Task planning method and device, task processing method and device and computer equipment

Abstract

The invention relates to a task planning method and device, a task processing method and device, computer equipment, a computer readable storage medium and a computer program product. The task planning method comprises the following steps: when a user task is received, carrying out initial planning on the user task to obtain a planning step; under the condition that the planning bifurcation step exists in the planning step, a plurality of professional agents are selected based on the planning bifurcation step; through a plurality of professional agents, respectively planning feasible subsequent paths of the bifurcation step, and obtaining candidate planning schemes which are generated by the plurality of professional agents respectively and aim at the bifurcation step; performing quantitative evaluation on the plurality of candidate planning schemes to obtain respective scheme evaluation results of the plurality of candidate planning schemes; and generating a task planning scheme of the user task based on the respective scheme evaluation results of the plurality of candidate planning schemes. By adopting the method, the task planning decision accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a task planning method, apparatus, computer equipment, computer-readable storage medium, and computer program product, as well as a task processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] With the development of computer technology, large language models have emerged. Large language models refer to a new generation of artificial intelligence models with large-scale natural language understanding and generation capabilities, capable of handling tasks using planning abilities. The planning ability of large language models refers to their ability to automatically decompose executable steps and design solutions for complex problems.

[0003] In related technologies, the planning mode of large language models is that the large language model is responsible for the "think-action" cycle of the entire task. When encountering a planning branching step, the large language model generates a one-time planning scheme based on its internal knowledge.

[0004] However, the quality of the planning scheme for the bifurcation steps in the related technologies is greatly affected by the current state and built-in preferences of the large language model, resulting in unstable output and a tendency to get stuck in local optima rather than the global optimal path, leading to low accuracy in task planning decisions for user tasks. Summary of the Invention

[0005] Therefore, it is necessary to provide a task planning method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of task planning decisions, as well as a task processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product, in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a task planning method, including:

[0007] Upon receiving a user task, an initial plan is performed on the user task to obtain the planning steps;

[0008] If a planning bifurcation step exists in the planning process, multiple specialized intelligent agents are selected based on the planning bifurcation step.

[0009] Through the multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, and candidate planning schemes for the planning fork step generated by each of the multiple specialized intelligent agents are obtained.

[0010] Quantitatively evaluate each of the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme.

[0011] Based on the evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated.

[0012] Secondly, this application also provides a task planning apparatus, comprising:

[0013] The first planning module is used to perform initial planning on the user task when a user task is received, and to obtain planning steps.

[0014] The first selection module is used to select multiple specialized intelligent agents based on the planning bifurcation step when there is a planning bifurcation step in the planning step.

[0015] The second planning module is used to plan feasible subsequent paths for the planning fork step by the multiple professional intelligent agents, and obtain the candidate planning schemes for the planning fork step generated by each of the multiple professional intelligent agents.

[0016] The first quantitative evaluation module is used to perform quantitative evaluation on multiple candidate planning schemes respectively, and obtain the scheme evaluation results of each of the multiple candidate planning schemes.

[0017] The first scheme generation module is used to generate a task planning scheme for the user task based on the scheme evaluation results of the multiple candidate planning schemes.

[0018] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0019] Upon receiving a user task, an initial plan is performed on the user task to obtain the planning steps;

[0020] If a planning bifurcation step exists in the planning process, multiple specialized intelligent agents are selected based on the planning bifurcation step.

[0021] Through the multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, and candidate planning schemes for the planning fork step generated by each of the multiple specialized intelligent agents are obtained.

[0022] Quantitatively evaluate each of the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme.

[0023] Based on the evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated.

[0024] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0025] Upon receiving a user task, an initial plan is performed on the user task to obtain the planning steps;

[0026] If a planning bifurcation step exists in the planning process, multiple specialized intelligent agents are selected based on the planning bifurcation step.

[0027] Through the multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, and candidate planning schemes for the planning fork step generated by each of the multiple specialized intelligent agents are obtained.

[0028] Quantitatively evaluate each of the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme.

[0029] Based on the evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated.

[0030] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0031] Upon receiving a user task, an initial plan is performed on the user task to obtain the planning steps;

[0032] If a planning bifurcation step exists in the planning process, multiple specialized intelligent agents are selected based on the planning bifurcation step.

[0033] Through the multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, and candidate planning schemes for the planning fork step generated by each of the multiple specialized intelligent agents are obtained.

[0034] Quantitatively evaluate each of the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme.

[0035] Based on the evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated.

[0036] The aforementioned task planning method, apparatus, computer equipment, computer-readable storage medium, and computer program product, upon receiving a user task, perform initial planning for the user task to obtain planning steps. This initial planning is completed. If a planning fork step exists within the planning steps, multiple specialized intelligent agents are selected based on the planning fork step. These agents plan feasible subsequent paths for each planning fork step, resulting in candidate planning schemes generated by each agent for that fork step. When encountering a planning fork, multiple specialized intelligent agents are introduced to perform parallel planning for the fork step, utilizing collective intelligence to reduce the uncertainty and randomness of the planning scheme. By quantitatively evaluating each of the multiple candidate planning schemes, evaluation results are obtained for each scheme. This quantitative evaluation method allows for objective evaluation of multiple candidate planning schemes, and ultimately, a task planning scheme for the user task can be generated based on the evaluation results of each of the multiple candidate planning schemes. Throughout the process, when encountering planning bifurcation during the planning process, the uncertainty and randomness of the solution planning can be reduced by using multiple professional intelligent agents to plan in parallel and to quantitatively evaluate the candidate solutions generated by each of the multiple professional intelligent agents. This improves the quality of task planning decisions at the planning bifurcation step and thus improves the accuracy of task planning decisions for user tasks.

[0037] Sixthly, this application provides a task processing method, including:

[0038] Upon receiving a user task, an initial plan is performed on the user task to obtain the planning steps;

[0039] If a planning bifurcation step exists in the planning process, multiple specialized intelligent agents are selected based on the planning bifurcation step.

[0040] Through the multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, and candidate planning schemes for the planning fork step generated by each of the multiple specialized intelligent agents are obtained.

[0041] Quantitatively evaluate each of the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme.

[0042] Based on the evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated.

[0043] The user task is executed according to the task planning scheme, the task execution result is output, and the task execution result is fed back to the sender of the user task.

[0044] Seventhly, this application also provides a task processing apparatus, comprising:

[0045] The third planning module is used to perform initial planning on the user task when a user task is received, and to obtain planning steps.

[0046] The second selection module is used to select multiple specialized intelligent agents based on the planning bifurcation step when there is a planning bifurcation step in the planning step.

[0047] The fourth planning module is used to plan feasible subsequent paths for the planning fork step by the multiple professional intelligent agents, and obtain the candidate planning schemes for the planning fork step generated by each of the multiple professional intelligent agents.

[0048] The second quantitative evaluation module is used to perform quantitative evaluation on multiple candidate planning schemes respectively, and obtain the evaluation results of each of the multiple candidate planning schemes.

[0049] The second scheme generation module is used to generate a task planning scheme for the user task based on the scheme evaluation results of the multiple candidate planning schemes.

[0050] The task execution module is used to execute the user task according to the task planning scheme, output the task execution result, and feed back the task execution result to the sender of the user task.

[0051] Eighthly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described task processing method.

[0052] Ninthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described task processing method.

[0053] In a tenth aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described task processing method.

[0054] The aforementioned task processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product, upon receiving a user task, perform initial planning for the user task to obtain planning steps. This initial planning is completed. If a planning fork step exists within the planning steps, multiple specialized intelligent agents are selected based on the planning fork step. These agents plan feasible subsequent paths for each planning fork step, resulting in candidate planning schemes generated by each agent for that fork step. When encountering a planning fork, multiple specialized intelligent agents are introduced to perform parallel planning for the fork step, utilizing collective intelligence to reduce the uncertainty and randomness of the planning scheme. By quantitatively evaluating each of the multiple candidate planning schemes, evaluation results are obtained for each scheme. This quantitative evaluation method allows for objective evaluation of the multiple candidate planning schemes. Based on the evaluation results of each candidate scheme, a task planning scheme for the user task is generated. The user task is executed according to the task planning scheme, and the task execution result is output and fed back to the sender of the user task. Throughout the process, when encountering planning bifurcation during the planning process, the uncertainty and randomness of the planning scheme can be reduced by using multiple professional intelligent agents to plan in parallel and to quantitatively evaluate the candidate solutions generated by each professional intelligent agent. This improves the quality of task planning decisions at the planning bifurcation step, thereby increasing the accuracy of task planning decisions for user tasks. User tasks are executed according to the task planning scheme, and the task execution results are output and feedback is provided, enabling accurate processing of user tasks. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a diagram illustrating the application environment of the task planning method in one embodiment;

[0057] Figure 2 This is a flowchart illustrating a task planning method in one embodiment;

[0058] Figure 3 This is a flowchart illustrating the calculation of confidence level in one embodiment;

[0059] Figure 4 This is a schematic diagram of a user decision-making interface in one embodiment;

[0060] Figure 5This is a schematic diagram comparing the evaluation data of the schemes in one embodiment;

[0061] Figure 6 This is a schematic diagram of the user decision-making interface in another embodiment;

[0062] Figure 7 This is a schematic diagram of the process for obtaining model evaluation scores in one embodiment;

[0063] Figure 8 This is a schematic diagram of the process for obtaining rule evaluation scores in one embodiment;

[0064] Figure 9 This is an application environment diagram of a task processing method in one embodiment;

[0065] Figure 10 This is a flowchart illustrating a task processing method in one embodiment;

[0066] Figure 11 This is a flowchart illustrating the task planning method in another embodiment;

[0067] Figure 12 This is a schematic diagram of the user decision-making interface in yet another embodiment;

[0068] Figure 13 This is a core architecture diagram of a task planning method in one embodiment;

[0069] Figure 14 This is a flowchart illustrating the workflow of a specialized intelligent agent pool in one embodiment.

[0070] Figure 15 Here is a flowchart of the voting consensus process in one embodiment;

[0071] Figure 16 Here is a flowchart of the feedback optimization process in one embodiment;

[0072] Figure 17 Here is a flowchart of the task planning system in one embodiment;

[0073] Figure 18 This is a structural block diagram of a task planning device in one embodiment;

[0074] Figure 19 This is a structural block diagram of a task processing device in one embodiment;

[0075] Figure 20 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0076] 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.

[0077] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0078] In order to clearly describe the technical solution of this application and facilitate understanding of the technical solution of this application, the key concepts involved in this application will be explained below.

[0079] 1. Planning steps.

[0080] A planning step refers to a standardized, executable task unit extracted from a user task using the reasoning and planning capabilities of a large language model. It is represented in a structured form and typically includes key information such as task title, task description, and dependencies (execution order). Because it is a step in the overall planning process, it is called a planning step.

[0081] 2. Plan the branching steps.

[0082] A forking step is a planning step where multiple feasible follow-up paths exist, meaning there are multiple strategy options. A feasible follow-up path is a logically sound and executable solution for the forking step, based on its task objectives and context; in other words, a planning scheme for the forking step. It's understandable that different feasible follow-up paths differ in methodology, perspective, or specific content, thus constituting different strategy choices.

[0083] 3. Professional intelligent agent.

[0084] A specialized intelligent agent refers to an intelligent agent that focuses on a specific domain or perspective, and can generate independent solutions or plans for specific problems. In this embodiment, the specialized intelligent agent can plan independent solutions, i.e., candidate planning schemes, for each planning branching step.

[0085] 4. Confidence level.

[0086] Confidence level refers to a quantitative or qualitative measure of the degree of certainty regarding the reliability, rationality, and effectiveness of a recommended planning scheme. Understandably, a higher confidence level indicates stronger reliability, rationality, and effectiveness, meaning the scheme can be trusted. Conversely, a lower confidence level indicates weaker reliability, rationality, and effectiveness, meaning the scheme cannot be trusted.

[0087] The task planning method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another network server. When a user task is received from terminal 102, server 104 performs initial planning for the user task, obtaining planning steps. If there are planning fork steps within the planning steps, multiple specialized intelligent agents are selected based on these fork steps. Each specialized intelligent agent plans a feasible subsequent path for the planning fork step, resulting in candidate planning schemes generated by each agent for that fork step. These candidate planning schemes are then quantitatively evaluated to obtain their respective evaluation results. Based on these evaluation results, a task planning scheme for the user task is generated.

[0088] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle systems, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0089] In one exemplary embodiment, such as Figure 2 As shown, a task planning method is provided. This embodiment illustrates the method applied to a server, but it is understood that the method can also be applied to a terminal, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes steps 202 to 210. Wherein:

[0090] Step 202: Upon receiving a user task, perform initial planning for the user task to obtain the planning steps.

[0091] In this context, a user task refers to a query request issued by a user to obtain specific information, typically accompanied by a user question. A user question is a question entered by the user to query or retrieve specific content. For example, a user question can be multiple keywords, a question, or a description in natural language.

[0092] The planning step refers to the standardized, executable task unit extracted from the user task through the reasoning and planning capabilities of the large language model. It is represented in a structured form and typically includes key information such as the task title, task description, and dependencies (execution order). Because it is a step in the overall planning process, it is called a planning step.

[0093] For example, when a user task is received, the server performs an initial planning of the user task to obtain the planning steps.

[0094] Step 204: If there is a planning bifurcation step in the planning steps, select multiple specialized intelligent agents based on the planning bifurcation step.

[0095] In this context, "planning bifurcation" refers to a scenario in task planning where multiple feasible subsequent paths or strategy choices are pre-identifiable, and a single model cannot directly determine the optimal solution. A planning bifurcation step refers to a planning step with multiple feasible subsequent paths, i.e., multiple strategy choices. A feasible subsequent path refers to a logically valid and executable solution direction for the planning bifurcation step, based on the task objective and context of the planning step; in other words, a planning scheme for the planning bifurcation step. It is understandable that different feasible subsequent paths differ in methodology, perspective, or specific content, thus constituting different strategy choices.

[0096] In this context, a specialized intelligent agent refers to an agent focused on a specific domain or perspective, capable of generating independent solutions or plans for specific problems. In this embodiment, the specialized intelligent agent can plan independent solutions, i.e., candidate plans, for each planning branching step. It is understood that the specialized intelligent agent is built upon a large language model, which refers to a new generation of artificial intelligence models possessing large-scale natural language understanding and generation capabilities.

[0097] For example, the specific domains or perspectives mentioned here could be search, analysis, programming, design, creativity, etc., and the specialized intelligent agents could be strategic intelligent agents, creative intelligent agents, domain expert intelligent agents, etc. Strategic intelligent agents are designed with task processing prompts prioritizing the path to achieving the goal, resource allocation, and risk avoidance, favoring robust and executable solutions. Creative intelligent agents are designed with task processing prompts designed to encourage divergent thinking, break conventional constraints, and strive to discover novel and non-obvious solutions. Domain expert intelligent agents are configured with detailed knowledge bases and thinking patterns for specific domains (such as shopping and programming), primarily used to provide professional and authoritative solutions.

[0098] For example, during the initial planning of a user task, the server will synchronously pre-identify the planning fork steps in the planning process. If there are planning fork steps in the planning process, the server will select multiple specialized agents based on the planning fork steps so that the planning fork steps can be planned by multiple specialized agents.

[0099] In practical applications, the server can guide a pre-trained large language model to perform initial planning for a user task using initial planning prompts, thus obtaining planning steps. These initial planning prompts may include descriptions guiding the identification of planning fork steps. Based on these descriptions, the pre-trained large language model can simultaneously pre-identify the planning fork steps during the initial planning process.

[0100] In a specific application, the description of guided identification of bifurcation planning steps can be a description of multiple strategy options, such as determining whether relevant information is explicitly provided in the user task from the perspectives of the first, second, and third aspects. To further illustrate, taking a shopping scenario as an example, the description of guided identification of bifurcation planning steps could be: determining whether alternative products are explicitly provided in the user task from the perspectives of user needs, usage scenarios, and constraints. Based on this description, the pre-trained large language model can determine whether determining whether alternative products are explicitly provided in the user task is a bifurcation planning step, and can make this judgment based on user needs, usage scenarios, and constraints.

[0101] In a specific application, the description guiding the identification of planning fork steps can specifically be a description of the method for identifying planning fork steps. For example, you can combine historical tasks of the same type to synchronously identify planning fork steps in the pre-identified planning process. Based on this description, the pre-trained large language model can, given the task type of the user's task, obtain task planning schemes of historical tasks of the same type, and combine these historical task planning schemes to synchronously identify planning fork steps in the pre-identified planning process. Specifically, if a planning fork step exists in the task planning scheme of a historical task, then a similar step in the planning process can be identified as a planning fork step.

[0102] To further illustrate, taking the shopping scenario as an example, if the task planning scheme of the historical task provides a plan for whether the user task clearly provides alternative products based on the user's needs, usage scenarios, and conditional constraints, then determining whether the user task clearly provides alternative products is a planning branching step, which can be done by considering the user's needs, usage scenarios, and conditional constraints.

[0103] Step 206: Through multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, resulting in candidate planning schemes generated by each specialized intelligent agent for the planning fork step.

[0104] For example, the server can use multiple specialized intelligent agents to plan feasible subsequent paths for the planning fork step, thereby obtaining candidate planning schemes generated by each specialized intelligent agent for the planning fork step.

[0105] In practical applications, the server will send a unified context and instructions to multiple specialized intelligent agents, enabling each agent to plan feasible subsequent paths for the forking step based on the unified context and instructions, thus obtaining candidate planning schemes generated by each agent for the forking step.

[0106] Step 208: Quantitatively evaluate the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme.

[0107] Quantitative evaluation refers to the objective analysis and comparison of candidate planning schemes through clearly defined indicators and quantitative methods. The evaluation result refers to the objective reflection of the merits of the candidate planning schemes after quantitative evaluation. For example, the evaluation result can be a scheme evaluation score. Another example is the scheme evaluation priority.

[0108] For example, when multiple professional intelligent agents generate candidate planning schemes for the planning bifurcation step, the server will perform quantitative evaluation on each of the multiple candidate planning schemes to achieve an objective evaluation of the multiple candidate planning schemes and obtain the scheme evaluation results of each of the multiple candidate planning schemes.

[0109] In practical applications, the server can quantitatively evaluate multiple candidate planning schemes based on predefined quantitative evaluation rules, obtaining rule evaluation scores for each scheme and using these scores as the final evaluation result. Alternatively, the server can use a large language model, based on predefined quantitative evaluation dimensions, to quantitatively evaluate multiple candidate planning schemes, obtaining model evaluation scores for each scheme and using these scores as the final evaluation result.

[0110] In practical applications, the server can also perform quantitative evaluation on multiple candidate planning schemes based on predefined quantitative evaluation rules to obtain the rule evaluation scores of each candidate planning scheme. Then, through a large language model, based on predefined quantitative evaluation dimensions, it can perform quantitative evaluation on multiple candidate planning schemes to obtain the model evaluation scores of each candidate planning scheme. Finally, based on the rule evaluation scores and model evaluation scores of each candidate planning scheme, the scheme evaluation results of each candidate planning scheme are obtained.

[0111] Step 210: Based on the evaluation results of multiple candidate planning schemes, generate a task planning scheme for the user task.

[0112] Among them, the task planning scheme refers to the complete plan that can achieve the user's task objectives after systematic planning and reasoning.

[0113] For example, based on the evaluation results of multiple candidate planning schemes, the server can determine the recommended planning scheme for the planning branching step from the multiple candidate planning schemes, and then generate the task planning scheme for the user task based on the recommended planning scheme.

[0114] In practical applications, the evaluation results can include evaluation scores. The server can then determine the candidate planning scheme with the highest evaluation score from among multiple candidate planning schemes, using this score as the recommended planning scheme for the planning fork step. The evaluation results can also include evaluation priorities. The server can then determine the candidate planning scheme with the highest evaluation priority from among multiple candidate planning schemes, using this priority as the recommended planning scheme for the planning fork step.

[0115] In practical applications, once the recommended planning scheme for the planning fork step is determined, the server will request the user to make a decision based on the recommended planning scheme, that is, let the user choose the first planning scheme for the planning fork step. After obtaining the user's decision result, the server will generate the task planning scheme for the user's task based on the user's decision result.

[0116] In a specific application, the method of requesting a user decision based on the recommended planning scheme can be as follows: allowing the user to confirm whether to accept the recommended planning scheme as the first planning scheme for the planning fork step; allowing the user to select the first planning scheme for the planning fork step from multiple candidate planning schemes including the recommended planning scheme; or allowing the user to confirm whether to reject multiple candidate planning schemes as the first planning scheme for the planning fork step. Furthermore, once the first planning scheme for the planning fork step is obtained, the server will integrate it into the user's task planning scheme to generate the user's task planning scheme.

[0117] The aforementioned task planning method, upon receiving a user task, performs initial planning to obtain planning steps. This initial planning is completed. When a planning fork occurs, multiple specialized agents are selected based on this fork. These agents plan feasible subsequent paths for each fork, resulting in candidate planning schemes generated by each agent. This allows for parallel planning of the fork by multiple agents, leveraging collective intelligence to reduce uncertainty and randomness. Quantitative evaluation of each candidate scheme yields its own evaluation result, enabling objective assessment. Based on these evaluations, a task planning scheme for the user task is generated. Throughout this process, when encountering planning forks, the parallel planning by multiple agents and the quantitative evaluation of their candidate solutions reduce uncertainty and randomness, improving the quality of task planning decisions at fork steps and ultimately increasing the accuracy of task planning decisions for the user task.

[0118] In an exemplary embodiment, a task planning scheme for the user task is generated based on the evaluation results of multiple candidate planning schemes, including:

[0119] Based on the evaluation results of multiple candidate planning schemes, the recommended planning scheme for the planning bifurcation step is determined;

[0120] The confidence level of the recommended planning scheme is determined based on the evaluation results of the recommended planning scheme and the evaluation results of the remaining planning schemes among multiple candidate planning schemes.

[0121] Based on the confidence level of the recommended planning scheme, request the user to make a decision, obtain the user's decision result, and generate a task planning scheme for the user's task based on the user's decision result.

[0122] Confidence level refers to a quantitative or qualitative measure of the degree of certainty regarding the reliability, rationality, and effectiveness of a recommended planning scheme. Understandably, a higher confidence level indicates stronger reliability, rationality, and effectiveness, meaning the scheme can be trusted. Conversely, a lower confidence level indicates weaker reliability, rationality, and effectiveness, meaning the scheme cannot be trusted.

[0123] For example, based on the evaluation results of multiple candidate planning schemes, the server selects a recommended planning scheme for the planning fork step from among the candidate planning schemes. After obtaining the recommended planning scheme, the server needs to evaluate its reliability. At this point, the server determines the confidence level of the recommended planning scheme based on its evaluation results and the evaluation results of the remaining planning schemes among the multiple candidate planning schemes. Then, based on the confidence level of the recommended planning scheme, the server requests the user to make a decision, allowing the user to select the first planning scheme for the planning fork step. Based on the user's decision, the server generates a task planning scheme for the user's task.

[0124] In practical applications, when requesting a user to make a decision based on the confidence level of a recommended planning scheme, the server compares the confidence level of the recommended scheme with a predefined confidence threshold. If the confidence level of the recommended scheme is greater than the predefined confidence threshold, the recommended scheme is considered reliable, and the server will request the user to confirm whether to use the recommended scheme as the first planning scheme for the forking step. If the confidence level of the recommended scheme is less than or equal to the predefined confidence threshold, the recommended scheme is considered unreliable, and the server will request the user to select the first planning scheme for the forking step from multiple candidate schemes.

[0125] In practical applications, upon receiving the user's decision result, if the user chooses to confirm the recommended planning scheme as the first planning scheme for the planning fork step, or selects the first planning scheme for the planning fork step from multiple candidate planning schemes, the server can integrate the user-selected first planning scheme for the planning fork step into the user task's task planning scheme to generate the user task's task planning scheme. If the user's decision result indicates that the user rejects multiple candidate planning schemes as the first planning scheme for the planning fork step, the server needs to obtain the user's instructions from the user's decision result, and then perform secondary task planning based on the obtained user instructions to generate the user task's task planning scheme.

[0126] In this embodiment, by utilizing the evaluation results of multiple candidate planning schemes, the recommended planning scheme for the planning bifurcation step can be determined. By utilizing the evaluation results of the recommended planning scheme and the evaluation results of the remaining planning schemes among the multiple candidate planning schemes, the confidence level of the recommended planning scheme can be calculated. Then, the confidence level of the recommended planning scheme can be used to request user decision-making. Based on the user's decision-making result, a task planning scheme for the user task can be generated. The entire process, through the mode of evaluation first and then user decision-making, generates a task planning scheme, which can improve the accuracy of task planning decisions for the user task.

[0127] In an exemplary embodiment, the scheme evaluation result includes a scheme evaluation score; based on the scheme evaluation result of the recommended planning scheme and the scheme evaluation results of the remaining planning schemes among multiple candidate planning schemes, the confidence level of the recommended planning scheme is determined, including:

[0128] Based on the evaluation scores of the remaining planning schemes among the candidate planning schemes, the evaluation scores of the schemes to be compared are determined.

[0129] Calculate the evaluation score of the recommended planning scheme and the score difference between the evaluation scores of the schemes to be compared;

[0130] Calculate the confidence level of the recommended planning scheme based on the score difference.

[0131] For example, the evaluation results include an evaluation score for the chosen option, such as... Figure 3As shown, when calculating the confidence level of the recommended planning scheme, the server first determines the evaluation score of the scheme to be compared based on the evaluation scores of the remaining planning schemes among multiple candidate planning schemes. Then, it calculates the evaluation score of the recommended planning scheme and the score difference between the evaluation scores of the schemes to be compared. Using the score difference and a predefined confidence level calculation formula, the confidence level of the recommended planning scheme is calculated. The predefined confidence level calculation formula can be configured according to the actual application scenario. It can be understood that the confidence level of the recommended planning scheme can be a value between 0 and 1, with a confidence level closer to 0 indicating lower confidence and a confidence level closer to 1 indicating higher confidence.

[0132] In specific applications, such as Figure 3 As shown, when determining the evaluation score of the scheme to be compared, the server can sort the evaluation scores of the remaining planning schemes among multiple candidate planning schemes, and select the highest evaluation score as the evaluation score of the scheme to be compared. Alternatively, the server can calculate the average evaluation score of the remaining planning schemes among multiple candidate planning schemes, and use the resulting average evaluation score as the evaluation score of the scheme to be compared.

[0133] In practical applications, a predefined formula for calculating confidence level is used: Confidence Level = (Highest Score / 10) * (Δscore / Highest Score). Here, the highest score refers to the evaluation score of the recommended planning scheme, and Δscore refers to the evaluation score of the scheme being compared. In essence, this method determines the confidence level of the recommended planning scheme by comparing its evaluation score with the evaluation scores of the remaining schemes among the candidate schemes. A larger difference indicates a higher confidence level and greater reliability, while a smaller difference indicates a lower confidence level and less reliability.

[0134] In this embodiment, when the scheme evaluation result includes scheme evaluation scores, the scheme evaluation score to be compared is determined based on the scheme evaluation scores of the remaining planning schemes among multiple candidate planning schemes. The confidence level of the recommended planning scheme can be accurately calculated by using the scheme evaluation scores of the recommended planning scheme and the score difference between the scheme evaluation scores to be compared.

[0135] In an exemplary embodiment, a user decision is requested based on the confidence level of the recommended planning scheme, the user's decision result is obtained, and a task planning scheme for the user's task is generated based on the user's decision result, including:

[0136] If the confidence level of the recommended planning scheme is greater than a predefined confidence threshold, a first user decision request is generated; the first user decision request is used to indicate that the recommended planning scheme should be used as the first planning scheme for the planning fork step.

[0137] Obtain the user's decision result. If the user's decision result indicates acceptance of the recommended planning scheme, use the recommended planning scheme as the first planning scheme for the planning fork step.

[0138] Based on the first planning scheme of the planning fork step and the second planning scheme of the remaining steps in the planning step, a task planning scheme for the user task is generated.

[0139] For example, if the confidence level of the recommended planning scheme is greater than a predefined confidence threshold, it indicates that the recommended planning scheme is credible. The server will then generate a first user decision request to instruct the user to confirm whether to adopt the recommended planning scheme as the first planning scheme for the planning fork step. After receiving the first user decision request, the user's terminal will display a user decision interface based on the request. The user can provide feedback on their decision through interactive operations on the user decision interface. After obtaining the user decision result based on the first user decision request, the server can determine whether the user chooses to accept, modify, or reject the recommended planning scheme. If the user's decision indicates acceptance, the server will adopt the recommended planning scheme as the first planning scheme for the planning fork step and further generate a task planning scheme for the user's task based on the first planning scheme for the planning fork step and the second planning schemes for the remaining steps in the planning process. The predefined confidence threshold can be configured according to the actual application scenario. For example, the predefined confidence threshold can be 0.75.

[0140] In practical applications, when generating the first user decision request, the server generates a "planning summary" for the recommended planning scheme, highlighting its core ideas, main advantages, and potential risks to assist the user's decision-making. In a specific application, the user decision interface might look like this: Figure 4 As shown, this includes details of the recommended planning scheme (i.e., the core idea), its main advantages and potential risks, a comparison of evaluation data for multiple candidate planning schemes, multiple selection options, and a feedback area. Among these, for example... Figure 4As shown, the various selection options can include "Confirm this plan," "Modify this plan," and "Reject this plan." "Confirm this plan" indicates that the user accepts the recommended planning plan; "Modify this plan" indicates that the user modifies the recommended planning plan; and "Reject this plan" indicates that the user rejects the recommended planning plan. Furthermore, if the user chooses to modify the recommended planning plan, they can enter their suggestions in the feedback area; if the user chooses to reject the recommended planning plan, they can enter user commands for secondary planning in the feedback area.

[0141] Furthermore, the multiple candidate planning schemes are categorized into three candidate planning schemes: Scheme A, Scheme B, and Scheme C. Scheme C is an example of the recommended planning scheme, such as... Figure 5 As shown, the evaluation data comparison of this scheme can be presented in tabular form, including the scores of multiple candidate planning schemes on multiple evaluation indicators. For example, Figure 5 As shown, the various evaluation indicators may include feasibility score, innovation score, user experience score, cost estimate (X, which can be a cost unit), and time efficiency (hours), etc.

[0142] Understandably, by viewing the details (i.e., core ideas), main advantages, and potential risks of the recommended planning scheme displayed in the user decision-making interface, users can make accurate decisions about whether the recommended planning scheme is suitable. By comparing the scheme evaluation data displayed in the user decision-making interface, users can intuitively compare multiple candidate planning schemes and make accurate decisions about whether the recommended planning scheme is suitable.

[0143] In this embodiment, the confidence level of the recommended planning scheme can be used to accurately determine whether the recommended planning scheme is reliable. If the confidence level of the recommended planning scheme is greater than a predefined confidence level threshold, a first user decision request is generated, allowing the user to further confirm whether to use the recommended planning scheme as the first planning scheme for the planning fork step. Based on the user's confirmation of using the recommended planning scheme as the first planning scheme for the planning fork step, a task planning scheme for the user task is generated based on the first planning scheme for the planning fork step and the second planning schemes for the remaining steps in the planning process. By using the confidence level of the recommended planning scheme to allow the user to directly confirm the first planning scheme for the planning fork step, the task planning scheme can be generated, which can improve the accuracy of the user task's task planning decision.

[0144] In one exemplary embodiment, the task planning method further includes:

[0145] When the recommended planning scheme is modified based on the user's decision result, the first planning scheme for the planning bifurcation step is generated.

[0146] Based on the first planning scheme of the planning fork step and the second planning scheme of the remaining steps in the planning step, a task planning scheme for the user task is generated.

[0147] For example, when the user's decision result indicates a modification to the recommended planning scheme, the server extracts the user's feedback from the decision result, modifies the recommended planning scheme based on the feedback, generates a first planning scheme for the planning fork step, and further generates a task planning scheme for the user task based on the first planning scheme for the planning fork step and the second planning scheme for the remaining steps in the planning process. Specifically, the user's feedback can be the modification suggestions entered by the user through the user decision interface.

[0148] In practical applications, when modifying the recommended planning scheme based on the modification opinions, the server can input the modification opinions and the recommended planning scheme into the pre-trained large language model, so that the pre-trained large language model can modify the recommended planning scheme based on the modification opinions and output the first planning scheme of the planning fork step.

[0149] In this embodiment, when the user's decision result characterizes the modification of the recommended planning scheme, modifying the recommended planning scheme based on the user's decision result can utilize user feedback to determine the first planning scheme for the planning fork step, ensuring the correctness of the planning decision for the planning fork step. Then, based on the first planning scheme for the planning fork step and the second planning scheme for the remaining steps in the planning step, a task planning scheme for the user task can be generated.

[0150] In one exemplary embodiment, the task planning method further includes:

[0151] If the user's decision result indicates rejection of the recommended planning scheme, obtain the user's instructions from the user's decision result;

[0152] Based on the obtained user instructions, secondary task planning is performed to generate a task planning scheme for the user task.

[0153] For example, if the user's decision result indicates rejection of the recommended planning scheme, the server will extract the user's feedback instructions from the user's decision result, and perform secondary task planning based on the obtained user instructions to generate a task planning scheme for the user task.

[0154] In practical applications, secondary task planning is performed based on the obtained user instructions, that is, task planning is re-performed. At this time, the server can combine the obtained user instructions to re-plan the user task and generate a task planning scheme for the user task.

[0155] Understandably, when a user chooses to reject the recommended planning scheme, it means that the user is not satisfied with the recommended planning scheme. At this time, the user can provide feedback containing the user's instructions to help the server understand the user's needs. Once the server receives the user's instructions, it can analyze the user's instructions to understand the user's needs and use the user's instructions to achieve accurate secondary planning of the user's task.

[0156] In a specific application, when performing secondary task planning, the server still first performs initial planning to obtain planning steps. Then, if there are planning fork steps in the planning steps, multiple specialized intelligent agents are selected based on the planning fork steps. Each of these specialized intelligent agents plans a feasible subsequent path for the planning fork step, resulting in candidate planning schemes generated by each specialized intelligent agent for the planning fork step. The multiple candidate planning schemes are then quantitatively evaluated to obtain the scheme evaluation results for each of the multiple candidate planning schemes. Based on the scheme evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated.

[0157] In this embodiment, when the user's decision result indicates rejection of the recommended planning scheme, the user's instructions are obtained from the user's decision result, and secondary task planning is performed based on the obtained user instructions to generate a task planning scheme for the user's task. This can utilize user feedback to achieve accurate generation of task planning schemes and ensure the correctness of task planning decisions.

[0158] In one exemplary embodiment, the task planning method further includes:

[0159] When the user's decision result indicates rejection of the recommended planning scheme, a secondary decision request is generated based on the remaining planning schemes among multiple candidate planning schemes; the secondary decision request is used to indicate the first planning scheme for selecting the planning bifurcation step from the remaining planning schemes.

[0160] Obtain the user's secondary decision result. If the user's secondary decision result indicates that any candidate planning scheme is selected, determine the first planning scheme for the planning bifurcation step based on the selected candidate planning scheme.

[0161] Based on the first planning scheme of the planning fork step and the second planning scheme of the remaining steps in the planning step, a task planning scheme for the user task is generated.

[0162] For example, if the user's decision indicates rejection of the recommended planning scheme, the server generates a secondary decision request based on the remaining planning schemes among multiple candidate planning schemes. This request instructs the user to select the first planning scheme for the planning fork step from the remaining planning schemes. Upon receiving the secondary decision request, the user's terminal displays a user decision interface based on the request. The user can provide feedback on their secondary decision through interactive operations on the user decision interface. After obtaining the user's secondary decision feedback based on the secondary decision request, the server determines whether the user has selected any candidate planning scheme. If the user's secondary decision indicates selection of any candidate planning scheme, the server determines the first planning scheme for the planning fork step based on the selected candidate planning scheme. Furthermore, based on the first planning scheme for the planning fork step and the second planning scheme for the remaining steps in the planning process, the server generates a task planning scheme for the user's task.

[0163] In practical applications, when generating a secondary decision request, the server generates a "planning summary" for the remaining planning schemes, highlighting their core ideas, main advantages, and potential risks to assist the user in making a secondary decision. In a specific application, taking the remaining planning schemes as an example involving multiple schemes, the user decision interface might look like this: Figure 6 As shown, the interface displays the "planning summary" (i.e., scheme details (core ideas), main advantages, and potential risks) of the remaining planning schemes, including the scheme with the highest evaluation score or the highest evaluation priority; a comparison of scheme evaluation data for multiple candidate planning schemes; multiple selection options; and a feedback area. Understandably, while the user decision-making interface displays the "planning summary" of the remaining planning scheme with the highest evaluation score or the highest evaluation priority, users can view the "planning summaries" of other remaining planning schemes through interactive operations on the user decision-making interface.

[0164] Among these options, users can choose from the current planning scheme displayed on the user decision-making interface. Furthermore, such as... Figure 6 As shown, the multiple selection options can specifically include "Select this plan," "Modify this plan," and "Reject this plan." "Select this plan" indicates that the user chooses the current planning plan and does not need to modify it; "Modify this plan" indicates that the user chooses the current planning plan and needs to modify it; and "Reject this plan" indicates that the user rejects the current planning plan. Furthermore, if the user chooses to modify the current planning plan, they can enter their suggestions in the feedback area; if the user chooses to reject the current planning plan, they can enter user commands for secondary planning in the feedback area.

[0165] In practical applications, when determining the first planning scheme for the planning fork step based on the selected candidate planning schemes, if the user's secondary decision indicates that any candidate planning scheme is selected and no modification is needed, the server will determine the selected candidate planning scheme as the first planning scheme for the planning fork step. If the user's secondary decision indicates that any candidate planning scheme is selected and modification is needed, the server will extract the user's feedback on modification suggestions from the user's secondary decision results and modify the selected candidate planning scheme based on the modification suggestions to generate the first planning scheme for the planning fork step. Specifically, the user's feedback on modification suggestions can be the modification suggestions entered by the user through the user decision interface.

[0166] In a specific application, when modifying the selected candidate planning scheme based on the modification opinions, the server can input the modification opinions and the selected candidate planning scheme into the pre-trained large language model, so that the pre-trained large language model can modify the selected candidate planning scheme based on the modification opinions and output the first planning scheme of the planning fork step.

[0167] In this embodiment, this approach allows users to choose from the remaining planning schemes among multiple candidate schemes when they reject the recommended planning scheme. It enables the determination of the first planning scheme for the planning fork step by utilizing user decisions. Furthermore, based on the first planning scheme for the planning fork step and the second planning scheme for the remaining steps in the planning process, a task planning scheme for the user's task can be generated. This approach also enables the accurate generation of the task planning scheme by utilizing user feedback, ensuring the correctness of the task planning decision.

[0168] In an exemplary embodiment, a user decision is requested based on the confidence level of the recommended planning scheme, the user's decision result is obtained, and a task planning scheme for the user's task is generated based on the user's decision result, including:

[0169] If the confidence level of the recommended planning scheme is less than or equal to a predefined confidence threshold, a second user decision request is generated; the second user decision request is used to indicate the first planning scheme for selecting the planning bifurcation step from multiple candidate planning schemes.

[0170] Obtain the user's decision result. If the user's decision result indicates that any candidate planning scheme is selected, determine the first planning scheme for the planning fork step based on the selected candidate planning scheme.

[0171] Based on the first planning scheme of the planning fork step and the second planning scheme of the remaining steps in the planning step, a task planning scheme for the user task is generated.

[0172] For example, if the confidence level of the recommended planning scheme is less than or equal to a predefined confidence threshold, indicating that the recommended planning scheme is unreliable, the server will generate a second user decision request to instruct the user to select the first planning scheme for the planning fork step from multiple candidate planning schemes. Upon receiving the second user decision request, the user's terminal will display a user decision interface based on the request. The user can provide feedback on their decision through interactive operations on the user decision interface. After obtaining the user decision result based on the second user decision request, the server can determine whether the user has selected any candidate planning scheme. If the user decision indicates selection of any candidate planning scheme, the server determines the first planning scheme for the planning fork step based on the selected candidate planning scheme, and further generates a task planning scheme for the user's task based on the first planning scheme for the planning fork step and the second planning schemes for the remaining steps in the planning process. The predefined confidence threshold can be configured according to the actual application scenario. For example, the predefined confidence threshold can be 0.75.

[0173] In practical applications, when generating a second user decision request, the server generates a "planning summary" for each of the multiple candidate planning schemes, highlighting its core ideas, main advantages, and potential risks to assist the user's decision-making. In a specific application, the user decision interface first displays the "planning summary" of the recommended planning scheme (i.e., scheme details (i.e., core ideas), main advantages, and potential risks), a comparison of scheme evaluation data for multiple candidate planning schemes, multiple selection options, and a feedback area. Users can view the "planning summaries" of the remaining planning schemes from the multiple candidate schemes through interactive operations on the user decision interface. The multiple selection options allow users to choose the current planning scheme displayed on the user decision interface.

[0174] Furthermore, the multiple selection options can specifically include "Select this plan," "Modify this plan," and "Reject this plan." "Select this plan" indicates that the user chooses the current planning plan and does not need to modify it; "Modify this plan" indicates that the user chooses the current planning plan and needs to modify it; and "Reject this plan" indicates that the user rejects the current planning plan. Even further, if the user chooses to modify the current planning plan, they can enter their suggestions in the feedback area; if the user chooses to reject the current planning plan, they can enter user commands for secondary planning in the feedback area.

[0175] In practical applications, when determining the first planning scheme for the planning fork step based on the selected candidate planning schemes, if the user's decision indicates that any candidate planning scheme is selected and no modification is needed, the server will determine the selected candidate planning scheme as the first planning scheme for the planning fork step. If the user's decision indicates that any candidate planning scheme is selected and modification is needed, the server will extract the user's feedback on modification suggestions from the user's decision results and modify the selected candidate planning scheme based on the modification suggestions to generate the first planning scheme for the planning fork step. Specifically, the user's feedback on modification suggestions can be the modification suggestions entered by the user through the user decision interface.

[0176] In a specific application, when modifying the selected candidate planning scheme based on the modification opinions, the server can input the modification opinions and the selected candidate planning scheme into the pre-trained large language model, so that the pre-trained large language model can modify the selected candidate planning scheme based on the modification opinions and output the first planning scheme of the planning fork step.

[0177] In this embodiment, this approach allows for the direct provision of multiple candidate planning schemes to the user for selection when the confidence level of the recommended planning scheme is less than or equal to a predefined confidence threshold. It enables the determination of the first planning scheme for the planning bifurcation step using user decisions. Furthermore, based on the first planning scheme for the planning bifurcation step and the second planning scheme for the remaining steps in the planning process, a task planning scheme for the user's task can be generated. This approach utilizes user feedback to accurately generate the task planning scheme, ensuring the correctness of the task planning decision.

[0178] In one exemplary embodiment, the task planning method further includes:

[0179] When the user's decision result represents the rejection of multiple candidate planning solutions, the user's instructions are obtained from the user's decision result;

[0180] Based on the obtained user instructions, secondary task planning is performed to generate a task planning scheme for the user task.

[0181] For example, if the user's decision result indicates rejection of multiple candidate planning schemes, the server will extract the user's feedback instructions from the user's decision result, and perform secondary task planning based on the obtained user instructions to generate a task planning scheme for the user task.

[0182] In practical applications, secondary task planning is performed based on the obtained user instructions, that is, task planning is re-performed. At this time, the server can combine the obtained user instructions to re-plan the user task and generate a task planning scheme for the user task.

[0183] Understandably, when a user chooses to reject multiple candidate planning schemes, it means that the user is not satisfied with the multiple candidate planning schemes. At this time, the user can provide feedback containing the user's instructions to the server to understand the user's needs. After obtaining the user's instructions, the server can analyze the user's instructions to understand the user's needs and use the user's instructions to achieve accurate secondary planning of the user's task.

[0184] In a specific application, when performing secondary task planning, the server still first performs initial planning to obtain planning steps. Then, if there are planning fork steps in the planning steps, multiple specialized intelligent agents are selected based on the planning fork steps. Each of these specialized intelligent agents plans a feasible subsequent path for the planning fork step, resulting in candidate planning schemes generated by each specialized intelligent agent for the planning fork step. The multiple candidate planning schemes are then quantitatively evaluated to obtain the scheme evaluation results for each of the multiple candidate planning schemes. Based on the scheme evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated.

[0185] In this embodiment, when the user's decision result indicates rejection of multiple candidate planning schemes, the user's instructions are obtained from the user's decision result, and secondary task planning is performed based on the obtained user instructions to generate a task planning scheme for the user's task. This can utilize user feedback to achieve accurate generation of task planning schemes and ensure the correctness of task planning decisions.

[0186] In an exemplary embodiment, multiple candidate planning schemes are quantitatively evaluated to obtain evaluation results for each candidate planning scheme, including:

[0187] Based on predefined quantitative evaluation rules, multiple candidate planning schemes are quantitatively evaluated to obtain their respective rule evaluation scores.

[0188] Using a large language model, based on predefined quantitative evaluation dimensions, multiple candidate planning schemes are quantitatively evaluated to obtain their respective model evaluation scores.

[0189] For each candidate planning scheme, the scheme evaluation result is obtained based on the rule evaluation score and model evaluation score of the candidate planning scheme.

[0190] Among them, quantitative evaluation rules refer to the evaluation rules that must be followed when quantitatively evaluating candidate planning schemes, and can be configured according to actual application scenarios. For example, quantitative evaluation rules can be predefined scheme scoring rules, which predefine multiple quantitative evaluation indicators, the scoring scale of each quantitative evaluation indicator, and the weight allocation of multiple quantitative evaluation indicators. Rule evaluation score refers to the evaluation score determined based on the quantitative evaluation rules for the candidate planning scheme.

[0191] The quantitative evaluation dimensions refer to the evaluation dimensions set when quantitatively evaluating candidate planning schemes, and can be configured according to the actual application scenario. For example, specific quantitative evaluation dimensions may include scheme feasibility, scheme execution efficiency, consistency between the scheme and the overall goal, scheme innovation, and scheme risk assessment. The model evaluation score refers to the evaluation score output by the large language model based on the quantitative evaluation dimensions when quantitatively evaluating candidate planning schemes.

[0192] For example, during the quantitative evaluation, the server performs quantitative evaluation on multiple candidate planning schemes based on predefined quantitative evaluation rules, obtaining rule evaluation scores for each candidate planning scheme. Then, using a large language model, based on predefined quantitative evaluation dimensions, it performs quantitative evaluation on multiple candidate planning schemes, obtaining model evaluation scores for each candidate planning scheme. Finally, for each candidate planning scheme, the scheme evaluation result can be obtained based on the rule evaluation score and model evaluation score of the candidate planning scheme.

[0193] In practical applications, for each candidate planning scheme, when using predefined quantitative evaluation rules to quantitatively evaluate the candidate planning scheme, the server can use multiple quantitative evaluation indicators in the quantitative evaluation rules, the scoring scale of each quantitative evaluation indicator, and the weight allocation of multiple quantitative evaluation indicators to score the candidate planning scheme and obtain the rule evaluation score of the candidate planning scheme.

[0194] In practical applications, for each candidate planning scheme, when the candidate planning scheme is quantitatively evaluated by the large language model, the prompt word template of the large language model will require it to summarize and score the candidate planning scheme based on the predefined quantitative evaluation dimensions. The score generated is the model evaluation score of the candidate planning scheme.

[0195] When requiring a large language model to provide scores, a scoring range must be specified, which can be configured according to the actual application scenario. For example, the scoring range can be from 1 to 10 points. It should be noted that this scoring range is the same as the scoring range used when using predefined quantitative evaluation rules to quantitatively evaluate candidate planning schemes; that is, the scoring range used for quantitative evaluation using quantitative evaluation rules and for quantitative evaluation using a large language model should be consistent.

[0196] In a specific application, such as Figure 7 As shown, for each candidate planning scheme, when quantitatively evaluating the candidate planning scheme using a large language model, if there is only one predefined quantitative evaluation dimension, the large language model will score the candidate planning scheme based on that single quantitative evaluation dimension, thus obtaining the model evaluation score of the candidate planning scheme. If there are multiple predefined quantitative evaluation dimensions, the large language model can score the candidate planning scheme based on each of the multiple quantitative evaluation dimensions, obtaining the quantitative evaluation dimension score corresponding to each of the multiple quantitative evaluation dimensions, and then obtaining the model evaluation score of the candidate planning scheme based on the quantitative evaluation dimension scores corresponding to each of the multiple quantitative evaluation dimensions.

[0197] In a specific application, such as Figure 7 As shown, after obtaining the scores for each of the multiple quantitative evaluation dimensions, the server can acquire the dimensional weights for each of these dimensions. Using these dimensional weights, the server performs a weighted calculation on the scores for each of the multiple quantitative evaluation dimensions to obtain the model evaluation score for the candidate planning scheme. The dimensional weights for each of the multiple quantitative evaluation dimensions can be configured according to the actual application scenario; the dimensional weights for different quantitative evaluation dimensions can be the same or different.

[0198] In practical applications, for each candidate planning scheme, the server can calculate a scheme evaluation score based on the rule evaluation score and model evaluation score of the candidate planning scheme, and then use the scheme evaluation score as the scheme evaluation result. The server can also determine the scheme evaluation priority of the candidate planning scheme based on the rule evaluation score and model evaluation score, and then use the scheme evaluation priority as the scheme evaluation score.

[0199] In a specific application, when determining the evaluation priority of candidate planning schemes, the server can combine the rule evaluation scores and model evaluation scores of the candidate planning schemes with predefined priority ranking rules to determine the evaluation priority. The predefined priority ranking rules can be configured according to the actual application scenario. For example, a predefined priority ranking rule could be to first sort by rule evaluation scores, and then, for at least two candidate planning schemes with the same rule evaluation score, sort by model evaluation score. Alternatively, a predefined priority ranking rule could be to first sort by model evaluation scores, and then, for at least two candidate planning schemes with the same model evaluation score, sort by rule evaluation scores.

[0200] In this embodiment, by combining quantitative evaluation of quantitative evaluation rules and quantitative evaluation of large language models, an objective evaluation of candidate planning schemes can be achieved, resulting in accurate evaluation results of candidate planning schemes.

[0201] In an exemplary embodiment, based on predefined quantitative evaluation rules, multiple candidate planning schemes are quantitatively evaluated to obtain rule evaluation scores for each candidate planning scheme, including:

[0202] For each candidate planning scheme, the candidate planning scheme is quantitatively evaluated based on multiple quantitative evaluation indicators in the predefined quantitative evaluation rules, and the quantitative evaluation scores corresponding to each of the multiple quantitative evaluation indicators are obtained.

[0203] Based on the evaluation scores of the various quantitative evaluation indicators, the rule evaluation scores of the candidate planning schemes are obtained.

[0204] Among them, quantitative evaluation indicators refer to the predefined evaluation indicators used when quantitatively evaluating candidate planning schemes based on quantitative evaluation rules. For example, quantitative evaluation indicators may specifically include the number of steps in the scheme, the estimated time consumption of the scheme, and the cost of the scheme. The quantitative indicator evaluation score refers to the evaluation score determined based on the quantitative evaluation indicators when quantitatively evaluating candidate planning schemes.

[0205] For example, for each candidate planning scheme, the server will perform quantitative evaluation on the candidate planning scheme based on multiple quantitative evaluation indicators in the predefined quantitative evaluation rules, obtain the quantitative indicator evaluation scores corresponding to each of the multiple quantitative evaluation indicators, and then obtain the rule evaluation score of the candidate planning scheme based on the quantitative indicator evaluation scores corresponding to each of the multiple quantitative evaluation indicators.

[0206] In practical applications, for each quantitative evaluation indicator, when quantitatively evaluating candidate planning schemes based on the quantitative evaluation indicator, the server obtains the scoring scale of the quantitative evaluation indicator, and then uses the scoring scale to quantitatively evaluate the candidate planning schemes, obtaining the corresponding quantitative evaluation indicator score. The scoring scale defines a specific scoring method, which can be configured according to the actual application scenario. For example, if the specific scoring method defined in the scoring scale can be a deduction item, then the quantitative evaluation method can be to start from the full score of the quantitative evaluation indicator, deduct points from the candidate planning schemes based on the deduction items, and obtain the corresponding quantitative evaluation indicator score.

[0207] In a specific application, such as Figure 8 As shown, the quantitative evaluation indicators can specifically include the number of steps in the solution. The server can determine the quantitative indicator evaluation score corresponding to the number of steps in the solution by comparing the number of steps in the solution with the average number of steps in similar solutions. The average number of steps in similar solutions can be obtained statistically. Further, as... Figure 8 As shown, the quantitative evaluation indicators can specifically include the estimated time consumption of the solution. The server can then determine the quantitative indicator evaluation score corresponding to the estimated time consumption by comparing it with the average time consumption of similar solutions. The average time consumption of similar solutions can be obtained through statistical analysis. Furthermore, as... Figure 8 As shown, specific quantitative evaluation indicators can include solution cost. The server can determine the solution cost by calculating the ratio of search time to execution time in the candidate planning solutions, and then determine the corresponding quantitative indicator evaluation score for that solution cost. Finally, as... Figure 8 As shown, after obtaining the quantitative evaluation scores for the number of steps in the plan, the estimated time of the plan, and the cost of the plan, the server will obtain the rule evaluation score of the candidate planning plan based on these three quantitative evaluation scores.

[0208] Understandably, a higher time consumption ratio indicates a higher solution cost, and a higher solution cost corresponds to a lower evaluation score. For example, the server can determine the evaluation score corresponding to the solution cost based on a predefined time consumption ratio and score correspondence table. In this table, different time consumption ratios correspond to different evaluation scores, and the correlation is negative. This time consumption ratio and score correspondence table can be configured according to the actual application scenario.

[0209] In practical applications, the full score of each quantitative evaluation indicator may be different. The sum of the full scores of multiple quantitative evaluation indicators is the full score of the quantitative evaluation rule. Based on the quantitative evaluation scores of each of the multiple quantitative evaluation indicators, the rule evaluation score of the candidate planning scheme can be obtained by directly accumulating the quantitative evaluation scores of each of the multiple quantitative evaluation indicators.

[0210] In practical applications, the full score of each quantitative evaluation indicator can be the same. Thus, the full score of each of the multiple quantitative evaluation indicators can be the same as the full score of the quantitative evaluation rule. Based on the quantitative evaluation scores corresponding to each of the multiple quantitative evaluation indicators, the server will obtain the weight allocation of the multiple quantitative evaluation indicators. Based on the weight allocation of the multiple quantitative evaluation indicators, the quantitative evaluation scores corresponding to each of the multiple quantitative evaluation indicators will be weighted and calculated to obtain the rule evaluation score of the candidate planning scheme.

[0211] In this embodiment, multiple quantitative evaluation indicators can be used to quantitatively evaluate candidate planning schemes from multiple perspectives, thereby achieving an objective evaluation of candidate planning schemes based on quantitative evaluation rules and obtaining accurate rule evaluation scores for candidate planning schemes.

[0212] In an exemplary embodiment, for each candidate planning scheme, an evaluation result is obtained based on the rule evaluation score and model evaluation score of the candidate planning scheme, including:

[0213] For each candidate planning scheme, the rule evaluation score and model evaluation score of the candidate planning scheme are weighted and calculated based on the predefined rule evaluation weight and model evaluation weight to obtain the scheme evaluation result of the candidate planning scheme;

[0214] Task planning methods also include:

[0215] Given any possible decision outcome, adjust the rule evaluation weights and model evaluation weights based on the obtained decision outcome.

[0216] The rule evaluation weight refers to the weight set for quantitative evaluation based on quantitative evaluation rules, and can be configured according to the actual application scenario. The model evaluation weight refers to the weight set for quantitative evaluation based on a large language model, and can also be configured according to the actual application scenario. It should be noted that when configuring, the sum of the rule evaluation weight and the model evaluation weight can be set to 1. For example, if the rule evaluation weight is 0.3, then the model evaluation weight is 0.7, meaning that the result of quantitative evaluation based on the large language model is the primary reference.

[0217] For example, for each candidate planning scheme, the server will perform a weighted calculation of the rule evaluation score and model evaluation score of the candidate planning scheme based on the predefined rule evaluation weight and model evaluation weight to obtain the scheme evaluation score of the candidate planning scheme, and then use the scheme evaluation score of the candidate planning scheme as the scheme evaluation result of the candidate planning scheme.

[0218] For example, by utilizing the evaluation results of multiple candidate planning schemes, a first planning scheme for the planning fork step can be determined from among them. Based on this, the server determines the confidence level of the first planning scheme and, based on that confidence level, requests the user to make a decision, granting the user the right to choose the first planning scheme for the planning fork step. In the event of any decision result, the server can generate a task planning scheme for the user's task based on the obtained decision result. Simultaneously, the server adjusts the rule evaluation weights and model evaluation weights based on the obtained decision result to optimize them. The obtained decision result can be the user's decision result, or a combination of the user's initial decision and a secondary decision result.

[0219] In practical applications, if the obtained decision results indicate acceptance of the recommended planning scheme, it means that the current rule evaluation weights and model evaluation weights are relatively reasonable. The server can appropriately increase the larger weight among the rule evaluation weights and model evaluation weights, and decrease the smaller weight among the rule evaluation weights and model evaluation weights.

[0220] In practical applications, if the obtained decision results represent either modifying the recommended planning scheme (i.e., using the recommended planning scheme as the basic scheme for the planning bifurcation step) or selecting any remaining planning scheme from multiple candidate planning schemes, it indicates that the current rule evaluation weights and model evaluation weights are somewhat unreasonable. The server can increase the smaller weight of the rule evaluation weights and model evaluation weights and decrease the larger weight of the rule evaluation weights and model evaluation weights.

[0221] In practical applications, if the obtained decision results indicate that the recommended planning scheme is rejected or multiple candidate planning schemes are rejected, it means that the current rule evaluation weights and model evaluation weights are relatively unreasonable, and the server needs to significantly adjust the rule evaluation weights and model evaluation weights.

[0222] In this embodiment, by using rule evaluation weights and model evaluation weights to perform weighted calculations on the rule evaluation scores and model evaluation scores, the evaluation results of candidate planning schemes can be accurately determined. In the case of obtaining any decision result, by adjusting the rule evaluation weights and model evaluation weights according to the obtained decision result, the rule evaluation weights and model evaluation weights can be optimized using the user's decision result, thereby optimizing the quantitative evaluation of candidate planning schemes and enhancing the objectivity of quantitative evaluation.

[0223] In an exemplary embodiment, when a planning bifurcation step exists in the planning process, multiple specialized agents are selected based on the planning bifurcation step, including:

[0224] When there are planning fork steps in the planning process, determine the task domain to which the planning fork step belongs;

[0225] Based on the task domain, the metadata tags of multiple candidate agents, and predefined scheduling strategies, agent matching is performed to select multiple specialized agents from the multiple candidate agents.

[0226] Task planning methods also include:

[0227] If any decision result is obtained, the scheduling strategy is adjusted according to the obtained decision result.

[0228] The task domain refers to the specific knowledge category, professional discipline, or activity type to which the user task belongs. It defines the scope of core knowledge, methodologies, and tools required to solve the user task, and can specifically be one of the predefined domains. Predefined domains can be configured according to actual application scenarios. For example, predefined domains may specifically include search, analysis, programming, and design.

[0229] Metadata tags are labels used to describe the capabilities and characteristics of candidate agents. For example, metadata tags can specifically describe the role definition of a candidate agent. For instance, metadata tags could be labels such as "strategist," "creative," or "domain expert."

[0230] The scheduling strategy refers to the policy used to describe how to schedule agents, which can be the matching relationship between the agent's metadata tags and the task domain. For example, the scheduling strategy can take the form of: metadata tag 1 - task domain 1, task domain 2; metadata tag 2 - task domain 2, task domain 3; metadata tag 3 - task domain 3, task domain 4.

[0231] For example, when multiple specialized agents need to be introduced to plan the fork step in parallel, the server first determines the task domain to which the user task belongs, and then matches agents according to the task domain, the metadata tags of the multiple candidate agents, and the predefined scheduling strategy, so as to select multiple specialized agents from the multiple candidate agents to plan the fork step in parallel.

[0232] For example, by selecting multiple specialized intelligent agents to perform parallel planning for the bifurcation step, the server can obtain multiple candidate planning schemes and, through quantitative evaluation, obtain the evaluation results of each candidate scheme. Using these evaluation results, a first planning scheme for the bifurcation step can be determined. Based on this, the server determines the confidence level of the first planning scheme and, based on that confidence level, requests the user to make a decision, granting the user the right to choose the first planning scheme for the bifurcation step. In the event of any decision result, the server can generate a task planning scheme for the user's task based on that decision. Simultaneously, the server adjusts the scheduling strategy based on the obtained decision results to optimize the scheduling strategy. The obtained decision results can be the user's decision result, or a combination of the user's initial decision and a secondary decision.

[0233] In practical applications, the task domain of a user task can be obtained through semantic analysis using a large language model. Based on the determined task domain, the server determines the matching metadata tags from a predefined scheduling strategy. These matching metadata tags are then matched with the metadata tags of multiple candidate agents to select several specialized agents from the candidate agents to perform parallel planning of the bifurcation steps.

[0234] In practical applications, if the obtained decision result indicates acceptance of the recommended planning scheme, it means the current scheduling strategy is relatively reasonable, and the server does not need to adjust the scheduling strategy. If the obtained decision result indicates modification of the recommended planning scheme (that is, the recommended planning scheme as the basic scheme for the planning fork step) or selection of any remaining planning scheme from multiple candidate planning schemes, it means the current scheduling strategy is somewhat unreasonable. The server can appropriately adjust the matching relationship between metadata tags and task domains in the scheduling strategy, such as adding or removing task domains matched by each metadata tag in the scheduling strategy. If the obtained decision result indicates rejection of the recommended planning scheme or rejection of multiple candidate planning schemes, it means the current scheduling strategy is relatively unreasonable, and the server needs to significantly adjust the matching relationship between metadata tags and task domains in the scheduling strategy, such as reducing the task domains matched by each metadata tag in the scheduling strategy.

[0235] In this embodiment, by determining the task domain to which the user task belongs, agent matching can be performed using the task domain, the metadata tags of multiple candidate agents, and predefined scheduling strategies. This enables accurate selection of multiple specialized agents, allowing for parallel planning of the bifurcation steps by multiple specialized agents. Collective intelligence is then used to reduce the uncertainty and randomness of task planning decisions. In the event of any decision result, the scheduling strategy can be optimized by adjusting the obtained decision result, thereby achieving further accurate selection of multiple specialized agents.

[0236] In an exemplary embodiment, multiple specialized intelligent agents plan feasible subsequent paths for the planning fork step, resulting in candidate planning schemes for the user task generated by each specialized intelligent agent, including:

[0237] By using multiple specialized intelligent agents, and based on the task processing prompts of each agent, feasible subsequent paths for the branching steps are planned, resulting in candidate planning schemes for the user task generated by each agent.

[0238] Task planning methods also include:

[0239] Given any decision result, adjust the task processing prompts for each of the multiple specialized intelligent agents based on the obtained decision result.

[0240] Task processing prompts refer to the prompts set for specialized intelligent agents to handle tasks, primarily guiding them in their execution. For example, task processing prompts may include processing steps and agent role definitions. Specifically, processing steps might instruct the agent where to obtain data and how to process it. Agent role definitions allow for the assignment of roles to the specialized intelligent agents. It's understandable that the role definitions for different agents will be completely different, and consequently, the task processing prompts for different specialized intelligent agents will also be entirely different.

[0241] For example, based on the identification of multiple specialized intelligent agents, the server will, through these agents and based on their respective task processing prompts, plan feasible subsequent paths for the branching steps, resulting in candidate planning schemes for the user task generated by each agent. It is understood that, since the task processing prompts can differ, the candidate planning schemes generated by the multiple agents for the user task can be completely different.

[0242] For example, based on this, the server can obtain the evaluation results of multiple candidate planning schemes through quantitative evaluation. Using the evaluation results of these multiple candidate planning schemes, the server can determine the first planning scheme for the planning fork step from among them. Based on this, the server will determine the confidence level of the first planning scheme and, based on that confidence level, request the user to make a decision, giving the user the right to choose the first planning scheme for the planning fork step. In the case of obtaining any decision result, the server can generate a task planning scheme for the user's task based on the obtained decision result. Simultaneously, the server will adjust the task processing prompts of multiple specialized intelligent agents based on the obtained decision result to optimize the task processing prompts. The obtained decision result can be the user's decision result, or a combination of the user's initial decision result and a secondary decision result.

[0243] In practical applications, if the obtained decision result indicates acceptance of the recommended planning scheme, it means the current scheduling strategy is reasonable, and the server does not need to adjust the task processing prompts. If the obtained decision result indicates modification of the recommended planning scheme (i.e., using the recommended planning scheme as the basic scheme for planning branching steps) or selection of any remaining planning scheme from multiple candidate planning schemes, it means the current task processing prompts are somewhat unreasonable, and the server can appropriately adjust the task processing prompts, such as adding or removing processing steps in the task processing prompts. If the obtained decision result indicates rejection of the recommended planning scheme or rejection of multiple candidate planning schemes, it means the current task processing prompts are relatively unreasonable, and the server needs to significantly adjust the task processing prompts, such as redesigning the processing steps in the task processing prompts or redefining the agent roles.

[0244] In this embodiment, for each specialized agent, task processing prompts can be used to guide the agent in accurately planning the forking steps. Given any decision result, these results can be used to adjust the task processing prompts for each of the specialized agents, thereby optimizing the prompts and enabling them to plan the forking steps more accurately.

[0245] The task processing method provided in this application embodiment can be applied to, for example, Figure 9In the application environment shown, terminal 902 communicates with server 904 via a network. A data storage system can store the data that server 904 needs to process. The data storage system can be set up independently, integrated into server 904, or placed in the cloud or on other network servers. When a user task is received from terminal 902, server 904 performs initial planning for the user task, obtaining planning steps. If there are planning fork steps, multiple specialized agents are selected based on these fork steps. Each specialized agent plans a feasible subsequent path for the planning fork step, resulting in candidate planning schemes generated by each agent for the fork step. These candidate schemes are then quantitatively evaluated to obtain their respective evaluation results. Based on these evaluation results, a task planning scheme for the user task is generated. The user task is executed according to the task planning scheme, the task execution result is output, and the result is fed back to terminal 902.

[0246] Among them, terminal 902 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can include virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 904 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0247] In one exemplary embodiment, such as Figure 10 As shown, a task processing method is provided. This embodiment illustrates the method applied to a server, but it is understood that the method can also be applied to a terminal, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes steps 1002 to 1012. Wherein:

[0248] Step 1002: Upon receiving a user task, perform initial planning for the user task to obtain the planning steps.

[0249] For example, when a user task is received, the server performs an initial planning of the user task to obtain the planning steps.

[0250] Step 1004: If there is a planning bifurcation step in the planning steps, select multiple specialized intelligent agents based on the planning bifurcation step.

[0251] For example, during the initial planning of a user task, the server will synchronously pre-identify the planning fork steps in the planning process. If there are planning fork steps in the planning process, the server will select multiple specialized agents based on the planning fork steps so that the planning fork steps can be planned by multiple specialized agents.

[0252] Step 1006: Through multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, resulting in candidate planning schemes generated by each specialized intelligent agent for the planning fork step.

[0253] For example, the server can use multiple specialized intelligent agents to plan feasible subsequent paths for the planning fork step, thereby obtaining candidate planning schemes generated by each specialized intelligent agent for the planning fork step.

[0254] In practical applications, the server will send a unified context and instructions to multiple specialized intelligent agents, enabling each agent to plan feasible subsequent paths for the forking step based on the unified context and instructions, thus obtaining candidate planning schemes generated by each agent for the forking step.

[0255] Step 1008: Quantitatively evaluate the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme.

[0256] For example, when multiple professional intelligent agents generate candidate planning schemes for the planning bifurcation step, the server will perform quantitative evaluation on each of the multiple candidate planning schemes to achieve an objective evaluation of the multiple candidate planning schemes and obtain the scheme evaluation results of each of the multiple candidate planning schemes.

[0257] Step 1010: Based on the evaluation results of multiple candidate planning schemes, generate a task planning scheme for the user task.

[0258] For example, based on the evaluation results of multiple candidate planning schemes, the server can determine the recommended planning scheme for the planning branching step from the multiple candidate planning schemes, and then generate the task planning scheme for the user task based on the recommended planning scheme.

[0259] Step 1012: Execute the user task according to the task planning scheme, output the task execution result, and send the task execution result back to the sender of the user task.

[0260] For example, based on the obtained task planning scheme, the server will execute the user task according to the task planning scheme, output the task execution result, and feed back the task execution result to the sender of the user task to complete the processing of the user task.

[0261] In practical applications, taking user tasks as query tasks as an example, based on the task planning scheme, the server will execute the query task according to the task planning scheme, output the query results, and send the query results back to the sender of the query task.

[0262] The above task processing method, upon receiving a user task, performs initial planning to obtain planning steps. If a planning fork occurs within these steps, multiple specialized agents are selected based on the fork. These agents plan feasible subsequent paths for each fork, resulting in candidate planning schemes generated by each agent. This allows for parallel planning of the fork by multiple agents, leveraging collective intelligence to reduce uncertainty and randomness. Quantitative evaluation of each candidate scheme yields its own evaluation result, enabling objective assessment. Based on these evaluation results, a task planning scheme for the user task is generated. The user task is then executed according to the plan, and the execution result is output and fed back to the sender of the user task. Throughout the process, when encountering planning bifurcation during the planning process, the uncertainty and randomness of the planning scheme can be reduced by using multiple professional intelligent agents to plan in parallel and to quantitatively evaluate the candidate solutions generated by each professional intelligent agent. This improves the quality of task planning decisions at the planning bifurcation step, thereby increasing the accuracy of task planning decisions for user tasks. User tasks are executed according to the task planning scheme, and the task execution results are output and feedback is provided, enabling accurate processing of user tasks.

[0263] In an exemplary embodiment, taking the task planning method of this application applied to a task planning system based on a large language model as an example, the task planning scheme of this application will be described. Figure 11 As shown, the user inputs a task: "Designate a week-long travel plan on the theme of XX, including visits to well-known companies, networking with industry professionals, and unique cultural activities." The master agent in the task planning system performs initial planning, simultaneously pre-identifying "which companies to visit" as a planning branching step (there are many companies to choose from, and opening hours, geographical location, and user interest matching must be considered). The master agent will initiate a multi-professional agent voting process based on a consensus mechanism. For this planning branching step, multiple professional agents will plan feasible subsequent paths for this branching step, resulting in candidate planning schemes generated by each professional agent. Figure 11As shown, there can be three professional agents. Agent A's candidate plan A suggests the most geographically compact and time-efficient tour route; Agent B's candidate plan B suggests focusing on visiting companies where you have connections to arrange internal exchanges; and Agent C's candidate plan C suggests adding more innovative cultural landmarks and activities. Based on these three candidate plans, the task planning system initiates a voting and decision-making process, performing weighted voting (i.e., quantitative evaluation) on each of the three candidate plans to obtain their respective evaluation results. Then, a human feedback mode is activated, allowing the user to make a decision. At this point, the task planning system will display a user selection prompt, allowing the user to make a decision. Taking candidate plan C as the recommended plan, and the confidence level of the recommended plan being greater than a predefined confidence threshold, requiring user confirmation to use the recommended plan as the first plan for the planning branching step, the user selection prompt could be: "We have pre-identified 'which companies to visit' as a key branching step and generated a recommended plan. The current optimal plan is to add more innovative cultural landmarks and activities. Do you need to modify it?" Users can choose to accept the recommended travel plan, modify and supplement it, or reject it. If a user modifies and supplements the plan by "wanting to add a visit to XXX", the task planning system will modify the recommended travel plan, integrate the "wanting to add a visit to XXX" activity into the recommended travel plan, and generate a complete travel plan based on the user's choice, and then initiate subsequent execution.

[0264] The master agent is a centralized agent responsible for generating initial plans, pre-identifying branching steps, coordinating processes, aggregating opinions from multiple specialized agents, and making the final decision. Specialized agents are sub-agents focused on specific domains or perspectives (such as search, analysis, programming, or creativity), scheduled by the master agent, and generating independent solutions or plans for the pre-identified branching steps in the initial planning. The voting consensus mechanism is a decision-making strategy that gathers opinions from multiple independent agents and selects the best solution based on preset rules (such as majority voting, weighted scoring, and consistency judgment) to reduce the uncertainty and randomness of a single source. The human feedback mode refers to an interactive mode where, after evaluating multiple options in the initial planning stage, the choice is given to the user, or when the automated system encounters decision points with high uncertainty, high risk, or inability to reach a consensus, the process is proactively interrupted and the choice is given to the user.

[0265] In practical applications, when displaying recommended solutions to users, the user decision-making interface can specifically be as follows: Figure 12As shown, the system includes a "Planning Summary" (i.e., scheme details, core ideas, main advantages, and potential risks) for candidate planning scheme C, a comparison of the scheme evaluation data of the three candidate planning schemes, multiple selection options, and a feedback area. The multiple selection options specifically include "Confirm this scheme," "Modify this scheme," and "Reject this scheme." "Confirm this scheme" indicates that the user accepts the recommended planning scheme, "Modify this scheme" indicates that the user chooses to modify the recommended planning scheme, and "Reject this scheme" indicates that the user chooses to reject the recommended planning scheme. Furthermore, if the user chooses to modify the recommended scheme, they can enter modification suggestions in the feedback area (e.g., "We hope to add visits to XXX"). If the user chooses to reject the recommended scheme, they can enter user instructions for secondary planning in the feedback area. The scheme evaluation data comparison can be presented in tabular form, including the scores of each of the three candidate planning schemes on multiple evaluation indicators. For example, as shown... Figure 12 As shown, the various evaluation indicators may include feasibility score, innovation score, user experience score, cost estimate (X, which can be a cost unit), and time efficiency (hours), etc.

[0266] Understandably, by viewing the details (i.e., the core idea), main advantages, and potential risks of the recommended planning schemes displayed in the user decision-making interface, users can make accurate decisions about the suitability of the recommended schemes. Through the comparison of scheme evaluation data displayed in the user decision-making interface, users can intuitively compare multiple candidate planning schemes and make accurate decisions. In other words, users have the right to know and the right to choose at key decision points, and the final plan is more in line with their personalized needs, while experiencing the collaboration rather than control of artificial intelligence.

[0267] In related technologies, task planning systems are mostly based on single-model decision-making or simple linear processes. Typical examples include: First, the single-agent planning mode, where a large language model is responsible for the entire task's "think-act" cycle. When encountering a fork in the road, this large language model generates a solution based on its internal knowledge, and the decision quality depends entirely on the current state and preferences of this single model, lacking comparison and verification mechanisms. Second, the ensemble prompt mode, which requires the large language model to generate multiple possible solutions at once through a single prompt word and performs a simple comparison. This method increases diversity to some extent, but it is essentially still a "self-game" of a single model, lacking a structured evaluation and selection mechanism, and the final decision may still have randomness.

[0268] In summary, the relevant technologies suffer from the following drawbacks: First, the decision-making source is singular: relying on a single model or a one-time generation, the model-generated plan is heavily influenced by its initial state and built-in preferences, resulting in unstable output, susceptibility to bias, and a tendency to get trapped in local optima rather than the globally optimal path. Second, there is a lack of quantitative evaluation: the selection of solutions lacks transparent and quantifiable evaluation standards, making it more like a "black box" selection process. Third, it cannot effectively utilize human wisdom: at complex or critical decision points, the system cannot proactively seek human assistance. Fourth, the process is rigid: the decision-making process is fixed and cannot be dynamically adjusted according to the difficulty of the decision and the context. Fifth, the failure handling strategy is simplistic: when encountering execution obstacles, the system can usually only attempt a limited number of retries or simple replanning, lacking the ability to utilize collective wisdom to find breakthrough solutions from multiple perspectives, and failing to fully utilize the predictive capabilities of the initial planning stage to mitigate decision risks in advance.

[0269] Based on this, this application proposes a task planning method and task processing method applicable to task planning systems. This method involves pre-identifying planning bifurcation steps during the initial planning stage, determining the optimal solution through voting by multiple professional intelligent agents and human interaction, and then executing the task to improve decision-making accuracy. It can be integrated as a decision enhancement module into various AI agent products. Users perceive a smarter, more reliable AI assistant that fully respects user opinions in key decisions.

[0270] The core breakthrough of the task planning method in this application lies in constructing a dynamic closed-loop negative feedback system of "pre-identification - multiple solutions - selection before execution". Specifically, its key technical points include: First, pre-identification of bifurcation steps and parallel reasoning of multiple specialized agents: In the initial planning stage, the master agent actively identifies planning bifurcation steps and simultaneously calls multiple specialized agents with different expertise and thinking patterns to generate diverse solutions for the planning bifurcation step. Second, a rule-based voting consensus mechanism: A set of configurable voting and scoring rules is designed to quantitatively evaluate and rank the solutions generated by multiple specialized agents in multiple dimensions, select the best solution, and automatically filter out the best solution with the highest confidence. Third, pre-integration of human feedback: After completing the evaluation of multiple solutions in the initial planning stage, the system automatically presents clear options and their advantages and disadvantages analysis, allowing the user to make the final decision. After the decision is made, task execution is started, realizing pre-integration of human-machine collaboration. Fourth, the dynamic workflow engine: The task planning system can pre-identify the complexity, historical accuracy, and user participation willingness of the planning branching steps based on the decision type, and intelligently select the pre-decision mode of "initial planning multiple options - evaluation - user selection - execution" or the exception handling mode during execution. In other words, the task planning system can dynamically switch between the multiple options generation - evaluation - user selection mode in the initial planning stage and the exception handling mode during execution.

[0271] In practical applications, the task planning method of this application is applied to a task planning system, and its core architecture is as follows: Figure 13 As shown, it includes a main control agent, a pool of specialized agents, a voting consensus module, and a human-computer interaction module. The following section will combine... Figure 13 Each module will be introduced separately.

[0272] In a specific application, such as Figure 13 As shown, the master control agent mainly includes a task coordinator and an expert scheduler. The task coordinator is mainly used to receive user tasks, perform initial planning, synchronize the planning fork steps in the pre-identified planning steps, and detect subsequent execution processes. The expert scheduler is mainly used to select and call a group (usually 3-5) of the most relevant professional agents from a predefined pool of professional agents based on the domain attributes (i.e., task domain) of the pre-identified planning fork steps in the initial planning (such as search, analysis, programming, and design). The scheduling strategy can be based on metadata tag matching.

[0273] In a specific application, such as Figure 13 As shown, the pool of specialized intelligent agents includes multiple specialized intelligent agents that are invoked. Figure 13The example uses n instances, namely professional agent 1, professional agent 2, ..., professional agent n. Its implementation mechanism mainly consists of a set of pre-configured instances of a large language model with different system prompts (i.e., task processing prompts) and areas of expertise (specifically, different instances of the same model or different models can be called), and each professional agent has a unique role definition. For example... Figure 13 As shown, the workflow of each specialized intelligent agent is as follows: receiving the unified context and scheduling instructions sent by the master intelligent agent, generating the complete solution for the planning fork step independently and in parallel, and returning it to the master intelligent agent, that is, returning multiple candidate solutions to the master intelligent agent.

[0274] For example, specialized intelligent agents can be categorized into strategic intelligent agents, creative intelligent agents, and domain expert intelligent agents. Strategic intelligent agents prioritize goal-achieving paths, resource allocation, and risk avoidance in their task prompts, favoring robust and executable planning solutions. Creative intelligent agents encourage divergent thinking and breaking conventional constraints, aiming to discover novel and non-obvious planning solutions. Domain expert intelligent agents are configured with detailed knowledge bases and thinking patterns for specific domains (such as shopping or programming), primarily providing professional and authoritative planning solutions.

[0275] In a specific application, taking three specialized intelligent agents as an example—a strategy agent, a creative agent, and a domain expert agent—the workflow of the specialized agent pool can be as follows: Figure 14 As shown, the master control agent can issue strategic tasks, creative tasks, and domain-specific tasks to the strategic agent, creative agent, and domain expert agent respectively through task parsing and scheduling. After executing the strategic task, the strategic agent will generate solution A; after executing the creative task, the creative agent will generate solution B; and after executing the domain-specific task, the domain expert agent will generate solution C. The candidate planning solutions generated by the three professional agents will be aggregated into the voting consensus module for processing.

[0276] In a specific application, such as Figure 13 As shown, the voting consensus module includes a scheme evaluator and a confidence calculator. The voting consensus process of this module can be specifically described as follows: Figure 15 As shown, the scheme evaluator is mainly used to receive candidate planning schemes (such as...) generated by all specialized agents for the planning bifurcation step. Figure 15The diagram shows N candidate planning schemes, which are evaluated according to preset, configurable rules. The core evaluation mechanism can be as follows: First, a rule engine is used for scoring, i.e., rules are set for quantifiable indicators (i.e., quantitative evaluation indicators, such as number of steps, estimated time, and cost) to assign scores (the scoring range can be 1-10). Second, a large language model is used as a judge. A high-performance, neutral large language model (such as GPT-4 (Generative Pre-trained Transformer 4)) is invoked as the "judge," and its prompt word template requires it to summarize and score each scheme based on the following quantitative evaluation dimensions (the scoring range can be 1-10). These quantitative evaluation dimensions can specifically include feasibility, efficiency, consistency with the overall goal, innovativeness, and risk assessment. The final score is a weighted average of the rule engine score (i.e., rule evaluation score) and the large language model's evaluation score (i.e., model evaluation score). Figure 15 As shown, the rule evaluation weight can be 0.3, and the model evaluation weight can be 0.7, to obtain the total score for each solution. Based on the total score of each solution, the best solution, i.e., the highest-scoring solution, can be selected by sorting the scores. The confidence calculator mainly calculates the confidence score of the highest-scoring solution. If the confidence score is higher than a preset threshold (i.e., a predefined confidence threshold, such as...), then... Figure 15 If the confidence level is 0.75, then the highest-scoring solution is taken as the candidate optimal solution (i.e., the recommended solution) for the planning bifurcation step. After user confirmation, it is integrated into the overall plan. If the confidence level is lower than the preset threshold, the solution is directly submitted to the user for selection. That is, if the confidence level is higher than the preset threshold, only the recommended solution can be submitted to the user for confirmation; if the confidence level is lower than the preset threshold, multiple candidate solutions can be submitted to the user for selection.

[0277] The confidence level can be calculated as follows: first, calculate the score difference Δscore between the highest-scoring scheme and the second-highest-scoring scheme, and then calculate the confidence level using the confidence level calculation formula (e.g., confidence = (highest score / 10) * (Δscore / highest score)).

[0278] In a specific application, such as Figure 13As shown, the human-computer interaction module mainly includes a decision point generator, a user interface, and a result parser. The decision point generator is primarily used to organize multiple solutions into structured, user-friendly options for the pre-identified planning bifurcation steps in the initial planning. It generates a concise "execution summary" for the highest-scoring option, highlighting its core idea, main advantages, and potential risks. The user interface primarily presents the system's recommended best solution (i.e., the highest-scoring solution, or the recommended planning solution) or all candidate planning solutions to the user through a graphical interface. The result parser primarily feeds back the user's selections or free inputs (user selection results, i.e., the user's decision results, which may include user decision results and secondary decision results) to the system, transforming them into the final solution for the planning bifurcation step, i.e., the first planning solution for the planning bifurcation step, and integrating it into the overall task plan, i.e., the user's task planning solution. Simultaneously, this selection result can be used as feedback data to optimize the weights or prompts of the corresponding specialized intelligent agents.

[0279] In a specific application, the feedback optimization flowchart can be specifically as follows: Figure 16 As shown, based on the user's decision, the task planning system records feedback data (i.e., the user's decision result, specifically approval / modification / rejection). The system analyzes the feedback data and optimizes the task processing prompts of the professional intelligent agent, adjusts voting weights (i.e., rule evaluation weights and model evaluation weights), and improves the scheduling strategy, thereby obtaining an optimized task planning system that can improve the quality of decision-making when executing new tasks.

[0280] In one exemplary embodiment, the complete workflow of the task planning system from task input to final completion can be specifically as follows: Figure 17 As shown, it includes the following steps:

[0281] 1. Task Input: The user submits a user task to the system, which is received by the main control agent.

[0282] 2. Initial planning and bifurcation pre-identification: The master control agent formulates the initial plan P0 and simultaneously pre-identifies the planning bifurcation steps in the planning process.

[0283] 3. Multi-agent decision-making: For the pre-identified planning bifurcation steps, multiple specialized agents are scheduled to generate candidate planning schemes in parallel.

[0284] 4. Voting and Consensus: The voting and consensus module quantitatively evaluates all candidate planning schemes and calculates the confidence level of the best scheme, which is also the confidence level of the recommended planning scheme.

[0285] 5. Human feedback intervention: The system presents the user with recommended planning schemes or all candidate planning schemes and analysis for the planning fork steps.

[0286] Specifically, if the confidence level of the recommended planning scheme is greater than a predefined confidence threshold, the recommended planning scheme with bifurcation steps can be presented to the user for confirmation. The specific format could be as follows: Figure 17 The message "Key fork steps have been pre-identified, recommended planning scheme X..." indicates that if the confidence level of the recommended planning scheme is less than or equal to a predefined confidence threshold, all candidate planning schemes for the fork steps can be presented to the user for confirmation.

[0287] It is understandable that the subsequent processing will differ depending on the user's choice.

[0288] Taking a recommended planning scheme with a confidence level greater than a predefined confidence threshold as an example, the user has the following options: Approve: The system integrates the recommended planning scheme into the overall plan. Modify: The user can adjust the recommended planning scheme, and the system adopts the modified version and integrates it into the overall plan. Reject: The user completely rejects the recommended planning scheme and can input new instructions or directions. The system then re-performs the initial planning and planning bifurcation pre-identification based on the new instructions.

[0289] 6. Plan Integration: Integrate the user-confirmed forking step plan (i.e., the first planning plan for the forking steps) into the overall task plan to form the final executable plan P1, which is the task planning plan for the user task.

[0290] 7. Task Execution: The system executes the task according to the final plan P1 and monitors the execution process. In other words, if the user's task planning scheme requires further execution, the system will execute the task according to the final plan P1. Upon completion, the system will output the task execution result and provide feedback to the user. For example, a task planning scheme requiring further execution could be a scheme that provides task processing methods; in this case, the task can be further executed according to the task processing methods (such as data collection channels, data processing methods, etc.) outlined in the task planning scheme.

[0291] It should be noted that the task planning method in this application, by introducing a decision-making mechanism of "initial planning fork pre-identification + multi-agent voting consensus + pre-human feedback," produces the following significant beneficial effects compared to traditional single-agent planning systems:

[0292] 1. Significantly improve decision quality and reliability: Through a hybrid evaluation mechanism of "rules + big language model evaluation", multiple candidate planning schemes for the planning bifurcation step are quantitatively scored in multiple dimensions, which effectively reduces the bias and randomness of a single model, selects more reliable planning schemes from the root, and avoids rework caused by the scheme not meeting expectations during execution, thus greatly improving the success rate of the task.

[0293] 2. Significantly optimizes human-computer interaction efficiency and experience:

[0294] Clear roles: The collaborative boundaries of "system pre-identification - generation of multiple professional intelligent agents - human decision-making" are clearly defined. Users feel that they are approving an intelligent draft rather than being presented with a difficult problem. The experience is improved from "passive decision-making" to "proactive management", which enhances their trust in the system.

[0295] Simplifying complexity: By presenting only a single recommended solution and its rationale, instead of forcing users to make difficult choices among multiple complex options, the cognitive load and decision fatigue of users are greatly reduced. At the same time, the decision-making process is brought forward, avoiding efficiency losses caused by temporary interruptions during execution.

[0296] 3. Enhance system robustness and interpretability: When faced with ambiguous requirements or high-risk decisions, the system can provide the best draft and its reasoning basis that has been demonstrated from multiple perspectives in the initial planning stage. The decision-making process changes from a "black box" to an "interpretable white box", which is more transparent and traceable.

[0297] 4. Continuous optimization capability: The system can record the user's final approval decision (approval, modification of content, reason for rejection). This high-quality human feedback data can be used to continuously optimize the prompts, voting weights, or scheduling strategies of the professional intelligent agent, making the system smarter with use.

[0298] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0299] Based on the same inventive concept, this application also provides a task planning apparatus for implementing the task planning method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more task planning apparatus embodiments provided below can be found in the limitations of the task planning method described above, and will not be repeated here.

[0300] In one exemplary embodiment, such as Figure 18As shown, a task planning device is provided, comprising: a first planning module 1802, a first selection module 1804, a second planning module 1806, a first quantitative evaluation module 1808, and a first scheme generation module 1810, wherein:

[0301] The first planning module 1802 is used to perform initial planning on the user task when a user task is received, and obtain the planning steps.

[0302] The first selection module 1804 is used to select multiple specialized intelligent agents based on the planning bifurcation step when there is a planning bifurcation step in the planning process.

[0303] The second planning module 1806 is used to plan feasible subsequent paths for the planning fork step by multiple professional intelligent agents, and obtain candidate planning schemes for the planning fork step generated by each of the multiple professional intelligent agents.

[0304] The first quantitative evaluation module 1808 is used to perform quantitative evaluation on multiple candidate planning schemes respectively, and obtain the evaluation results of each candidate planning scheme.

[0305] The first scheme generation module 1810 is used to generate a task planning scheme for the user task based on the scheme evaluation results of multiple candidate planning schemes.

[0306] The aforementioned task planning device, upon receiving a user task, performs initial planning to obtain planning steps. When a planning fork occurs, it selects multiple specialized agents to plan feasible subsequent paths for that fork, generating candidate planning schemes for each agent. This allows for parallel planning of the fork by multiple agents, leveraging collective intelligence to reduce uncertainty and randomness. Quantitative evaluation of each candidate scheme yields an objective assessment, leading to the generation of a task planning scheme for the user task. Throughout this process, when encountering planning forks, the parallel planning by multiple agents and the quantitative evaluation of their candidate solutions reduce uncertainty and randomness, improving the quality of task planning decisions at fork points and ultimately increasing the accuracy of task planning decisions for the user task.

[0307] In an exemplary embodiment, the first scheme generation module is further configured to determine a recommended planning scheme for the planning bifurcation step based on the scheme evaluation results of each of the multiple candidate planning schemes, determine the confidence level of the recommended planning scheme based on the scheme evaluation results of the recommended planning scheme and the scheme evaluation results of the remaining planning schemes among the multiple candidate planning schemes, request the user to make a decision based on the confidence level of the recommended planning scheme, obtain the user decision result, and generate a task planning scheme for the user task based on the user decision result.

[0308] In an exemplary embodiment, the scheme evaluation result includes a scheme evaluation score. The first scheme generation module is further configured to determine the scheme evaluation score to be compared based on the scheme evaluation scores of the remaining planning schemes in the candidate planning schemes, calculate the scheme evaluation score of the recommended planning scheme, and the score difference between the scheme evaluation scores to be compared, and calculate the confidence level of the recommended planning scheme based on the score difference.

[0309] In an exemplary embodiment, the first scheme generation module is further configured to generate a first user decision request when the confidence level of the recommended planning scheme is greater than a predefined confidence level threshold. The first user decision request is used to indicate confirmation that the recommended planning scheme is used as the first planning scheme for the planning fork step. The module obtains the user decision result. If the user decision result indicates acceptance of the recommended planning scheme, the module uses the recommended planning scheme as the first planning scheme for the planning fork step. Based on the first planning scheme for the planning fork step and the second planning scheme for the remaining steps in the planning step, the module generates a task planning scheme for the user task.

[0310] In an exemplary embodiment, the first scheme generation module is further configured to modify the recommended planning scheme based on the user decision result when the user decision result characterization modifies the recommended planning scheme, generate a first planning scheme for the planning fork step, and generate a task planning scheme for the user task based on the first planning scheme for the planning fork step and the second planning scheme for the remaining steps in the planning steps.

[0311] In an exemplary embodiment, the first scheme generation module is further configured to obtain user instructions from the user decision results when the user decision results indicate rejection of the recommended planning scheme, perform secondary task planning based on the obtained user instructions, and generate a task planning scheme for the user task.

[0312] In an exemplary embodiment, the first scheme generation module is further configured to generate a secondary decision request based on the remaining planning schemes among multiple candidate planning schemes when the user's decision result indicates rejection of the recommended planning scheme. The secondary decision request is used to instruct the user to select the first planning scheme for the planning fork step from the remaining planning schemes. The user's secondary decision result is obtained. When the user's secondary decision result indicates selection of any candidate planning scheme, the user determines the first planning scheme for the planning fork step based on the selected candidate planning scheme. Based on the first planning scheme for the planning fork step and the second planning scheme for the remaining steps in the planning steps, the user's task planning scheme is generated.

[0313] In an exemplary embodiment, the first scheme generation module is further configured to generate a second user decision request when the confidence level of the recommended planning scheme is less than or equal to a predefined confidence level threshold; the second user decision request is configured to instruct the selection of a first planning scheme for the planning bifurcation step from multiple candidate planning schemes, obtain user decision results, and, when the user decision results instruct the selection of any candidate planning scheme, determine the first planning scheme for the planning bifurcation step based on the selected candidate planning scheme, and generate a task planning scheme for the user task based on the first planning scheme for the planning bifurcation step and the second planning schemes for the remaining steps in the planning steps.

[0314] In an exemplary embodiment, the first scheme generation module is further configured to obtain user instructions from the user decision results when the user decision results indicate that multiple candidate planning schemes are rejected, perform secondary task planning based on the obtained user instructions, and generate a task planning scheme for the user task.

[0315] In an exemplary embodiment, the first quantitative evaluation module is further configured to perform quantitative evaluation on multiple candidate planning schemes based on predefined quantitative evaluation rules to obtain rule evaluation scores for each candidate planning scheme; to perform quantitative evaluation on multiple candidate planning schemes based on predefined quantitative evaluation dimensions using a large language model to obtain model evaluation scores for each candidate planning scheme; and to obtain a scheme evaluation result for each candidate planning scheme based on the rule evaluation score and model evaluation score of the candidate planning scheme.

[0316] In an exemplary embodiment, the first quantitative evaluation module is further configured to perform quantitative evaluation on each candidate planning scheme based on multiple quantitative evaluation indicators in the predefined quantitative evaluation rules, to obtain the quantitative indicator evaluation scores corresponding to each of the multiple quantitative evaluation indicators, and to obtain the rule evaluation score of the candidate planning scheme based on the quantitative indicator evaluation scores corresponding to each of the multiple quantitative evaluation indicators.

[0317] In an exemplary embodiment, the first quantitative evaluation module is further configured to perform weighted calculation of the rule evaluation score and model evaluation score of each candidate planning scheme based on predefined rule evaluation weights and model evaluation weights to obtain the scheme evaluation result of the candidate planning scheme, and adjust the rule evaluation weights and model evaluation weights according to the obtained decision result when any decision result is obtained.

[0318] In an exemplary embodiment, the first selection module is further configured to determine the task domain to which the planning fork step belongs when there is a planning fork step in the planning steps, perform agent matching based on the task domain, the metadata tags of each of the multiple candidate agents and the predefined scheduling strategy, select multiple professional agents from the multiple candidate agents, and adjust the scheduling strategy based on the obtained decision result when any decision result is obtained.

[0319] In an exemplary embodiment, the second planning module is further configured to plan feasible subsequent paths for the planning fork steps based on the task processing prompts of each of the multiple professional intelligent agents, thereby obtaining candidate planning schemes for the user task generated by each of the multiple professional intelligent agents, and adjusting the task processing prompts of each of the multiple professional intelligent agents according to the obtained decision results when any decision result is obtained.

[0320] In one exemplary embodiment, such as Figure 19 As shown, a task processing device is provided, including: a third planning module 1902, a second selection module 1904, a fourth planning module 1906, a second quantitative evaluation module 1908, a second scheme generation module 1910, and a task execution module 1912, wherein:

[0321] The third planning module 1902 is used to perform initial planning on the user task when a user task is received, and to obtain the planning steps.

[0322] The second selection module 1904 is used to select multiple specialized intelligent agents based on the planning bifurcation step when there is a planning bifurcation step in the planning process.

[0323] The fourth planning module 1906 is used to plan feasible subsequent paths for the planning fork step by multiple specialized intelligent agents, and obtain candidate planning schemes for the planning fork step generated by each of the specialized intelligent agents.

[0324] The second quantitative evaluation module 1908 is used to perform quantitative evaluation on multiple candidate planning schemes respectively, and obtain the evaluation results of each candidate planning scheme.

[0325] The second scheme generation module 1910 is used to generate a task planning scheme for the user task based on the scheme evaluation results of multiple candidate planning schemes.

[0326] The task execution module 1912 is used to execute user tasks according to the task planning scheme, output the task execution results, and feed back the task execution results to the sender of the user task.

[0327] The aforementioned task processing device, upon receiving a user task, performs initial planning for the user task, obtaining planning steps. It can complete the initial planning. If a planning fork step exists within the planning steps, it selects multiple specialized intelligent agents based on the planning fork step. These agents plan feasible subsequent paths for each planning fork step, resulting in candidate planning schemes generated by each agent for that fork step. When encountering a planning fork, it can introduce multiple specialized intelligent agents to perform parallel planning for the fork step, utilizing collective intelligence to reduce the uncertainty and randomness of the planning scheme. By quantitatively evaluating each of the multiple candidate planning schemes, it obtains evaluation results for each scheme. This quantitative evaluation method allows for objective assessment of the multiple candidate planning schemes. Based on the evaluation results of each candidate scheme, it can generate a task planning scheme for the user task, execute the user task according to the task planning scheme, output the task execution result, and feed the task execution result back to the sender of the user task. Throughout the process, when encountering planning bifurcation during the planning process, the uncertainty and randomness of the planning scheme can be reduced by using multiple professional intelligent agents to plan in parallel and to quantitatively evaluate the candidate solutions generated by each professional intelligent agent. This improves the quality of task planning decisions at the planning bifurcation step, thereby increasing the accuracy of task planning decisions for user tasks. User tasks are executed according to the task planning scheme, and the task execution results are output and feedback is provided, enabling accurate processing of user tasks.

[0328] The modules in the aforementioned task planning and task processing devices can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0329] In one exemplary embodiment, a computer device is provided, which can be a server or a terminal. Taking the computer device as a server as an example, its internal structure diagram can be as follows: Figure 20As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores metadata tags and other data for multiple candidate intelligent agents. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a task planning method.

[0330] Those skilled in the art will understand that Figure 20 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0331] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0332] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0333] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0334] It should be noted that the data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are all data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0335] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0336] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0337] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A task planning method, characterized in that, The method includes: Upon receiving a user task, an initial plan is performed on the user task to obtain the planning steps; If a planning bifurcation step exists in the planning process, multiple specialized intelligent agents are selected based on the planning bifurcation step. Through the multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, and candidate planning schemes for the planning fork step generated by each of the multiple specialized intelligent agents are obtained. Quantitatively evaluate each of the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme. Based on the evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated.

2. The method according to claim 1, characterized in that, The step of generating a task planning scheme for the user task based on the evaluation results of the multiple candidate planning schemes includes: Based on the evaluation results of each of the multiple candidate planning schemes, a recommended planning scheme for the planning bifurcation step is determined; The confidence level of the recommended planning scheme is determined based on the scheme evaluation results of the recommended planning scheme and the scheme evaluation results of the remaining planning schemes among the multiple candidate planning schemes. Based on the confidence level of the recommended planning scheme, request the user to make a decision, obtain the user's decision result, and generate the task planning scheme for the user's task based on the user's decision result.

3. The method according to claim 2, characterized in that, The scheme evaluation result includes a scheme evaluation score; determining the confidence level of the recommended planning scheme based on the scheme evaluation result of the recommended planning scheme and the scheme evaluation results of the remaining planning schemes among the multiple candidate planning schemes includes: Based on the evaluation scores of the remaining planning schemes among the candidate planning schemes, the evaluation score of the scheme to be compared is determined; Calculate the evaluation score of the recommended planning scheme and the score difference between the evaluation scores of the schemes to be compared; Based on the score difference, the confidence level of the recommended planning scheme is calculated.

4. The method according to claim 2, characterized in that, The process of requesting user decision based on the confidence level of the recommended planning scheme, obtaining user decision results, and generating a task planning scheme for the user task based on the user decision results includes: If the confidence level of the recommended planning scheme is greater than a predefined confidence threshold, a first user decision request is generated; the first user decision request is used to indicate that the recommended planning scheme should be used as the first planning scheme for the planning fork step. Obtain the user's decision result. If the user's decision result indicates acceptance of the recommended planning scheme, use the recommended planning scheme as the first planning scheme for the planning fork step. Based on the first planning scheme of the planning fork step and the second planning scheme of the remaining steps in the planning step, a task planning scheme for the user task is generated.

5. The method according to claim 4, characterized in that, The method further includes: When the user decision result characterizes the recommended planning scheme, the recommended planning scheme is modified based on the user decision result to generate the first planning scheme for the planning fork step; Based on the first planning scheme of the planning fork step and the second planning scheme of the remaining steps in the planning step, a task planning scheme for the user task is generated.

6. The method according to claim 4, characterized in that, The method further includes: If the user's decision result indicates rejection of the recommended planning scheme, then obtain the user's instruction from the user's decision result; Based on the obtained user instructions, secondary task planning is performed to generate the task planning scheme for the user task.

7. The method according to claim 4, characterized in that, The method further includes: If the user's decision result indicates rejection of the recommended planning scheme, a secondary decision request is generated based on the remaining planning schemes among the multiple candidate planning schemes; the secondary decision request is used to indicate the selection of the first planning scheme for the planning bifurcation step from the remaining planning schemes. Obtain the user's secondary decision result. If the user's secondary decision result indicates that any candidate planning scheme is selected, determine the first planning scheme for the planning fork step based on the selected candidate planning scheme. Based on the first planning scheme of the planning fork step and the second planning scheme of the remaining steps in the planning step, a task planning scheme for the user task is generated.

8. The method according to claim 2, characterized in that, The process of requesting user decision based on the confidence level of the recommended planning scheme, obtaining user decision results, and generating a task planning scheme for the user task based on the user decision results includes: If the confidence level of the recommended planning scheme is less than or equal to a predefined confidence threshold, a second user decision request is generated; the second user decision request is used to indicate the selection of the first planning scheme for the planning bifurcation step from the plurality of candidate planning schemes; Obtain the user's decision result, and if the user's decision result indicates that any candidate planning scheme is selected, determine the first planning scheme for the planning fork step based on the selected candidate planning scheme; Based on the first planning scheme of the planning fork step and the second planning scheme of the remaining steps in the planning step, a task planning scheme for the user task is generated.

9. The method according to claim 8, characterized in that, The method further includes: If the user decision result indicates rejection of the multiple candidate planning schemes, obtain the user instruction from the user decision result; Based on the obtained user instructions, secondary task planning is performed to generate the task planning scheme for the user task.

10. The method according to any one of claims 1 to 9, characterized in that, The step of quantitatively evaluating multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme includes: Based on predefined quantitative evaluation rules, multiple candidate planning schemes are quantitatively evaluated to obtain the rule evaluation scores of each candidate planning scheme. Using a large language model, based on predefined quantitative evaluation dimensions, the multiple candidate planning schemes are quantitatively evaluated to obtain their respective model evaluation scores. For each candidate planning scheme, the scheme evaluation result is obtained based on the rule evaluation score and model evaluation score of the candidate planning scheme.

11. The method according to claim 10, characterized in that, The predefined quantitative evaluation rules are used to quantitatively evaluate multiple candidate planning schemes respectively, obtaining the rule evaluation scores for each of the multiple candidate planning schemes, including: For each candidate planning scheme, the candidate planning scheme is quantitatively evaluated based on multiple quantitative evaluation indicators in the predefined quantitative evaluation rules, and the quantitative evaluation scores corresponding to each of the multiple quantitative evaluation indicators are obtained. The rule evaluation score of the candidate planning scheme is obtained based on the evaluation scores of the quantitative indicators corresponding to each of the multiple quantitative evaluation indicators.

12. The method according to claim 10, characterized in that, For each candidate planning scheme, based on the rule evaluation score and model evaluation score of the candidate planning scheme, the scheme evaluation result of the candidate planning scheme is obtained, including: For each candidate planning scheme, the rule evaluation score and model evaluation score of the candidate planning scheme are weighted and calculated based on predefined rule evaluation weights and model evaluation weights to obtain the scheme evaluation result of the candidate planning scheme; The method further includes: If any decision result is obtained, the rule evaluation weight and the model evaluation weight are adjusted according to the obtained decision result.

13. The method according to any one of claims 1 to 9, characterized in that, In the case where there is a planning bifurcation step in the planning process, multiple specialized intelligent agents are selected based on the planning bifurcation step, including: If a planning fork step exists in the planning steps, determine the task domain to which the planning fork step belongs; Based on the task domain, the metadata tags of each of the multiple candidate agents, and the predefined scheduling strategy, agent matching is performed, and multiple specialized agents are selected from the multiple candidate agents. The method further includes: If any decision result is obtained, the scheduling strategy is adjusted according to the obtained decision result.

14. The method according to any one of claims 1 to 9, characterized in that, The step of planning feasible subsequent paths for the planning fork step using the multiple specialized intelligent agents to obtain candidate planning schemes for the user task generated by each of the multiple specialized intelligent agents includes: Through the multiple professional intelligent agents, based on the task processing prompts of each of the multiple professional intelligent agents, feasible subsequent paths of the planning fork step are planned respectively, and candidate planning schemes for the user task generated by each of the multiple professional intelligent agents are obtained. The method further includes: In the event of obtaining any decision result, the task processing prompts of the multiple specialized intelligent agents are adjusted according to the obtained decision result.

15. A task processing method, characterized in that, The method includes: Upon receiving a user task, an initial plan is performed on the user task to obtain the planning steps; If a planning bifurcation step exists in the planning process, multiple specialized intelligent agents are selected based on the planning bifurcation step. Through the multiple specialized intelligent agents, feasible subsequent paths for the planning fork step are planned respectively, and candidate planning schemes for the planning fork step generated by each of the multiple specialized intelligent agents are obtained. Quantitatively evaluate each of the multiple candidate planning schemes to obtain the evaluation results of each candidate planning scheme. Based on the evaluation results of the multiple candidate planning schemes, a task planning scheme for the user task is generated. The user task is executed according to the task planning scheme, the task execution result is output, and the task execution result is fed back to the sender of the user task.

16. A task planning device, characterized in that, The device includes: The first planning module is used to perform initial planning on the user task when a user task is received, and to obtain planning steps. The first selection module is used to select multiple specialized intelligent agents based on the planning bifurcation step when there is a planning bifurcation step in the planning step. The second planning module is used to plan feasible subsequent paths for the planning fork step by the multiple professional intelligent agents, and obtain the candidate planning schemes for the planning fork step generated by each of the multiple professional intelligent agents. The first quantitative evaluation module is used to perform quantitative evaluation on multiple candidate planning schemes respectively, and obtain the scheme evaluation results of each of the multiple candidate planning schemes. The first scheme generation module is used to generate a task planning scheme for the user task based on the scheme evaluation results of the multiple candidate planning schemes.

17. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 15.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 15.

19. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 15.

Images