of safety alarms triggered, number of collisions and over-limit violations, and degree of trajectory deviation.

[0111] Specifically, quantifiable performance indicators are extracted from the first performance result. Then, multiple evaluation dimensions are selected, and each performance indicator is uniformly mapped to the same scoring range for scoring. The scoring results of each dimension are then combined to obtain the first evaluation result.

[0112] Step c2: Obtain the historical execution results of the first execution action under the first control rule set, compare the historical execution results with the first execution result, and obtain the second evaluation result.

[0113] Historical execution results refer to the historical successful use cases, historical failed use cases, and rule paths corresponding to the first execution action.

[0114] Specifically, the historical execution results of the first action under the first control rule set are obtained from the historical use case database of the intelligent agent control system. Then, the historical rule paths corresponding to the historical execution results and the current rule paths corresponding to the first execution results are extracted, and the sequence similarity between the two rule paths is calculated to obtain the path difference degree. Based on the path difference degree, the second evaluation result is obtained.

[0115] Step c3: Based on the preset simulation model and the first control rule set, simulate the first execution action to obtain the third evaluation result.

[0116] Specifically, the system calls the preset simulation model in the intelligent agent control system and loads the first set of control rules. The task execution parameters are used as input data. The first execution action is performed in the preset simulation model. The system detects whether there is a collision risk, whether the kinematic constraints are exceeded, or whether a logical conflict occurs. Based on the detection results, the system outputs a risk prediction score and obtains the third evaluation result.

[0117] Step c4: In response to the evaluation operation for the first execution result, obtain the fourth evaluation result.

[0118] Evaluation operations refer to the interactive behavior of operators (such as field engineers and remote monitoring personnel) to conduct qualitative evaluations of the initial execution results.

[0119] Specifically, based on the input interface used for human evaluation in the intelligent agent control system, the operator inputs a natural language description through this interface to complete the evaluation of the first execution result. Then, by converting the unstructured natural language description into a quantitative score through a large language model, a fourth evaluation result is obtained.

[0120] Step c5: Based on the fusion of the first evaluation result, the second evaluation result, the third evaluation result, and the fourth evaluation result, determine the evaluation result of the first execution result.

[0121] Specifically, the first evaluation result, the second evaluation result, the third evaluation result, and the fourth evaluation result are weighted and fused using preset weights and credibility coefficients to obtain the evaluation result of the first execution result.

[0122] The credibility coefficient refers to the quantitative estimate of the authenticity and reliability of a certain evaluation result in the final evaluation result. The credibility coefficient of each evaluation result can be set according to factors such as the number of historical samples, the operator's permission level, and the accuracy of the simulation environment.

[0123] By acquiring the execution metrics corresponding to the first execution result and evaluating them from multiple dimensions, a first evaluation result is obtained. This allows for a quantitative evaluation of the execution metrics corresponding to the first execution result from multiple dimensions, avoiding the one-sidedness caused by a single metric evaluation. Next, the historical execution results of the first execution action under the first control rule set are acquired, and the historical execution results are compared with the first execution result to obtain a second evaluation result. This verifies whether the current rule has degraded compared to historical performance, avoiding the problem of losing historical capabilities due to over-adaptation to current failed use cases during rule evolution. Then, based on a preset simulation model and the first control rule set, the first execution action is simulated to obtain a third evaluation result. This allows for advance deduction of the rules in the first control rule set, achieving risk screening before rule updates. Following the evaluation operation on the first execution result, a fourth evaluation result is obtained, enabling automatic evaluation and human experience judgment to complement each other. Finally, based on the fusion of the first, second, third, and fourth evaluation results, the evaluation result of the first execution result is determined, avoiding the influence of the limitations or biases of a single evaluation dimension on the evaluation conclusion and improving the reliability of the evaluation result.

[0124] Step S305: When the first execution action is triggered again, the embodied agent is controlled to execute the first execution action based on the updated first control rule set. For details, please refer to [link to relevant documentation]. Figure 1 Step S105 of the illustrated embodiment will not be described again here.

[0125] The control method for embodied intelligent agents provided in this invention, when the evaluation result indicates that the first execution result has failed, backtracks the rule path of the first execution result to locate the second sub-rule. The second sub-rule is the first sub-rule that caused the first execution result to fail. This allows for precise identification of the first sub-rule corresponding to the failure from the multi-level structure of the first control rule set, shortening the response time from rule failure to triggering correction. Based on the action threshold corresponding to the second sub-rule, the threshold range of the second sub-rule is determined, and the rule parameters of the first control rule set are updated based on the threshold range. This constrains the range of rule parameters within a controllable range, avoiding the problem of other rule functions degrading or triggering hard security risks due to blind modification, and ensuring the stability of rule updates. If the second sub-rule is not located from the rule path of the first execution result, the first execution result is simulated based on the task execution parameters to obtain the simulated execution result corresponding to the first execution result. This proactively explores the correlation between task execution failure and various task execution parameters, avoiding abandoning rule evolution due to path backtracking failure. Based on the evaluation results of the simulation execution, the target parameters that caused the first execution result to fail are determined. The task execution parameters include the target parameters. Based on the causal relationship between the target parameters and the first execution result, new sub-rules are determined. This can accurately identify the influencing factors of task execution failure at the level of task execution parameters, and transform the causal relationship between the influencing factors and the execution result into explicit control rules. Then, the new sub-rules are updated to the first control rule set, which can effectively fill the decision blind spots of the existing first control rule set in specific scenarios.

[0126] This embodiment also provides a control device for an embodied intelligent agent, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0127] This embodiment provides a control device for an embodied intelligent agent, such as Figure 4 As shown, it includes: The parameter acquisition module 401 is used to acquire task execution parameters, which are used to characterize the task execution environment. The action determination module 402 is used to compare the task execution parameters with the first control rule set to determine the first execution action. The first control rule set includes multiple parallel independent root nodes, each root node includes multiple parallel dimensional branches, and each dimensional branch includes multiple serial first sub-rules. The first sub-rules are used to characterize the mapping relationship between the task execution parameters and the first execution action. The execution result determination module 403 is used to control the embodied intelligent agent to perform the first execution action and obtain the first execution result; The parameter update module 404 is used to update the rule parameters of the first control rule set based on the evaluation result of the first execution result; The action execution module 405 is used to control the embodied intelligent agent to execute the first execution action based on the updated first control rule set of rule parameters when the first execution action is triggered again.

[0128] In some alternative implementations, the action determination module 402 includes: The rule comparison unit is used to compare the task execution parameters with multiple first sub-rules corresponding to each parallel dimension branch, and determine at least one first target sub-rule that the task execution parameters satisfy under each parallel dimension branch.

[0129] The target action acquisition unit is used to acquire the target execution action corresponding to each first target sub-rule.

[0130] The action determination unit is used to determine the first action from each target action based on the priority of the root node corresponding to each parallel dimension branch.

[0131] In some alternative implementations, the parameter update module 404 includes: The path backtracking unit is used to backtrack the rule path of the first execution result and locate the second sub-rule if the evaluation result indicates that the first execution result failed. The second sub-rule is the first sub-rule that caused the first execution result to fail.

[0132] The threshold interval determination unit is used to determine the threshold interval of the second sub-rule based on the action threshold corresponding to the second sub-rule.

[0133] The rule parameter update unit is used to update the rule parameters of the first control rule set based on the threshold range.

[0134] The result simulation unit is used to simulate the first execution result based on the task execution parameters if the second sub-rule cannot be located from the rule path of the first execution result, so as to obtain the simulated execution result corresponding to the first execution result.

[0135] The target parameter determination unit is used to determine the target parameters that cause the first execution result to fail based on the evaluation results of the simulation execution results. The task execution parameters include the target parameters.

[0136] The rule addition unit is used to determine new sub-rules based on the causal relationship between the target parameters and the first execution result.

[0137] The rule set addition unit is used to update the first control rule set with newly added sub-rules.

[0138] In some optional implementations, the threshold interval determination unit includes: The threshold sampling subunit is used to perform first sampling and second sampling on the action threshold according to a preset step size to obtain multiple sampling thresholds.

[0139] The action simulation subunit is used to simulate the first execution action for each sampling threshold, and obtain the second execution result of the first execution action under each sampling threshold.

[0140] The interval determination sub-unit is used to determine the threshold interval based on the execution result type corresponding to the second execution result.

[0141] In some optional implementations, the rule parameter update unit includes: The sub-interval acquisition sub-unit is used to acquire, within the threshold interval, the first sub-interval representing the first execution result indicating that all execution actions were successfully executed, the second sub-interval representing the partial execution success, and the third sub-interval representing the complete failure of execution.

[0142] The sub-interval update sub-unit is used to update the parameter interval corresponding to the rule parameter to the first sub-interval; or, to update the parameter interval corresponding to the rule parameter to the first sub-interval, the second sub-interval, and the third sub-interval.

[0143] In some alternative embodiments, the above-described apparatus may further include: The results evaluation module is used to evaluate the first execution result.

[0144] Specifically, the results evaluation module includes: The first result acquisition unit is used to acquire the execution indicators corresponding to the first execution result, and to evaluate the execution indicators in multiple dimensions to obtain the first evaluation result.

[0145] The second result acquisition unit is used to acquire the historical execution results of the first execution action under the first control rule set, compare the historical execution results with the first execution result, and obtain the second evaluation result.

[0146] The third result acquisition unit is used to simulate the first execution action based on the preset simulation model and the first control rule set, and obtain the third evaluation result.

[0147] The fourth result acquisition unit is used to acquire a fourth evaluation result in response to the evaluation operation on the first execution result.

[0148] The result determination unit is used to determine the evaluation result of the first execution result based on the fusion result of the first evaluation result, the second evaluation result, the third evaluation result, and the fourth evaluation result.

[0149] The control device for the embodied intelligent agent provided in this embodiment of the invention can execute the control method for the embodied intelligent agent provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0150] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0151] The following is a detailed reference. Figure 5 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 501, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 502 or a program loaded from memory 508 into random access memory (RAM) 503. RAM 503 also stores various programs and data required for the operation of the electronic device. The processor 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0152] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0153] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a memory 508, or installed from a ROM 502. When the computer program is executed by the processor 501, it performs the functions defined in the control method of the embodied intelligent agent according to embodiments of the present invention.

[0154] Figure 5The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0155] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the control method of the embodied intelligent agent shown in the above embodiments is implemented.

[0156] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0157] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A control method for an embodied intelligent agent, characterized in that, The method includes: Obtain task execution parameters, which are used to characterize the task execution environment; The task execution parameters are compared with the first control rule set to determine the first execution action. The first control rule set includes multiple parallel independent root nodes, each root node includes multiple parallel dimensional branches, and each dimensional branch includes multiple serial first sub-rules. The first sub-rules are used to characterize the mapping relationship between the task execution parameters and the first execution action. Control the embodied intelligent agent to execute the first execution action and obtain the first execution result; Based on the evaluation result of the first execution result, update the rule parameters of the first control rule set; When the first execution action is triggered again, the embodied intelligent agent is controlled to execute the first execution action based on the first control rule set updated by the rule parameters.

2. The control method for an embodied intelligent agent according to claim 1, characterized in that, The step of comparing the task execution parameters with the first control rule set to determine the first execution action includes: The task execution parameters are compared with multiple first sub-rules corresponding to each parallel dimension branch to determine at least one first target sub-rule that the task execution parameters satisfy under each parallel dimension branch. Obtain the target execution action corresponding to each of the first target sub-rules; Based on the priority of the root node corresponding to each of the parallel dimension branches, the first execution action is determined from each of the target execution actions.

3. The control method for an embodied intelligent agent according to claim 1, characterized in that, The step of updating the rule parameters of the first control rule set based on the evaluation result of the first execution result includes: If the evaluation result indicates that the first execution result has failed, then the rule path of the first execution result is traced back to locate the second sub-rule, which is the first sub-rule that caused the first execution result to fail. Based on the action threshold corresponding to the second sub-rule, determine the threshold range of the second sub-rule; The rule parameters of the first control rule set are updated based on the threshold range.

4. The control method for an embodied intelligent agent according to claim 3, characterized in that, The step of determining the threshold range of the second sub-rule based on the action threshold corresponding to the second sub-rule includes: The action threshold is sampled first and second according to a preset step size to obtain multiple sampling thresholds; the first sampling is performed in the direction that decreases the action threshold; the second sampling is performed in the direction that increases the action threshold. The first execution action is simulated for each of the sampling thresholds to obtain the second execution result of the first execution action under each of the sampling thresholds; The threshold range is determined based on the execution result type corresponding to the second execution result.

5. The control method for an embodied intelligent agent according to claim 3 or 4, characterized in that, The step of updating the rule parameters of the first control rule set based on the threshold range includes: Within the threshold range, the execution result is used to identify a first sub-range where all executions of the first execution action are successful, a second sub-range where some executions are successful, and a third sub-range where all executions fail. Update the parameter range corresponding to the rule parameter to the first sub-range; or, The parameter range corresponding to the rule parameter is updated to the first sub-range, the second sub-range, and the third sub-range.

6. The control method for an embodied intelligent agent according to claim 3, characterized in that, Also includes: If the second sub-rule cannot be located from the rule path of the first execution result, then the first execution result is simulated based on the task execution parameters to obtain the simulated execution result corresponding to the first execution result; Based on the evaluation results of the simulation execution results, the target parameters that cause the first execution result to fail are determined, and the task execution parameters include the target parameters; Based on the causal relationship between the target parameters and the first execution result, a new sub-rule is determined; Update the newly added sub-rule to the first control rule set.

7. The control method for an embodied intelligent agent according to claim 3, characterized in that, The evaluation of the first execution result includes: Obtain the execution metrics corresponding to the first execution result, and evaluate the execution metrics from multiple dimensions to obtain the first evaluation result; Obtain the historical execution results of the first execution action under the first control rule set, compare the historical execution results with the first execution result, and obtain the second evaluation result; Based on the preset simulation model and the first control rule set, the first execution action is simulated to obtain the third evaluation result; In response to the evaluation operation on the first execution result, a fourth evaluation result is obtained; The evaluation result of the first execution result is determined based on the fusion result of the first evaluation result, the second evaluation result, the third evaluation result, and the fourth evaluation result.

8. A control device for an embodied intelligent agent, characterized in that, The device includes: The parameter acquisition module is used to acquire task execution parameters, which are used to characterize the task execution environment. An action determination module is used to compare the task execution parameters with a first control rule set to determine a first execution action. The first control rule set includes multiple parallel independent root nodes, each root node includes multiple parallel dimensional branches, and each dimensional branch includes multiple serial first sub-rules. The first sub-rules are used to characterize the mapping relationship between the task execution parameters and the first execution action. The execution result determination module is used to control the embodied intelligent agent to execute the first execution action and obtain the first execution result; The parameter update module is used to update the rule parameters of the first control rule set based on the evaluation result of the first execution result; The action execution module is used to control the embodied intelligent agent to execute the first execution action based on the first control rule set updated by the rule parameters when the first execution action is triggered again.

9. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the control method of the embodied intelligent agent as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the control method of the embodied intelligent agent as described in any one of claims 1 to 7.

11. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the control method for an embodied intelligent agent as described in any one of claims 1 to 7.

Images