Multi-agent dynamic allocation and cooperation method driven by i-star demand model
Through the multi-agent dynamic allocation and collaboration method driven by the i-star demand model, the problem of inefficiency of multi-agent collaboration under static mode is solved, dynamic role allocation and task scheduling are realized, and the flexibility of multi-agent collaboration is improved. Scalability, adapt to the needs of complex scenarios.
Patent Information
- Application Number
- CN202411117617.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-12
AI Technical Summary
The existing multi-agent collaboration methods mainly adopt static methods, making it difficult to flexibly respond to dynamically changing environment and task requirements, resulting in inefficient collaboration and waste of resources.
Using the i-star demand model-driven method, the agent is divided into planning agents and action agents, and dynamic role allocation and task scheduling is realized through the role allocation table and task pool, supporting the rapid switching of agent actors and improving multi-integrators Flexibility and scalability of collaboration.
It realizes dynamic role allocation and task scheduling of multi-agents, improves the flexibility and scalability of multi-agents collaboration, reduces the overhead of modifying collaboration methods, and adapts to complex and changeable application scenarios.
Smart Images

Figure CN118941047A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent agent control technology, in particular to a multi-agent dynamic allocation and collaboration method driven by an i-star demand model. Background Art
[0002] In today's technological development, the i-star demand model is a representative and widely used demand analysis tool. It can clearly describe the goals, tasks, resources and relationships of each subject role in a specific scenario.
[0003] In recent years, intelligent agents have developed rapidly. Multi-agent collaboration, as an important research field, aims to enable multiple agents to work together to achieve common goals or complete complex tasks. However, in terms of multi-agent collaboration, most of them adopt a static agent approach. This approach has certain limitations and it is difficult to flexibly respond to dynamically changing environments and task requirements.
[0004] In actual application scenarios, it is often necessary to dynamically schedule and coordinate multiple agents based on real-time situations. For example, in a retail store model represented by a coffee shop, the dynamic and diverse needs of customers require multiple agents to respond quickly and adjust their collaboration methods; in a complex logistics distribution system, the quantity, destination, and transportation conditions of goods may change at any time, requiring dynamic adjustment of the collaborative strategies of the agents.
[0005] However, the existing static seating method often cannot make such dynamic adjustments in a timely and effective manner, resulting in problems such as low collaboration efficiency and waste of resources. Therefore, there is an urgent need for a collaborative method that can dynamically schedule multiple agents to better adapt to various complex and changing application scenarios and improve the effectiveness and flexibility of multi-agent collaboration. Summary of the invention
[0006] In view of this, the purpose of the present invention is to provide a multi-agent dynamic allocation and collaboration method driven by an i-star demand model. The method is driven by an i-star demand model, and the agents are divided into a planning agent for dynamically allocating roles and scheduling tasks and various specific action agents for playing abstract roles. The planning agent dynamically allocates roles to each action agent and schedules tasks to the abstract role, and each action agent receives the task according to the role it plays. The present invention can dynamically perform role allocation and task scheduling for multiple agents, and plays a guiding role in the collaboration of multiple agents.
[0007] In order to achieve the above object, the present invention provides the following technical solutions: An i-star demand model driven multi-agent dynamic allocation and collaboration method, characterized in that the method comprises: an agent, a role allocation table for dynamically configuring the roles played by each action agent, and a task pool for assisting each action agent to dynamically collaborate to achieve an external goal; The intelligent agents are based on the idea of role playing and are driven by the i-star demand model to perform dynamic collaboration; the intelligent agents include: a planning agent, which is used to dynamically allocate roles and schedule tasks; and several action agents, which are used to play specific actions of abstract roles; The role allocation table is dynamically generated by the planning agent in combination with the action agent information and the defined role constraints to allocate roles; The task pool is dynamically generated by the planning agent based on the i-star demand model according to the external input target to schedule the task process, and has the ability to continuously manage the data in the task pool; the task pool includes: a collaboration scheme set for storing multiple collaboration scheme information, including the execution order information of each task in the collaboration scheme; a task queue set for storing the specific task information required to be performed by each role, including the resource limitation information for performing specific tasks; wherein: The role allocation process takes the agent information of each action agent stored in the planning agent and the defined role constraints as input, and outputs the final role allocation result to the role allocation table, specifically including: S1-1, calculate the comprehensive score of the role, input the agent information of each action agent, and output the comprehensive score of each action agent on each optional role; S1-2, determine the assigned roles, input the comprehensive scores of each action agent on each role, and output the roles assigned to each action agent; according to the specific actual situation, an agent can be assigned multiple roles, a single role, or no role; S1-3, check constraints, input the roles assigned to each action agent and the defined role constraints, and obtain the constraint check result; if passed, output the final role assignment result; if not passed, return to S1-2 and re-determine the final assigned roles of each action agent; The scheduling task process takes the external goals given by external users and the i-star demand model stored in the planning agent as input, and outputs the formed overall collaboration plan and each specific task to the task pool, specifically including: S2-1, identify all tasks, input the external goals and i-star demand model, and output all tasks required to achieve the external goals; S2-2, add task resource dependency, input all the tasks to be executed and the i-star demand model, identify the resources required for each task to be executed from the i-star demand model, and add resource dependency constraints for each task; S2-3, form an overall collaboration plan, input all the tasks to be executed and the i-star demand model, identify the roles required for the execution of each task from the i-star demand model; then, identify the predecessor tasks that need to be completed before the execution of each task from the i-star demand model based on the dependency relationship, build the order relationship of task execution, and form an overall collaboration plan; S2-4, parallel execution, specifically including: S2-4a, publish the collaboration plan, input the formed overall collaboration plan, encapsulate and add a serial number, and publish the collaboration plan to the collaboration plan set of the task pool; S2-4b, publish specific tasks, input the formed overall collaboration plan, and publish the tasks in sequence to the corresponding role task queues in the task queue set of the task pool according to the execution roles of each task in the plan; The ability to manage data in the task pool includes: whenever the planning agent publishes a task to the task pool, the task pool notifies the action agent players of each role according to the execution role of the published task to remind the action agent to collect new tasks from the task queue of the role it plays; whenever the action agent collects a task from the task pool or completes a task, the task pool modifies the status information of the task in the collaboration plan set to reflect the collaboration progress between the action agents in real time.
[0008] Further, the S1-1 is specifically: S1-11, calculating the role ability matching score, inputting the agent ability information in the action agent information, and comparing it with the ability information of each role in turn, and calculating the ability matching score of each role to measure whether the action agent is capable of performing each role; S1-12, calculating the past task completion rate, inputting the past task completion information of the action agent when playing each role, and calculating the past task completion rate of the action agent on each role, so as to estimate the task completion rate of the agent when performing tasks in each role; S1-13. Calculate the role's comprehensive score by weighting the role's ability matching score and past task completion rate to obtain the role's comprehensive score.
[0009] Further, the S2-1 is specifically: S2-11, match the target, input the external target and the i-star demand model, and output the target element in the i-star demand model that is most similar to the external target; S2-12, disassemble the target, input the matching target element and i-star requirement model, and disassemble it into multiple sub-target elements according to the "and refinement relationship" and "or refinement relationship", until all sub-target elements cannot be disassembled and become atomic target elements, and then output all atomic target elements; among them, the disassembly based on the "and refinement relationship" needs to retain all sub-target elements, while the disassembly based on the "or refinement relationship" only needs to randomly retain a sub-target element; S2-13, identify the tasks, input all atomic target elements and i-star requirement model, take the target element as the root node, perform depth-first search, and output all tasks with relationships as all tasks that need to be executed.
[0010] It can be seen from the above technical solution that compared with the prior art, the present invention provides a multi-agent dynamic allocation and collaboration method of an i-star demand model, which has the following beneficial effects: dynamically assigning roles and scheduling tasks to multiple agents based on the i-star demand model, supporting rapid switching of agent player instances of a certain role, and improving the flexibility and scalability of multi-agent collaboration. Business personnel can modify the business flow by modifying the i-star demand model without performing related modification operations on the action agent instance, reducing the overhead of modifying the multi-agent collaboration method. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] Figure 1 It is a schematic diagram of the method of the present invention; Figure 2 A flow chart of the process of assigning roles of the present invention; Figure 3 A flowchart of the scheduling task process of the present invention; Figure 4 This is a schematic diagram of the i-star demand model for a simple coffee shop scenario in an embodiment of the present invention; Figure 5 This is the task pool result of a simple coffee shop scene in an embodiment of the present invention. DETAILED DESCRIPTION
[0012] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0013] The method schematic diagram of the present invention is shown in Figure 1In the present invention, the intelligent agents include two categories, namely, a planning agent for dynamically allocating roles and scheduling tasks and several specific action agents for playing abstract roles.
[0014] The present invention has a role allocation table for dynamically configuring the roles played by each action agent. The role allocation table scores the action agent from two dimensions: the degree of ability matching between the action agent and the optional role and the action agent's past task completion rate, and then weights the two dimensions to obtain a comprehensive score, and then allocates roles to the action agent based on the comprehensive score. The role allocation table is dynamically generated by the planning agent in the process of allocating roles in combination with the action agent information and the defined role constraints.
[0015] The present invention includes a task pool for assisting various action agents to dynamically collaborate to achieve external goals. The task pool includes a collaboration plan set and a task queue set. The collaboration plan set includes multiple collaboration plans, which are dynamically generated by different external goals. Each collaboration plan includes a collaboration plan sequence number, a plan goal, a release date, and an activity network diagram containing task execution sequence information; wherein, each internal node in the activity network diagram represents a specific task, and the node attributes include the task name, task status, task start time, task execution time, the role required to execute the task, and the role player; wherein the task status is "not received", "received" or "completed". The task queue set includes the task queues to be executed by each role in the i-star demand model. The action agent can receive and execute the corresponding task from the corresponding task queue according to its own role, and report to the task pool that the task has been completed after completing the task. The task pool has the ability to manage data. Whenever the planning agent publishes a task to the task pool, the task pool notifies the action agent of each role according to the execution role of the published task to remind the action agent to receive new tasks from the task queue of the role played; whenever the action agent receives a task from the task pool or completes a task, the task pool modifies the status information of the task in the collaborative solution set to reflect the collaborative progress between the action agents in real time. The task pool is dynamically generated by the planning agent based on the i-star demand model according to the external goals given by external users to schedule tasks, and continuously manages the data in the task pool.
[0016] and Figure 1 The process of allocating roles corresponds to the process of scheduling tasks. The embodiment of the present invention also discloses the specific processes of the two execution processes. Figure 2 As shown, the following steps are included: S1-1, calculate the comprehensive score of the role, input the agent information of each action agent, and output the comprehensive score of each action agent on each optional role; S1-2, determine the assigned roles, input the comprehensive scores of each action agent on each role, and output the roles assigned to each action agent; according to the specific actual situation, an agent can be assigned multiple roles, a single role, or no role; S1-3, check constraints, input the roles assigned to each action agent and the defined role constraints, and obtain the constraint check result; if passed, output the final role assignment result; if not passed, return to S1-2 and re-determine the final assigned roles of each action agent; Furthermore, S1-1 is specifically: S1-11, calculating the role ability matching score, inputting the agent ability information in the action agent information, and comparing it with the ability information of each role in turn, and calculating the ability matching score of each role to measure whether the action agent is capable of performing each role; S1-12, calculating the past task completion rate, inputting the past task completion information of the action agent when playing each role, and calculating the past task completion rate of the action agent on each role, so as to estimate the task completion rate of the agent when performing tasks in each role; S1-13. Calculate the role's comprehensive score by weighting the role's ability matching score and past task completion rate to obtain the role's comprehensive score.
[0017] The process flow chart of scheduling tasks is as follows: Figure 3 As shown, the following steps are included: S2-1, identify all tasks, input external goals and i-star demand model, and output all tasks that need to be performed to achieve external goals; S2-2, add task resource dependency, input all the tasks to be executed and the i-star demand model, identify the resources required for each task to be executed from the i-star demand model, and add resource dependency constraints for each task; S2-3, form an overall collaboration plan, input all the tasks to be executed and the i-star demand model, identify the roles required for the execution of each task from the i-star demand model; then, identify the predecessor tasks that need to be completed before the execution of each task from the i-star demand model based on the dependency relationship, build the order relationship of task execution, and form an overall collaboration plan; S2-4, the following steps are steps that can be executed in parallel, specifically including: S2-4a, publish the collaboration plan, input the formed overall collaboration plan, encapsulate and add a serial number, and publish the collaboration plan to the collaboration plan set of the task pool; S2-4b, publish specific tasks, input the formed overall collaboration plan, and publish the tasks in sequence to the corresponding role task queues in the task queue set of the task pool according to the execution roles of each task in the plan; Furthermore, S2-1 is specifically: S2-11, match the target, input the external target and the i-star demand model, and output the target element in the i-star demand model that is most similar to the external target; S2-12, disassemble the target, input the matching target element and i-star requirement model, disassemble the target element into multiple sub-target elements according to the "and refinement relationship" and "or refinement relationship", until all sub-target elements cannot be disassembled and become atomic target elements, and then output all atomic target elements; among them, the disassembly based on the "and refinement relationship" needs to retain all sub-target elements, while the disassembly based on the "or refinement relationship" only needs to randomly retain a sub-target element; S2-13, identify the tasks, input all atomic target elements and i-star requirement model, take the target element as the root node, perform depth-first search, and output all tasks with relationships as all tasks that need to be executed. Example
[0018] In an embodiment, based on the i-star demand model of a simple coffee shop scenario, dynamic collaboration guidance is provided for several action agents, including a desktop robot and a general humanoid robot. Figure 4This is the i-star requirement model for a simple coffee shop scenario, which analyzes the various roles in the scenario in detail. In the simple coffee shop scenario, there are two abstract roles that can be selected by the action agent, including the barista and the waiter. Among them, the barista's responsibility is to prepare latte coffee, ensure the quality of coffee, and meet the customer's taste requirements; while the waiter's responsibility is to receive customers, deliver food products, provide latte coffee related services, and ensure customer satisfaction. There may be multiple element nodes in the role, among which the main nodes include three types, target elements (solid rounded rectangles), task elements (solid hexagons), and resource elements (solid rectangles). The target element describes the state condition with clear achievement standards that the role expects to achieve; the task element describes the specific operations taken by the role to achieve the target element; the resource element describes the physical or information entity that the role will use in the process of performing the task. There may be multiple relationships between nodes in the i-star requirement model, among which the main relationships include three types, refinement relationship, use relationship, and dependency relationship. Among them, the refinement relationship represents the abstract relationship between target / task elements and target / task elements, so as to decompose an abstract target / task element into one or more more specific target / task elements. Furthermore, the refinement relationship includes two types: and refinement relationship and or refinement relationship. Among them, the and refinement relationship indicates that the parent goal / task is satisfied only when all sub-goals / tasks are satisfied, which is represented by a straight line with a short dash at the end; the or refinement relationship indicates that the parent goal / task is satisfied when at least one sub-goal / task is satisfied, which is represented by a straight arrow. The use relationship indicates the resources that a role needs to use when performing a specific operation, which is represented by a straight line with a solid circle at the end; the dependency relationship indicates that a role needs to rely on other roles to realize its own elements, which is represented by a curve with two "D", where the two "D"s are the specific content of the dependency. In this embodiment, the barista has the goal of "making a cup of latte"; the waiter has the goal of "providing latte-related services". Further, the goal of "providing latte-related services" can be refined into the goal of "delivering latte to customers" and the goal of "understanding customers' needs for latte". In order to meet these goals, each role needs to perform more specific task operations. For example, the goal of "understanding customers' demand for latte coffee" is met by the task of "obtaining customer demand", and there is a refinement relationship between them. When performing tasks, various resources need to be used. For example, the task of "delivering latte coffee" requires the use of the "latte coffee" resource. There are dependencies between the nodes of different roles. For example, the "latte coffee" resource used by the "delivering latte coffee" task in the waiter depends on the "making latte coffee" task in the barista.
[0019] According to the two execution processes defined in the present invention, dynamic collaborative guidance is provided for multiple action agents, including desktop robots and general humanoid robots. First, the role allocation process is performed: The comprehensive scores of the roles are calculated, and based on the capability information of the desktop robot and the general humanoid robot, the required capabilities of each role and the capabilities of each action agent are evaluated in turn.
[0020] The barista's role responsibilities include preparing lattes, ensuring quality and satisfying customers' taste needs. Ability requirements include gripping, turning knobs, pressing, picking up, placing, and observing; The waiter role responsibilities include greeting customers, delivering food and beverages, and providing latte-related services to ensure customer satisfaction. Ability requirements include moving, communicating, grasping, pressing, picking up, placing, and observing.
[0021] The desktop robot has the ability to grasp, turn knobs, press, pick up, place, push, pull, and observe; General humanoid robots have the ability to move, communicate, grasp, turn knobs, press, pick up, place, push and pull, rotate the body, and observe.
[0022] The calculation shows that the desktop robot's ability matching score for the barista role is 91.9, which can well meet the barista's ability requirements, while the ability matching score for the waiter role is only 30.7, which cannot well meet the waiter's ability requirements; the ability matching score for the general humanoid robot for the barista role is 92.0, which can well meet the barista's ability requirements, and the ability matching score for the waiter role is 93.1, which can well meet the waiter's ability requirements. At the same time, the past task completion rate of each action agent in playing different roles is calculated. When the desktop robot played the role of barista in the past, the completion rate of the task was 0.74, but since the desktop robot has never played the role of waiter, it has no score; the general humanoid robot has never played the role of barista, so it has no score, and when the general humanoid robot played the role of waiter in the past, the completion rate of the task was 0.89. Finally, through weighted calculation, the comprehensive scores of the desktop robot and the general humanoid robot in the barista role and waiter role were obtained respectively. The comprehensive score of the desktop robot in barista was 79.2, and the comprehensive score in waiter was 30.7, while the comprehensive score of the general humanoid robot in barista was 77.4, and the comprehensive score in waiter was 87.5.
[0023] Decide on the role assignment. After assigning roles to each action agent based on the comprehensive scores of the desktop robot and the general humanoid robot, for example, if 70 points is used as the score threshold for role assignment, then the desktop robot can be assigned the barista role, and the general humanoid robot can be assigned the barista role and the waiter role.
[0024] Check the constraints and check whether the roles assigned to each agent meet the constraints. The role constraints of the simple cafe scene are as follows: 1. Each role must be played by at least one agent 2. The waiter role must have the ability to communicate with customers 3. At least one of the comprehensive scores of the waiter role must be no less than 70 points 4. At least one of the comprehensive scores of the barista role must be no less than 70 points The current role assignment is that the desktop robot plays the role of a barista, with a comprehensive score of 79.2; the general humanoid robot is assigned to the barista role and the waiter role, with a comprehensive score of 77.4 for the barista role and 87.5 for the waiter role. After passing the inspection, the current role assignment result is output as the final role assignment result to the role assignment table and displayed.
[0025] At this point, the role assignment process is complete.
[0026] Then, proceed with the task scheduling process: Identify all tasks and match the target elements that are most similar to the external goals in the i-star demand model. For example, if "the customer needs a cup of latte now" is the external goal of the external user, the most similar target element in the i-star demand model is the waiter's "provide latte-related services" goal. Then, according to the i-star demand model, the matched goals are disassembled. The "provide latte-related services" goal can be disassembled into "understand the customer's demand for latte" and "deliver latte to customers" sub-goals based on the refinement relationship. Finally, identify tasks based on all sub-goal elements. For example, the "deliver latte to customers" goal can identify the "deliver latte" task. Continuing the depth-first search, the "make latte" task and the "obtain customer demand" task can be further identified.
[0027] Add task resource dependencies. According to the i-star requirement model, add resource dependency constraints for all identified tasks. For example, the "Making Latte" task requires the use of "Latte Making Requirements" resources, "Milk" resources, "Coffee Powder" resources, and "Coffee Machine" resources.
[0028] To form an overall collaboration plan, according to the i-star demand model, identify the roles required for the execution of each task from the i-star demand model. For example, the execution role of the "obtain customer needs" task is the waiter; then, according to the dependency relationship in the i-star demand model, identify the predecessor tasks that need to be completed before the execution of each task, build the sequential relationship of task execution, and form an overall collaboration plan. For example, the "latte making requirements" resource in the barista depends on the "obtain customer needs" task in the waiter, and the "making latte" task in the barista uses the "latte making requirements" resource. Therefore, the predecessor task of the "making latte" task is the "obtain customer needs" task. Then, execute the "obtain customer needs" task first, and then execute the "making latte" task, and so on, to obtain an activity network diagram, in which the task status of each internal node is "not received", the task start time and task execution time are both "NaN", and the role player is "none".
[0029] Then execute the following steps in parallel: Publish the collaboration plan, encapsulate the overall collaboration plan, add the plan number "1", set the plan goal to "the customer needs a cup of latte now", and record the release date as "2024 / 07 / 16", and publish the collaboration plan to the collaboration plan collection of the task pool; Release specific tasks. According to the execution roles of tasks in the i-star demand model, release the tasks in sequence to the corresponding role task queues in the task queue set of the task pool. For example, if the execution role of the "obtain customer demand" task is waiter, then add the "obtain customer demand" task to the waiter task queue.
[0030] At this point, the task scheduling process ends, and the final task pool is obtained as follows: Figure 5 shown.
[0031] Whenever the planning agent publishes a task to the task pool, the task pool notifies the action agent player of each role to go to the corresponding role task queue to pick up the new task based on the execution role of the published task. For example, if the planning agent publishes the "obtain customer needs" task, the task pool will notify the general humanoid robot to remind it to go to the waiter task queue to pick up the "obtain customer needs" task.
[0032] Whenever an action agent receives a task from the task pool or completes a task, the task pool modifies the status information of the task in the collaboration scheme set to reflect the progress of collaboration between action agents in real time. For example, at 15:25, the general humanoid robot receives the "obtain customer needs" task, and the task pool will modify the node attributes of the task named "obtain customer needs" in the collaboration scheme with serial number 1, including changing the task status to "received", the task start time to "15:25", and the role player to "general humanoid robot"; similarly, 3 minutes later, the general humanoid robot completes the "obtain customer needs" task, and the task pool will modify the node attributes of the task named "obtain customer needs" in the collaboration scheme with serial number 1, including changing the task status to "completed" and the task execution time to "3min".
[0033] The details of each embodiment in this specification are described in a progressive manner. The details of each embodiment focus on the differences from the details of other embodiments. The same or similar parts between the details of each embodiment can be referenced to each other.
[0034] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An i-star demand model driven multi-agent dynamic allocation and collaboration method, characterized in that: The method includes: an agent, a role allocation table for dynamically configuring the roles played by each action agent, and a task pool for assisting each action agent to dynamically collaborate to achieve an external goal; The intelligent agents are based on the idea of role playing and are driven by the i-star demand model to perform dynamic collaboration; the intelligent agents include: a planning agent, which is used to dynamically allocate roles and schedule tasks; and several action agents, which are used to play specific actions of abstract roles; The role allocation table is dynamically generated by the planning agent in combination with the action agent information and the defined role constraints to allocate roles; The task pool is dynamically generated by the planning agent based on the i-star demand model according to the external input target to schedule the task process, and has the ability to continuously manage the data in the task pool; the task pool includes: a collaboration scheme set for storing multiple collaboration scheme information, including the execution order information of each task in the collaboration scheme; a task queue set for storing the specific task information required to be performed by each role, including the resource limitation information for performing specific tasks; wherein: The role allocation process takes the agent information of each action agent stored in the planning agent and the defined role constraints as input, and outputs the final role allocation result to the role allocation table, specifically including: S1-1, calculate the comprehensive score of the role, input the agent information of each action agent, and output the comprehensive score of each action agent on each optional role; S1-2, determine the assigned roles, input the comprehensive scores of each action agent on each role, and output the roles assigned to each action agent; according to the specific actual situation, an agent can be assigned multiple roles, a single role, or no role; S1-3, check constraints, input the roles assigned to each action agent and the defined role constraints, and obtain the constraint check result; if passed, output the final role assignment result; if not passed, return to S1-2 and re-determine the final assigned roles of each action agent; The scheduling task process takes the external goals given by external users and the i-star demand model stored in the planning agent as input, and outputs the formed overall collaboration plan and each specific task to the task pool, specifically including: S2-1, identify all tasks, input the external goals and i-star demand model, and output all tasks required to achieve the external goals; S2-2, add task resource dependency, input all the tasks to be executed and the i-star demand model, identify the resources required for each task to be executed from the i-star demand model, and add resource dependency constraints for each task; S2-3, form an overall collaboration plan, input all the tasks to be executed and the i-star demand model, identify the roles required for the execution of each task from the i-star demand model; then, identify the predecessor tasks that need to be completed before the execution of each task from the i-star demand model based on the dependency relationship, build the order relationship of task execution, and form an overall collaboration plan; S2-4, parallel execution, specifically including: S2-4a, publish the collaboration plan, input the formed overall collaboration plan, encapsulate and add a serial number, and publish the collaboration plan to the collaboration plan set of the task pool; S2-4b, publish specific tasks, input the formed overall collaboration plan, and publish the tasks in sequence to the corresponding role task queues in the task queue set of the task pool according to the execution roles of each task in the plan; The ability to manage data in the task pool includes: whenever the planning agent publishes a task to the task pool, the task pool notifies the action agent players of each role according to the execution role of the published task to remind the action agent to collect new tasks from the task queue of the role it plays; whenever the action agent collects a task from the task pool or completes a task, the task pool modifies the status information of the task in the collaboration plan set to reflect the collaboration progress between the action agents in real time.
2. The multi-agent dynamic allocation and collaboration method according to claim 1, characterized in that: The S1-1 is specifically: S1-11, calculating the role ability matching score, inputting the agent ability information in the action agent information, and comparing it with the ability information of each role in turn, and calculating the ability matching score of each role to measure whether the action agent is capable of performing each role; S1-12, calculating the past task completion rate, inputting the past task completion information of the action agent when playing each role, and calculating the past task completion rate of the action agent on each role, so as to estimate the task completion rate of the agent when performing tasks in each role; S1-13. Calculate the role's comprehensive score by weighting the role's ability matching score and past task completion rate to obtain the role's comprehensive score.
3. The multi-agent dynamic allocation and collaboration method according to claim 1, characterized in that: The S2-1 is specifically: S2-11, match the target, input the external target and the i-star demand model, and output the target element in the i-star demand model that is most similar to the external target; S2-12, disassemble the target, input the matching target element and i-star requirement model, and disassemble it into multiple sub-target elements according to the "and refinement relationship" and "or refinement relationship", until all sub-target elements cannot be disassembled and become atomic target elements, and then output all atomic target elements; among them, the disassembly based on the "and refinement relationship" must retain all sub-target elements, while the disassembly based on the "or refinement relationship" can randomly retain a sub-target element; S2-13, identify the tasks, input all atomic target elements and i-star requirement model, take the target element as the root node, perform depth-first search, and output all tasks with relationships as all tasks that need to be executed.
Citation Information
Cited By
Multi-agent-based task solving method and system and multi-agent control equipment
CN120124906A
Task execution method and system
CN122411533A
