Metatask management method and device

Abstract

The invention provides a meta-task management method and device, and the method comprises the steps: S10, receiving a task list, extracting task information, and obtaining real-time available resources; s20, filtering real-time available resources, screening out all our roles and opposite roles for the current task list, and allocating the our roles and the opposite roles to each task by using a resource allocation algorithm in combination with the extracted task information to form a meta-task list; and S30, executing the meta-task list, and monitoring the execution condition at the same time. According to the method, real-time response to the dynamic environment is realized by fusing the global state and the historical execution information, local optimal traps caused by neglecting global state change and inter-task collaboration are avoided, and it is ensured that meta-task execution is consistent with a global target. By monitoring the execution condition of the meta-task, the monitoring result can be incorporated into a reference system for subsequent distribution, the strategy adjustment period is shortened, and the timeliness and accuracy of decision making are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent technology, and in particular to a meta-task management method and apparatus. Background Technology

[0002] In existing strategic interaction scenarios (such as MOBA (Multiplayer Online Battle Arena) game decision-making), meta-tasks are currently an important decision-making form used in this field. Compared with traditional operation-level decision-making forms, meta-tasks abstract and combine single operations to form a higher-level set of operations. In actual decision-making, especially when facing a large number of roles at the same time, it can effectively reduce the workload of users and agents, effectively increase the freedom of agent / user operations, and effectively reduce the probability of agent / user decision-making errors.

[0003] In existing meta-task management systems, highly abstract meta-tasks often have fixed execution roles, objectives, and operation modes. This leads to the neglect of dynamic global state changes and coordination with other meta-tasks during execution, making it difficult to achieve a globally optimal solution. Furthermore, in some systems with multi-level planning capabilities, the task allocation method for a specific role often considers only a single factor such as distance, lacking a comprehensive consideration of the task execution cost for that role. Moreover, in existing meta-task management methods, meta-task dispatch is based on the instantaneous global state, lacking feedback on the dynamic global state impact of different meta-tasks. This affects the execution effect of meta-tasks, leading to failure to achieve expectations. The lack of key decision-making elements in the entire strategy interaction process further hinders the achievement of the intended goals.

[0004] Currently, hierarchical task decomposition and assignment is a common decision-making method in strategic interaction scenarios. Hierarchical task decomposition and assignment refers to a task allocation method where, after a human or large model outputs a macro-level intention, the large model performs secondary allocation or a rule-based intelligent agent performs further strategy decomposition into specific executable actions or agreed-upon meta-tasks, and then the tasks are communicated and assigned to each role.

[0005] Most existing methods utilize large models for secondary allocation. The main approach involves a module receiving a specific policy intent, decomposing the policy using a large model, and then, based on the decomposition, defining task distribution content and providing it to the LLM (Large Language Model). The LLM then generates a detailed allocation plan for specific roles based on this information. Another approach uses rule-based agents to generate detailed plans. Rule-based meta-task management methods typically receive a list of policies decomposed from the large model and use rules to generate detailed plans. These rules often consider factors such as role capabilities and distance.

[0006] The above method has the following drawbacks: The method of using large models for secondary allocation often results in the inability to accurately optimize and adjust the output strategy due to the black-box nature of large models. At the same time, the excessively long decision-making time makes it impossible to maintain real-time performance in the actual interaction process, thus reducing the energy efficiency of the agent.

[0007] Detailed solutions are generated using rule-based intelligent agents. Rule-based intelligent agents often offer superior real-time performance. However, most existing solutions only consider simple distance and capability factors when calculating costs, neglecting complex factors such as global state and policy costs. Furthermore, rule-based intelligent agents often focus on simpler, more direct micro-level operations rather than meta-task levels; the agents lack autonomy during execution, making success rates unreliable.

[0008] Finally, existing solutions all lack monitoring modules at the meta-task or micro-operation level after task allocation, making it impossible to provide correct feedback on the global state of task progress. This results in untimely adjustments to macro-strategies and reduces the efficiency of the agent in game decision-making. Summary of the Invention

[0009] The technical problem to be solved by the present invention is how to improve the rationality of meta-task dispatch. The present invention proposes a meta-task management method and device.

[0010] The meta-task management method according to an embodiment of the present invention includes: S10: Receive the task list and extract task information, while obtaining real-time available resources; S20 filters real-time available resources, selects all friendly and opposing roles for the current task list, and uses a resource allocation algorithm to assign friendly and opposing roles to each task based on the extracted task information, forming a meta-task list. S30 executes the meta-task list and monitors its execution status.

[0011] The meta-task management method according to embodiments of the present invention achieves real-time response to dynamic environments by integrating global state and historical execution information, avoiding the trap of local optima caused by ignoring changes in global state and inter-task coordination, and ensuring that meta-task execution is consistent with the global objective. Furthermore, the present invention collects meta-task execution data in real time and monitors the execution status of meta-tasks. The monitoring results can be incorporated into the reference system for subsequent task assignment, shortening the strategy adjustment cycle and improving the timeliness and accuracy of decision-making.

[0012] According to some embodiments of the present invention, in step S10, the extracted task information includes: start time, end time, meta-task type, our role type, our role region, and opposing role list information; At the same time, real-time available resources are extracted from the global status interface and formatted and organized.

[0013] In some embodiments of the present invention, in step S20, when filtering real-time available resources, the selected friendly characters simultaneously meet the following conditions: the character can complete the tasks in the task list, the character has sufficient resources or moves to complete the tasks in the task list, the character is alive, and the character is available. The selected opposing characters must simultaneously meet the following conditions: the character is within the field of vision, the character is alive, the character is in the opposing character type list in the mission list, and the character can be damaged.

[0014] According to some embodiments of the present invention, the resource allocation algorithm in step S20 includes: S21, obtain the integer programming matrix by executing the capability integer programming algorithm; S22, the task cost matrix is ​​obtained through the execution cost optimization and overall planning algorithm; S23, multiply the integer programming matrix and the task cost matrix to obtain the comprehensive cost matrix, calculate the minimum cost path based on the comprehensive cost matrix, and allocate our roles according to the minimum cost path.

[0015] In some embodiments of the present invention, step S22, executing the cost optimization and overall planning algorithm, includes: Calculate the percentage of damage dealt to the opposing character; Standardize distance costs globally; Numericalize the target's damageability / location status; The summation is performed on the percentage of damage obtained, the distance cost after global standardization, and the target's damage / destination value after numerical processing.

[0016] A meta-task management device according to an embodiment of the present invention includes: The task parsing module is used to receive the task list and extract task information, while also obtaining real-time available resources; The resource allocation module is used to filter real-time available resources, select all friendly and opposing roles for the current task list, and combine the extracted task information with the resource allocation algorithm to assign friendly and opposing roles to each task, forming a meta-task list. The meta-task monitoring module is used to monitor the execution status of the meta-task list when it is executed.

[0017] The meta-task management device according to embodiments of the present invention achieves real-time response to dynamic environments by integrating global state and historical execution information, avoiding the trap of local optima caused by ignoring changes in global state and inter-task coordination, and ensuring that meta-task execution is consistent with the global objective. Furthermore, the present invention collects meta-task execution data in real time and monitors the meta-task execution status, allowing the monitoring results to be incorporated into the subsequent dispatch reference system, shortening the strategy adjustment cycle and improving the timeliness and accuracy of decision-making.

[0018] According to some embodiments of the present invention, the task information extracted by the task parsing module includes: start time, end time, meta-task type, our role type, our role region, and opposing role list information; At the same time, real-time available resources are extracted from the global status interface and formatted and organized.

[0019] In some embodiments of the present invention, when the resource allocation module is used to filter real-time available resources, the selected friendly characters simultaneously meet the following conditions: the character can complete the tasks in the task list, the character has sufficient resources or skills to complete the tasks in the task list, the character is alive, and the character is available. The selected opposing characters must simultaneously meet the following conditions: the character is within the field of vision, the character is alive, the character is in the opposing character type list in the mission list, and the character can be damaged.

[0020] According to some embodiments of the present invention, the resource allocation algorithm executed by the resource allocation module includes: The integer programming matrix is ​​obtained by executing the capability integer programming algorithm; The task cost matrix is ​​obtained by using an optimization algorithm for execution cost. Multiply the integer programming matrix and the task cost matrix to obtain the comprehensive cost matrix. Calculate the minimum cost path based on the comprehensive cost matrix and allocate our roles according to the minimum cost path.

[0021] In some embodiments of the present invention, the apparatus further includes: The dynamic resource pool is used to connect with the global state interface. It is responsible for extracting available or all resources from the global state interface, organizing them into a preset format, and sending all available resources to the resource allocation module for further processing after receiving a request. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a meta-task management device according to an embodiment of the present invention; Figure 2 This is a flowchart of a resource filtering algorithm according to an embodiment of the present invention. Detailed Implementation

[0023] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0024] The steps described in the specification and the flowcharts in the accompanying drawings of this invention are not necessarily to be strictly followed according to the step numbers; the execution order of the steps can be changed. Furthermore, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be broken down into multiple steps.

[0025] The meta-task management method according to an embodiment of the present invention includes: S10: Receive the task list and extract task information, and simultaneously obtain real-time available resources; according to some embodiments of the present invention, in step S10, the extracted task information includes: start time, end time, meta-task type, our role type, our role area, and opponent's role list information; at the same time, extract real-time available resources from the global status interface and format and organize them.

[0026] S20 filters real-time available resources, selects all friendly and opposing roles for the current task list, and uses a resource allocation algorithm to assign friendly and opposing roles to each task based on the extracted task information, forming a meta-task list. In some embodiments of the present invention, in step S20, when filtering real-time available resources, the selected friendly characters simultaneously meet the following conditions: the character can complete the tasks in the task list, the character has sufficient resources or moves to complete the tasks in the task list, the character is alive, and the character is available. The selected opposing characters must simultaneously meet the following conditions: the character is within the field of vision, the character is alive, the character is in the opposing character type list in the mission list, and the character can be damaged.

[0027] According to some embodiments of the present invention, the resource allocation algorithm in step S20 includes: S21, obtain the integer programming matrix by executing the capability integer programming algorithm; S22, the task cost matrix is ​​obtained through the execution cost optimization and overall planning algorithm; S23, multiply the integer programming matrix and the task cost matrix to obtain the comprehensive cost matrix, calculate the minimum cost path based on the comprehensive cost matrix, and allocate our roles according to the minimum cost path.

[0028] In some embodiments of the present invention, step S22, executing the cost optimization and overall planning algorithm, includes: Calculate the percentage of damage dealt to the opposing character; Standardize distance costs globally; Numericalize the target's damageability / location status; The summation is performed on the percentage of damage obtained, the distance cost after global standardization, and the target's damage / destination value after numerical processing.

[0029] It should be noted that in the above-mentioned cost optimization algorithm, some data can be calculated based on historical execution data, such as the percentage of damage dealt to the opposing character. This invention introduces a multi-factor cost analysis algorithm, which optimizes allocation at the micro level by combining multi-dimensional data such as character skills, resource consumption, and threats, thereby improving the economy and success rate of task execution.

[0030] S30 executes the meta-task list and monitors its execution status.

[0031] The meta-task management device according to an embodiment of the present invention includes: a task parsing module, a resource allocation module, and a meta-task monitoring module.

[0032] The task parsing module is used to receive the task list and extract task information, and at the same time obtain real-time available resources. According to some embodiments of the present invention, the task information extracted by the task parsing module includes: start time, end time, meta-task type, our role type, our role region, and opponent list information. At the same time, it extracts real-time available resources from the global status interface and formats and organizes them.

[0033] The resource allocation module is used to filter real-time available resources, select all friendly and opposing roles for the current task list, and combine the extracted task information with the resource allocation algorithm to assign friendly and opposing roles to each task, forming a meta-task list. In some embodiments of the present invention, when the resource allocation module is used to filter real-time available resources, the selected friendly characters simultaneously meet the following conditions: the character can complete the tasks in the task list, the character has sufficient resources or skills to complete the tasks in the task list, the character is alive, and the character is available. The selected opposing characters must simultaneously meet the following conditions: the character is within the field of vision, the character is alive, the character is in the opposing character type list in the mission list, and the character can be damaged.

[0034] According to some embodiments of the present invention, the resource allocation algorithm executed by the resource allocation module includes: The integer programming matrix is ​​obtained by executing the capability integer programming algorithm; The task cost matrix is ​​obtained by using an optimization algorithm for execution cost. Multiply the integer programming matrix and the task cost matrix to obtain the comprehensive cost matrix. Calculate the minimum cost path based on the comprehensive cost matrix and allocate our roles according to the minimum cost path.

[0035] The meta-task monitoring module is used to monitor the execution status of the meta-task list during its execution. It's worth noting that this module can incorporate the execution status of meta-tasks into the subsequent dispatching reference system, forming a closed loop of "dispatch-execution-feedback-optimization," shortening the strategy adjustment cycle and improving the timeliness and accuracy of decision-making.

[0036] In some embodiments of the present invention, the apparatus further includes: The dynamic resource pool is used to connect with the global state interface. It is responsible for extracting available or all resources from the global state interface, organizing them into a preset format, and sending all available resources to the resource allocation module for further processing after receiving a request.

[0037] The present invention has the following beneficial effects: This invention addresses three core shortcomings of existing meta-task management technologies through a hierarchical dispatch mechanism based on dynamic global state fusion and a resource allocation algorithm based on multi-factor cost analysis: 1. Insufficient awareness of dynamic global state: Existing methods dispatch meta-tasks based on fixed execution patterns and instantaneous global states, resulting in insufficient global coordination capabilities. This invention integrates global state and historical execution information through a macro-level meta-task layer, enabling real-time response to dynamic environments. This avoids the trap of local optima caused by ignoring changes in global state and inter-task collaboration, ensuring that meta-task execution is consistent with the global objective.

[0038] 2. Single-factor task allocation: Traditional methods rely solely on single factors such as distance and damage capability to allocate tasks, without comprehensively considering execution costs (such as overall risk and character skill depletion). This invention introduces a multi-factor cost analysis algorithm, which optimizes allocation at the micro level by combining multi-dimensional data such as character skills, resource consumption, and threats, thereby improving the economy and success rate of task execution.

[0039] 3. Lack of Feedback Mechanism: Existing solutions lack dynamic monitoring and strategy adjustment after the execution of meta-tasks, leading to a lag in macro-level decision-making. This invention collects meta-task execution data in real time through a hierarchical monitoring system and incorporates it into the reference system for subsequent assignments, forming a closed loop of "assignment-execution-feedback-optimization," shortening the strategy adjustment cycle and improving the timeliness and accuracy of decision-making.

[0040] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should be understood that the following description is merely exemplary and should not be construed as a specific limitation of the present invention.

[0041] This method for managing meta-tasks employs a hierarchical meta-task dispatching approach based on dynamic global state. By integrating global state and historical meta-task execution information, it comprehensively coordinates global meta-tasks using macro-level meta-tasks. Then, it refines the coordination of similar / local meta-tasks using a resource dynamic allocation algorithm with comprehensive multi-factor cost analysis, addressing issues such as neglecting dynamic global state changes, insufficient task allocation factors, and coordination with other meta-tasks during execution. Simultaneously, this method establishes a robust hierarchical meta-task monitoring system, incorporating historical information into the macro-level meta-task dispatching reference range to arrive at more reasonable task allocation results. Combining these methods, a global Nash equilibrium is rapidly achieved during the decision-making process.

[0042] The meta-task management method of the present invention, based on dynamic resource allocation and real-time global status monitoring, includes a dynamic resource pool, a task parsing module, a resource allocation module, and a meta-task monitoring module.

[0043] like Figure 1 As shown, in this method, the task parsing module receives tasks and extracts the start time, end time, meta-task type, friendly role type, friendly role region, and the list of opposing roles (target role name / ID or region name / ID). After extraction, it is input into the resource allocation module. Simultaneously, the dynamic resource pool refreshes resources, obtains real-time available resource information through relevant interfaces, formats and organizes it, and sends the resource information to the resource allocation module. The resource information is divided into available friendly resources and reachable opposing resources in the current task list. The resource allocation module first filters available resources, selecting all roles that can be used in the current task list based on the role ability library, and also selecting all opposing roles that meet the requirements of the task list. Finally, using the resource allocation algorithm, a suitable executor is assigned to each task, forming the final meta-task list. The meta-task list is sent to subsequent execution mechanisms for execution and also to the monitoring module, which uses a monitoring algorithm to monitor the execution status of the meta-tasks.

[0044] The following is a detailed introduction to each module: I. Task Analysis Module: The task parsing module is used to parse the preliminary task list, which reflects the fuzzy strategy intent of human users or large models, and extract information such as start time, end time, meta-task type, our role type, our role region, and the list of opposing roles (role name / ID or region name / ID). Based on the start and end times, it determines whether the current task requires resource allocation. If resource allocation is required within the execution time, the meta-task type, our role type, our role region, and the list of opposing roles (role name / ID or region name / ID) are passed to the resource allocation module. The information passed can be in any format, as long as it is parsed by the corresponding parsing module.

[0045] II. Dynamic Resource Pool: The dynamic resource pool is primarily responsible for resource management. This module connects to the global status interface, extracts available or all friendly / opposite resources from the global status interface, organizes them into relevant formats, and sends all available resources to the resource allocation module for further processing upon receiving a request.

[0046] Taking strategy-based interactive games as an example, this module reads global state information and obtains and organizes available resource information in the following format. The content can be a subset or the entire set of the following structure, or the following content is a subset of the information returned by this module: [ { "name": name, # Character Name "classname": clipnum(name), # Character type "type": get_team_type(name, self.side, obs), # Note that it has been converted to a numeric type, based on the team leader's type. "side": self.side, # camp "location": get_location(self.side, name, obs), # Role location "area":get_area(self.side, get_location(self.side, name, obs), obs), # The area where the character is located. "detection_range": get_detection_range(self.side, name, obs), # Maximum visible range of the character "status": 0, ## Whether it is available 'damage': get_damage(name, self.side, obs), # Character's damage output "sensors": get_sensor(name, self.side, obs), # Character perception ability "alive": get_alive_status(name, self.side, obs), # Character survival status "max_life":len(RED_TEAM_INFO[name]), # Character's health points "heading":get_direction(name, self.side, obs), # Current direction the character is facing "current_speed": get_speed(name, self.side, obs), # Character speed "as_target":"", ## Is an enemy character currently dealing damage to this character? "current_weather":"sunny", "current_time":"day", "target":get_target(name, self.side, obs) # The enemy character within the field of view }, ... , "team_num" = self.init_units_len # Number of character teams after forming an alliance "team_damage_ratio" = damage_ratio # Ratio of damage to the team after forming an alliance "single_units_num" = k # Number of characters "single_damage_ratio" = 1 - len(check_single_unit_alive) / k # Character-level damage ratio ] [ { "name": name, # Opponent's character name "type": get_team_type(name, side_, obs), # Opponent's character type "classname": clipnum(name), # Opponent character type "side": side_, # Opposite character faction "location": get_location(side_, name, obs), # Opposite character's position "area":get_area(side_, get_location(side_, name, obs), obs), # area "alive":get_alive_status(name, side_, obs), # The opposing character is alive "status": 0, ## Whether it is available "detection_range": get_detection_range(side_, name, obs), # The range of vision of the opposing character } ... , "team_num" = self.init_units_len # Number of teams after the opposing team forms an alliance "team_damage_ratio" = damage_ratio # Team damage ratio after the opposing team forms an alliance "single_units_num" = k # Number of opposing characters "single_damage_ratio" = 1 - len(check_single_unit_alive) / k # The ratio of damage to the opposing character ] The above are merely examples based on strategy-based interactive games. The related formats, naming, parameter compositions, or similar formats, naming methods, and parameter compositions are all innovative designs of this invention.

[0047] III. Resource Allocation Module: The resource allocation module is responsible for screening and filtering the resources transferred from the dynamic resource pool, and then using the resource allocation algorithm to allocate the corresponding resources to specific roles or game facilities for specific tasks.

[0048] 1. Resource screening; The resource filtering section filters resources from the dynamic resource pool. For friendly resources, the reference parameters are: whether they can complete the tasks in the task list, whether they have sufficient resources or skills to complete the tasks, whether they are alive, and whether they are usable. For enemy resources, the reference parameters are: whether they are within sight (within the range of their skills), whether they are alive, whether they are in the enemy character type list in the task list, and whether they can be damaged. Finally, usable resources for both friendly and enemy sides are categorized and sent to the resource allocation algorithm. The specific algorithm flow is as follows. Figure 2 As shown.

[0049] The ability to complete a task and whether a character can be damaged are recorded in a predefined character ability library. The library determines whether an ability can be used for the corresponding task based on the character's type and the type of the opposing character. The format of the character ability library is as follows: "Move": { # Ability Type "entities": [ # Executable roles] 'Li Bai' ... 'Arthur', ], "mission_type": "move", # Ability type code "targets": [ # Executable mission objectives; movement / blink mission objectives are area strings; attack commands are opponent character name strings.] "Opposite Crystal Area", ... "Top lane turret area 1" ], "attack_scope": None } The above format, naming, parameter composition, algorithm flow, or similar formats, naming methods, parameter composition, and algorithm flow all constitute innovative designs of this invention. Task types include, but are not limited to, movement, damage, search area, or other task types determined according to the meta-task delineation method.

[0050] 2. Resource allocation algorithm; The resource allocation algorithm is divided into an execution capability integer programming algorithm and an execution cost optimization algorithm that sums three values: character damage capability (the percentage of damage that can be dealt to the opponent), distance cost (globally normalized to 0-1), and target damage / destination status (execution cost is 0 if damage or destination is possible, otherwise cost is 1). Finally, there is an algorithm that multiplies the results of the two algorithms. This method can comprehensively consider the global situation and make optimal decisions.

[0051] 1) Execution capability integer programming algorithm: Table 1 Integer Programming Matrix Table 1 shows the list of available / all characters in our resource pool on the vertical axis and the list of tasks that need to be completed on the horizontal axis. The numerical elements represent whether our current character can complete the task, assigned values ​​according to the character ability library, with ∞ representing impossible completion and 1 representing possible completion. Constraints include the degree of matching between entity type and meta-task type, and the matching requirements between character type and task type in the macro-meta-task.

[0052] 2) Execution cost optimization and overall planning algorithm: Table 2 Task Cost Matrix Table 2's horizontal and vertical axes represent the same concepts as above. The C value represents the cost derived from the expected benefit of this role performing this task in the current global state. Calculation conditions include: Character Ability (Percentage of damage dealt to the target character): ; Where C represents the current character's damage cost, and the numerator represents the damage that the current attacker can inflict on the opponent. This represents the number of moves the attacker can currently use for the mission. The numerator represents the damage that a move can inflict on the opponent. The numerator represents the sum of the damage capabilities of all moves currently possessed by the attacker that can inflict damage on the opponent. This represents the total health of the current target. After obtaining the above value, all the obtained numerator values ​​are included in the original cost set of damage ability. After the cost of all combinations of the task cost matrix has been solved, the original cost set of damage ability is uniformly standardized and distributed to 0-1.

[0053] Distance cost (globally normalized to 0-1): ; in, To solve for the distance function, For the attacker's position, Let's define the target location. First, calculate the distance between all combinations in the cost matrix, then standardize the distance between all combinations to 0-1.

[0054] Target availability: The cost of striking is 0 if the target can be attacked or if the target can be moved / blinked; otherwise, the cost is 1.

[0055] Finally, the three values ​​are summed and filled into the task cost matrix.

[0056] 3) Find the optimal solution by combining all methods: The elements of the two matrices are multiplied. Multiplication here refers to multiplying corresponding elements in the two matrices. For example, multiplying the 1 in the first row and first column of Table 1 with C11 in the first row and first column of Table 2, and so on. After multiplying the corresponding values, a new matrix is ​​formed, namely the comprehensive cost matrix. During the multiplication operation, the "∞" value in the integer programming matrix of Table 1 is taken as... After multiplying the two matrices to obtain the comprehensive cost matrix, the result is used to perform the Hungarian algorithm or other minimum cost combination algorithms to find the minimum cost path and obtain the optimal attack plan. It should be noted that in the obtained comprehensive cost matrix, the vertical axis represents the list of available / all characters in our resource pool, and the horizontal axis represents the multiplication of the list of tasks that need to be completed. The values ​​in the comprehensive cost matrix are the comprehensive cost values ​​of the corresponding characters performing the corresponding tasks. The minimum cost path is the plan corresponding to the minimum sum of the comprehensive cost values ​​of our characters selected for each task in the task list. This is used to allocate the corresponding characters to perform the tasks.

[0057] Optional task resource allocation modes: Global allocation: Global resources are reallocated at each round of decision points; Local allocation: Each round of decision points allocates currently available resources; IV. Meta-task monitoring module: enter: observation, task_name, allocation_result, start_time, end_time, llm_task_id Status, task name, sequence of micro-meta-tasks under this macro-meta-task, start time, end time, macro-meta-task ID; Function: Generate a corresponding PID (monitoring task ID) and a corresponding monitoring class instance, monitor the task status of the monitoring role, and return the information to the parent module, structure, or role that needs it. The monitored task status includes: 0: Standby; 1: In progress; 2: Success; # Exploration mission covers 70% of the area, or one or more characters reach the target area (target point) in a movement / teleport mission, or one or more targets die in each micro-level mission objective in a damage mission; 3: Failure # Otherwise, it is considered a failure; In other words, the meta-task monitoring module can create monitoring tasks based on the relevant information of the meta-task to be monitored. The meta-task monitoring module can obtain situational data information in real time and monitor the execution of meta-tasks in real time based on the real-time situational information.

[0058] Through the above-mentioned technical improvements, this invention achieves global optimal allocation of meta-tasks in dynamic environments. While ensuring real-time performance, it expands the task execution cost from a single dimension to a multi-dimensional comprehensive evaluation and improves the system's adaptive capability through historical data feedback. Ultimately, it quickly achieves global Nash equilibrium in the game, which is significantly better than the limitations of traditional large-scale model black-box decision-making or rule-based intelligent agents.

[0059] Through the description of specific embodiments, a more in-depth and specific understanding should be gained of the technical means and effects adopted by the present invention to achieve the intended purpose. However, the accompanying drawings are only for reference and illustration and are not intended to limit the present invention.

Claims

1. A meta-task management method, characterized by, Comprising: S10, receiving a task list and extracting task information, while obtaining real-time available resources; S20, filtering real-time available resources, screening all my characters and opposite characters for the current task list, and combining the extracted task information with the resource allocation algorithm to assign my characters and opposite characters to each task to form a meta-task list; S30, executing the meta-task list, while monitoring the execution of the meta-task list.

2. The metatask management method of claim 1, wherein, In step S10, the extracted task information includes: start time, end time, meta-task type, my character type, my character area, and opposite character list information; At the same time, extract real-time available resources from the global state interface and format them.

3. The metatask management method of claim 1, wherein, In step S20, when filtering real-time available resources, the screened my characters meet the following conditions simultaneously: the character can complete the tasks in the task list, the character has sufficient resources or tricks to complete the tasks in the task list, the character is alive, and the character is available; The screened opposite characters meet the following conditions simultaneously: the character is within the field of view, the character is alive, the character is in the opposite character type list of the task list, and the character can be damaged.

4. The metatask management method of claim 1, wherein, The resource allocation algorithm in step S20 includes: S21, obtain an integer programming matrix by executing an execution capability integer programming algorithm; S22, obtain a task cost matrix by executing a cost optimization algorithm; S23, multiply the integer programming matrix and the task cost matrix to obtain a comprehensive cost matrix, calculate the minimum cost path based on the comprehensive cost matrix, and assign my characters according to the minimum cost path.

5. The metatask management method of claim 4, wherein, In step S22, the cost optimization algorithm includes: Calculate the damage percentage to the opposite character; Perform global standardization processing on the distance cost; Numerical processing is performed on the target damageable / go-to situation; Sum the three values of the obtained damage percentage, globally standardized distance cost, and numerically processed target damageable / go-to.

6. A meta-task management apparatus characterized by comprising: Comprising: A task analysis module for receiving a task list and extracting task information while obtaining real-time available resources; A resource allocation module for filtering real-time available resources, screening all my characters and opposite characters for the current task list, and combining the extracted task information with the resource allocation algorithm to assign my characters and opposite characters to each task to form a meta-task list; A meta-task monitoring module for monitoring the execution of the meta-task list when executing the meta-task list.

7. The meta-task management apparatus according to claim 6, characterized by The task analysis module is used to extract task information including: start time, end time, meta-task type, my character type, my character area, and opposite character list information; At the same time, extract real-time available resources from the global state interface and format them.

8. The meta-task management apparatus according to claim 6, characterized by The resource allocation module is used to filter real-time available resources, and the screened my characters meet the following conditions simultaneously: the character can complete the tasks in the task list, the character has sufficient resources or tricks to complete the tasks in the task list, the character is alive, and the character is available; The screened opposite role satisfies the following conditions simultaneously: the role is in the field of view, the role survives, the role is in the opposite role type list of the task list, and the role can be damaged.

9. The meta-task management apparatus according to claim 6, characterized by The resource allocation module executes a resource allocation algorithm, which includes: An integer programming matrix is obtained by executing an integer programming algorithm; A task cost matrix is obtained by executing a cost optimization algorithm; The integer programming matrix and the task cost matrix are multiplied to obtain a comprehensive cost matrix, the minimum cost path is calculated based on the comprehensive cost matrix, and the self role is distributed according to the minimum cost path.

10. The meta-task management apparatus according to claim 6, characterized by The device further includes: A dynamic resource pool is used to communicate with the global state interface, responsible for extracting available or all resources from the global state interface, arranging them into a preset format, and sending all available resources to the resource allocation module for subsequent processing after receiving a request.

Images