Unmanned system task planning method, device and equipment based on large model

Abstract

The invention discloses an unmanned system task planning method, device and equipment based on a large model, and relates to the field of task planning, and the method comprises the steps: carrying out the semantic analysis of a task demand of an unmanned system, and obtaining a task semantic set; mapping the task semantic set and a predetermined task planning constraint into an initial planning model of a planning domain definition language by adopting a large model; performing constraint enhancement on the initial planning model based on a graph neural network to obtain an enhanced planning model; solving the enhanced planning model with maximization of the execution revenue of the unmanned system as a target to obtain a task planning result of the unmanned system; the task planning result comprises an action sequence of each unmanned device in the unmanned system. According to the invention, the construction efficiency and precision of the task planning model are improved, and then the coordination, precision and overall efficiency of task execution of the unmanned system are improved.

Description

Technical Field

[0001] This application relates to the field of mission planning, and in particular to a method, apparatus and equipment for mission planning of unmanned systems based on a large model. Background Technology

[0002] With the widespread adoption of large-scale unmanned systems such as large satellite constellations (e.g., Starlink, BeiDou, Qianfan) and multi-intelligent robot systems (unmanned mobile robots, collaborative robot systems), real-time intelligent task planning for these systems is crucial for efficiently fulfilling user-defined tasks. As the complexity of task environments increases, requiring multiple agents to cooperate efficiently simultaneously, the task planning problem for these complex large-scale unmanned systems becomes increasingly difficult to solve quickly. While traditional Planning Domain Definition Language (PDDL) techniques can provide a logically robust and standardized method for task planning applicable to different domains, they also require extensive expert knowledge in specific vertical domains and time-consuming pre-definition of actions and objectives for tasks in different domains. This makes it extremely difficult to model complex large-scale unmanned systems for rapid task planning under the guidance of natural language.

[0003] With the rapid development of Large Language Model (LLM) technology, new ideas have been provided for task planning algorithms. Large model technology possesses powerful semantic understanding and generation capabilities, and can serve as a bridge to easily convert user input requirements into a generalized PDDL (Programmable Detailed Language Description) description for task planning models. The GPT-4 large model can be used to generate a generalized task planner. However, large models face the challenge of standardizing predicate actions in autonomous recognition tasks, and the consistency of PDDL planning models obtained from multiple experiments is poor. Furthermore, the validity of generated predicates cannot be guaranteed in real-time correction of the generated PDDL. Finally, for many complex unmanned systems, the actions represented by the decomposed predicates have strong logical sequence constraints. For example, in a robot system performing an assembly task, the action of "fixing" the assembly target needs to be completed before the action of "tightening" the assembly link. The problem of action sequence constraints in generating PDDL planning models using LLM still needs further development. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, and device for unmanned system mission planning based on a large model, which can improve the efficiency and accuracy of unmanned system mission planning.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a task planning method for unmanned systems based on a large model, including: Semantic analysis is performed on the task requirements of unmanned systems to obtain a task semantic set; A large model is used to map the task semantic set and the pre-determined task planning constraints into an initial planning model defined by the planning domain definition language; The initial planning model is constrained and enhanced using a graph neural network to obtain an enhanced planning model. With the goal of maximizing the execution benefits of the unmanned system, the enhanced planning model is solved to obtain the task planning results of the unmanned system; the task planning results include the action sequence of each unmanned device in the unmanned system.

[0006] Secondly, this application provides a large-scale model-based unmanned system mission planning device, comprising: The semantic parsing module is used to perform semantic parsing on the task requirements of unmanned systems to obtain a task semantic set; The planning mapping module is used to map the task semantic set and the pre-determined task planning constraints into an initial planning model defined by the planning domain definition language using a large model. The constraint enhancement module is used to enhance the constraints of the initial planning model based on a graph neural network to obtain an enhanced planning model. The model solving module is used to solve the augmented planning model with the goal of maximizing the execution benefits of the unmanned system, and obtain the task planning results of the unmanned system; the task planning results include the action sequence of each unmanned device in the unmanned system.

[0007] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described large-model-based unmanned system task planning method.

[0008] According to the specific embodiments provided in this application, this application has the following technical effects: By performing semantic parsing on task requirements, core task information can be accurately extracted to form a task semantic set, ensuring an accurate understanding of complex task requirements; by using a large model, the task semantic set and preset task planning constraints are efficiently mapped to an initial planning model defined by the planning domain definition language, realizing the rapid transformation of task requirements into a standardized planning model and improving the adaptability and efficiency of model construction; by enhancing the constraints of the initial planning model through graph neural networks, the correlation and effectiveness of various constraints can be deeply explored and strengthened, significantly improving the rationality and reliability of the planning model; by solving the task planning result containing the action sequences of each unmanned device with the goal of maximizing the execution benefits of the unmanned system, the overall benefit of task execution is optimized, and a clear and explicit execution basis is provided for each unmanned device, effectively improving the coordination, accuracy and overall efficiency of unmanned system task execution, and applicable to the task planning needs of unmanned systems in various complex scenarios. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0010] Figure 1 This is an application environment diagram of a large-model-based unmanned system task planning method in one embodiment of this application.

[0011] Figure 2 This is a flowchart illustrating a task planning method for unmanned systems based on a large model, provided as an embodiment of this application.

[0012] Figure 3 This is a schematic diagram of the functional modules of an unmanned system mission planning device based on a large model, provided as an embodiment of this application.

[0013] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0014] 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.

[0015] This application proposes a large-model-based task planning method for unmanned systems, such as satellite constellations, multi-agent robot systems, and land-based and sea-based unmanned equipment systems. The method uses large-model technology to organize the received natural language user requirements into task objectives, ultimately outputting a mathematical model of the task objectives in the planning domain definition language. The mathematical model of the task objectives includes the definition and symbols of the target objects and a set of constraints corresponding to the task planning mathematical model, expressed as logical sequences.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] The large-model-based unmanned system task planning method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up independently, integrated into server 102, or placed in the cloud or on another server. Terminal 101 can send the task requirements and task planning constraints of the unmanned system to server 102. Based on the task requirements and task planning constraints, server 102 uses a large model and graph neural network to generate the task planning results for the unmanned system, and server 102 can feed back the task planning results to terminal 101.

[0018] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0019] In one exemplary embodiment, such as Figure 2 As shown, a task planning method for unmanned systems based on a large model is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps 201 to 206.

[0020] Step 201: Semantic parsing is performed on the task requirements of the unmanned system to obtain a task semantic set.

[0021] Specifically, the task requirements proposed by users are transformed into a computable and planarable task semantic structure, providing a unified semantic foundation for subsequent formal modeling.

[0022] In one specific application example, the mission requirement is an Earth observation satellite constellation. The unmanned system comprises multiple satellites.

[0023] The task requirements include natural language text. and / or images The natural language text consists of a user-inputted Chinese text task description. For example: "Schedule several remote sensing satellites to achieve high-frequency revisit imaging of region A within 24 hours." The images include a target area map, coverage diagram, and orbital diagram.

[0024] This application employs a multimodal large model to perform semantic parsing of the task requirements of unmanned systems, obtaining a task semantic set, the formula of which is: .in, This is a task semantic set used to describe "the observation target that needs to be maximized," including the target region, target point set, and expected number of observations. This is the semantic inference function for a multimodal large language model.

[0025] The task semantic set includes an action set, a predicate set, and a task target set.

[0026] Furthermore, this application also provides an example of maximizing satellite observation: the user input is "to image 20 key targets within region A as many times as possible within 24 hours," and after processing in step 201, the output is... .in, For ground targets n At least the expected number of observations, Representing ground targets n At least expected to be observed The subsequent planning objective is to maximize the actual number of observations completed under constraints. The output task semantic set includes actions such as imaging, attitude maneuvering, data storage, and downlink; predicates include satellite availability, target observability, and sufficient storage; and task objectives include multi-time-segment imaging requirements for region A and constraints on revisit frequency or time intervals.

[0027] Step 202: Using a large model, the task semantic set and pre-determined task planning constraints are mapped to an initial planning model in a planning domain definition language. The task planning constraints include maximum revisit interval, energy limit, etc.

[0028] Specifically, the task semantic set output in step 201 is automatically transformed into a planning domain and planning problem model that conforms to the PDDL specification, and time and resource constraints in the task are explicitly introduced.

[0029] In a specific application example, the initial planning model includes a planning domain and a planning problem. The planning domain includes action definitions, predicate definitions, and basic constraints. The planning problem includes the unmanned equipment, the initial state, and the task objective state.

[0030] The core mapping formula of the large model is ;in, For the set of actions in the PDDL planning domain, For large model-driven semantic-to-PDDL mapping functions, For task semantic set, This refers to a collection of satellites and their capability parameters (orbit, payload, energy).

[0031] For example: regarding a certain satellite Mapping PDDL actions: imaging And generate corresponding predicates, including: target observable, energy sufficient, storage available, etc.; among which, For the first m One satellite, For the first n One ground target, t This provides a time frame, and thus a formal condition for whether the imaging action can be performed at that time.

[0032] Step 203: The initial planning model is constrained and enhanced based on a graph neural network to obtain an enhanced planning model.

[0033] In a specific application example, to address the complex time-resource-action coupling constraints in Earth observation satellite constellation mission planning, a constraint relationship model based on a Graph Neural Network (GNN) is constructed. This model automatically infers and completes the implicit constraints and injects them into the initial planning model of PDDL. The data used to construct the constraint relationship model includes satellite mission attribute data: number of satellites, payload capacity, observable time window, and energy and storage resource model.

[0034] Specifically, step 203 includes steps 31 to 34.

[0035] Step 31: Construct a constraint relationship diagram based on the initial planning model. .in, For the constraint relationship diagram, For a set of nodes, Let it be the set of edges.

[0036] The nodes in the constraint graph represent actions. The characteristics of a node include the action type, the unmanned device executing the action, the action's executable time window, and the resource vector occupied by the action. The expression for a node is: .in, v i For the first i 1 node a i For action type (such as imaging, attitude maneuvering). s i For the satellite to perform this action, For the time window in which the action can be executed, r i This is the resource vector (energy, storage, payload, etc.) used by the action.

[0037] The edges in the constraint graph represent the execution relationship between two actions. .in, For the first i The node and the first j Edges between nodes v j For the first j 1 node Representative definition i The node and the first j Each node forms a constraint relationship edge.

[0038] Step 32: Use a graph neural network to update the features of each node in the constraint relationship graph, and use a multilayer perceptron classifier to determine the constraint relationships between nodes to obtain a constraint set.

[0039] The formula for updating node features is: ;in, For the first k The first in the layer i The feature vector of each node In order to be with the first i A set of neighboring nodes with which a node has a constraint relationship. W c The constraint weight matrix, W s Let σ be the node's own weight matrix, and σ() be the non-linear activation function.

[0040] Through formula Output the constraint relationships between nodes. To predict the first i The node and the first j Does there exist a constraint between the nodes? MLP() is a multilayer perceptron classifier. This is a feature concatenation operation. The final output constraint. .

[0041] Step 33: Transform the constraint set into action mutual exclusion constraints, time window constraints, resource contention constraints, and action pre- and post-dependency constraints. Specifically, each constraint is transformed into a mutual exclusion predicate, time window constraint, and resource contention constraint in PDDL.

[0042] Specifically, the mutual exclusion constraint of actions is as follows: .in, For satellite s The set of executable observation actions, a For a specific imaging action, This refers to the allowed observation time window for this imaging action. For indicator functions, it means t Whether the time falls within the time window of the action is 1 if yes and 0 if no; the mutual exclusion constraint of the action means that the same satellite can only perform one imaging action at any time.

[0043] For example, if a satellite can simultaneously observe two ground targets between 10:00 and 10:30, the GNN will identify a temporal conflict between these two imaging actions and generate a mutual exclusion constraint to prevent the planner from incorrectly scheduling two imaging operations at the same time. Action mutual exclusion constraints are a prerequisite for maximizing the number of observations.

[0044] Time window constraint is ;in, For satellite s For the target o The set of observable time windows.

[0045] Step 34: Based on the action mutual exclusion constraint, the time window constraint, the resource competition constraint, the action pre- and post-dependency constraint, and the initial planning model, the enhanced planning model is obtained.

[0046] Step 204: With the objective of maximizing the execution benefits of the unmanned system, solve the augmented planning model to obtain the task planning results for the unmanned system. The task planning results include the action sequence of each unmanned device in the unmanned system.

[0047] In the mission of the Earth observation satellite constellation, the objective function solved by the model is: .in, The mission planning results include the action sequences of each satellite. J The objective function value, The total number of ground targets. For ground targets n The number of times the planned observation attempts, This indicates the satellite's imagery of the target; if the satellite... Successfully completed the target If the image is captured in a single image, then the value is 1; otherwise, the value is 0.

[0048] Specifically, under the premise of satisfying constraints such as time, energy, and mutual exclusion, the planner automatically selects which satellite to image which target in which time window in order to maximize the total number of observations.

[0049] Step 205: Perform simulation verification on the task planning results to determine the execution result loss of the task planning results.

[0050] Specifically, by combining the simulation environment parameters of Earth observation (orbit parameters, ground target distribution, imaging requirement time span), the feasibility and performance level of the generated mission planning results in actual Earth observation mission planning are verified through the simulation system, and a set of evaluation indicators is output, including the number of completed observations, average revisit time, resource utilization rate, and whether there are planning failures or conflicts.

[0051] In a specific application example, a large model is used to perform semantic analysis and diagnosis on the task planning results, forming feedback to the aforementioned steps 201 to 204, and constructing a closed-loop modeling mechanism for continuous optimization.

[0052] The formula for the execution result loss is: .in, For ground targets n The actual number of observations completed according to the plan. The function is positive, meaning that a penalty is incurred only if the objective is not met. The execution result loss represents the loss due to insufficient observation of the planning results.

[0053] Step 206: If the execution result loss does not meet the set threshold requirement, adjust the mapping rules of the large model and the constraint enhancement process of the graph neural network according to the execution result loss until the execution result loss meets the set threshold requirement, and obtain the final task planning result.

[0054] For example, if certain key targets have not been sufficiently observed for a long time, the loss function drives the updating of GNN constraint weights and adjusts the PDDL mapping strategy of LLM to ensure that these targets are fully observed in the next round of planning.

[0055] This application addresses the mission planning problem of Earth observation satellite constellations by constructing a five-stage mission planning method: semantic analysis, formal modeling, constraint reasoning, simulation verification, and feedback optimization. Each step is interconnected and progressively builds upon the previous one, forming a closed-loop feedback mechanism that enables the mission planning model to continuously evolve and optimize, thereby significantly improving the automation and engineering applicability of complex observation mission planning.

[0056] In summary, the beneficial effects of this application include at least the following: (1) Improved efficiency of mission planning. This application introduces the key step of predicate definition in the establishment of PDDL model with the help of large model technology, and performs intelligent analysis and generation, which improves the efficiency of predicate generation for mission planning of complex satellite systems, and can generate interpretable text, which helps to quickly learn and use the technology.

[0057] (2) The logical constraints in task planning modeling are supplemented and the versatility is improved. This application improves the generated task planning model by introducing graph neural network embedding large model and PDDL technology, which includes action logic constraints, thereby improving the versatility of the technology in engineering practice.

[0058] Based on the same inventive concept, this application also provides an apparatus for implementing the method described above. The solution provided by this apparatus is similar to the solution described in the above method; therefore, specific limitations in one or more apparatus embodiments provided below can be found in the limitations of the method described above, and will not be repeated here.

[0059] In one exemplary embodiment, such as Figure 3 As shown, a large-scale model-based unmanned system mission planning device is provided, which includes the following functional modules.

[0060] The semantic parsing module 301 is used to perform semantic parsing on the task requirements of the unmanned system to obtain a task semantic set.

[0061] The planning mapping module 302 is used to map the task semantic set and the pre-determined task planning constraints into an initial planning model in the planning domain definition language using a large model.

[0062] The constraint enhancement module 303 is used to enhance the constraints of the initial planning model based on a graph neural network to obtain an enhanced planning model.

[0063] The model solving module 304 is used to solve the augmented planning model with the objective of maximizing the execution benefits of the unmanned system, thereby obtaining the task planning results of the unmanned system. The task planning results include the action sequence of each unmanned device in the unmanned system.

[0064] The simulation verification module 305 is used to perform simulation verification on the task planning results and determine the execution result loss of the task planning results.

[0065] The result optimization module 306 is used to adjust the mapping rules of the large model and the constraint enhancement process of the graph neural network according to the execution result loss when the execution result loss does not meet the set threshold requirement, until the execution result loss meets the set threshold requirement, and then obtain the final task planning result.

[0066] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores the task requirements and task planning constraints of the unmanned system. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a task planning method for an unmanned system based on a large model.

[0067] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0068] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0069] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0070] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0071] In this application, all actions to acquire signals, information, or data are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with the authorization granted by the owner of the relevant device.

[0072] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0073] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0074] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0075] This document uses specific examples 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 methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A task planning method for unmanned systems based on a large model, characterized in that, The method includes: Semantic analysis is performed on the task requirements of unmanned systems to obtain a task semantic set; A large model is used to map the task semantic set and the pre-determined task planning constraints into an initial planning model defined by the planning domain definition language; The initial planning model is constrained and enhanced using a graph neural network to obtain an enhanced planning model. With the goal of maximizing the execution benefits of the unmanned system, the enhanced planning model is solved to obtain the task planning results of the unmanned system; the task planning results include the action sequence of each unmanned device in the unmanned system.

2. The unmanned system task planning method based on a large model according to claim 1, characterized in that, The task requirements include natural language text and / or images.

3. The unmanned system task planning method based on a large model according to claim 1, characterized in that, Semantic parsing of the task requirements of unmanned systems yields a task semantic set, including: A multimodal large model is used to perform semantic parsing on the task requirements of unmanned systems, resulting in a task semantic set.

4. The unmanned system task planning method based on a large model according to claim 1, characterized in that, The task semantic set includes an action set, a predicate set, and a task target set.

5. The unmanned system task planning method based on a large model according to claim 1, characterized in that, The initial planning model includes a planning domain and a planning problem; the planning domain includes action definitions, predicate definitions, and basic constraints; the planning problem includes unmanned equipment, initial state, and task objective state.

6. The unmanned system task planning method based on a large model according to claim 1, characterized in that, The initial planning model is constrained and enhanced using a graph neural network to obtain an enhanced planning model, including: A constraint relationship graph is constructed based on the initial planning model; the nodes in the constraint relationship graph represent actions, and the characteristics of the nodes include action type, unmanned equipment executing the action, action execution time window, and resource vector occupied by the action; the edges in the constraint relationship graph represent the execution relationship between two actions; A graph neural network is used to update the features of each node in the constraint relationship graph, and a multilayer perceptron classifier is used to determine the constraint relationships between the nodes to obtain a constraint set. The constraint set is transformed into action mutual exclusion constraints, time window constraints, resource competition constraints, and action pre- and post-dependency constraints. Based on the mutual exclusion constraint of actions, the time window constraint, the resource competition constraint, the pre- and post-action dependency constraint, and the initial planning model, an enhanced planning model is obtained.

7. The unmanned system task planning method based on a large model according to claim 1, characterized in that, The method further includes: The task planning results are simulated and verified to determine the execution result loss of the task planning results; If the execution result loss does not meet the set threshold requirement, the mapping rules of the large model and the constraint enhancement process of the graph neural network are adjusted according to the execution result loss until the execution result loss meets the set threshold requirement, and the final task planning result is obtained.

8. The unmanned system task planning method based on a large model according to claim 1, characterized in that, The mission requirement is a constellation of Earth observation satellites; the unmanned system includes multiple satellites.

9. A mission planning device for unmanned systems based on a large model, characterized in that, The device executes the unmanned system mission planning method based on a large model as described in any one of claims 1-8, and the device comprises: The semantic parsing module is used to perform semantic parsing on the task requirements of unmanned systems to obtain a task semantic set; The planning mapping module is used to map the task semantic set and the pre-determined task planning constraints into an initial planning model defined by the planning domain definition language using a large model. The constraint enhancement module is used to enhance the constraints of the initial planning model based on a graph neural network to obtain an enhanced planning model. The model solving module is used to solve the augmented planning model with the goal of maximizing the execution benefits of the unmanned system, and obtain the task planning results of the unmanned system; the task planning results include the action sequence of each unmanned device in the unmanned system.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the unmanned system mission planning method based on a large model as described in any one of claims 1-8.

Images