Unmanned cluster command and control method, electronic equipment and storage medium

Abstract

The invention provides an unmanned cluster command and control method, electronic equipment and a computer readable storage medium. The method comprises the steps of performing decision planning on a to-be-commanded cooperative task of an unmanned cluster to obtain a to-be-scheduled target unmanned unit of the to-be-commanded cooperative task and an execution decision scheme of a target execution task of the target unmanned unit; determining a candidate tool set of the target execution task based on the current task state of the target execution task and the execution decision scheme; arranging the candidate tool set by taking maximization of the calling revenue value of the candidate tool set as a target to obtain an execution tool sequence of the target execution task; performing instruction generation processing based on the execution tool sequence to obtain a control instruction for executing the decision scheme; and sending a control instruction to the target unmanned unit to control the target unmanned unit to execute the target execution task. According to the method, the tool calling decision can be automatically completed, the reliability of the tool calling decision is improved, and the automation degree and efficiency of task execution are improved.

Description

Unmanned swarm command and control methods, electronic equipment and storage media Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to an unmanned swarm command and control method, electronic equipment, and computer-readable storage medium. Background Technology

[0002] With the continuous improvement of intelligence and digitalization, unmanned swarms, as a complex system composed of a large number of unmanned units (such as drones, unmanned vehicles, and unmanned boats) with certain perception, computing and execution capabilities, have shown great application potential in logistics and transportation, environmental monitoring, emergency rescue and precision agriculture in recent years.

[0003] As the level of intelligence of unmanned swarms continues to improve, the complexity of their tasks, the dynamism of the environment, and the need for multi-target collaboration are increasing, which places higher demands on the motion control precision and tool call adaptability of unmanned units.

[0004] However, the execution control of traditional unmanned clusters often relies on preset tool calling rules or manually specified interface adaptation schemes. This leads to the following problems in real-world scenarios: on the one hand, preset tool calling rules are difficult to cover all scenario combinations, resulting in low reliability of tool calling decisions; on the other hand, traditional solutions require manual planning of tool calling processes and configuration of interface parameters in advance. When the task objective is adjusted or a new tool type is added, the adaptation rules need to be rewritten, resulting in low automation of task execution. Summary of the Invention

[0005] This application provides an unmanned swarm command and control method, electronic device, and computer-readable storage medium, which can automatically complete tool invocation decisions, improve the reliability of tool invocation decisions, and enhance the automation and efficiency of task execution.

[0006] In a first aspect, this application provides a command and control method for an unmanned swarm, the method comprising: performing decision planning on a task to be commanded and coordinated by an unmanned swarm, obtaining a target unmanned unit to be scheduled for the task to be commanded and coordinated, and an execution decision scheme for the target unmanned unit's target execution task, wherein the unmanned swarm includes multiple unmanned units; determining a set of candidate tools for the target execution task based on the current task status of the target execution task and the execution decision scheme; arranging the set of candidate tools with the objective of maximizing the invocation benefit value of the set of candidate tools, obtaining a sequence of execution tools for the target execution task; performing instruction generation processing based on the sequence of execution tools to obtain control instructions for the execution decision scheme; and sending the control instructions to the target unmanned unit to control the target unmanned unit to execute the target execution task.

[0007] Secondly, this application also provides an electronic device, which includes a processor and a memory, wherein the memory stores a computer program, and the processor executes any of the unmanned swarm command and control methods provided in this application when it calls the computer program in the memory.

[0008] Thirdly, this application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to execute the unmanned swarm command and control method described above.

[0009] In this application, firstly, by determining a set of candidate tools for the target task based on its current task status and execution decision scheme, tools matching the target task can be automatically selected. With the goal of maximizing the invocation benefit of the candidate tool set, the set is orchestrated to obtain a sequence of execution tools for the target task. This allows for automatic planning and invocation of tools to generate control commands. Even when facing multiple target tasks simultaneously, the tool generation control commands can still be automatically adapted. This enables unmanned clusters to automatically complete tool invocation decisions when facing complex tasks (such as continuous execution of multiple actions or collaborative operation of multiple devices) without rewriting adaptation rules, thus improving the automation level of task execution to a certain extent. Secondly, by orchestrating the candidate tool set to maximize the invocation benefit of the candidate tool set to obtain a sequence of execution tools, the sequence can be automatically adapted to the scenario requirements to generate execution tool sequences with higher invocation benefits, thereby improving the reliability of tool invocation decisions. Thirdly, by orchestrating the candidate tool set with the goal of maximizing the calling benefit value of the candidate tool set to obtain the execution tool sequence, a better tool execution sequence can be automatically orchestrated to achieve the connection from "decision" to "execution", thereby improving the automation and efficiency of task execution. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 is a structural block diagram of an electronic device provided in an embodiment of this application; Figure 2 is a framework diagram of a cluster command and control device provided in an embodiment of this application; Figure 3 is a flowchart of an unmanned cluster command and control method provided in an embodiment of this application; Figure 4 is a framework illustration of the tool arrangement and usage process provided in this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0014] In the description of the embodiments of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0015] To enable any person skilled in the art to implement and use this application, the following description is provided. In this description, details are set forth for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be implemented without using these specific details. In other instances, well-known processes will not be described in detail to avoid obscuring the description of the embodiments of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in the embodiments of this application.

[0016] Traditional unmanned swarm execution control relies heavily on preset tool invocation rules or manually specified interface adaptation schemes, lacking dynamic response capabilities to changes in task scenarios, device status, and environment. In practical applications, when unmanned swarms face complex tasks (such as the sequential execution of multiple actions or collaborative operation of multiple devices) or dynamic environments (such as fluctuations in device status or sudden interference), they encounter the following problems: Firstly, preset tool invocation rules cannot cover all scenario combinations, resulting in low reliability of tool invocation decisions; secondly, traditional solutions require manual planning of tool invocation processes and configuration of interface parameters in advance. When task objectives are adjusted or new tool types are added, adaptation rules need to be rewritten, leading to low automation of task execution.

[0017] This application provides an unmanned swarm command and control method, an electronic device, and a computer-readable storage medium. The electronic device can be a mobile phone, computer, etc. The execution entity of the unmanned swarm command and control method can be a swarm command and control device or an electronic device. In some embodiments, the swarm control device can be implemented in software and loaded into the electronic device. For example, the swarm control device can be loaded into the memory of the electronic device as a computer program, and the processor of the electronic device can call the computer program in the memory of the electronic device to realize the specific functions of the swarm control device. In some embodiments, the swarm control device can also be implemented in hardware. For example, independent hardware implementation modules can be set for the preset planning system, preset coordination system, and preset execution system of the swarm control device to realize the specific functions of the swarm control device.

[0018] Firstly, based on the current task status of the target execution task and the execution decision scheme, a set of candidate tools for the target execution task is determined. With the goal of maximizing the invocation benefit value of the candidate tool set, the candidate tool set is arranged to obtain an execution tool sequence for the target execution task. Instruction generation processing is performed based on the execution tool sequence to obtain control instructions for the execution decision scheme. The control instructions are sent to the target unmanned unit to control the target unmanned unit to execute the target execution task. This allows for automatic planning and invocation of tool generation control instructions. Even when facing multiple target execution tasks simultaneously, the tool generation control instructions can still be automatically adapted. This enables the unmanned cluster to automatically complete tool invocation decisions when facing complex tasks (such as continuous execution of multiple actions or collaborative operation of multiple devices) without rewriting adaptation rules, thus improving the automation level of task execution to a certain extent. It can automatically adapt to scenario requirements to generate execution tool sequences with higher invocation benefit values, thereby improving the reliability of tool invocation decisions.

[0019] Secondly, based on the current task status of the target local task and the overall decision-making scheme, a first set of candidate delivery tools for the target local task is determined; with the goal of maximizing the call benefit value of the first set of delivery tools, the first set of delivery tools is arranged to obtain a planning tool sequence for the target local task; based on the planning tool sequence, instruction generation processing is performed to obtain the delivery instruction of the overall decision-making scheme; the delivery instruction of the overall decision-making scheme is sent to the preset coordination system to control the preset coordination system to start the decision planning of the target local task; the tool can be automatically planned and called to generate delivery instructions, and even when facing multiple target local tasks that need to be delivered at the same time, the tool can still be automatically adapted to generate delivery instructions, so that when the unmanned cluster faces complex tasks (such as the continuous execution of multiple actions, the collaborative operation of multiple devices) or dynamic environments (such as device status fluctuations, sudden interference), it can automatically complete the tool call decision without rewriting the adaptation rules, which improves the automation level of the overall decision-making scheme delivery to a certain extent, and can automatically adapt to the scenario requirements to generate a planning tool sequence with higher call benefit value, thereby improving the reliability of tool call decision.

[0020] Thirdly, based on the current task status of the target execution task and the coordination decision scheme, a second set of candidate delivery tools for the target execution task is determined; with the goal of maximizing the call benefit value of the second set of delivery tools, the second set of delivery tools is arranged to obtain a coordination tool sequence for the target execution task; based on the coordination tool sequence, instruction generation processing is performed to obtain the delivery instruction of the coordination decision scheme; the delivery instruction of the coordination decision scheme is sent to the preset execution system to control the preset execution system to start the decision planning of the target execution task; the tool can be automatically planned and called to generate delivery instructions, and even when facing multiple target execution tasks that need to be delivered at the same time, the tool can still automatically adapt to generate delivery instructions, so that when the unmanned cluster faces complex tasks (such as the continuous execution of multiple actions, the collaborative operation of multiple devices) or dynamic environments (such as device status fluctuations, sudden interference), it can automatically complete the tool call decision without rewriting the adaptation rules, which improves the automation level of the coordination decision scheme delivery to a certain extent, and can automatically adapt to the scenario requirements to generate a coordination tool sequence with higher call benefit value, thereby improving the reliability of the tool call decision.

[0021] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0022] Figure 1 is a schematic block diagram of an electronic device provided in an embodiment of this application.

[0023] As shown in Figure 1, the electronic device 100 includes a processor 101 and a memory 102, which are connected by a bus 103, such as a PCIe (Peripheral Component Interconnect Express) bus.

[0024] Specifically, processor 101 provides computing and control capabilities to support the operation of the entire electronic device 100. Processor 101 can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0025] Specifically, the memory 102 can be a Flash chip, a read-only memory (ROM) disk, an optical disk, a USB flash drive, or a portable hard drive, etc.

[0026] Those skilled in the art will understand that the structure shown in FIG1 is merely a block diagram of a portion of the structure related to the embodiments of this application, and does not constitute a limitation on the electronic device to which the embodiments of this application are applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0027] The processor 101 is used to run a computer program stored in the memory 102, and implements any of the unmanned swarm command and control methods provided in the embodiments of this application when executing the computer program.

[0028] As shown in Figure 2, the cluster command and control device 200 may include a preset planning system 201, a preset coordination system 202, and a preset execution system 203.

[0029] The pre-defined planning system 201 includes one or more planning agents. Each planning agent is driven by a large language model to achieve reasoning and decision-making. Each planning agent can independently perform overall planning for the task to be commanded and coordinated based on the global situational information of the task, and obtain an overall decision scheme. Furthermore, the pre-defined planning system 201 can include multiple planning agents, and can further filter the multiple overall decision schemes obtained by the multiple planning agents independently, and use them as the final overall decision scheme for the task to be commanded and coordinated.

[0030] The pre-defined coordination system 202 includes multiple coordination agents. Each coordination agent is driven by a large language model to achieve reasoning and decision-making. Each coordination agent can independently perform coordination planning for the target local task based on the local situation information of the target local task, and obtain a coordination decision scheme. Furthermore, the multiple coordination decision schemes obtained independently by the multiple coordination agents can be further filtered to become the final coordination decision scheme for the target local task.

[0031] The pre-defined execution system comprises multiple execution agents, each driven by a large language model to achieve reasoning and decision-making. Each agent can independently plan the execution of the target unmanned unit's task based on local situational information, resulting in an execution decision scheme for the target unmanned unit. Furthermore, the multiple execution decision schemes independently planned by the agents can be further filtered to become the final execution decision scheme for the target unmanned unit. The execution decision scheme planned by the pre-defined execution system is then distributed to the target unmanned unit in the unmanned cluster for execution, thereby enabling the unmanned cluster to complete the coordinated task to be commanded.

[0032] Therefore, it can be seen that by setting up a three-level hierarchical structure of a preset planning system 201, a preset coordination system 202, and a preset execution system, the cluster command and control device decomposes the task to be commanded and coordinated into decision-making problems such as overall decision-making scheme, coordination decision-making scheme, and execution decision-making scheme. This can, to a certain extent, solve the problem of low decision-making efficiency caused by the exponential growth of computing and communication burden, improve the decision-making efficiency of unmanned cluster command schemes, and improve the problem of low coordination efficiency of unmanned clusters to a certain extent.

[0033] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the electronic equipment described above can be referred to the corresponding process in the following embodiments of the unmanned swarm command and control method, and will not be repeated here.

[0034] The following will use the cluster command and control device shown in Figure 2 as the execution subject of the unmanned cluster command and control method as an example to describe the unmanned cluster command and control method provided in this application embodiment in detail. For the sake of simplification and ease of description, the execution subject will be omitted in the subsequent method embodiments.

[0035] Please refer to Figure 3, which is a flowchart illustrating an unmanned swarm command and control method provided in an embodiment of this application. The unmanned swarm command and control method includes steps 301-305, wherein: 301, decision planning is performed on the unmanned swarm's tasks to be commanded and coordinated, resulting in the target unmanned unit to be scheduled for the tasks to be commanded and coordinated, and the execution decision scheme for the target unmanned unit's target execution task.

[0036] The unmanned cluster includes multiple unmanned units.

[0037] Among them, tasks requiring command and control refer to tasks that need to be completed collaboratively by a swarm of unmanned aerial vehicles (UAVs) in a coordinated manner. For example, in an environmental monitoring scenario, a task requiring command and control could be "to use 20 UAVs equipped with multispectral sensors and 5 unmanned surface vessels equipped with water quality analyzers to conduct ecological monitoring of a lake and a surrounding 10 square kilometers." In a logistics and transportation scenario, a task requiring command and control could be "to use 10 unmanned delivery vehicles to complete the material delivery task for 30 companies in an industrial park." Furthermore, in an emergency rescue scenario, a task requiring command and control could be "to use 8 rescue UAVs equipped with thermal imagers and 4 unmanned vehicles equipped with life detectors to conduct emergency rescue in a 3 square kilometer disaster area caused by a landslide triggered by heavy rain in a mountainous region."

[0038] For example, step 301 may specifically include the following steps 3011 to 3014: 3011, obtaining the unmanned swarm's collaborative tasks to be commanded.

[0039] In some embodiments, the cluster control device can establish a connection with external devices (such as mobile phones, computers, etc.) to receive cluster control tasks sent by the external devices and treat these tasks as unmanned cluster coordination tasks to be commanded. For example, a user can input a cluster control task such as "Use 10 unmanned delivery vehicles to complete the material delivery task for 30 companies in an industrial park" through an external device. The external device sends the user-input cluster control task to the cluster control device, which treats the received cluster control task as a coordination task to be commanded. Thus, the cluster control device can obtain the coordination task to be commanded, "Use 10 unmanned delivery vehicles to complete the material delivery task for 30 companies in an industrial park".

[0040] In some embodiments, the cluster control device can be integrated into an electronic device (such as a computer). Users can input cluster control tasks, such as "using 5 unmanned surface vessels equipped with water quality detectors to conduct ecological monitoring of a lake and a surrounding 10 square kilometers area", through an electronic device with the cluster control device integrated with it. The cluster control device will take the cluster control task input by the user as a task to be commanded and coordinated, so that the cluster control device can obtain the task to be commanded and coordinated.

[0041] 3012. Perform overall planning on the task to be commanded and coordinated to obtain an overall decision-making scheme for the task to be commanded and coordinated.

[0042] For example, the overall planning of the task to be commanded and coordinated can be carried out by a target planning agent associated with the task to be commanded and coordinated in a preset planning system, so as to obtain the overall decision scheme of the task to be commanded and coordinated.

[0043] Among them, the target planning agent is the planning agent associated with the task to be commanded and coordinated in the preset planning system.

[0044] There are multiple ways to implement step 3012. For example, it includes the following methods (1) to (2): (1) In some embodiments, the target planning agent directly performs cyclic reasoning planning on the task to be commanded and coordinated to obtain the overall decision scheme. At this time, the overall decision scheme can be determined by executing the following steps 3012-1A to 3012-4A by the target planning agent. Steps 3012-1A to 3012-4A are as follows: 3012-1A: Based on the initial state of the target planning agent, the current global environment information of the target planning agent and the initial coordination message of the first external agent of the target planning agent, the initial decision scheme of the target planning agent is obtained.

[0045] 3012-2A. Interact with the target planning agent according to its initial decision scheme to obtain the initial interaction action and the initial interaction result of the target planning agent.

[0046] 3012-3A. Based on the initial state of the target planning agent, the initial interaction action of the target planning agent, and the initial interaction result of the target planning agent, the updated state of the target planning agent and the updated global environment information of the target planning agent are obtained.

[0047] 3012-4A. Based on the updated state of the target planning agent, the updated collaborative message of the first external agent, and the updated global environment information, a planning process is performed to obtain the overall decision scheme.

[0048] (2) In some embodiments, if the task to be commanded and coordinated meets the preset parallel reasoning conditions, the preset planning and reasoning branches are invoked to perform overall planning and fusion to obtain an overall decision scheme; if the preset parallel reasoning conditions are not met, the overall decision scheme is obtained by referring to the steps 3012-1A to 3012-4A. At this time, the overall decision scheme can be determined by executing the following steps 3012-1B to 3012-3B by the target planning agent. Steps 3012-1B to 3012-3B are as follows: 3012-1B, evaluate the parallel reasoning conditions of the task to be commanded and coordinated.

[0049] 3012-2B When the task to be commanded and coordinated meets the preset parallel reasoning conditions, the preset planning and reasoning branches are invoked to perform overall planning and fusion to obtain the overall decision scheme of the task to be commanded and coordinated.

[0050] For example, step 3012-2B specifically includes: (1) reasoning about the task to be commanded and coordinated based on the current global environment information of the target planning agent to obtain the planning reasoning information of the target planning agent, specifically as follows: obtaining the current global environment information of the target planning agent; obtaining the first historical decision information of the target planning agent; obtaining the first historical decision information of the target planning agent; and performing reasoning based on the task to be commanded and coordinated, the current global environment information, the first retrieval information, and the first historical decision information to obtain the planning reasoning information. (2) calling each planning reasoning branch to perform reasoning based on the planning reasoning information to obtain the decision scheme of each planning reasoning branch, specifically as follows: obtaining the weight coefficient of each planning reasoning branch; obtaining the constraint condition of each planning reasoning branch; obtaining the retrieval information of each planning reasoning branch; and performing reasoning based on the planning reasoning information, the weight coefficient of each planning reasoning branch, the constraint condition of each planning reasoning branch, and the retrieval information of each planning reasoning branch to obtain the decision scheme of each planning reasoning branch. (3) Obtain the fusion weights of each planning reasoning branch, specifically as follows: Based on the decision schemes of each planning reasoning branch, obtain the quality assessment index of each planning reasoning branch; determine the fusion weights of each planning reasoning branch according to the quality assessment indexes of each planning reasoning branch and the preset weight mapping strategy. (4) Perform weighted fusion of the decision schemes of each planning reasoning branch according to the fusion weights of each planning reasoning branch to obtain the planning decision scheme.

[0051] 3012-3B. When the task to be commanded and coordinated does not meet the preset parallel reasoning conditions, perform cyclic reasoning planning on the task to be commanded and coordinated to obtain the overall decision scheme.

[0052] The implementation of step 3012-3B, "perform cyclic reasoning planning on the task to be commanded and coordinated to obtain the overall decision scheme", is similar to the implementation of steps 3012-1A to 3012-4A. For details, please refer to the relevant descriptions, which will not be repeated here.

[0053] Furthermore, in order to ensure that the preset planning system can successfully send the overall decision plan of the task to be commanded and coordinated to the lower-level preset coordination system, and control the preset coordination system to start the decision planning of multiple target local tasks included in the overall decision plan, thereby ensuring the further coordination planning of the target local tasks, the method may also include the following steps A1~A4: A1. Based on the current task status of the target local task and the overall decision plan, determine the first set of sending tools for the target local task candidates.

[0054] The first set of delivery tools refers to the set of tools that can be used to deliver the target local task.

[0055] For example, step 11 may include: extracting requirements based on the current task status of the target local task and the overall decision scheme to obtain first distribution requirement information for multiple sub-tasks; matching the first distribution requirement information of each sub-task with a preset tool set to obtain candidate tools for each sub-task; and obtaining a first distribution tool set based on the candidate tools for each sub-task.

[0056] The first requirement information is used to indicate the requirement for issuing the target local task.

[0057] A2. With the goal of maximizing the call benefit value of the first distribution tool set, the first distribution tool set is arranged to obtain the planning tool sequence of the target local task.

[0058] For example, step A2 may include: performing orchestration processing based on the first distribution tool set to obtain each first tool sequence of the first distribution tool set; constructing a target optimization function for the first distribution tool set, wherein the target optimization function includes a sequence validity parameter, a sequence execution cost parameter, and a sequence reliability parameter, and the target optimization function is used to indicate the call benefit value; with the goal of maximizing the target optimization function, outputting the planning tool sequence based on the sequence validity, sequence execution cost, and sequence reliability of each first tool sequence.

[0059] The first tool sequence refers to the tool sequence obtained by arranging and processing the first set of issued tools.

[0060] Among them, the planning tool sequence is a sequence of tools arranged to generate instructions for issuing overall decision-making plans.

[0061] A3. Based on the planning tool sequence, perform instruction generation processing to obtain the overall decision-making scheme issuance instruction.

[0062] A4. Send the instruction to the preset coordination system to issue the overall decision-making scheme, so as to control the preset coordination system to start the decision-making and planning of the target local task.

[0063] The implementation of steps A1 to A4 is similar to that of steps 302 to 305. For details, please refer to the relevant explanations in this article. For the sake of simplicity, they will not be repeated here.

[0064] Thus, through steps A1 to A4, tools can be automatically planned and invoked to generate and issue commands. Even when multiple target local tasks need to be issued simultaneously, the tools can still be automatically adapted to generate and issue commands. This allows unmanned clusters to automatically complete tool invocation decisions when facing complex tasks (such as the continuous execution of multiple actions and the collaborative operation of multiple devices) or dynamic environments (such as fluctuations in device status and sudden interference), without having to rewrite adaptation rules. This improves the automation level of the overall decision-making scheme issuance to a certain extent. It can automatically adapt to scenario requirements to generate planning tool sequences with higher call benefits, thereby improving the reliability of tool invocation decisions.

[0065] 3013. Perform coordination planning on the target local task to obtain the coordination decision scheme for the target local task.

[0066] The coordination decision scheme is used to instruct the target unmanned unit to be scheduled for the target local task in the unmanned cluster and the target execution task of the target unmanned unit.

[0067] For example, a target coordination agent associated with the target local task in a preset coordination system can be used to coordinate and plan the target local task, thereby obtaining a coordination decision scheme for the target local task.

[0068] The target coordinating agent is the coordinating agent associated with the target local task in the preset coordination system.

[0069] There are multiple ways to implement step 3013. For example, it includes the following methods (1) to (2): (1) In some embodiments, the target coordinating agent directly performs cyclic reasoning planning on the target local task to obtain a coordination decision scheme. At this time, the target coordinating agent executes the following steps 3013-1A to 3013-4A to determine the coordination decision scheme of the target local task. Steps 3013-1A to 3013-4A are as follows: 3013-1A: Plan according to the initial state of the target coordinating agent, the current local environment information of the target coordinating agent and the initial coordination message of the second external agent of the target coordinating agent to obtain the initial decision scheme of the target coordinating agent.

[0070] 3013-2A. Interact with the target coordinating agent according to its initial decision-making scheme to obtain the target coordinating agent's initial interaction action and initial interaction result.

[0071] 3013-3A. Based on the initial state of the target coordinating agent, the initial interaction action of the target coordinating agent, and the initial interaction result of the target coordinating agent, the updated state of the target coordinating agent and the updated local environment information of the target coordinating agent are obtained.

[0072] 3013-4A. Based on the updated state of the target coordinating agent, the updated collaborative message of the second external agent, and the updated local environment information, a planning process is performed to obtain the coordination decision scheme.

[0073] (2) In some embodiments, if the target local task meets the preset parallel reasoning conditions, the preset planning reasoning branches are invoked to perform coordinated planning and fusion to obtain a coordinated decision scheme; if the preset parallel reasoning conditions are not met, the coordinated decision scheme is obtained by referring to the steps 3013-1A to 3013-4A. At this time, the coordinated decision scheme can be determined by executing the following steps 3012-1B to 3012-3B by the target planning agent. Steps 3013-1B to 3013-3B are as follows: 3013-1B, evaluate the parallel reasoning conditions of the target local task.

[0074] For example, step 3013-2B specifically includes: (1) reasoning about the target local task based on the current local environment information of the target coordinating agent to obtain the coordination reasoning information of the target coordinating agent, wherein the target coordinating agent is a coordinating agent associated with the target local task in the preset coordination system; specifically as follows: determining the feasibility score of each coordination reasoning branch based on the decision scheme of each coordination reasoning branch; determining the efficiency score of each coordination reasoning branch based on the decision scheme of each coordination reasoning branch; determining the orderliness score of each coordination reasoning branch based on the decision scheme of each coordination reasoning branch; determining the quality evaluation index of each coordination reasoning branch based on the feasibility score, efficiency score, orderliness score, and preset quality evaluation function of each coordination reasoning branch. (2) Based on the coordinated reasoning information, call each coordinated reasoning branch to perform reasoning to obtain the decision scheme of each coordinated reasoning branch; specifically as follows: obtain the weight coefficient of each coordinated reasoning branch; obtain the constraint condition of each coordinated reasoning branch; obtain the retrieval information of each coordinated reasoning branch; perform reasoning based on the coordinated reasoning information, the weight coefficient of each coordinated reasoning branch, the constraint condition of each coordinated reasoning branch and the retrieval information of each coordinated reasoning branch to obtain the decision scheme of each coordinated reasoning branch. (3) Obtain the fusion weight of each coordinated reasoning branch; specifically as follows: based on the decision scheme of each coordinated reasoning branch, obtain the quality assessment index of each coordinated reasoning branch; determine the fusion weight of each coordinated reasoning branch based on the quality assessment index of each coordinated reasoning branch and the preset weight mapping strategy. (4) Perform weighted fusion of the decision schemes of each coordinated reasoning branch based on the fusion weight of each coordinated reasoning branch to obtain the coordinated decision scheme.

[0075] 3013-3B. When the target local task does not meet the preset parallel reasoning conditions, perform cyclic reasoning planning on the target local task to obtain a coordination decision scheme for the target local task.

[0076] Thus, for each target local task planned in step 3012, coordination planning can be carried out by referring to steps 3013-1A to 3013-4A or 3013-1B to 3013-3B to obtain a coordination decision scheme for each target local task.

[0077] Furthermore, in order to ensure that the preset planning system can successfully distribute the coordination decision scheme of the target local task to the lower-level preset execution system, and control the preset execution system to start the decision planning of multiple target execution tasks included in the coordination decision scheme, thereby ensuring the further coordination planning of the target execution task, the method may also include the following steps B1~B4: B1, Based on the current task status of the target execution task and the coordination decision scheme, determine the second distribution tool set of the target execution task candidates.

[0078] The second set of delivery tools refers to the set of tools that can be used to deliver the target execution task.

[0079] For example, step B1 may include: extracting requirements based on the current task status of the target execution task and the execution decision scheme to obtain second distribution requirement information for multiple sub-tasks; matching the second distribution requirement information of each sub-task with a preset tool set to obtain candidate tools for each sub-task; and obtaining a second distribution tool set based on the candidate tools for each sub-task.

[0080] The second requirement information is used to instruct the target to issue the requirement for executing the task.

[0081] B2. With the goal of maximizing the call benefit value of the second distribution tool set, the second distribution tool set is arranged to obtain the coordination tool sequence for the target execution task.

[0082] The second tool sequence refers to the tool sequence obtained by orchestrating the second set of distributed tools.

[0083] Among them, the coordination tool sequence is a sequence of tools arranged to generate instructions for issuing coordination decision-making schemes.

[0084] For example, step B2 may include: performing orchestration processing based on the second distribution tool set to obtain each second tool sequence of the second distribution tool set; constructing a target optimization function for the second distribution tool set, wherein the target optimization function includes a sequence validity parameter, a sequence execution cost parameter, and a sequence reliability parameter, and the target optimization function is used to indicate the call benefit value; with the goal of maximizing the target optimization function, outputting the coordination tool sequence based on the sequence validity, sequence execution cost, and sequence reliability of each second tool sequence.

[0085] B3. Based on the coordination tool sequence, perform instruction generation processing to obtain the issuance instruction for the coordination decision scheme.

[0086] B4. Send the instruction to the preset execution system to issue the coordination decision scheme, so as to control the preset execution system to start the decision planning of the target execution task.

[0087] The implementation of steps B1 to B4 is similar to that of steps 302 to 305. For details, please refer to the relevant explanations in this article. For the sake of simplicity, they will not be repeated here.

[0088] Thus, through steps B1 to B4, tools can be automatically planned and invoked to generate and issue instructions. Even when multiple target tasks need to be issued simultaneously, the tools can still automatically adapt to generate and issue instructions. This allows unmanned clusters to automatically complete tool invocation decisions when facing complex tasks (such as the continuous execution of multiple actions and the collaborative operation of multiple devices) or dynamic environments (such as fluctuations in device status and sudden interference), without having to rewrite adaptation rules. This improves the automation level of coordination decision-making scheme issuance to a certain extent, and can automatically adapt to scenario requirements to generate coordination tool sequences with higher call benefits, thereby improving the reliability of tool invocation decisions.

[0089] 3014. Perform execution planning on the target execution task to obtain the execution decision scheme for the target execution task.

[0090] For example, the execution plan for the target execution task can be obtained by using a target execution agent associated with the target unmanned unit in a preset execution system.

[0091] Among them, the target execution agent is the execution agent associated with the target unmanned unit in the preset execution system.

[0092] There are multiple ways to implement step 3014. For example, it includes the following methods (1) to (2): (1) In some embodiments, the target execution agent directly performs cyclic reasoning planning on the target execution task to obtain the execution decision scheme. At this time, the execution decision scheme can be determined by the target execution agent performing the following steps 3014-1A to 3014-4A. Steps 3014-1A to 3014-4A are as follows: 3014-1A: Plan according to the initial state of the target execution agent, the current individual environment information of the target execution agent and the initial collaborative message of the third external agent of the target execution agent to obtain the initial decision scheme of the target execution agent.

[0093] 3014-2A. Interact with the target execution agent according to its initial decision-making scheme to obtain the target execution agent's initial interaction action and initial interaction result.

[0094] 3014-3A. Based on the initial state of the target execution agent, the initial interaction action of the target execution agent, and the initial interaction result of the target execution agent, update the state of the target execution agent and the updated individual environment information of the target execution agent.

[0095] 3014-4A. Based on the updated state of the target execution agent, the updated collaborative message of the third external agent, and the updated individual environment information, a planning process is performed to obtain the execution decision scheme.

[0096] (2) In some embodiments, if the target execution task meets the preset parallel inference conditions, the preset execution inference branches are invoked to perform execution planning and fused to obtain an execution decision scheme; if the preset parallel inference conditions are not met, the execution decision scheme is obtained by referring to steps 3014-1A to 3014-4A for cyclic inference planning. At this time, the execution decision scheme can be determined by executing the following steps 3014-1B to 3014-3B by the target execution agent. Steps 3014-1B to 3014-3B are as follows: 3014-1B, evaluate the parallel inference conditions of the target execution task.

[0097] For example, in step 3014-1B: the task attributes of the target execution task can be obtained, including target level characteristics, environmental dynamic characteristics, multi-target conflict characteristics, and collaboration scale characteristics; it is determined whether the task attributes meet preset parallel inference rules, wherein the preset parallel inference rules are that the target execution task can be decomposed into at least two sub-target levels, the frequency of environmental dynamic changes exceeds a preset threshold, there are at least two mutually conflicting execution targets, or the number of unmanned units to be collaborated reaches a preset collaboration scale; if the task attributes meet the preset parallel inference rules, it is determined that the target execution task meets the preset parallel inference conditions; if the task attributes do not meet the preset parallel inference rules, it is determined that the target execution task does not meet the preset parallel inference conditions. In this way, parallel inference can be selectively started or not started, avoiding the waste of computing resources caused by blind parallelism, and also preventing the limitations of serial loop inference planning in dealing with complex tasks. When the task is complex (multiple sub-targets, environmental dynamics, etc.), parallel inference improves decision robustness, and when the task is simple, serial loop inference planning improves efficiency, thereby optimizing the allocation of decision resources and improving the adaptability and overall effect of unmanned swarm command and decision-making. 3013-2B When the target local task meets the preset parallel reasoning conditions, the preset coordination reasoning branches are invoked to perform coordination planning and fusion to obtain the coordination decision scheme of the target local task.

[0098] 3014-2B When the target execution task meets the preset parallel reasoning conditions, the preset execution reasoning branches are called to perform execution planning and fusion to obtain the execution decision scheme of the target execution task.

[0099] For example, step 3042B specifically includes: (1) reasoning about the target execution task based on the current individual environment information of the target execution agent to obtain the execution reasoning information of the target execution agent, specifically as follows: obtaining the current individual environment information of the target execution agent; obtaining the third historical decision information of the target execution agent; obtaining the third historical decision information of the target execution agent; performing reasoning based on the target execution task, the current individual environment information, the third retrieval information and the third historical decision information to obtain the execution reasoning information. (2) calling each execution reasoning branch to perform reasoning based on the execution reasoning information to obtain the decision scheme of each execution reasoning branch, specifically as follows: obtaining the weight coefficient of each execution reasoning branch; obtaining the constraint condition of each execution reasoning branch; obtaining the retrieval information of each execution reasoning branch; performing reasoning based on the execution reasoning information, the weight coefficient of each execution reasoning branch, the constraint condition of each execution reasoning branch and the retrieval information of each execution reasoning branch to obtain the decision scheme of each execution reasoning branch. (3) Obtain the fusion weight of each execution inference branch, specifically as follows: Based on the decision scheme of each execution inference branch, obtain the quality assessment index of each execution inference branch; determine the fusion weight of each execution inference branch according to the quality assessment index of each execution inference branch and the preset weight mapping strategy. (4) Perform weighted fusion of the decision schemes of each execution inference branch according to the fusion weight of each execution inference branch to obtain the execution decision scheme.

[0100] 3014-3B. When the target execution task does not meet the preset parallel reasoning conditions, perform cyclic reasoning planning on the target execution task to obtain the execution decision scheme.

[0101] Thus, for each target execution task planned in step 3013, the execution decision scheme for each target execution task can be obtained by referring to the execution planning methods of steps 3014-1A to 3014-4A or 3014-1B to 3014-3B.

[0102] 302. Based on the current task status of the target execution task and the execution decision scheme, determine the candidate tool set for the target execution task.

[0103] The candidate tool set refers to the set of candidate tools for the target task.

[0104] In some embodiments, step 302 may specifically include the following steps 3021 to 3023: 3021, extracting requirements based on the current task status of the target task and the execution decision scheme to obtain execution requirement information for multiple sub-tasks.

[0105] The execution requirement information for each subtask is used to indicate, but is not limited to, the task objective information, task constraint information, and task tool type of the subtask.

[0106] For example, suppose the target local task is "25 Team A drones generate heart-shaped waypoints in the left half of the area and coordinate obstacle avoidance," and the target execution task is "Drone 01 flies according to the planned waypoints, avoiding the shed obstacle at X195-Y355." The target execution task is broken down into several sub-tasks: Sub-task 1 "Generate the target waypoints for Drone 01," Sub-task 2 "Detect obstacles in the flight path," and Sub-task 3 "Plan the flight sequence of Drone 01." Preset toolkit: T1 (heart-shaped coordinate calculation tool), T2 (laser obstacle detection tool), T3 (ultrasonic obstacle detection tool), T4 (coordinated timing planning tool), and T5 (ordinary coordinate calculation tool). The current state of the target execution task: Drone 01's initial position (X180, Y340, Z50), with a shed obstacle (12 meters high) at X195-Y355 in the left half of the area, real-time wind speed of 2.5, and needs to reach its destination within 30 seconds. The requirements were extracted to obtain the execution requirements information for subtask 1: "Generate a 3D waypoint adapted to the left half of the heart shape, with an accuracy of ≤0.5 meters", the execution requirements information for subtask 2: "Detect fixed obstacles within a 10-meter range, with an accuracy of ≥95%", and the execution requirements information for subtask 3: "Plan the flight sequence from the initial position to the target waypoint, adapting to the requirement of arriving within 30 seconds".

[0107] 3022. Match the execution requirements of each subtask with a preset set of tools to obtain candidate tools for each subtask.

[0108] For example, step 3022 may specifically include the following steps a1 to a3: a1. Matching based on the execution requirement information of each subtask to obtain the semantic matching degree and scenario matching degree between each tool in the preset tool set and each subtask.

[0109] For example, taking step 3021 as an example scenario, for subtask 1, based on the execution requirement information of subtask 1, "generate 3D waypoints adapted to the left half of the heart shape, with an accuracy ≤ 0.5 meters", matching is performed to obtain the semantic matching degree and scene matching degree between each tool and subtask 1, as shown in Table 1 below: Table 1

[0110] Similarly, for subtask 2, based on the execution requirement information of subtask 2, "detect fixed obstacles within a 10-meter range with an accuracy of ≥95%", matching is performed to obtain the semantic matching degree and scene matching degree between each tool and subtask 2.

[0111] Similarly, for subtask 3, based on the execution requirement information of subtask 3, "planning the flight sequence from the initial position to the target waypoint, adapting to the requirement of arriving within 30 seconds", the semantic matching degree and scenario matching degree between each tool and subtask 3 are obtained.

[0112] a2. Based on the semantic matching degree and scenario matching degree between each tool and each subtask, as well as the execution efficiency of each tool and the historical execution data of each tool, determine the probability of each tool being selected for each subtask.

[0113] For example, firstly, a preset relevance function can be invoked to determine the relevance of each tool to each subtask under the current task state s of the target execution task, based on the semantic and scene matching degrees between each tool and each subtask, as well as the execution efficiency and historical execution data of each tool. Then, a preset tool selection function is invoked to determine the probability of each tool being selected by each subtask under the current task state s of the target execution task, based on the relevance of the tool to each subtask. For example, if the preset relevance function is shown in Formula 1 and the preset tool selection function is shown in Formula 2, then the relevance function shown in Formula 1 is invoked first, based on the tool T... i sub-tasks The semantic matching degree and scene matching degree, as well as the tool T i Execution efficiency, tools T i The historical execution data is used to calculate the tool T. i sub-tasks The correlation (as shown in Formula 1) As shown; then, call the tool selection function shown in Formula 2, according to tool T i sub-tasks The correlation is calculated to obtain the subtasks under the current task state s of the target task. Selection tool T i The probability (as in Formula 2) (As shown). In this way, semantic matching, capability assessment, and usability checks can be performed on each tool, and finally, tools with strong semantic matching, capability matching, and usability can be selected as candidate tools.

[0114] Formula 1 In Formula 1, Representation tool T i sub-tasks The correlation, , , , Representing tool T respectively i sub-tasks Scene matching degree, tool T i sub-tasks semantic matching degree, tool T i Historical execution data, tools T i Execution efficiency , , , There are four adjustable hyperparameters, which together constitute the weighting system of the correlation assessment model.

[0115] Formula 1 defines the "correlation" function between a tool and a subtask in a specific task scenario. This function forms the basis for calculating the tool selection probability and aims to comprehensively evaluate the tool's applicability from multiple dimensions. Formula 1 corresponds to the calculation of tool T. i Functions related to subtasks.

[0116] Specifically, parameters The scenario matching weight coefficient is used to adjust the importance of "scenario matching between tools and subtasks" in the total relevance score. Scenario matching assesses the applicability of the tool in the current task state (such as environment and resources).

[0117] parameter The semantic matching degree weight coefficient is used to adjust the importance of "semantic matching degree between tool and subtask" in the total relevance score. The semantic matching degree measures the textual similarity between the tool's functional description and the subtask requirements.

[0118] parameter The historical execution data weighting coefficient is used to adjust the importance of "the tool's historical execution data" in the overall relevance score. Historical data usually refers to the tool's past success rate, call frequency, etc., reflecting its reliability.

[0119] parameter The execution efficiency weighting coefficient is used to adjust the importance of "tool execution efficiency" in the overall relevance score. Execution efficiency focuses on performance indicators such as tool response time and computational cost.

[0120] By adjusting the values ​​of these four weighting coefficients (which typically sum to 1), the tool selection strategy can be optimized based on different command and decision-making preferences. For example, in tasks requiring high reliability, the weighting coefficients can be increased. The value can be increased in tasks requiring high efficiency; The value of .

[0121] Formula 2 In Formula 2, This represents a subtask in the current task state s of the target task. Selection tool T i The probability, Representation tool T i sub-tasks The correlation is given by k, which represents the number of tools in the preset toolset.

[0122] a3. Based on the probability of selecting each tool for each subtask, determine the candidate tool for each subtask.

[0123] For example, based on the current task state s of the target task, subtasks Selection tool T i Based on the probability, tools with probabilities greater than a preset probability threshold are selected as subtasks. Candidate tools are then obtained. Similarly, candidate tools for each subtask can be obtained, and the set of candidate tools for all subtasks constitutes the set of candidate tools for the target execution task.

[0124] Thus, steps a1-a3 obtain the semantic and scenario matching degrees between each tool and the subtask through matching, and determine the selection probability by combining tool execution efficiency and historical execution data, thereby determining candidate tools for the subtask. Tools can be evaluated from multiple dimensions, including semantic adaptation, scenario adaptation, execution efficiency, and historical reliability, avoiding the limitations of single-dimensional selection. This ensures that candidate tools align with the core needs of the subtask, while quantifying probabilities enhances the scientific rigor of tool selection and reduces subjective judgment bias. High-quality candidate tools for the subtask lay the foundation for subsequent orchestration of the optimal execution sequence, further improving the accuracy of unmanned swarm tool invocation decisions and the reliability of task execution.

[0125] 3023. Based on the candidate tools for each subtask, the candidate tool set is obtained.

[0126] For example, the candidate tool set determined through steps a1 to a3 is {T1 (cardioid coordinate calculation tool), T2 (laser obstacle detection tool), T4 (cooperative temporal planning tool)}, which provides a foundation for subsequent tool orchestration.

[0127] Thus, by combining the current task status and execution decision plan of the target task in steps 3021-3023 to extract the sub-task execution requirements, and then matching and filtering candidate tools with the preset tool set according to the requirements of each sub-task, a candidate tool set is finally obtained. This allows complex execution tasks to be broken down into sub-tasks with precisely matched tools, avoiding blind tool selection. This ensures that the candidate tools are highly compatible with the requirements of each sub-task, and improves the adaptability of the overall tool set through precise matching at the sub-task level. This provides a high-quality foundation for subsequent tool orchestration, further enhancing the targeting and reliability of tool invocation decisions, and helping to improve the automation effect of unmanned cluster task execution.

[0128] 303. With the goal of maximizing the invocation benefit value of the candidate tool set, the candidate tool set is arranged to obtain the execution tool sequence for the target execution task.

[0129] Among them, the execution tool sequence is a sequence of control instructions that are arranged to generate execution decision schemes.

[0130] The call benefit value is used to indicate the execution effect of each tool in the candidate tool set after it is called.

[0131] In some embodiments, step 303 may specifically include the following steps 3031 to 3033: 3031, performing an arrangement process based on the candidate tool set to obtain each initial tool sequence of the candidate tool set.

[0132] The initial tool sequence refers to the tool sequence obtained by orchestrating the candidate tool set, such as the tool sequence. This represents a sequence containing m tools.

[0133] For example, the candidate tool set contains three tools: T1 (cardioid coordinate calculation tool), T2 (laser obstacle detection tool), and T4 (cooperative temporal planning tool). All feasible permutations and combinations (excluding logically invalid sequences) yield three initial tool sequences: Sequence S1: T2 (laser obstacle detection tool) → T1 (cardioid coordinate calculation tool) → T4 (cooperative temporal planning tool); Sequence S2: T1 (cardioid coordinate calculation tool) → T2 (laser obstacle detection tool) → T4 (cooperative temporal planning tool); Sequence S3: T1 (cardioid coordinate calculation tool) → T4 (cooperative temporal planning tool) → T2 (laser obstacle detection tool).

[0134] For example, step 3032 may specifically include the following steps b1 to b3: b1, determine the pre-execution score of each tool in the candidate tool set based on the pre-execution state of each tool in the candidate tool set and the pre-execution state of the target execution task.

[0135] b2. Determine the post-execution score of each tool in the candidate tool set based on the post-execution status of each tool in the candidate tool set and the post-execution status of the target execution task.

[0136] b3. Based on the pre-execution score and post-execution score of each tool in the candidate tool set, determine the insertion tool for the current orchestration sequence of the candidate tool set.

[0137] b4. Until the current orchestration sequence meets the requirements of the target execution task, the current orchestration sequence shall be used as the initial tool sequence.

[0138] For example, referring to Formula 3, to perform dynamic orchestration planning on the candidate tool set, first, an empty sequence is initialized as the current orchestration sequence; during the dynamic orchestration planning process, a preset dynamic programming algorithm is called (as shown in Formula 3). On the one hand, based on each tool T... i The pre-execution state s and the pre-execution state of the target task Determine each tool T i The pre-execution score (i.e., the score at time t); on the other hand, based on each tool T i The post-execution state s' and the post-execution state of the target task. Determine each tool T i The post-execution score (i.e., the score at time t+1); then, based on each tool T i Pre-execution scoring (as in Formula 3) (as shown) and each tool T i Post-execution scoring (as in Formula 3) As shown), calculate each tool T. i Insertion score in the current orchestration sequence; based on each tool T i Based on the insertion score of the current orchestration sequence, the tool with the highest insertion score is selected from the candidate tool set and added to the current orchestration sequence. This process is repeated, selecting the tool with the highest insertion score from the candidate tool set and adding multiple tools to the current orchestration sequence sequentially, until the current orchestration sequence meets the requirements of the target execution task. At this point, the current orchestration sequence is used as the initial tool sequence. This process is repeated to obtain multiple initial tool sequences that meet the requirements.

[0139] Formula 3 In Formula 3, , , Each tool T represents a different tool. i Insertion score in the current orchestration sequence, for each tool T iPre-execution score (i.e., score at time t), each tool T i The post-execution score (i.e., the score at time t+1), s' is for each tool T i Post-execution status The post-execution state of the task for the target, s for each tool T i Pre-execution state The parameters represent the state before the target task is executed. This represents a discount factor for future earnings.

[0140] Equation 3 involves calculating an "insertion score" for inserting a single tool into the current sequence during the dynamic orchestration planning process of the tool sequence. This insertion score is used to decide which tool to add to the tool sequence being built at each step. In the context of dynamic programming and reinforcement learning, the parameters of Equation 3... This represents a discount factor for future returns, used to balance the importance of "pre-execution score" (immediate returns) and "post-execution score" (potential future returns). Parameters The range of its value is [0, 1]. When When the value approaches zero, the decision-making model focuses more on immediate gains and tends to choose the tool that currently appears to be the best, which is a "short-sighted" strategy. Conversely, when... As the value approaches 1, the decision-making model focuses more on long-term gains, choosing tools that may not be optimal in the present but are more beneficial for future steps in order to obtain a better final sequence—a "far-sighted" strategy. Therefore, It plays a key role in Equation 3, ensuring that the algorithm is forward-looking when constructing the tool sequence, thereby finding the globally optimal or near-optimal solution.

[0141] Thus, by combining steps b1-b3 with the pre- and post-execution states of the tools and target tasks, the pre- and post-execution scores of each tool are determined. Based on these scores, tools for insertion into the current orchestration sequence are selected until the task requirements are met, resulting in an initial tool sequence. This allows for the evaluation of tool adaptability across the entire task lifecycle, avoiding the bias of relying solely on a single state. This ensures both the adaptability of the inserted tools and the sequence, and through iterative optimization of the orchestration logic, makes the initial tool sequence more aligned with the full lifecycle requirements of the task execution. This provides a high-quality foundation for subsequent selection of the optimal sequence using the target optimization function, further improving the rationality of tool orchestration and the reliability of unmanned cluster task execution.

[0142] 3032. Construct the objective optimization function for the candidate tool set.

[0143] The objective optimization function includes a sequence validity parameter, a sequence execution cost parameter, and a sequence reliability parameter, and the objective optimization function is used to indicate the call benefit value.

[0144] For example, the objective optimization function can be represented by the following formula 4: Formula 4 In Formula 4, Representing tool sequence Overall score , , These represent the sequence validity parameter (indicating the effectiveness of the tool sequence for the task), the sequence execution cost parameter (indicating the total execution cost of the tool sequence), and the sequence reliability parameter (indicating the reliability of the tool sequence), respectively. , These represent the sequence validity weight coefficient and the sequence execution cost weight coefficient, respectively.

[0145] Formula 4 is used to comprehensively score the multiple candidate "initial tool sequences" that have been generated in order to select the final "execution tool sequence". , These are two key weight coefficients in the objective optimization function, used to achieve multi-objective optimization.

[0146] parameter This represents the sequence effectiveness weighting coefficient, used to adjust the importance of "the effectiveness of the tool sequence for the task" in the overall score. Effectiveness measures the extent to which the sequence can successfully achieve the task objective.

[0147] parameter This represents the sequence execution cost weighting coefficient, used to adjust the importance of the "total execution cost of the tool sequence" in the overall score. Cost can be a comprehensive measure of time, computing resources, energy consumption, etc.

[0148] In the decision-making process for control of unmanned swarms, trade-offs need to be made among several conflicting objectives, such as high effectiveness ( Corresponding objectives), low cost ( (corresponding target) and high reliability. By setting... and The value of can define different decision preferences. For example, a high ,Low The configuration implies "no expense spared, success at all costs"; while a low ,high The configuration represents "the lower the cost, the better, provided the task is basically completed." Therefore, these two parameters are key to realizing command intentions and making refined decision-making trade-offs.

[0149] 3033. With the goal of maximizing the objective optimization function, output the execution tool sequence based on the sequence validity, sequence execution cost, and sequence reliability of each initial tool sequence.

[0150] For example, as shown in Equation 4, based on the sequence validity, sequence execution cost, and sequence reliability of the initial tool sequence S, the function value of the objective optimization function under the initial tool sequence S is calculated, which serves as the invocation benefit value corresponding to the initial tool sequence S. Based on the invocation benefit value corresponding to each initial tool sequence, the initial tool sequence with the largest invocation benefit value is output as the execution tool sequence for the target execution task. Thus, it is possible to arrange the candidate tool set with the objective of maximizing the invocation benefit value of the candidate tool set, thereby obtaining the execution tool sequence for the target execution task.

[0151] Thus, steps 3031-3033, by orchestrating the candidate tool set, obtain initial tool sequences and construct a target optimization function containing parameters for sequence effectiveness, execution cost, and reliability. The goal is to maximize this function to output the execution tool sequence. This comprehensively quantifies the benefits of tool sequence invocation, avoiding the one-sidedness of single-dimensional evaluation. It ensures that the execution tool sequence efficiently achieves the task objective while also considering cost control and execution reliability, automatically selecting the optimal sequence. Intelligent orchestration of tool invocation can be achieved without manual intervention, further enhancing the efficiency of the connection between "decision-making" and "execution," and improving the automation level and overall effectiveness of unmanned cluster task execution.

[0152] 304. Based on the execution tool sequence, perform instruction generation processing to obtain the control instructions for the execution decision scheme.

[0153] For example, taking the scenario of "25 Team A drones generating waypoints in the left half of a heart shape and coordinating obstacle avoidance" as an example, assuming the execution tool sequence S1: T2 → T1 → T4, i.e., "laser obstacle detection tool → heart shape coordinate calculation tool → collaborative timing planning tool", the output results of the execution tool sequence are analyzed, and combined with the hardware interface specifications of UAV 01 (supporting API command calls), structured control commands are generated. The command format includes "timestamp, command type, parameter details, and execution priority", thereby obtaining the control commands for the execution decision scheme.

[0154] 305. Send the control command to the target unmanned unit to control the target unmanned unit to execute the target execution task.

[0155] For example, in the cluster command and control device, the target execution agent in the preset execution layer, based on the semantic-level agent communication protocol (adapted to the communication module of UAV 01), sends control instructions for the execution decision scheme of the target execution task to the corresponding target unmanned unit, so that the target unmanned unit executes the target task according to the execution decision scheme, thereby enabling multiple target unmanned units of each target local task to complete their corresponding target execution tasks, and thus realize the execution of each target local task; and so on, multiple target local tasks can be completed separately, and after the execution of multiple target local tasks is completed, the execution of the task to be commanded and coordinated is realized.

[0156] Please refer to Figure 4, which is a schematic diagram illustrating the framework of the tool orchestration process provided in this application. As shown in Figure 4, firstly, referring to the "Task Analysis Module" in Figure 4, step 3021 achieves task parsing, requirement identification, and constraint extraction. Then, referring to the "Tool Discovery and Screening Module" in Figure 4, steps 3022-3023 achieve semantic matching, capability assessment, and usability checks to screen a set of candidate tools. Next, referring to the "Intelligent Orchestration Optimization Module" in Figure 4, step 303 performs dynamic programming to obtain the optimal tool sequence as the execution tool sequence. Following this, referring to the "Execution Control Module" in Figure 4, step 304 uses the execution tool sequence to generate control instructions for the execution decision scheme, and step 305 issues these instructions to implement tool execution calls, achieving parallel execution, status monitoring, and exception handling to obtain execution results. Finally, referring to the "Learning Optimization Module" in Figure 4, the execution results are further used for tool effectiveness evaluation, experience storage, and strategy updates, thereby improving the performance of subsequent tool orchestration, forming a feedback loop, and improving the dynamic orchestration effect of tools.

[0157] As can be seen from the above, firstly, by determining the candidate tool set for the target task based on its current task status and execution decision scheme, tools matching the target task can be automatically selected. With the goal of maximizing the invocation benefit of the candidate tool set, the set is orchestrated to obtain the execution tool sequence for the target task. This allows for automatic planning and invocation of tools to generate control commands. Even when facing multiple target tasks simultaneously, the tool generation control commands can still be automatically adapted. This enables unmanned clusters to automatically complete tool invocation decisions when facing complex tasks (such as continuous execution of multiple actions or collaborative operation of multiple devices) without rewriting adaptation rules, thus improving the automation level of task execution to a certain extent. Secondly, by orchestrating the candidate tool set to maximize its invocation benefit to obtain the execution tool sequence, the system can automatically adapt to scenario requirements and generate execution tool sequences with higher invocation benefits, thereby improving the reliability of tool invocation decisions. Thirdly, by orchestrating the candidate tool set with the goal of maximizing the invocation benefit of the candidate tool set to obtain the execution tool sequence, a better tool execution sequence can be automatically orchestrated to achieve the connection from "decision-making" to "execution" of the target execution task, thereby improving the automation and efficiency of task execution. Fourthly, by orchestrating the first distribution tool set with the goal of maximizing the invocation benefit of the first distribution tool set to obtain the planning tool sequence for the target local task, a better tool execution sequence can be automatically orchestrated to achieve the connection from "decision-making" to "distribution and execution" of the target local task, thereby improving the automation and efficiency of the target local task execution. Fifthly, by orchestrating the second distribution tool set with the goal of maximizing the invocation benefit of the second distribution tool set to obtain the coordination tool sequence for the target execution task, a better tool execution sequence can be automatically orchestrated to achieve the connection from "decision-making" to "distribution and execution" of the target execution task, thereby improving the automation and efficiency of the target execution task execution.

[0158] Those skilled in the art will understand that all or part of the steps in the above-described unmanned swarm command and control method can be accomplished by instructions, or by controlling related hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0159] Therefore, embodiments of this application provide a computer-readable storage medium storing a plurality of computer programs that can be loaded by a processor to execute any of the unmanned swarm command and control methods provided in the embodiments of this application. For example, the computer program can be loaded by a processor to execute the steps in the unmanned swarm command and control method.

[0160] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0161] In the above embodiments of the unmanned swarm command and control method, computer-readable storage medium, and electronic device, the descriptions of each embodiment have different focuses. For parts not described in detail in a particular embodiment, please refer to the relevant descriptions of other embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes and beneficial effects of the computer-readable storage medium, electronic device, and their corresponding units described above can be referred to the description of the unmanned swarm command and control method in the above embodiments, and will not be repeated here.

[0162] The above provides a detailed description of an unmanned swarm command and control method, electronic device, and computer-readable storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for commanding and controlling unmanned swarms, characterized in that, The method includes: performing decision planning on a task to be commanded and coordinated by an unmanned swarm to obtain a target unmanned unit to be scheduled for the task to be commanded and coordinated, and an execution decision scheme for the target unmanned unit's target execution task, wherein the unmanned swarm includes multiple unmanned units; determining a set of candidate tools for the target execution task based on the current task status of the target execution task and the execution decision scheme; arranging the set of candidate tools with the objective of maximizing the invocation benefit value of the set of candidate tools to obtain a sequence of execution tools for the target execution task; performing instruction generation processing based on the sequence of execution tools to obtain control instructions for the execution decision scheme; and sending the control instructions to the target unmanned unit to control the target unmanned unit to execute the target execution task.

2. The unmanned swarm command and control method according to claim 1, characterized in that, The step of determining a set of candidate tools for the target execution task based on the current task status and the execution decision scheme includes: extracting requirements based on the current task status and the execution decision scheme to obtain execution requirement information for multiple sub-tasks; matching the execution requirement information of each sub-task with a preset tool set to obtain candidate tools for each sub-task; and obtaining the set of candidate tools based on the candidate tools for each sub-task.

3. The unmanned swarm command and control method according to claim 2, characterized in that, The process of matching the execution requirements of each subtask with a preset set of tools to obtain candidate tools for each subtask includes: matching the execution requirements of each subtask to obtain the semantic matching degree and scenario matching degree between each tool in the preset set and each subtask; determining the probability of each tool being selected for each subtask based on the semantic matching degree and scenario matching degree between each tool and each subtask, as well as the execution efficiency and historical execution data of each tool; and determining candidate tools for each subtask based on the probability of each tool being selected for each subtask.

4. The unmanned swarm command and control method according to claim 1, characterized in that, The step of orchestrating the candidate tool set to maximize the invocation benefit value of the candidate tool set and obtaining the execution tool sequence for the target execution task includes: performing orchestration processing based on the candidate tool set to obtain each initial tool sequence of the candidate tool set; constructing a target optimization function for the candidate tool set, wherein the target optimization function includes a sequence validity parameter, a sequence execution cost parameter, and a sequence reliability parameter, and the target optimization function is used to indicate the invocation benefit value; and outputting the execution tool sequence based on the sequence validity, sequence execution cost, and sequence reliability of each initial tool sequence with the goal of maximizing the target optimization function.

5. The unmanned swarm command and control method according to claim 4, characterized in that, The orchestration process based on the candidate tool set to obtain initial tool sequences of the candidate tool set includes: determining the pre-execution score of each tool in the candidate tool set based on the pre-execution state of each tool in the candidate tool set and the pre-execution state of the target execution task; determining the post-execution score of each tool in the candidate tool set based on the post-execution state of each tool in the candidate tool set and the post-execution state of the target execution task; determining the insertion tool of the current orchestration sequence of the candidate tool set based on the pre-execution score and the post-execution score of each tool in the candidate tool set; and using the current orchestration sequence as the initial tool sequence until the current orchestration sequence meets the requirements of the target execution task.

6. The unmanned swarm command and control method according to claim 1, characterized in that, The step of making decisions and planning for the unmanned swarm's tasks to be commanded and coordinated, to obtain the target unmanned units to be scheduled for the tasks to be commanded and coordinated, and the execution decision schemes for the target execution tasks of the target unmanned units, includes: acquiring the tasks to be commanded and coordinated in the unmanned swarm; performing overall planning on the tasks to be commanded and coordinated, to obtain an overall decision scheme for the tasks to be commanded and coordinated, wherein the overall decision scheme includes multiple target local tasks to be coordinated by the tasks to be commanded and coordinated; performing coordination planning on the target local tasks, to obtain a coordination decision scheme for the target local tasks, wherein the coordination decision scheme is used to indicate the target unmanned units to be scheduled for the target local tasks in the unmanned swarm and the target execution tasks of the target unmanned units; and performing execution planning on the target execution tasks, to obtain the execution decision schemes for the target execution tasks.

7. The unmanned swarm command and control method according to claim 6, characterized in that, The unmanned swarm command and control method is applied to a swarm command and control device, which includes a preset coordination system. The method further includes: determining a first set of candidate delivery tools for the target local task based on the current task status of the target local task and the overall decision-making scheme; arranging the first set of delivery tools with the goal of maximizing the call benefit value of the first set of delivery tools to obtain a planning tool sequence for the target local task; performing instruction generation processing based on the planning tool sequence to obtain the delivery instruction for the overall decision-making scheme; and sending the delivery instruction for the overall decision-making scheme to the preset coordination system to control the preset coordination system to initiate the decision-making planning of the target local task.

8. The unmanned swarm command and control method according to claim 6, characterized in that, The unmanned swarm command and control method is applied to a swarm command and control device, which includes a preset execution system. The method further includes: determining a second set of candidate dispatch tools for the target execution task based on the current task status of the target execution task and the coordination decision scheme; arranging the second set of dispatch tools with the goal of maximizing the call benefit value of the second set of dispatch tools to obtain a coordination tool sequence for the target execution task; performing instruction generation processing based on the coordination tool sequence to obtain the dispatch instruction of the coordination decision scheme; and sending the dispatch instruction of the coordination decision scheme to the preset execution system to control the preset execution system to initiate the decision planning of the target execution task.

9. An electronic device, characterized in that, It includes a processor, a memory, and a cluster command and control device, wherein the memory stores a computer program, and the processor executes the unmanned cluster command and control method as described in any one of claims 1 to 8 when it invokes the computer program in the memory.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to execute the unmanned swarm command and control method according to any one of claims 1 to 8.

Images