A Robot Cooperative Task Planning Method and System Based on a Large Model

By establishing action dependencies and transitional coordination constraints in multi-agent collaborative planning, the problem of robot action switching mismatch during task changes is solved, and the continuity and coordination of action switching during task changes are achieved.

CN123103687APending Publication Date: 2026-11-17HEFEI EMBODIED DATA OPERATION MANAGEMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611575781.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-12
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing multi-agent collaborative planning, when a task is canceled or the plan is changed, the cross-robot action dependencies are not coordinated and adjusted during the exit process, resulting in a mismatch in the timing of robot action switching, which leads to interruption of collaborative execution or action conflict.

Method used

By establishing action dependencies between robots, exit actions and transitional coordination constraints are generated, the coordination task plan is replanned, control conditions provide action switching for robots, and the dependencies are updated based on exit status feedback to ensure that the coordination conditions continue into the exit phase.

Benefits of technology

It improves the continuity and coordination of action switching during multi-robot task changes, avoids execution interruptions and action conflicts, and enhances the coordination continuity and execution flexibility in task change scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123103687A_ABST
    Figure CN123103687A_ABST
Patent Text Reader

Abstract

This application provides a robot cooperative task planning method and system based on a large model, relating to the field of robot intelligent control technology. The method includes: acquiring task instructions, robot states, and action libraries; generating a cooperative task plan for multiple robots using a large model; establishing action dependencies between condition-providing robots, condition-dependent robots, and persistent conditions; generating exit actions and cooperative requirements upon detecting a task change; generating transitional cooperative constraints based on action dependencies and cooperative requirements; replanning the cooperative task plan using the large model; and controlling the action switching of the robot according to the transitional cooperative constraints; executing the replanned cooperative task plan; and updating the transitional cooperative constraints and action dependencies based on exit status feedback. This application can prevent the premature cancellation of persistent conditions after a task change, improving the continuity of action transitions and the reliability of cooperative execution during the exit process of multiple robots.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of robot intelligent control technology, and in particular to a robot collaborative task planning method and system based on a large model. Background Technology

[0002] In existing robot collaborative task planning, there are schemes that use large language models to parse and decompose task instructions, allocate tasks according to the capabilities and executable actions of different robots, and adjust subsequent plans based on robot execution feedback; some schemes, under a multi-agent collaborative architecture, coordinate task allocation and plan adjustment by exchanging task information, robot status and execution feedback among agents.

[0003] In multi-robot collaborative assembly, the actions of one robot may depend on another robot to maintain specific working conditions. When a task is canceled or the plan is changed, although the original task stops, the cross-robot action dependencies formed during the execution of the original task may still exist. The relevant robots cannot switch to the new task state synchronously at this time. If the relevant actions are directly canceled or replaced based on the result of task termination, the maintenance conditions may disappear before the dependency relationship is removed, resulting in the exit action losing its execution basis and further causing the interruption of robot action connection, failure of execution path or conflict of multiple robot actions.

[0004] Therefore, as can be seen from the above phenomena, the existing multi-agent collaborative planning mainly relies on the task state of each agent to update interaction information and execute actions. However, the removal of cross-robot action dependencies is also affected by the exit process of other robots, which makes the task state changes of a single agent asynchronous with the actual changes in the dependency relationship between multiple agents. Consequently, each agent adjusts its actions based on different state criteria, resulting in a mismatch in the timing of robot action switching, which leads to the interruption of collaborative execution or action conflicts. Summary of the Invention

[0005] This application provides a robot collaborative task planning method and system based on a large model, which solves the technical problem in the prior art that when a task is canceled or the plan is changed, the cross-robot action dependencies are not coordinated and adjusted with the exit process, which can easily lead to the premature cancellation of the continuous conditions required for exit and the mismatch of robot action switching timing, thereby causing the interruption of collaborative execution or action conflict.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, a robot cooperative task planning method based on a large model includes: Acquire task instructions, robot status, and motion library, and use the large model to generate collaborative task plans for multiple robots; Based on the collaborative task plan and the action library, action dependencies between different robots are established, including condition-providing robots, condition-dependent robots, and persistent conditions. When a task change is detected, an exit action is generated based on the robot state of the condition-dependent robot and the action library, and a collaborative requirement for the robot to provide the robot for the condition is formed based on the exit action. Transitional coordination constraints are generated based on the action dependencies and coordination requirements. The coordination task plan is replanned using a large model based on the transitional coordination constraints. The exit action is incorporated into the replanned coordination task plan. The robot's action switching is controlled according to the transitional coordination constraints. The replanned collaborative task plan is executed, the exit status feedback generated by the condition-dependent robot performing the exit action is collected, and the transitional collaborative constraints and the action dependencies are updated based on the exit status feedback.

[0007] Based on the above technical solution, in the robot collaborative task planning method based on a large model provided in this application, the cross-robot action dependencies formed during the multi-robot collaborative execution are characterized by establishing action dependencies between the condition-providing robot, the condition-dependent robot, and the continuous conditions according to the collaborative task plan and action library; by generating the exit action of the condition-dependent robot when the task changes, and forming collaborative requirements for the condition-providing robot based on the exit action, and generating transitional collaborative constraints in combination with the original action dependencies, the collaborative conditions that still need to be maintained in the exit stage can continue to act on the large model replanning and the action switching of the condition-providing robot; by executing the replanned collaborative task plan and updating the transitional collaborative constraints and action dependencies according to the exit status feedback, the collaborative conditions are prevented from being prematurely canceled after the task ends, thereby improving the continuity and collaboration of multi-robot action switching during task changes.

[0008] In conjunction with the first aspect above, in one possible implementation, the plurality of robots are each configured with an intelligent agent; the process of forming the collaborative requirement includes: The agent of the condition-dependent robot extracts the persistent conditions that the condition-providing robot needs to maintain during the execution of the exit action, and sends the exit action and the persistent conditions to the agent of the condition-providing robot. The agent of the robot provided by the conditions reads the action execution requirements for maintaining the continuous conditions from the action library according to the continuous conditions, compares the robot state of the robot provided by the conditions with the action execution requirements, and obtains the condition maintenance result; The condition maintenance result is sent to the agent of the condition-dependent robot, which then forms the collaborative requirement based on the persistent condition and the condition maintenance result.

[0009] In conjunction with the first aspect above, in one possible implementation, generating transitional collaboration constraints based on the action dependencies and the collaboration requirements includes: The execution conditions for the exit action are read from the action library, and the persistent conditions maintained by the robot by the conditions are extracted from the action dependencies. Match the execution conditions and the persistence conditions according to the job object, compare the state requirements of the execution conditions and the persistence conditions acting on the same job object, and mark the persistence condition as a continuing persistence condition when the state represented by the persistence condition meets the state requirements of the execution condition. From the execution conditions that have never been satisfied by the continued conditions, select the execution conditions that require the conditions to provide robot maintenance, and form new maintenance conditions; Based on the continued conditions, the new maintenance conditions, and the collaborative requirements, the conditions are used to provide the robot with the action states that need to be maintained and the actions that restrict switching, and the transitional collaborative constraints are generated.

[0010] By matching the execution conditions of the exit action with the persistence conditions in the original action dependencies according to the job object and state requirements, distinguishing between continued persistence conditions and newly added maintenance conditions, and forming action maintenance states and switching constraints accordingly, the problem of difficulty in connecting the original dependencies and exit requirements after task changes is solved, so that the transitional coordination constraints can accurately reflect the actual collaboration needs of the exit phase.

[0011] In conjunction with the first aspect above, in one possible implementation, the step of controlling the robot's action switching according to the transitional cooperative constraint control conditions includes: Obtain the target action assigned to the condition-providing robot from the replanned collaborative task plan; The conditions obtained from the transitional cooperative constraints provide the robot with the action state to be maintained and the action to be restricted from switching. The target action is matched with the action of the restriction switching, and the action state is determined after the target action is executed based on the action effect of the target action in the action library. When the target action does not match the action of the restriction switching, and the action state is maintained after the target action is executed, the condition-providing robot is controlled to execute the target action; When the target action matches the action of the restriction switching, or when the action state cannot be maintained after the target action is executed, the current execution action of the robot provided by the condition is maintained, and the execution of the target action is delayed.

[0012] By extracting conditions from the replanning plan to provide the robot's target action, and combining the action maintenance state and restricted switching action in the transitional cooperative constraints, the impact of the target action on the state after execution is judged. This solves the problem of the robot prematurely executing new actions and breaking the exit conditions, and achieves coordinated control between the exit process and the timing of robot action switching.

[0013] In conjunction with the first aspect above, in one possible implementation, updating the transitional coordination constraint and the action dependency based on the exit state feedback includes: The exit action is divided into multiple exit action units arranged in execution order, and a completion status and duration condition are configured for each exit action unit. Update the completion status of each exit action unit according to the exit status feedback, and distinguish between exit action units that have been completed and those that have not been completed. Obtain the persistence condition of the incomplete exit action unit and retain the transitional coordination constraint used to maintain the persistence condition; Cancel the action switching restrictions in the transition coordination constraints that are associated with the completed exit action unit and are not continued to be used by the incomplete exit action unit, and update the action dependency relationship.

[0014] By dividing the exit action into multiple sequentially executed action units and continuously updating the completion status of each action unit using exit status feedback, the system retains only the continuing conditions required by the incomplete exit action units and gradually releases the failure constraints, thus solving the problem of excessively long constraint maintenance time caused by unified release after the overall exit is completed, and improving the timeliness of robot action recovery and collaborative flexibility.

[0015] In conjunction with the first aspect above, in one possible implementation, updating the transitional coordination constraint and the action dependency based on the exit state feedback includes: Collect exit status feedback from multiple condition-dependent robots that have established action dependency relationships with the robot that provides the same condition; Based on the feedback of each exit status, update the action dependency relationship between the condition-providing robot and each condition-dependent robot, and mark the updated action dependency relationship as a retained state or a released state. Filter action dependencies that are associated with the same persistence condition. If there is an action dependency in the retained state among the filtered action dependencies, retain the action switching restrictions for the persistence condition in the transitional coordination constraint. When all the selected action dependencies are in the released state, the action switching restriction for the continuous condition in the transitional coordination constraint is removed.

[0016] By updating the action dependencies between the robot providing the same condition and multiple robots that depend on the same condition, and determining the action switching restrictions based on whether the dependencies associated with the same persistent condition are still in a reserved state, the problem of prematurely releasing common conditions upon the completion of a single robot's exit is solved, thus preventing other robots from losing the required working conditions during their exit process.

[0017] In conjunction with the first aspect mentioned above, one possible implementation also includes: The execution conditions for the exit action are read from the action library. When the robot state represented by the exit state feedback does not meet the execution conditions, the exit action is regenerated based on the robot state and the action library. Based on the regenerated exit action, new collaborative requirements are formed. The transitional collaborative constraints are updated according to the action dependencies and the new collaborative requirements. The collaborative task plan is then replanned again using a large model based on the updated transitional collaborative constraints. The regenerated exit action is then incorporated into the replanned collaborative task plan.

[0018] In conjunction with the first aspect mentioned above, one possible implementation also includes: If a task change occurs again during the execution of the replanned collaborative task schedule, the new task instructions are received and the robot status is read. The transitional collaborative constraints, updated based on the exit status feedback, will continue to be used as replanning constraints. The collaborative task plan is replanned again using the large model based on the new task instructions, the robot state, and the replanning constraints; The replanned collaborative task plan is executed, and the conditions for robot action switching are continued to be controlled according to the replanning constraints.

[0019] In conjunction with the first aspect mentioned above, one possible implementation also includes: When the condition maintenance result indicates that the condition-providing robot cannot maintain the persistent condition, a candidate robot is selected from multiple robots; Based on the continuous conditions, the action execution requirements are obtained from the action library, and the robot state of each candidate robot is compared with the action execution requirements; From the candidate robots that meet the action execution requirements, determine the alternative condition-providing robot and establish the action dependency relationship between the alternative condition-providing robot and the condition-dependent robot; The transitional coordination constraints are updated based on the established action dependencies.

[0020] Secondly, a robot cooperative task planning system based on a large model includes: The task planning module is used to generate collaborative task plans for multiple robots based on task instructions, robot states, and motion library calls to a large model. The dependency management module is used to establish action dependencies based on the collaborative task plan and the continuous conditions of the actions. The action dependencies include the condition-providing robot, the condition-dependent robot, and the continuous conditions. The exit collaboration module is used to identify the condition-dependent robots affected by the task change when the task changes, generate an exit action based on the robot state of the condition-dependent robot and the action library, and form a collaboration requirement for the condition-providing robot based on the exit action. The replanning control module is used to generate transitional coordination constraints based on the action dependencies and coordination requirements, call the large model to replan the coordination task plan based on the transitional coordination constraints, so that the replanned coordination task plan includes the exit action, and control the conditions to provide robot action switching according to the transitional coordination constraints. The execution feedback module is used to issue the replanned collaborative task plan, collect the exit status feedback generated by the condition-dependent robot performing the exit action, and update the transitional collaborative constraints and the action dependencies based on the exit status feedback.

[0021] This application provides a robot cooperative task planning method and system based on a large model, which has the following beneficial effects: 1. By establishing action dependencies between condition-providing robots, condition-dependent robots, and continuous conditions, and combining exit actions with the formation of collaborative requirements and transitional collaborative constraints after task changes, the action dependencies formed by the original task can continue to the exit stage, and simultaneously affect the large model replanning and action switching of condition-providing robots. This avoids the synchronous cancellation of collaborative conditions after the task logic terminates, thereby reducing execution interruptions and action conflicts caused by the loss of gripping, support, or posture conditions during the exit process.

[0022] 2. By interactively confirming the continuous conditions and maintenance capabilities required for exiting among robot agents, and distinguishing between continued continuous conditions and newly added maintenance conditions based on the execution conditions of the exit action, action maintenance states and switching constraints are formed accordingly. This ensures that the transitional coordination constraints inherit the original cooperation relationship and reflect the new cooperation requirements added in the exit phase. This avoids coordination mismatch caused by only continuing the original plan or only adjusting actions according to the new plan, and improves the accuracy of multi-robot action connection during task change.

[0023] 3. By dynamically updating transitional collaborative constraints and action dependencies through exit status feedback, and combining the execution progress of the exit action unit and the dependency status of multiple condition-dependent robots on the same continuous condition, action switching restrictions are retained or removed. This ensures that conditions still required for the exit process are not released prematurely, while restrictions that have lost their effect can be canceled in a timely manner. This balances robot exit safety with the speed of subsequent task recovery, improving collaborative continuity and execution flexibility in continuous task change scenarios.

[0024] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0025] Figure 1 A flowchart illustrating a robot cooperative task planning method based on a large model, provided for an embodiment of this application; Figure 2 In a robot collaborative task planning method based on a large model provided in this application embodiment, a flowchart for forming collaborative requirements is shown. Figure 3 A flowchart illustrating the generation of transitional cooperative constraints and action switching in a robot cooperative task planning method based on a large model provided in this application embodiment; Figure 4 The flowchart of cooperative constraint update in a robot cooperative task planning method based on a large model provided in this application embodiment; Figure 5A structural block diagram of a robot cooperative task planning system based on a large model is provided for embodiments of this application; Detailed Implementation In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0026] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0027] Example 1: like Figure 1 As shown, this embodiment provides a robot collaborative task planning method based on a large model, applicable to scenarios where multiple robots cooperate to complete a task. The method includes the following steps: S101. Obtain task instructions, robot status and motion library, and use the large model to generate a collaborative task plan for multiple robots.

[0028] The task instructions describe the work objectives and content that multiple robots need to complete; the robot status reflects the current position, posture, action execution status, and interaction with the work object of each robot; the action library records the actions that the robot can perform and the conditions required to perform the actions, saves the control programs that the robot can execute according to the action identifier, and records the work object to which the action is applicable, the start conditions, the continuous conditions that need to be maintained during the execution, the action effect, and the completion requirements; the collaborative task plan describes the actions undertaken by different robots, the work objects to which the actions are performed, and the execution order and coordination relationship between the actions.

[0029] In a further implementation, task instructions are received, and the robot status is read from the control system of each robot; the actions that each robot can perform and the conditions for performing the actions are read from the action library, and the task instructions, robot status, and action information in the action library are organized into the planning input of the large model; the large model identifies the work content that needs to be completed according to the task instructions, and decomposes the work content into actions to be performed by different robots in combination with the robot status and executable actions, and arranges the execution order and coordination relationship of the actions; the planning results generated by the large model are organized into a collaborative task plan, and the actions and work objects assigned to each robot are recorded.

[0030] S102. Based on the collaborative task plan and the action library, establish action dependencies between different robots. The action dependencies include condition-providing robots, condition-dependent robots, and persistent conditions.

[0031] Among them, the condition-providing robot is the robot that undertakes the task of maintaining continuous conditions, and the condition-dependent robot is the robot that needs to use continuous conditions when performing actions. Continuous conditions refer to the working conditions that need to be continuously maintained by another robot during the execution of an action by a certain robot. Action dependency relationship is used to record the association between the condition-providing robot, the condition-dependent robot and the continuous conditions, so that the provider and user of the continuous conditions can be clearly identified.

[0032] In a further implementation, the actions assigned to each robot in the collaborative task plan are read, and the conditions that need to be maintained during the execution of the actions are read from the action library; combined with the work objects and robot division of labor in the collaborative task plan, the conditions that need to be maintained by another robot, and the robot undertaking the task of maintaining the conditions are identified; when a condition that needs to be maintained during the execution of a certain action cannot be maintained by the robot performing the action itself, and the action undertaken by another robot in the collaborative task plan can maintain the condition, the condition is identified as a continuous condition that needs to be maintained across robots.

[0033] Robots that undertake the task of maintaining conditions are recorded as condition-providing robots, robots that perform dependent actions are recorded as condition-dependent robots, and conditions that need to be maintained continuously are written into the action dependency relationship.

[0034] By establishing the cooperative relationship between robots as a continuous provision and use relationship, this step provides a data foundation for identifying the collaborative support that still needs to be maintained after task changes, reducing omissions caused by judging whether a robot can switch actions solely based on task allocation results.

[0035] S103. When a task change is detected, an exit action is generated based on the robot state of the condition-dependent robot and the action library, and a collaborative requirement for providing the robot to the condition is formed based on the exit action.

[0036] Task changes include task cancellation and adjustments to the original collaborative task plan; exit actions refer to the actions required for a condition-dependent robot to exit the operational state formed by the original task; collaborative requirements are used to express the support needed from the condition-providing robot during the execution of the exit action, including the operational conditions that need to be maintained and the robot providing support.

[0037] In a further implementation, upon receiving a task cancellation instruction or task adjustment information, the actions affected by the task change are identified in conjunction with the collaborative task plan, and the action dependencies are read to obtain the condition-dependent robots that execute the affected actions and the associated condition-providing robots; the current robot state of the condition-dependent robots is read to identify the actions that the condition-dependent robots are executing and their interactions with the work objects.

[0038] Based on the current robot state, the exit actions and execution conditions applicable to the current work state are read from the action library, and the selected actions are organized into exit actions required to exit the current work state; the conditions that need to be maintained by the condition-providing robot during the execution of the exit action are read, and the conditions that need to be maintained are associated with the condition-providing robot to form the collaborative requirements.

[0039] S104. Generate transitional coordination constraints based on the action dependencies and coordination requirements, replan the coordination task plan using a large model based on the transitional coordination constraints, incorporate the exit action into the replanned coordination task plan, and control the conditions to provide robot action switching according to the transitional coordination constraints.

[0040] Among them, the transitional coordination constraint is used to express the conditions that the robot needs to maintain and the restricted action switching during the transition from the original task execution state to the task-changed execution state.

[0041] In a further implementation, the condition-providing robot, condition-dependent robot, and persistent conditions recorded in the action dependency relationship are read, and the support content required for exiting the action in the collaboration requirement is read; based on the condition-providing robot and the work object pointed to by the collaboration requirement, the conditions that need to be maintained during the exit are identified, the condition maintenance requirements are transformed into the action states that the condition-providing robot needs to maintain, and the actions that will change the action states are identified in combination with the action library, forming a restriction on action switching; the action states that need to be maintained and the action switching restrictions are organized into the transitional collaboration constraints.

[0042] The transitional coordination constraints, the exit action, and the task requirements after the task change are submitted to the large model, and the coordination task plan is replanned in combination with the current robot state. During the replanning process, the exit action is arranged as the action that the condition-dependent robot needs to perform, and the condition-providing robot's cooperative action and subsequent actions are arranged according to the transitional coordination constraints.

[0043] When the robot issues an action switching instruction to the conditions, the transitional coordination constraint is read, and it is checked whether the action to be executed is restricted by the action switching. For action switching that is restricted, the relevant action switching is delayed in accordance with the requirement of maintaining the transitional coordination constraint retention condition. For actions that meet the transitional coordination constraint, they are organized and executed in accordance with the replanned coordination task plan.

[0044] It should be noted that the transitional coordination constraint is formed by simultaneously retaining the persistent conditions that still need to be used in the exit process of the original action dependency relationship, and combining them with the new coordination requirements generated by the exit action; therefore, the new task requirements after the task change cannot directly cover the cross-robot action dependencies that have not yet been resolved, and the timing of the robot's action switching is subject to the common constraint of the actual exit requirements.

[0045] S105. Execute the replanned collaborative task plan, collect the exit status feedback generated by the condition-dependent robot performing the exit action, and update the transitional collaborative constraints and the action dependency relationship based on the exit status feedback.

[0046] Among them, exit status feedback refers to the status information formed when the condition-dependent robot actually performs the exit action, which is used to reflect the execution process of the exit action and the interaction changes between the condition-dependent robot and the work object.

[0047] In a further implementation, the replanned collaborative task plan is distributed to each robot, the condition-dependent robot is organized to perform an exit action, and the action switching of the condition-providing robot is controlled according to the transitional collaborative constraints; during the execution, the execution status of the exit action and the robot status are read from the control system of the condition-dependent robot to form the exit status feedback.

[0048] The exit status feedback is compared with the execution requirements of the exit action to identify the actual progress of the exit action. The collaborative requirements are combined to determine whether the exit process still requires the condition-providing robot to maintain the continuous condition. When the exit process still requires the use of the continuous condition, the transitional collaborative constraints and action dependencies used to maintain the continuous condition are retained. When the exit status feedback indicates that the exit process no longer requires a certain continuous condition, the transitional collaborative constraints involving that continuous condition are updated, and the action dependencies involving that continuous condition between the condition-providing robot and the condition-dependent robot are released.

[0049] The updated transitional coordination constraints are used for subsequent action switching control, and the updated action dependencies are retained for subsequent execution feedback processing to read. Each update is based on the newly collected exit status feedback and the currently valid action dependencies, so that the retention and release of constraints are adjusted according to the actual exit process.

[0050] Example 2: Based on Example 1, this example further illustrates the formation process of collaborative requirements and the alternative handling when the condition-providing robot cannot maintain continuous conditions.

[0051] Multiple robots are each equipped with an intelligent agent. Through information exchange between the intelligent agents, the continuous conditions required for the exit action are confirmed, and it is determined whether the robot can continue to undertake the corresponding maintenance task under the current conditions.

[0052] It should be noted that the intelligent agent in this embodiment is a software execution unit corresponding to the robot, running in the robot control device or a computing device that is communicatively connected to the robot control system, and establishing an identification association with the corresponding robot; the intelligent agent can read the robot status and motion library information of the corresponding robot, and exchange information generated during task execution with the intelligent agents of other robots through a collaborative communication network.

[0053] like Figure 2 As shown, after the exit action is generated, the agent of the condition-dependent robot extracts the persistent conditions that need to be maintained during the exit and sends the condition maintenance request to the agent of the condition-providing robot; the agent of the condition-providing robot combines the action library and the current robot state to determine whether the persistent conditions can be maintained and returns the condition maintenance result.

[0054] For an exit action that has already been generated, the agent of the condition-dependent robot extracts the persistent conditions that need to be maintained by another robot during the exit from the action library, and identifies the robot that undertakes the task of maintaining the conditions by combining the established action dependencies; the exit action and persistent conditions are sent to the agent of the condition-providing robot, and the content of the message retains the work objects and state requirements involved in the persistent conditions; when there are multiple persistent conditions, each persistent condition is listed separately for the receiver to judge item by item.

[0055] During the sending and feedback process, the corresponding job objects and status requirements of the persistent conditions are retained to ensure that multiple persistent conditions and their maintenance results remain in a one-to-one correspondence.

[0056] After receiving a request, the robot's agent, based on the work object and state requirements involved in the continuous conditions, filters actions from the action library that act on the same work object and whose effects can maintain the state requirements. It reads the corresponding action execution requirements and checks whether the current robot state meets the action execution requirements. The scope of the check is determined by the state items involved in the action execution requirements, and the comparison results are recorded separately for each continuous condition. For actions containing multiple necessary state requirements, if all necessary state requirements are met, it is recorded as "can be maintained"; if there are any unmet necessary state requirements, it is recorded as "cannot be maintained," and the unmet requirements are retained.

[0057] In a further implementation, when there are multiple actions in the action library that can maintain the same continuous condition, the execution requirements of each action are checked separately; if there is an action that meets the execution requirements, the continuous condition is recorded as being able to be maintained; if there is no action that meets the execution requirements, it is recorded as being unable to be maintained; the current robot state is compared with the execution requirements item by item by checking the state category or allowed range recorded in the action library.

[0058] After completing the judgment, the agent of the condition-providing robot sends the condition maintenance results of each continuous condition to the agent of the condition-dependent robot. After receiving the feedback, the agent of the condition-dependent robot checks the continuous conditions that need to be maintained for the exit action against the feedback results one by one, according to the work object and state requirements retained in the feedback, and forms a collaborative requirement for the condition-providing robot. For continuous conditions that can be maintained, the requirement that the original condition-providing robot undertake the maintenance task is retained in the collaborative requirement. For continuous conditions that cannot be maintained, the condition maintenance requirements required for the exit action are still retained, and the process of selecting an alternative condition-providing robot is initiated accordingly.

[0059] It should be noted that the condition maintenance result indicates whether the condition-providing robot can undertake the condition maintenance task; when the condition-providing robot cannot maintain the continuous condition, the requirement for the continuous condition still exists for the exit action.

[0060] For persistent conditions in collaborative tasks that cannot be maintained by the original conditions, alternative robots are selected from multiple robots participating in the collaborative operation. During the screening process, it is first checked whether the action library of each alternative robot contains actions that can maintain the persistent conditions. Then, the current state of the alternative robot with the required actions is compared with the action execution requirements. Alternative robots that do not have the required actions or whose current robot state does not meet the action execution requirements are not included in the alternative selection range.

[0061] Furthermore, an alternative condition-providing robot is selected from the candidate robots that meet the action execution requirements, and an action dependency relationship is established between the alternative condition-providing robot and the condition-dependent robot. When establishing the relationship, the condition-dependent robot is still the robot that performed the original exit action, the continuing condition is still the condition that has not yet been met in the collaborative requirements, and the robot that undertakes the condition maintenance task is recorded as the alternative condition-providing robot.

[0062] The newly established action dependencies are then used to update the transitional coordination constraints, ensuring that the conditions in the constraints maintain consistency between the task and the replacement robot division of labor.

[0063] In a further implementation, the condition maintenance action of the alternative condition providing robot is incorporated into the replanned collaborative task plan; after the alternative condition providing robot performs the condition maintenance action, it determines whether the continuous condition has been established based on its robot state and the state of the work object; after the continuous condition meets the requirements, the exit action dependent on the continuous condition is continued to be executed; if the continuous condition has not been established, the exit action dependent on the continuous condition is suspended.

[0064] Example 3: Based on Example 1, this example further illustrates the generation of transitional cooperative constraints and the provision of conditions for robot motion switching control.

[0065] like Figure 3 As shown, firstly, the execution conditions of the exit action and the persistence conditions in the action dependency relationship are read. By matching the job object and state attributes, the continued persistence conditions and the newly added maintenance conditions are formed. Then, the conditions that need to be met simultaneously during the exit are collected to obtain the state maintenance range. Combined with the action effect, the action that restricts the switching action is identified to form the transitional coordination constraint. For the target action assigned to the condition-providing robot after replanning, it is determined whether execution is allowed according to the transitional coordination constraint and the action effect of the target action, thereby realizing the control of the action switching of the condition-providing robot.

[0066] For state requirements expressed using numerical ranges in the action library, the collaborative requirements are implemented as the basis for action switching by calculating the boundary tightening amount of the continuous condition, the state maintenance range during the exit period, and the allowable range before the execution of the target action.

[0067] The execution conditions for exit actions are read from the action library, and the conditions for maintaining continuous conditions are extracted from the action dependencies. Based on the work objects and state attributes involved in the conditions, conditions that act on the same work object and describe the same state attribute are grouped together, the unit of measurement for state values ​​is unified, and the upper and lower limits of the allowable range of each condition are retained.

[0068] Among them, the state attribute represents the measurable state quantity that the condition requires to be maintained. The numerical calculation in this embodiment is applicable to the state requirements expressed by continuous intervals. The endpoints of the intervals come from the execution requirements recorded in the action library and the continuous conditions in the action dependency relationship.

[0069] For an exit condition that allows for range comparison, the allowed range is denoted as: The range of conditions that are being maintained is denoted as When comparing the two sets of ranges, calculate the amount by which the original persistence condition exceeds the exit allowable range in both the lower and upper limits: , , in, j is the exit condition number, and j is the persistence condition number; , Exit execution requirements from the action library , Persistent conditions derived from action dependency records; and These represent the ranges within which the original continuation conditions still need to be tightened in the lower and upper bound directions, respectively.

[0070] Based on the comparison results, the two continuous conditions with zero tightening amount are marked as continued continuous conditions, and the exit execution conditions satisfied by the continued continuous conditions are recorded; for continuous conditions with positive tightening amount, the boundary and tightening amount to be tightened are recorded, and the corresponding exit execution conditions are reserved as execution conditions that have not yet been satisfied; from the execution conditions that have not yet been satisfied, conditions that need to be maintained by the condition providing robot are selected to form new maintenance conditions, and the new maintenance conditions adopt the allowable range of the corresponding exit execution conditions.

[0071] It should be noted that the fact that the current robot state meets the exit requirements does not mean that all other states within the original range of the continuous conditions meet the exit requirements; therefore, this embodiment compares the range of the continuous conditions with the range of the exit execution conditions.

[0072] Based on the aforementioned continuation conditions, new maintenance conditions, and collaborative requirements, the maintenance requirements that the robot needs to undertake for each condition are compiled. For conditions that need to be met simultaneously during the exit period and act on the same state attribute of the same work object, the original range of the continuation conditions is retained, and the exit state requirements in the new maintenance conditions are used to calculate the common allowable range.

[0073] The upper bound of the measurement error is obtained based on the sensor accuracy parameters of the corresponding state attributes or the measurement accuracy recorded in the robot control system. To allow the state measurement error to be included in the judgment, the upper bound of the measurement error of the r-th state attribute is denoted as... And calculate the range within which the measured values ​​need to be maintained: , , in, This represents a set of condition numbers that must be met simultaneously during the exit period and apply to the same state attribute of the same job object; This represents the allowable range of the h-th condition after aggregation; for continued conditions, the range of continued conditions that have been confirmed to meet the exit requirements is adopted, and for newly added maintenance conditions, the range of exit execution conditions is adopted.

[0074] After calculation, This is denoted as the range within which the measured state value needs to be maintained; when When, retain the calculation result; when At the same time, record the conditions and scope conflicts involved in the calculation, and provide the conflict results for processing in the replanning process.

[0075] The scope conflict records at least the conflicting job object, state attribute, conditions involved in the calculation, and corresponding allowable range, and inputs them as constraint conflict information into the replanning process; for conflict conditions that can be avoided by adjusting the execution order, the large model adjusts the execution order of the corresponding exit actions and condition maintenance actions, and recalculates the adjusted state maintenance range; if there is still no effective state maintenance range after adjustment, the corresponding exit action is retained as a temporarily unexecuted state.

[0076] It should be noted that the same actual state needs to satisfy multiple conditions simultaneously. Therefore, the largest lower limit and the smallest upper limit among the conditions are used to form a common range. By subtracting the measurement error into the middle of the common range, the actual state within the error boundary still meets the exit requirements when the measured value is within the maintenance range.

[0077] Based on the state maintenance range and the action effects recorded in the action library, first identify actions that can be directly determined to break the state maintenance requirements after execution, and list the identified actions as actions with restriction switching; record the actions that provide conditions to the robot, the work object, the state maintenance range, and the actions with restriction switching together as transitional cooperative constraints.

[0078] For the newly added maintenance condition, the condition-providing robot is scheduled to perform the condition maintenance action in the collaborative task plan, and the achievement of the new maintenance condition is used as the execution condition for the exit action that depends on the condition. After the condition-providing robot performs the maintenance action, it checks whether the work object and the status requirements are met according to the robot status feedback. If they are met, the exit action that depends on the condition is executed.

[0079] In one optional implementation, the large model employs an autoregressive language model fine-tuned by instructions. The model includes an input representation part, a feature processing part, and an output part, used to generate a collaborative task plan based on task instructions, robot state, motion library information, and transitional collaborative constraints.

[0080] Load the model weights, configuration files, and word segmenter corresponding to the large model, organize the task planning requirements into system messages, and organize the robot status, action library information, exit actions, and transitional collaborative constraints into task input messages; use the dialogue templates that come with the model to convert the messages into input sequences, encode them through the word segmenter, and then input them into the model; the model generates an output collaborative task plan through autoregression, and then decodes the output sequence into planning text containing robot identifiers, action identifiers, work objects, and execution order.

[0081] This implementation uses pre-trained model weights to perform inference. When the task changes, the robot state, exit action, and transitional coordination constraints in the input message are updated. The model parameters remain unchanged during the planning process, and each replanning reflects the actual task changes by updating the input content.

[0082] It should be noted that the large model is responsible for understanding the task content and arranging actions. The state maintenance range and action switching restrictions are formed by the aforementioned calculation process and are input into the model as planning conditions. The target actions generated by the model continue to be subject to the execution judgment of transitional cooperative constraints, so that the model output is connected with the robot's motion control.

[0083] The target action assigned to the condition-providing robot is obtained from the replanned collaborative task plan. According to the action identifier and target object in the action library, it is matched with the action of the restriction switching. For the target action that does not match the action of the restriction switching, the robot's current state and action effect are further combined to determine whether the required state can be maintained after the target action is executed.

[0084] For a target action whose effect can be expressed as a state increment, the range of change of the r-th state is denoted as . When the action effect is a fixed increment, the upper and lower limits are the same value; when the action effect includes a range of allowed changes, the upper and lower limits of the increment recorded in the action library are used.

[0085] It should be noted that, for the state requirement that needs to be continuously maintained during the execution of the exit action, the action library also records the range of state changes during the execution of the target action; before the allowable range is judged before the target action enters the execution, the range of state changes is compared with the state maintenance range; only when the corresponding state is kept within the state maintenance range during the execution of the target action is it judged whether the state after the target action is completed meets the state maintenance requirement.

[0086] In a further implementation, the state increment of the target action is back-calculated to the allowable state range before execution: , , in, This represents the range that the r-th state measurement value is allowed to be before executing the target action a; the current state measurement value is denoted as... Only when satisfied Only then can it be determined that the target action, after being executed within the known increment range, still maintains the required state.

[0087] Based on the pre-execution allowable range, the current state measurement value of the robot is checked item by item. When the target action does not match the action of the restriction switch and all state measurement values ​​are within the allowable range, the robot is controlled to execute the target action. When the target action matches the action of the restriction switch, or when any state measurement value exceeds the allowable range, the current execution action of the robot is maintained and the execution of the target action is delayed. When the pre-execution allowable range is empty, the target action is included in the delayed execution branch.

[0088] For target actions that use the absolute state after execution to express the effect of the action, the range of the state after execution is directly compared with the range of the state that is maintained, and the same execution or delay processing is used to retain the delayed target action in the collaborative task plan. Combined with the currently effective transitional collaborative constraints, the action switching judgment continues.

[0089] Example 4: This embodiment further illustrates the transitional collaborative constraint update process driven by exit state feedback, as well as the constraint retention and release methods when multiple condition-dependent robots jointly use continuous conditions.

[0090] like Figure 4As shown, the exit action is divided into multiple exit action units, and each unit is associated with completion requirements and the persistent conditions that need to be maintained during execution. The completion status of each exit action unit is updated based on the exit status feedback, and it is checked whether the exit action units that have not yet been completed still use the corresponding persistent conditions. The retention or release status of the action dependency relationship is updated based on the actual usage of the persistent conditions. When the same persistent condition is used by multiple condition-dependent robots at the same time, the status of each action dependency relationship is further integrated to determine whether to continue to retain the action switching restriction for that persistent condition.

[0091] The exit action is divided into multiple exit action units arranged in execution order. Based on the execution requirements in the action library, completion requirements and initial completion status are set for each exit action unit, and the continuous conditions that need to be maintained during execution are associated. During the division, the action part that can independently identify the execution result is taken as the exit action unit, and the completion requirements of the action unit are associated with the actual execution feedback. During initialization, the action units that have not yet been completed are recorded as incomplete, and the association between the action unit and the condition-providing robot and the continuous conditions is preserved.

[0092] The completion status indicates whether the exit action unit has met the completion requirements; the incomplete exit action units include exit action units that have not yet started, and exit action units that have started but have not yet finished and still need to be executed.

[0093] Based on the collected exit status feedback, identify the exit action unit that is being executed, and compare the action execution status and robot status in the feedback with the completion requirements; when the completion requirements are met, update the exit action unit to the completed status; if execution still needs to continue, retain the incomplete status.

[0094] It should be noted that the issuance of the action command and the actual completion of the action unit are different states. The release of the continuous condition depends on whether the exit process still uses the continuous condition. Therefore, the completion state is updated based on the actual exit feedback to keep the constraint changes consistent with the robot's actual exit process.

[0095] Based on the completion status of each exiting action unit, check whether each persistent condition is still being used by incomplete action units; denot the condition-dependent robot as d, the condition-providing robot as p, and the persistent condition as s, and calculate the number of conditions used by the same condition-dependent robot that have not yet been completed: , in, This indicates the number of exit action units for the conditionally dependent robot d; This indicates the completion status of the k-th exit action unit; 1 indicates completion, and 0 indicates incompleteness. This indicates whether the k-th exit action unit needs robot p to maintain the continuous condition s; it is set to 1 if needed and 0 otherwise.

[0096] When the number of conditions used is greater than zero, the action switching restrictions for the continuous condition in the transitional coordination constraint are retained, and the action dependency relationship between the condition-providing robot and the condition-dependent robot involving the continuous condition is marked as retained; when the number of conditions used becomes zero, the action dependency relationship is marked as released, and it is further checked whether there are other condition-dependent robots using the continuous condition.

[0097] It should be noted that the completion of an exit action unit only indicates that the action unit has ended its use of the continuous condition; when subsequent action units still require the continuous condition, the condition-providing robot still needs to continue to undertake the maintenance task.

[0098] When a robot providing the same condition supports multiple robots that depend on the same condition, the completion status is updated based on the exit status of each robot that depends on the condition, and the number of conditions used is calculated. The action dependencies are grouped according to the status requirements of the robot providing the condition, the work object, and the continuous condition. Action dependencies that are maintained by the same robot providing the condition, act on the same work object, and have the same status requirements are considered to be using the same continuous condition. The grouped action dependencies are then checked to see if there are still any reserved states.

[0099] If there are action dependencies in the collection results that are in a retained state, the action switching restrictions for the persistent condition will continue to be retained; if all action dependencies are in a released state, the action switching restrictions for the persistent condition will be cancelled; the feedback from each condition-dependent robot will first be used to update the action dependencies involved in its own exit process, and then the restriction on the condition-providing robot will be decided based on the common usage situation.

[0100] In a further implementation, the same state attribute has different allowable ranges in different exit action units. Based on the updated completion state, incomplete exit action units that are maintained by the robot under the same conditions, act on the same work object, and still need to use the state attribute are collected. The state ranges required by these action units are read, and a common allowable range is formed with the maximum lower limit and the minimum upper limit. The boundary is adjusted according to the original measurement error upper limit to obtain the updated state maintenance range.

[0101] The action of restricting switching is re-identified based on the updated state retention range; for target actions not listed as restricted switching actions, action switching judgment is continued based on the current robot state and action effect; when the state attribute no longer exists as an exit action unit that has not yet been completed and needs to use the state attribute, the action switching restriction for that state attribute is cancelled.

[0102] The updated transitional coordination constraints and action dependencies are saved as the current valid results for subsequent exit feedback processing and condition-providing robot action switching judgments; the new feedback continues to act on exit action units that have not yet been completed, so that the completion status, condition usage and action switching constraints are continuously updated.

[0103] Through the above processing, the release of the persistent condition is first controlled by the actual completion progress of the exit action unit of a single condition-dependent robot, and then by the common usage state of multiple condition-dependent robots for the same persistent condition. This avoids both the excessively long condition maintenance time caused by releasing the condition uniformly at the completion time of the entire exit action, and the immediate release of the common persistent condition after a certain condition-dependent robot completes its work ahead of schedule, which would affect the exit process of other robots.

[0104] Example 5: This embodiment further illustrates the regeneration of the exit action, as well as the constraint inheritance and replanning processing when a task change occurs again during the exit execution period.

[0105] During the exit action of the condition-dependent robot, the execution conditions of the exit action are read from the action library, and the robot state in the exit status feedback is compared with the execution conditions. During the comparison, feedback information is extracted according to the state items involved in the execution conditions, and it is checked whether the action execution state, robot state and work object interaction state meet the execution requirements. For the conditions recorded according to the execution stage, the requirements that need to be met in the current execution stage are used for judgment.

[0106] The exit action is formed by combining existing exit action units in the action library. Specifically, based on the original executed action and the work object, the completion state that needs to be reached for exit is read, and action units that meet the start conditions are selected based on the current robot state. The state after the action unit is executed is calculated based on the action effect, and subsequent action units are matched with this state until the exit completion state is reached. The action sequence is searched by expanding layer by layer, and the existing path is retained for the same action state that is repeatedly reached. The exit action is selected from the sequence that can reach the exit completion state.

[0107] Among them, the execution conditions for the exit action limit the state basis that the action needs to have when it is actually executed; the condition check according to the execution stage is to link the state changes that have occurred with the current action requirements, so that the judgment basis changes with the exit process.

[0108] When the exit status feedback indicates that the robot's state does not meet the execution conditions, the unmet conditions are recorded, and the current robot state in the feedback is used as the starting point to select a new action from the action library that can continue to complete the exit. The current state is compared with the execution requirements of the candidate actions, and the action sequence is organized in combination with the exit content that has not yet been completed to form a regenerated exit action. The exit results that have been completed are retained as part of the current state, and the new action arrangement continues to unfold from the actual state of the robot.

[0109] Based on the regenerated exit action, extract the continuous conditions that the robot needs to maintain during execution from the action library to form new collaborative requirements; check the new collaborative requirements with the current action dependencies, identify the condition provisioning relationships that the new exit action still needs to use, and update the transitional collaborative constraints accordingly.

[0110] During the update process, the persistent conditions still used by the new exit action continue to maintain the entry conditions; the cooperative requirements changed by the new exit action are used to adjust the relevant constraints; based on the updated transitional cooperative constraints, the regenerated exit action and the current robot state are submitted to the large model, and the cooperative task plan is re-planned; the new exit action is arranged in the execution plan of the condition-dependent robot, and the cooperative actions and subsequent tasks of the condition-providing robot are adjusted; the execution is organized according to the re-planned cooperative task plan, and exit status feedback continues to be collected for subsequent constraint and action dependency updates.

[0111] When a task change occurs again during the execution of the replanning collaborative task plan, the new task instructions are received and the current status of each robot is read. At the same time, the transitional collaborative constraints that have been updated based on the exit status feedback are read, and the still valid condition maintenance requirements and action switching restrictions are used as the constraints for this replanning.

[0112] Based on the new task instructions, the current robot state, and the replanning constraints, the subsequent actions of each robot are rearranged using the large model; for robots that are still subject to action switching restrictions, the condition maintenance requirements are incorporated into the action arrangement; for actions that have been unrestricted, they are planned according to the new task requirements; the results of completed actions are retained in the current robot state, and actions that have not yet been executed are adjusted according to the results of the replanning.

[0113] When executing the collaborative task plan after replanning, the robot's action switching is provided according to the replanning constraints, and the exit status feedback of the condition-dependent robot is continuously collected; the constraint update results formed by the feedback are used for subsequent action judgment, so that the execution of the plan after the replanning of the task is connected with the actual exit process.

[0114] Example 6: like Figure 5As shown, based on Embodiments 1 to 5, this embodiment provides a robot collaborative task planning system based on a large model, including a task planning module, a dependency management module, an exit collaboration module, a replanning control module, and an execution feedback module. The task planning module generates collaborative task plans for multiple robots; the dependency management module establishes and updates action dependencies between different robots; the exit collaboration module generates exit actions and forms collaboration requirements when tasks change; the replanning control module generates transitional collaboration constraints, calls the large model to replan collaborative tasks, and provides control conditions for robot action switching; the execution feedback module collects exit status feedback and updates transitional collaboration constraints and action dependencies based on the exit process.

[0115] The system runs on a robot collaborative control device or a computing device that communicates with multiple robot control systems. It can read the robot status and motion library information of each robot and call a large model to complete collaborative task planning.

[0116] The task planning module receives task instructions, robot status, and motion library information, and calls the large model to generate a collaborative task plan for multiple robots. The dependency management module establishes motion dependencies between the condition-providing robot, the condition-dependent robot, and the continuous conditions based on the robot division of labor in the collaborative task plan and the continuous conditions recorded in the motion library.

[0117] When a task change is detected, the exit collaboration module generates an exit action based on the current state of the condition-dependent robot and the action library, and extracts the continuous conditions that the robot needs to maintain during the exit period, thus forming a collaboration requirement.

[0118] The replanning control module generates transitional coordination constraints based on the action dependencies and coordination requirements, and inputs the exit action, the current robot state, and the transitional coordination constraints into the large model to replan the collaborative task plan. Before the replanned target action is issued, the action switching judgment is still based on the transitional coordination constraints to avoid the robot changing the operation state that still needs to be maintained during the exit process in advance.

[0119] The execution feedback module collects exit status feedback generated by the condition-dependent robot's exit action and updates the action dependency relationship and transition coordination constraint accordingly. When the exit process still needs to use the continuous condition, the corresponding action switching restriction is retained. When the exit process no longer needs the continuous condition, the corresponding action dependency relationship and action switching restriction are released. For the case where multiple condition-dependent robots use the same continuous condition, the action switching restriction for the continuous condition is canceled after all the relevant action dependencies are released.

[0120] In the above embodiments, the robot collaborative task planning method and system based on a large model described in this application can be implemented through software, hardware, firmware, or any combination thereof. Functions such as task planning, action dependency management, collaborative requirement formation, transitional collaborative constraint generation, collaborative task plan replanning, and exit status feedback processing can be implemented by one or more processors executing corresponding computer programs. The robot agent can be implemented as a software execution unit running in a robot control device, edge computing device, or other computing device communicatively connected to the robot control system. The large model can be deployed on a local computing device, edge server, or remote server, and completes the generation and replanning of collaborative task plans through corresponding computing resources.

[0121] When implemented in software, some or all of the processing procedures of this application can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed by a processor, the processor performs one or more steps described in the embodiments of this application: robot state reading, action library invocation, action dependency establishment, exit action generation, collaborative requirement formation, transitional collaborative constraint generation, collaborative task plan replanning, robot action switching control, and exit state feedback update. The computer instructions can be stored in a computer-readable storage medium, including magnetic storage media, optical storage media, semiconductor storage media, or other data storage media capable of storing program instructions.

[0122] It should be noted that the task planning module, dependency management module, exit collaboration module, replanning control module, and execution feedback module, which are functionally divided in this application embodiment, are only used to illustrate the logical relationship between different processing functions and do not limit each module to be implemented by independent hardware. The above-mentioned multiple functions can be integrated into the same processor, control device, or computing device, or they can be implemented separately by multiple communicating processing devices; the same function can also be implemented by both software programs and hardware circuits. The deployment location, operating equipment, and communication method of the robot agent and the large model can be adjusted according to the actual architecture of the robot collaborative control system, as long as the information interaction and processing described in this application can be achieved.

[0123] The embodiments of this application are primarily used to illustrate the technical solutions of this application, and not to limit the scope of protection. The descriptions of the number of robots, the deployment locations of intelligent agents, the organization of the action library, the deployment of large models, the expression of state attributes, the communication connection methods, and the specific calculation processes in the embodiments are all used to explain the implementation methods of the technical solutions of this application. Without departing from the technical concept of this application, those skilled in the art can make corresponding adjustments or combinations of the above implementation methods according to the actual robot type, the work object, and the collaborative task.

[0124] In understanding and implementing this application, the terms "comprising" and "including" used in the specification do not exclude the existence of other components, processing steps, or functional units not explicitly listed; unless the context explicitly limits it to the singular, a technical object may also include one or more technical objects of the same type. Technical features described in different embodiments and dependent claims can be combined to form other implementations where there is no technical conflict.

[0125] The scope of protection of this application is determined by the contents of the claims. Any equivalent substitutions, formal adjustments, merging or splitting of functional modules, or other changes that do not alter the technical essence of this application made by a person skilled in the art based on the disclosure of this application after reading this specification and the drawings, should be determined in accordance with applicable law and in conjunction with the claims and their equivalents to determine whether they fall within the scope of protection of this application.

Claims

1. A large model-based robot collaborative task planning method, characterized in that, include: Acquire task instructions, robot status, and motion library, and use the large model to generate collaborative task plans for multiple robots; Based on the collaborative task plan and the action library, action dependencies between different robots are established, including condition-providing robots, condition-dependent robots, and persistent conditions. When a task change is detected, an exit action is generated based on the robot state of the condition-dependent robot and the action library, and a collaborative requirement for the robot to provide the robot for the condition is formed based on the exit action. Transitional coordination constraints are generated based on the action dependencies and coordination requirements. The coordination task plan is replanned using a large model based on the transitional coordination constraints. The exit action is incorporated into the replanned coordination task plan. The robot's action switching is controlled according to the transitional coordination constraints. The replanned collaborative task plan is executed, the exit status feedback generated by the condition-dependent robot performing the exit action is collected, and the transitional collaborative constraints and the action dependencies are updated based on the exit status feedback.

2. The method of claim 1, wherein, The multiple robots are each configured with an intelligent agent; the process of forming the collaborative requirement includes: The agent of the condition-dependent robot extracts the persistent conditions that the condition-providing robot needs to maintain during the execution of the exit action, and sends the exit action and the persistent conditions to the agent of the condition-providing robot. The agent of the robot provided by the conditions reads the action execution requirements for maintaining the continuous conditions from the action library according to the continuous conditions, compares the robot state of the robot provided by the conditions with the action execution requirements, and obtains the condition maintenance result; The condition maintenance result is sent to the agent of the condition-dependent robot, which then forms the collaborative requirement based on the persistent condition and the condition maintenance result.

3. The method of claim 1, wherein, The step of generating transitional collaboration constraints based on the action dependencies and collaboration requirements includes: The execution conditions for the exit action are read from the action library, and the persistent conditions maintained by the robot by the conditions are extracted from the action dependencies. Match the execution conditions and the persistence conditions according to the job object, compare the state requirements of the execution conditions and the persistence conditions acting on the same job object, and mark the persistence condition as a continuing persistence condition when the state represented by the persistence condition meets the state requirements of the execution condition. From the execution conditions that have never been satisfied by the continued conditions, select the execution conditions that require the conditions to provide robot maintenance, and form new maintenance conditions; Based on the continued conditions, the new maintenance conditions, and the collaborative requirements, the conditions are formulated to provide the robot with the action states that need to be maintained and the actions that restrict switching, and the transitional collaborative constraints are generated.

4. The method of claim 3, wherein, The step of providing robot action switching according to the transitional cooperative constraint control conditions includes: Obtain the target action assigned to the condition-providing robot from the replanned collaborative task plan; The conditions obtained from the transitional cooperative constraints provide the robot with the action state to be maintained and the action to be restricted from switching. The target action is matched with the action of the restriction switching, and the action state is determined after the target action is executed based on the action effect of the target action in the action library. When the target action does not match the action of the restriction switching, and the action state is maintained after the target action is executed, the condition-providing robot is controlled to execute the target action; When the target action matches the action of the restriction switching, or when the action state cannot be maintained after the target action is executed, the current execution action of the robot provided by the condition is maintained, and the execution of the target action is delayed.

5. The method of claim 4, wherein, The step of updating the transitional coordination constraints and the action dependencies based on the exit status feedback includes: The exit action is divided into multiple exit action units arranged in execution order, and a completion status and duration condition are configured for each exit action unit. Update the completion status of each exit action unit according to the exit status feedback, and distinguish between exit action units that have been completed and those that have not been completed. Obtain the persistence condition of the incomplete exit action unit and retain the transitional coordination constraint used to maintain the persistence condition; Cancel the action switching restrictions in the transition coordination constraints that are associated with the completed exit action unit and are not continued to be used by the incomplete exit action unit, and update the action dependency relationship.

6. The method of claim 1, wherein, The step of updating the transitional coordination constraints and the action dependencies based on the exit status feedback includes: Collect exit status feedback from multiple condition-dependent robots that have established action dependency relationships with the robot that provides the same condition; Based on the feedback of each exit status, update the action dependency relationship between the condition-providing robot and each condition-dependent robot, and mark the updated action dependency relationship as a retained state or a released state. Filter action dependencies that are associated with the same persistence condition. If there is an action dependency in the retained state among the filtered action dependencies, retain the action switching restrictions for the persistence condition in the transitional coordination constraint. When all the selected action dependencies are in the released state, the action switching restriction for the continuous condition in the transitional coordination constraint is removed.

7. The method of claim 1, wherein, Also includes: The execution conditions for the exit action are read from the action library. When the robot state represented by the exit state feedback does not meet the execution conditions, the exit action is regenerated based on the robot state and the action library. Based on the regenerated exit action, new collaborative requirements are formed. The transitional collaborative constraints are updated according to the action dependencies and the new collaborative requirements. The collaborative task plan is then replanned again using a large model based on the updated transitional collaborative constraints. The regenerated exit action is then incorporated into the replanned collaborative task plan.

8. The method of claim 1, wherein, Also includes: If a task change occurs again during the execution of the replanned collaborative task schedule, the new task instructions are received and the robot status is read. The transitional collaborative constraints, updated based on the exit status feedback, will continue to be used as replanning constraints. The collaborative task plan is replanned again using the large model based on the new task instructions, the robot state, and the replanning constraints; The replanned collaborative task plan is executed, and the conditions for robot action switching are continued to be controlled according to the replanning constraints.

9. The method of claim 2, wherein, Also includes: When the condition maintenance result indicates that the condition-providing robot cannot maintain the persistent condition, a candidate robot is selected from multiple robots; Based on the continuous conditions, the action execution requirements are obtained from the action library, and the robot state of each candidate robot is compared with the action execution requirements; From the candidate robots that meet the action execution requirements, determine the alternative condition-providing robot and establish the action dependency relationship between the alternative condition-providing robot and the condition-dependent robot; The transitional coordination constraints are updated based on the established action dependencies.

10. A large model based robot collaborative task planning system, applied to the method of any one of claims 1-9, characterized in that, include: The task planning module is used to generate collaborative task plans for multiple robots based on task instructions, robot states, and motion library calls to a large model. The dependency management module is used to establish action dependencies based on the collaborative task plan and the continuous conditions of the actions. The action dependencies include the condition-providing robot, the condition-dependent robot, and the continuous conditions. The exit collaboration module is used to identify the condition-dependent robots affected by the task change when the task changes, generate an exit action based on the robot state of the condition-dependent robot and the action library, and form a collaboration requirement for the condition-providing robot based on the exit action. The replanning control module is used to generate transitional coordination constraints based on the action dependencies and coordination requirements, call the large model to replan the coordination task plan based on the transitional coordination constraints, so that the replanned coordination task plan includes the exit action, and control the conditions to provide robot action switching according to the transitional coordination constraints. The execution feedback module is used to issue the replanned collaborative task plan, collect the exit status feedback generated by the condition-dependent robot performing the exit action, and update the transitional collaborative constraints and the action dependencies based on the exit status feedback.