Intelligent agent intercommunication cooperation method, leader node, participant node and system for intelligent agent internet

Abstract

The application provides an inter-agent communication cooperation method, a leader node, a participant node and a system for an agent internet, the method is executed by an agent currently acting as a leader node in the agent internet, target tasks corresponding to target task demand data received from a user terminal are generated based on the user terminal; the number of target tasks is determined to determine the target session for the target task; the target session includes at least one mode of group, the group includes the leader node and other agents currently acting as participant nodes in the agent internet; and the target task is executed by communicating and cooperating with the participant nodes in the target session to obtain the execution result data corresponding to the target task and return to the user terminal. The application provides a feasible and scalable way for the interaction, cooperation and communication of the agents in the agent internet to specify the interaction mode and the communication mechanism, to support reliable communication and task cooperation between the agents, and to support diversified applications of the agent internet.

Description

Technical Field

[0001] This application relates to the field of intelligent agent interaction technology, and in particular to intelligent agent communication and cooperation methods, leader nodes, participating nodes and systems for intelligent agent Internet. Background Technology

[0002] The Internet of Agents (IoA) refers to a network formed by connecting intelligent agents with autonomous perception, planning, decision-making, and execution capabilities via the internet using standardized communication protocols. The IoA enables intelligent agents to self-organize and self-negotiate to form a highly efficient collaborative network, dynamically adjusting resource allocation and collaboration methods according to user task requirements. This overcomes the limitations of single-agent capabilities and breaks free from the constraints of vendor-manufactured multi-agent frameworks, improving the overall capability and efficiency of intelligent agent systems. In the IoA, multiple intelligent agents need to collaborate to complete a specified complex task. During this collaboration, intelligent agents need to negotiate task details and synchronize task progress through reliable communication.

[0003] However, existing communication methods between intelligent agents lack standardization, have excessive uncertainty, and are not widely applicable. Therefore, there is an urgent need to design a standardized, reliable, and scalable communication and collaboration method between intelligent agents. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, leader node, participating node and system for inter-agent communication and cooperation in the Internet of Intelligent Agents, in order to eliminate or improve one or more defects existing in the prior art.

[0005] One aspect of this application provides a method for inter-agent communication and cooperation in an intelligent agent internet, executed by an intelligent agent currently acting as a leader node in the intelligent agent internet, the method comprising:

[0006] Generate corresponding target tasks based on user task requirement data received from the user terminal;

[0007] The target session for the target task is determined based on the number of target tasks; wherein the target session includes a group of at least one pattern, and the group includes an agent that acts as a leader node and other agents in the agent network that are currently participating nodes;

[0008] The system communicates and collaborates with the participating nodes in the target session to complete the target task, obtains the execution result data corresponding to the target task, and returns it to the user terminal.

[0009] In some embodiments of this application, determining the target session for the target tasks based on the number of target tasks includes:

[0010] If the user task requirement data corresponds to a target task, and there is currently an idle session in the intelligent agent Internet that contains a point-to-point mode group, then the idle session is reused as the target session corresponding to the target task.

[0011] If the user task requirement data corresponds to a target task, and there is currently no idle session in the agent internet that contains a point-to-point mode group, then an agent is selected in the agent internet as a participating node in the point-to-point mode group, the target session containing the point-to-point mode group is reconstructed, and a unique identifier for the target session is set.

[0012] In some embodiments of this application, determining the target session for the target tasks based on the number of target tasks includes:

[0013] If the user task requirement data corresponds to multiple target tasks, a directed acyclic graph (DAG) composed of each target task is constructed. The DAG includes an Object-Oriented Virtual Network (AOV) or an Object-Oriented Array (AOE). The AOV uses the target task as a vertex, and the directed edges connecting the vertices represent task dependencies. The AOE uses the execution process of the target task as directed edges, and the common point between the beginning and end of different directed edges in the AOE represents the event that all preceding dependent tasks corresponding to any target task have been completed.

[0014] Based on the directed acyclic graph, the corresponding target session requirement data is determined, wherein the target session requirement data includes: the target mode and number of groups required to constitute the target session, the number of agents as participating nodes required to constitute each group, and the idle resource rate of each agent as participating node; the target mode includes: multicast mode and hybrid mode, wherein the hybrid mode is composed of a combination of point-to-point mode and multicast mode;

[0015] If there is an idle session in the intelligent agent internet that meets the data requirements of the target session, then the idle session is reused as the target session corresponding to the target task;

[0016] If there is no idle session in the agent internet that meets the target session requirements, then according to the target session requirements, each agent in the agent internet is selected as a participating node in the group specified by the target session requirements, the target session containing the group is reconstructed, and a unique identifier for the target session is set.

[0017] In some embodiments of this application, communicating and cooperating with the participating nodes in the target session to complete the target task, obtaining the execution result data corresponding to the target task, and returning it to the user terminal includes:

[0018] If the target session contains a point-to-point mode group, the target task is sent to the participating nodes in the group so that the participating nodes perform a preset first task execution step to obtain the corresponding task completion data and return the task completion data.

[0019] The first task execution steps include:

[0020] If it is determined that the data carried by the target task contains the task execution data required to complete the target task, then the target task is executed based on the task execution data;

[0021] If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed with the leader node in the target session until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

[0022] In some embodiments of this application, communicating and cooperating with the participating nodes in the target session to complete the target task, obtaining the execution result data corresponding to the target task, and returning it to the user terminal includes:

[0023] If the target session contains a multicast group, then subscribe to the message queue and send request data to each participating node in the group to notify them of the message queue subscription, so that each participating node subscribes to the message queue according to the request data and returns request data to notify them of the successful message queue subscription.

[0024] The system receives request data from each participating node to notify that the message queue subscription has been successful. Based on the directed acyclic graph, it adds each target task to the message queue, so that the message queue forwards each target task to the designated participating node, so that the participating node executes a preset second task execution step to obtain the corresponding task completion data and returns the task completion data.

[0025] The second task execution steps include:

[0026] If it is determined that the data carried by the target task contains the task execution data required to complete the target task, then the target task is executed based on the task execution data;

[0027] If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed between the leader node in the target session and / or other participating nodes in the same group via the message queue until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

[0028] In some embodiments of this application, communicating and cooperating with the participating nodes in the target session to complete the target task, obtaining the execution result data corresponding to the target task, and returning it to the user terminal includes:

[0029] If the target session contains a hybrid mode group, the target task is sent to each of the participating nodes in the hybrid mode group that are connected to the leader node in a point-to-point mode, so that the participating nodes execute a preset first task execution step to obtain the corresponding task completion data and return the task completion data.

[0030] Furthermore, the system subscribes to message queues and sends request data to each participating node in the hybrid mode group that is connected to the leader node in multicast mode to notify them of the message queue subscription. This allows each participating node connected to the leader node in multicast mode to subscribe to the message queue according to the request data and return request data to notify them of successful message queue subscription. The system also receives request data from each participating node connected to the leader node in multicast mode to notify them of successful message queue subscription, adds each target task to the message queue according to the directed acyclic graph, and forwards each target task to the designated participating nodes. This allows each participating node connected to the leader node in multicast mode to execute a preset second task execution step to obtain corresponding task completion data and return the task completion data.

[0031] A second aspect of this application provides a leader node for the aforementioned inter-agent communication and cooperation method for an intelligent agent Internet.

[0032] A third aspect of this application provides a participating node for performing the following:

[0033] If a target task is received directly from the leader node in the group, a preset first task execution step is executed to obtain the corresponding task completion data and the task completion data is returned; wherein, the leader node is used to execute the intelligent agent inter-agent communication and cooperation method for intelligent agent Internet.

[0034] If a target task forwarded by a message queue is received from the group, a preset second task execution step is executed to obtain the corresponding task completion data and the task completion data is returned.

[0035] The first task execution steps include:

[0036] If it is determined that the data carried by the target task contains the task execution data required to complete the target task, then the target task is executed based on the task execution data;

[0037] If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed with the leader node in the target session until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

[0038] The second task execution steps include:

[0039] If it is determined that the data carried by the target task contains the task execution data required to complete the target task, then the target task is executed based on the task execution data;

[0040] If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed between the leader node in the target session and / or other participating nodes in the same group via the message queue until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

[0041] In some embodiments of this application, the participating node, during the execution of the target task based on the task execution data, is also configured to perform the following:

[0042] Determine whether collaboration with other intelligent agents is needed to complete the target task;

[0043] If not, the target task is executed locally based on the task execution data.

[0044] If so, the current participating node will also act as the new leader node, and a secondary session for the target task will be determined based on the number of target tasks. The leader node will communicate and cooperate with other participating nodes in the secondary session to complete the target task.

[0045] The fourth aspect of this application provides an inter-agent communication and cooperation system for an intelligent agent Internet, comprising: the leader node mentioned in the second aspect above, and the participating node mentioned in the third aspect above.

[0046] This application provides a method for inter-agent communication and collaboration in an intelligent agent internet. The method involves an intelligent agent currently acting as a leader node generating a corresponding target task based on user task request data received from a user terminal; determining a target session for each target task based on its quantity; wherein each target session includes at least one type of group, and the group includes the intelligent agent acting as the leader node and other intelligent agents currently acting as participating nodes in the intelligent agent internet; communicating and collaborating with the participating nodes in the target session to execute the target task; obtaining execution result data corresponding to the target task and returning it to the user terminal. This method provides a standardized, feasible, and scalable way to define interaction modes and communication mechanisms for inter-agent interaction and collaboration in an intelligent agent internet, supporting reliable communication and task collaboration between intelligent agents and enabling diverse applications of the intelligent agent internet.

[0047] Additional advantages, objectives, and features of this application will be set forth in part in the description which follows, and will in part become apparent to those skilled in the art upon review of the following description, or may be learned by practice of the application. The objectives and other advantages of this application can be realized and obtained by means of the structures specifically pointed out in the specification and drawings.

[0048] Those skilled in the art will understand that the purposes and advantages that can be achieved with this application are not limited to those specifically described above, and that the above and other purposes that this application can achieve will be more clearly understood from the following detailed description. Attached Figure Description

[0049] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, do not constitute a limitation thereof. The components in the drawings are not drawn to scale but are merely for illustrating the principles of this application. For ease of illustration and description of certain parts of this application, corresponding portions in the drawings may be enlarged, i.e., may appear larger relative to other components in an exemplary device actually manufactured according to this application. In the drawings:

[0050] Figure 1 This is a schematic diagram of the first process of an intelligent agent communication and cooperation method for intelligent agent Internet according to an embodiment of this application.

[0051] Figure 2 This is a schematic diagram of a point-to-point mode group in one embodiment of this application.

[0052] Figure 3 This is a schematic diagram of a group in a multicast mode according to an embodiment of this application.

[0053] Figure 4 This is a schematic diagram of a group of mixed modes in one embodiment of this application.

[0054] Figure 5 This is a schematic diagram of the intelligent agent interaction network structure in one example of this application.

[0055] Figure 6 This is a schematic diagram of a second process for an intelligent agent communication and cooperation method for an intelligent agent Internet according to an embodiment of this application.

[0056] Figure 7 This is a schematic diagram of the inter-agent communication process in a point-to-point interactive mode in a video example of this application.

[0057] Figure 8 This is a schematic diagram of the inter-agent communication process in a multicast mode interaction in a video example of this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain this application, but are not intended to limit it.

[0059] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the structures and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0060] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0061] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0062] In the following description, embodiments of the present application will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0063] It should be noted that the Internet of Intelligent Agents (IAI) is currently one of the cutting-edge research areas in IAI technology. However, solutions for achieving interaction and collaboration between IAI agents within the IAI are not yet fully mature. There are currently few applications related to multi-agent collaboration; most define the communication connection framework between IAI agents. Specific protocols that can be used to standardize communication and interaction between IAI agents include A2A (Agent to Agent), ACP (Agent Communication Protocol), and ANP (Agent Network Protocol). However, all of these existing protocols have some shortcomings to varying degrees.

[0064] Specifically, for interactive and collaborative scenarios between intelligent agents in the intelligent agent internet, current patent technologies related to multi-agent collaboration have the following shortcomings:

[0065] One existing multi-agent collaboration method and system only uses Large Language Model (LLM) for task decomposition and allocation, without addressing the details of inter-agent communication and collaboration. Another existing multi-agent communication collaboration method is mainly applied to agent communication in multi-agent reinforcement learning, which differs from the application scenario of this protocol. A third existing multi-agent collaboration system provides a framework for multi-agent collaboration to complete tasks, but does not define the implementation details of inter-agent interaction and communication, such as agent communication flow and message format, only mentioning the use of an interaction paradigm module to enable information exchange between agents through a unified language interaction interface. Yet another existing multi-agent interaction framework mainly focuses on the overall framework and flow of multi-agent interaction and collaboration without providing specific flow and implementation schemes for inter-agent communication.

[0066] Specifically, regarding the implementation schemes for interaction and communication between intelligent agents, the shortcomings of the existing intelligent agent communication protocols are mainly manifested in the following aspects:

[0067] (1) ACP is not yet perfect and lacks many key details. For example, regarding communication between agents, it only mentions sending messages via HTTPS without specifying the communication process. At the same time, the message content format definition for sending messages between agents in ACP is too simple and not clear enough.

[0068] (2) The A2A protocol is relatively complete and has relatively high operability, but its application scenarios are mainly for communication between two intelligent agents, and its versatility is slightly lacking for intelligent agent Internet scenarios.

[0069] (3) ANP primarily focuses on policy negotiation and resource optimization among agents, which differs slightly from the scenarios addressed in this protocol. During the collaborative communication process after negotiation, the natural language message format used by ANP is mainly determined by the negotiation process, lacking standardization and exhibiting excessive uncertainty. Its predefined message format is simple, making it difficult for agents to obtain sufficient information from historical messages. Furthermore, ANP's natural language message communication only adopts a request-response pair pattern, limiting its applicability.

[0070] Based on this, in order to address the problems of insufficient standardization, excessive uncertainty, and limited applicability of existing intelligent agent interaction communication methods, this application provides an intelligent agent communication and cooperation method for the intelligent agent Internet, a leader node for executing the intelligent agent communication and cooperation method for the intelligent agent Internet, related participating nodes, and an intelligent agent communication and cooperation system for the intelligent agent Internet. This system can provide the intelligent agent Internet with a standardized and flexible interactive communication scheme applicable to various types of task scenarios.

[0071] The following examples will provide a detailed description.

[0072] Based on this, embodiments of this application provide a method for inter-agent communication and cooperation in an intelligent agent Internet that can be executed by a leader node. See [link to relevant documentation]. Figure 1 The method for inter-agent communication and cooperation in the Internet of Intelligent Agents specifically includes the following:

[0073] Step 100: Generate the corresponding target task based on the user task requirement data received from the user terminal.

[0074] Step 200: Determine the target session for the target task based on the number of target tasks; wherein the target session includes at least one pattern of groups, and the group includes the agent itself as the leader node and other agents in the agent network currently acting as participating nodes.

[0075] Step 300: Communicate and cooperate with the participating nodes in the target session to complete the target task, obtain the execution result data corresponding to the target task, and return it to the user terminal.

[0076] It should be noted that in agent interaction protocols, agents collaborating towards a common task goal form a group. Agents within a group may play one of the following two roles:

[0077] (1) Leader node, also known as leader; the leader node is an intelligent agent that issues tasks to participating nodes (Partner) according to the task objectives and manages them. A group has one and only one leader node.

[0078] (2) Participating node (Partner), also known as participant; a participating node (Partner) is an intelligent agent that accepts a task from the leader node (Leader), executes the task, and returns the task execution result. There can be one or more participating nodes (Partner) in a group.

[0079] It is worth noting that, after accepting a task, an agent acting as a partner can, after analysis and planning, determine that collaboration with other agents is needed to complete the task. In this case, it can act as a leader to find other agents and create a new group, and then issue the task to the other partner agents in the new group.

[0080] In one or more embodiments of this application, the target session refers to the current session. A session is identified by a unique ID and is responsible for organizing, coordinating, and managing all agents participating in the task, as well as storing the interaction context between agents. A session contains at least one group of agents and corresponds to at least one complete user task completion process.

[0081] The lifecycle of a session is as follows: When a user makes a request to the leader, the leader first searches for a reusable session. If none is found, it declares a new session. Within this session, the leader publishes tasks to the discovered participating partners and maintains the session. Each participating partner completes its task within the session and submits the result to the leader. The leader determines whether the user's task is complete and returns the result to the user. If all agents within the session have completed their tasks, the leader can close the session.

[0082] It is understandable that the user end can refer to an app, an object instance, a client device, or other objects that can serve as an interaction interface between the user and the intelligent agent.

[0083] In one or more embodiments of this application, the target task refers to the task to be executed. The task is the basic working unit in the communication process defined in this application. It is identified by a unique ID belonging to a certain session and has a defined lifecycle and state.

[0084] A complex task should be divided into multiple simple subtasks by the leader node to create corresponding tasks, which are then handed over to the corresponding participating nodes for execution, updating, and maintenance. The leader node decides whether a task is to be terminated.

[0085] When an agent receives a task, it first checks whether the task is recorded locally. If not, it stores the task and executes it based on the message carried by the task. Otherwise, it updates the local task status.

[0086] In one or more embodiments of this application, a message is a basic communication unit in the communication process defined in this application. It is identified by a unique ID belonging to a certain task and records the sender and receiver of the message, as well as one or more data parts. Depending on the meaning of the data carried by the message, messages can be divided into two categories: one is communication between intelligent agents, and the other is the work results of participating nodes (Partners) regarding the task.

[0087] In one or more embodiments of this application, data is the basic content unit in the communication process defined in this application. It is sent along with a message and has various data types, with different parameters depending on the type of transmitted content. The agent can determine the type and meaning of the data based on its fields, thereby obtaining the received interactive information.

[0088] As can be seen from the above description, the intelligent agent communication and cooperation method for intelligent agent Internet provided in the embodiments of this application provides a standardized, feasible and highly scalable way to define the interaction mode and communication mechanism for intelligent agent interaction and cooperation communication in intelligent agent Internet, so as to support reliable communication and task cooperation between intelligent agents and support the diversified applications of intelligent agent Internet.

[0089] To further improve the reliability and standardization of inter-agent communication and collaboration in the intelligent agent Internet, this application provides an embodiment of an inter-agent communication and collaboration method for the intelligent agent Internet. The mode types include: point-to-point mode and target mode. The target mode further includes: multicast mode and hybrid mode. The hybrid mode is composed of a combination of point-to-point mode and multicast mode. That is, in a group within the intelligent agent Internet, the communication interaction between the leader node and participating nodes mainly has three modes. Specific details are as follows:

[0090] (1) Point-to-point mode

[0091] In peer-to-peer mode, the leader node in a group interacts directly with each participating node, while there is no interaction between participating nodes. This mode is suitable for scenarios with simple tasks and unrelated subtasks. The communication mechanism between the leader and participating nodes can employ three methods depending on the application scenario: request-response pairs, server-sent events (SSE), and push notifications. The structure of a peer-to-peer group is as follows: Figure 2 As shown.

[0092] (2) Multicast mode

[0093] In multicast mode, communication messages between the leader and partners are distributed through a message queue. This is suitable for scenarios with complex tasks, where agents may need to communicate with each other to execute subtasks, or where tasks need to be executed simultaneously by different agents. After the leader creates a session, it notifies the partners in the group to subscribe to messages in that session from the message queue. Then, communication messages between the leader and partners are sent to the message queue for distribution. For each message, the sender is the publisher and the receiver is the receiver. The message queue determines which agent to forward the message to based on the message's receiver attribute. The structure of a group in multicast mode is as follows: Figure 3 As shown.

[0094] (3) Hybrid mode

[0095] In hybrid mode, within a group, the leader node interacts with participating nodes both through point-to-point and multicast methods, dynamically determining the communication method to conserve server resources. The structure of a hybrid group is as follows: Figure 4 As shown.

[0096] In the Internet of Agents, multiple groups can interconnect and reuse message queue services within the network, forming a dynamic agent interaction network. In one example, the agent interaction network structure comprising multiple groups in a target session at a given time is as follows: Figure 5 As shown.

[0097] To optimize session resource utilization in single-objective task scenarios and avoid the overhead caused by repeatedly creating sessions, this application provides a method for inter-agent communication and cooperation in an intelligent agent Internet framework, see [link to relevant documentation]. Figure 6 Step 200 of the method for inter-agent communication and cooperation in the Internet of Intelligent Agents specifically includes the following:

[0098] Step 210: If the user task requirement data corresponds to a target task, and there is currently an idle session in the intelligent agent Internet that contains a point-to-point mode group, then the idle session is reused as the target session corresponding to the target task.

[0099] Step 220: If the user task requirement data corresponds to a target task, and there is currently no idle session in the agent internet that contains a point-to-point mode group, then select an agent in the agent internet as a participating node in the point-to-point mode group, reconstruct the target session containing the point-to-point mode group, and set a unique identifier for the target session.

[0100] Correspondingly, see Figure 6 Step 300 of the method for inter-agent communication and cooperation in the Internet of Intelligent Agents specifically includes the following:

[0101] Step 310: If the target session contains a point-to-point mode group, the target task is sent to the participating nodes in the group so that the participating nodes execute the preset first task execution steps to obtain the corresponding task completion data and return the task completion data.

[0102] The first task execution step performed by the participating nodes specifically includes the following:

[0103] (1) If it is determined that the data carried by the target task contains task execution data required to complete the target task, then the target task is executed based on the task execution data;

[0104] (2) If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed with the leader node in the target session until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

[0105] One feasible method for inter-agent communication is to use HTTP(S) as the transport protocol and to format all communication messages using JSON-RPC 2.0.

[0106] In one example, see Figure 7 The point-to-point mode is suitable for general agent collaborative communication scenarios where the task is simple and only requires the cooperation of one agent. Specifically, it can include the following:

[0107] (1) Task Issuance: After planning and analysis, the Leader creates a Task and sends a request to the Agent Discovery Server to find a suitable Partner to complete the Task. According to the task management mechanism, the Leader sends a request for the Task to the Partner through the Partner's access endpoint at an appropriate time. After receiving the Task, the Partner sends a task acceptance response to the Leader and begins to execute the Task.

[0108] (2) Task Execution: If the task provided by the leader node provides sufficient information, the partner node will directly execute the task until completion. If the partner node finds that the necessary information for completing the task is missing during execution, it will initiate a communication request to the leader node according to the task execution requirements. After receiving the request, the leader node will analyze the content and respond. Multiple rounds of dialogue will continue until the partner node deems the supplementary information sufficient, at which point the partner node will continue executing the task.

[0109] (3) Task Completion: Upon completion of the task, the participating node (Partner) submits an updated task (Task) to the leader node (Leader). Upon receiving the task (Task), the leader node immediately sends a response to the participating node (Partner) confirming receipt of the task result, ensuring that a timeout mechanism is not triggered and the participating node (Partner) resends the task (Task). Based on this task (Task), the leader node obtains the complete task process and result, determines whether the task is complete, and then sends the task completion result to the participating node (Partner). Upon receiving the notification from the leader node (Leader), the participating node (Partner) immediately returns a confirmation response and performs subsequent corresponding operations (such as releasing resources or continuing task execution according to the leader node's requirements).

[0110] To address the challenges of dependency management and complex session construction among multi-objective tasks, this application provides a method for inter-agent communication and collaboration in an embodiment of the intelligent agent internet, see [link to relevant documentation]. Figure 6 Step 200 of the method for inter-agent communication and cooperation in the Internet of Intelligent Agents further includes the following:

[0111] Step 230: If the user task requirement data corresponds to multiple target tasks, then construct a directed acyclic graph (DAG) composed of each target task, wherein the DAG includes an AOV network or an AOE network; the AOV network uses the target task as a vertex, and the directed edges connecting the vertices in the AOV network are used to represent task dependencies; the AOE network uses the execution process of the target task as a directed edge, and the common point between the arc head and arc tail of different directed edges in the AOE network is used to represent the event that all the preceding dependent tasks corresponding to any target task have been completed;

[0112] Step 240: Determine the corresponding target session requirement data based on the directed acyclic graph, wherein the target session requirement data includes: the target mode and number of groups required to constitute the target session, the number of agents as participating nodes required to constitute each group, and the idle resource rate of each agent as participating node; the target mode includes: multicast mode and hybrid mode, wherein the hybrid mode is composed of a combination of point-to-point mode and multicast mode.

[0113] Specifically, the task management mechanism is based on the task planning capabilities of the leader node and the status of the tasks. After the leader node plans and breaks down complex tasks, it generates a directed acyclic graph (DAG) composed of tasks. Based on this DAG, the dependencies between the tasks can be clearly determined, and the resource consumption of each task can be estimated based on the expected difficulty of each task and the performance of the corresponding partner nodes.

[0114] Tasks that meet the preconditions and have no dependencies can be executed in parallel. Once all tasks within the directed acyclic graph (DAG) have completed, the leader node can integrate the results and return the final execution result of the complex task to the user. This DAG allows for task management and resource scheduling based on the state of each task and its current resource usage, thereby improving resource utilization and task execution efficiency. For example, when network computing power is insufficient, tasks with low expected resource consumption and which are preconditions of most tasks in the DAG are prioritized for distribution to the corresponding partner nodes. Furthermore, controlling the input of context messages based on this DAG can alleviate the context decay problem caused by inputting irrelevant historical context into the agent due to the lack of a task management mechanism.

[0115] Step 250: If there is an idle session in the intelligent agent Internet that meets the data requirements of the target session, then the idle session is reused as the target session corresponding to the target task;

[0116] Step 260: If there is no idle session in the intelligent agent Internet that meets the target session requirements data, then according to the target session requirements data, select each intelligent agent in the intelligent agent Internet as a participating node in the group specified by the target session requirements data, reconstruct the target session containing the group, and set a unique identifier for the target session.

[0117] Correspondingly, see Figure 6 Step 300 of the method for inter-agent communication and cooperation in the Internet of Intelligent Agents specifically includes the following:

[0118] Step 320: If the target session contains a multicast group, subscribe to the message queue and send request data to each participating node in the group to notify them of the message queue subscription, so that each participating node subscribes to the message queue according to the request data and returns request data to notify them of the successful message queue subscription.

[0119] Step 330: Receive request data sent by each of the participating nodes to notify that the message queue subscription has been successful. Add each of the target tasks to the message queue according to the directed acyclic graph, so that the message queue forwards each of the target tasks to the designated participating nodes, so that the participating nodes execute the preset second task execution steps to obtain the corresponding task completion data and return the task completion data.

[0120] The second task execution step, which involves the participating nodes, specifically includes the following:

[0121] (1) If it is determined that the data carried by the target task contains task execution data required to complete the target task, then the target task is executed based on the task execution data;

[0122] (2) If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed between the leader node in the target session and / or other participating nodes in the same group through the message queue until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

[0123] In one example, see Figure 8 Multicast mode is suitable for complex collaborative communication scenarios where agents need to communicate with each other on complex tasks or where the same task needs to be executed by different agents simultaneously. Specifically, it can include the following:

[0124] (1) Task Issuance: After planning and analysis, the Leader creates a Task and sends a request to the Agent Discovery Server to find multiple Partners suitable for completing the Task. The Leader sends a request to each Partner through their respective access endpoints, along with a message queue port and information about all Partners participating in the Task, to create a group. Upon receiving the request, each Partner subscribes to the message queue and returns a response to the Leader, establishing connections with each other through the message queue. According to the task management mechanism, the Leader designates the task recipient in the Task at an appropriate time and sends it to the message queue, which then distributes it to the corresponding Partners. Upon receiving the Task to be executed, each Partner sends a task acceptance response to the Leader and begins executing the Task.

[0125] (2) Task Execution: If the task provided by the leader node provides sufficient information, the participating node directly executes the task until completion. If a participating node finds that the necessary information for completing the task is missing during execution, the participating node designates the leader node or another participating node as the receiver and initiates a communication request through a message queue. After receiving the request from the participating node, the receiver analyzes the content and responds. Multiple rounds of dialogue between the two parties continue until the participating node deems the supplementary information sufficient, at which point the participating node continues to execute the task.

[0126] (3) Task Completion: Upon completion of the task, each participating node (Partner) designates the Leader node as the recipient and submits an updated task to the Leader node via a message queue. Upon receiving the task, the Leader node immediately returns a confirmation response to the participating nodes (Partners) confirming receipt of the task result, ensuring that a timeout mechanism is not triggered and the participating nodes (Partners) do not resend the task. The Leader node obtains the complete task process and result based on this task and determines whether the task is complete, then sends the task completion result to the participating nodes (Partners). Upon receiving the notification from the Leader node (Partner), each participating node (Partner) immediately returns a confirmation response and performs subsequent corresponding operations. When all participating nodes (Partners) have completed their tasks, meaning there is no longer a need for further communication via the message queue, the Leader node designates all participating nodes (Partners) involved in the task as recipients and notifies them to unsubscribe from the message queue to close the group.

[0127] Correspondingly, see Figure 6 Step 300 of the method for inter-agent communication and cooperation in the Internet of Intelligent Agents specifically includes the following:

[0128] Step 340: If the target session contains a mixed-mode group, the target task is sent to each of the participating nodes in the mixed-mode group that are connected to the leader node in a point-to-point mode, so that the participating nodes execute a preset first task execution step to obtain the corresponding task completion data and return the task completion data.

[0129] Step 350: Subscribe to the message queue and send request data to each participating node in the hybrid mode group connected to the leader node in multicast mode to notify them of the message queue subscription, so that each participating node connected to the leader node in multicast mode subscribes to the message queue according to the request data and returns request data to notify them of successful message queue subscription; receive the request data sent by each participating node connected to the leader node in multicast mode to notify them of successful message queue subscription, add each target task to the message queue according to the directed acyclic graph, so that the message queue forwards each target task to the designated participating node, so that each participating node connected to the leader node in multicast mode executes the preset second task execution step to obtain the corresponding task completion data and returns the task completion data.

[0130] In one example, the hybrid mode is suitable for complex agent collaborative communication scenarios where the task is complex and the leader and partners may need to communicate and collaborate via both point-to-point and multicast modes. This mode is the most flexible and has the widest range of applications, and can specifically include the following:

[0131] (1) Task Issuance: After planning and analysis, the Leader creates a Task and sends a request to the Agent Discovery Server to find multiple Partners suitable for completing the Task. For Partners that need to connect via multicast, the Leader sends a request with the message queue port and information of all Partners to each Partner through their respective access endpoints to create a group. Upon receiving the request, each Partner subscribes to the message queue and returns a response to the Leader, establishing a connection with each other through the message queue. Then, according to the task management mechanism, the Leader sends the Task to the corresponding Partner at the appropriate time through the message queue or the Partner's access endpoint. Upon receiving the Task, the Partner sends a Task Acceptance Response to the Leader and begins executing the Task.

[0132] (2) Task execution: Each participating node (Partner) that receives a task executes the task and obtains additional information according to its own interaction mode and communication requirements.

[0133] (3) Task Completion: After a participating node (Partner) determines that the task has been completed, it submits an updated task (Task) to the leader node (Leader) according to its respective interaction mode. Upon receiving the task (Task), the leader node (Leader) immediately returns a response notifying the participating node (Partner) that the task result has been received, ensuring that the timeout mechanism is not triggered and the participating node (Partner) resends the task (Task). Based on this task (Task), the leader node (Leader) obtains the complete task process and result and determines whether the task is completed, and then sends the task completion determination result to the participating node (Partner). After receiving the notification from the leader node (Leader), the participating node (Partner) immediately returns a confirmation response and performs subsequent corresponding operations. When there is no longer a need for participating nodes (Partners) to communicate through the message queue, the leader node (Leader) designates all participating nodes (Partners) that have subscribed to the message queue as recipients and notifies them to unsubscribe from the message queue to close this group.

[0134] In other words, existing technologies defining the agent connection framework and the process of assigning tasks to multiple agents in one-to-many agent collaboration scenarios lack specific implementation details regarding the communication and collaboration processes and message format design between agents during task execution. In contrast, the method provided in this application provides a more detailed and explicit definition of the communication process and implementation details between agents, exhibiting higher standardization and feasibility, effectively avoiding communication failures caused by the lack of a unified message processing method among agents. Furthermore, existing technologies that only define the connection method between agents without specifying the communication process and message design, in contrast, define in detail the communication and collaboration process and information communication mechanism between agents after task allocation, filling the gap in the implementation scheme of inter-agent communication and collaboration. Additionally, the A2A protocol only focuses on communication between two independent agents without considering more agents, and does not provide communication design for task scenarios involving multiple agents collaborating. In comparison, this protocol offers well-adapted solutions for communication needs between multiple agents in different application scenarios, making it more versatile and usable.

[0135] It should also be noted that the intelligent agent used in the embodiments of this application is essentially a computational entity capable of autonomously perceiving the environment, making decisions, and executing tasks. Its core characteristics are autonomy, interactivity, responsiveness, and adaptability. The intelligent agent can be a software entity, existing in the form of a program, and achieves autonomous decision-making through algorithms and large models (such as LLMs), such as customer service chatbots and data analysis tools.

[0136] Based on this, this application also provides an embodiment of a leader node, which is specifically used to execute part or all of the content of the inter-agent communication and cooperation method for intelligent agent Internet in the foregoing embodiments.

[0137] The leader node embodiment provided in this application can be used to execute the processing flow of the embodiment of the inter-agent communication and cooperation method for intelligent agent Internet in the above embodiment. Its function will not be repeated here, but can be referred to the detailed description of the embodiment of the inter-agent communication and cooperation method for intelligent agent Internet in the above embodiment.

[0138] As can be seen from the above description, the leader node provided in the embodiments of this application provides a standardized, feasible, and highly scalable way to define interaction modes and communication mechanisms for intelligent agent interaction and collaborative communication in the intelligent agent Internet, so as to support reliable communication and task collaboration between intelligent agents and support the diverse applications of the intelligent agent Internet.

[0139] In addition, this application also provides an embodiment of a participating node, which is used to perform the following:

[0140] (1) If the target task is received directly from the leader node in the group, the first task execution step is executed to obtain the corresponding task completion data and the task completion data is returned; wherein, the leader node is used to execute the intelligent agent communication and cooperation method for intelligent agent Internet.

[0141] (2) If a target task forwarded by a message queue is received from the group, the preset second task execution step is executed to obtain the corresponding task completion data and the task completion data is returned;

[0142] The first task execution steps include:

[0143] (1) If it is determined that the data carried by the target task contains task execution data required to complete the target task, then the target task is executed based on the task execution data;

[0144] (2) If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed with the leader node in the target session until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

[0145] The second task execution steps include:

[0146] (1) If it is determined that the data carried by the target task contains task execution data required to complete the target task, then the target task is executed based on the task execution data;

[0147] (2) If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed between the leader node in the target session and / or other participating nodes in the same group through the message queue until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

[0148] In one embodiment of the participating node, the participating node is also configured to perform the following:

[0149] Determine whether collaboration with other intelligent agents is needed to complete the target task;

[0150] If not, the target task is executed locally based on the task execution data.

[0151] If so, the current participating node will also act as the new leader node, and a secondary session for the target task will be determined based on the number of target tasks. The leader node will communicate and cooperate with other participating nodes in the secondary session to complete the target task.

[0152] This application also provides an embodiment of an inter-agent communication and cooperation system for the Internet of Intelligent Agents, which includes the leader node and participating nodes mentioned in the above embodiments.

[0153] In summary, this application realizes multi-agent collaborative communication in groups through message queues, constructs intelligent agent groups with multiple communication methods, and standardizes the communication and collaboration process between multi-agents after task allocation.

[0154] This application applies to three intelligent agent interaction modes under different task requirements in the intelligent agent Internet, as well as the method proposed in this paper for dynamically managing tasks and allocating resources within a group through a task management mechanism led by a leader node.

[0155] In the intelligent agent internet, the method and process by which intelligent agents acting as leaders communicate with participating intelligent agents through a peer-to-peer collaborative group mainly includes three parts: task publishing, task execution, and task completion. Leaders and partners publish tasks, engage in multi-round dialogue negotiation, and submit tasks through request-response mechanisms.

[0156] In the intelligent agent internet, the method and process by which intelligent agents acting as leader nodes communicate with participating intelligent agents through a multicast collaborative group mainly includes three parts: task publishing, task execution, and task completion. The leader node notifies participating nodes via request-response to subscribe to a message queue, enabling communication between them. Subsequently, the leader and participating nodes publish tasks, engage in multi-round negotiation, and submit tasks through request-response. Once all participating nodes no longer need to communicate via the message queue, the leader node sends a notification to all participating nodes via the message queue to unsubscribe from the message queue.

[0157] In the intelligent agent internet, the method and process by which intelligent agents acting as leaders communicate with intelligent agents acting as partners through a hybrid mode to build intelligent agent collaborative groups combines point-to-point and multicast modes, and has the highest flexibility and applicability.

[0158] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned inter-agent communication and cooperation method for the Internet of Intelligent Agents.

[0159] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave.

[0160] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0161] In this application, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0162] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to the embodiments of this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for inter-agent communication and cooperation in an intelligent agent Internet, characterized in that, The method, executed by the agent currently acting as the leader node in the agent internet, includes: Generate corresponding target tasks based on user task requirement data received from the user terminal; The target session for the target task is determined based on the number of target tasks; wherein the target session includes a group of at least one pattern, and the group includes an agent that acts as a leader node and other agents in the agent network that are currently participating nodes; The system communicates and collaborates with the participating nodes in the target session to execute and complete the target task, obtains the execution result data corresponding to the target task, and returns it to the user terminal. Determining the target session for the target tasks based on the number of target tasks includes: If the user task requirement data corresponds to multiple target tasks, a directed acyclic graph (DAG) composed of each target task is constructed. The DAG includes an Object-Oriented Virtual Network (AOV) or an Object-Oriented Array (AOE). The AOV uses the target task as a vertex, and the directed edges connecting the vertices represent task dependencies. The AOE uses the execution process of the target task as directed edges, and the common point between the beginning and end of different directed edges in the AOE represents the event that all preceding dependent tasks corresponding to any target task have been completed. Based on the directed acyclic graph, the corresponding target session requirement data is determined, wherein the target session requirement data includes: the target mode and number of groups required to constitute the target session, the number of agents as participating nodes required to constitute each group, and the idle resource rate of each agent as participating node; the target mode includes: multicast mode and hybrid mode, wherein the hybrid mode is composed of a combination of point-to-point mode and multicast mode; If there is an idle session in the intelligent agent internet that meets the data requirements of the target session, then the idle session is reused as the target session corresponding to the target task; If there is no idle session in the agent internet that meets the target session requirements, then according to the target session requirements, each agent in the agent internet is selected as a participating node in the group specified by the target session requirements, the target session containing the group is reconstructed, and a unique identifier for the target session is set.

2. The inter-agent communication and cooperation method for the intelligent agent Internet according to claim 1, characterized in that, Determining the target session for the target tasks based on the number of target tasks includes: If the user task requirement data corresponds to a target task, and there is currently an idle session in the intelligent agent Internet that contains a point-to-point mode group, then the idle session is reused as the target session corresponding to the target task. If the user task requirement data corresponds to a target task, and there is currently no idle session in the agent internet that contains a point-to-point mode group, then an agent is selected in the agent internet as a participating node in the point-to-point mode group, the target session containing the point-to-point mode group is reconstructed, and a unique identifier for the target session is set.

3. The inter-agent communication and cooperation method for the intelligent agent Internet according to claim 2, characterized in that, Communicating and cooperating with the participating nodes in the target session to complete the target task, obtaining the execution result data corresponding to the target task, and returning it to the user terminal, including: If the target session contains a point-to-point mode group, the target task is sent to the participating nodes in the group so that the participating nodes perform a preset first task execution step to obtain the corresponding task completion data and return the task completion data. The first task execution steps include: If it is determined that the data carried by the target task contains the task execution data required to complete the target task, then the target task is executed based on the task execution data; If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed with the leader node in the target session until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

4. The inter-agent communication and cooperation method for the intelligent agent Internet according to claim 3, characterized in that, Communicating and cooperating with the participating nodes in the target session to complete the target task, obtaining the execution result data corresponding to the target task, and returning it to the user terminal, including: If the target session contains a multicast group, then subscribe to the message queue and send request data to each participating node in the group to notify them of the message queue subscription, so that each participating node subscribes to the message queue according to the request data and returns request data to notify them of the successful message queue subscription. The system receives request data from each participating node to notify that the message queue subscription has been successful. Based on the directed acyclic graph, it adds each target task to the message queue, so that the message queue forwards each target task to the designated participating node, so that the participating node executes a preset second task execution step to obtain the corresponding task completion data and returns the task completion data. The second task execution steps include: If it is determined that the data carried by the target task contains the task execution data required to complete the target task, then the target task is executed based on the task execution data; If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed between the leader node in the target session and / or other participating nodes in the same group via the message queue until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

5. The inter-agent communication and cooperation method for the intelligent agent Internet according to claim 4, characterized in that, Communicating and cooperating with the participating nodes in the target session to complete the target task, obtaining the execution result data corresponding to the target task, and returning it to the user terminal, including: If the target session contains a hybrid mode group, the target task is sent to each of the participating nodes in the hybrid mode group that are connected to the leader node in a point-to-point mode, so that the participating nodes execute a preset first task execution step to obtain the corresponding task completion data and return the task completion data. Furthermore, the system subscribes to message queues and sends request data to each participating node in the hybrid mode group that is connected to the leader node in multicast mode to notify them of the message queue subscription. This allows each participating node connected to the leader node in multicast mode to subscribe to the message queue according to the request data and return request data to notify them of successful message queue subscription. The system also receives request data from each participating node connected to the leader node in multicast mode to notify them of successful message queue subscription, adds each target task to the message queue according to the directed acyclic graph, and forwards each target task to the designated participating nodes. This allows each participating node connected to the leader node in multicast mode to execute a preset second task execution step to obtain corresponding task completion data and return the task completion data.

6. A leader node, characterized in that, The leader node is used to execute the inter-agent communication and cooperation method for the Internet of Agents as described in any one of claims 1 to 5.

7. A participating node, characterized in that, The participating nodes are used to perform the following: If a target task is received directly from the leader node in the group, a preset first task execution step is executed to obtain the corresponding task completion data and the task completion data is returned; wherein, the leader node is used to execute the inter-agent communication and cooperation method for intelligent agent Internet as described in any one of claims 1 to 5; If a target task forwarded by a message queue is received from the group, a preset second task execution step is executed to obtain the corresponding task completion data and the task completion data is returned. The first task execution steps include: If it is determined that the data carried by the target task contains the task execution data required to complete the target task, then the target task is executed based on the task execution data; If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed with the leader node in the target session until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data. The second task execution steps include: If it is determined that the data carried by the target task contains the task execution data required to complete the target task, then the target task is executed based on the task execution data; If it is determined that the data carried by the target task does not contain the information required to complete the target task, then data interaction is performed between the leader node in the target session and / or other participating nodes in the same group via the message queue until the task execution data required to complete the target task is obtained, and then the target task is executed based on the task execution data.

8. The participating node according to claim 7, characterized in that, During the execution of the target task based on the task execution data, the participating node is also used to perform the following: Determine whether collaboration with other intelligent agents is needed to complete the target task; If not, the target task is executed locally based on the task execution data. If so, the current participating node will also act as the new leader node, and a secondary session for the target task will be determined based on the number of target tasks. The leader node will communicate and cooperate with other participating nodes in the secondary session to complete the target task.

9. A communication and cooperation system between intelligent agents for the Internet of Intelligent Agents, characterized in that, include: The leader node as described in claim 6, and the participating node as described in claim 7 or 8.

Images