CN117744497A - Robot mission planning model construction system, method, equipment and storage medium - Google Patents

Abstract

The invention discloses a robot task planning model construction system, a method, equipment and a storage medium. The system comprises: the model module is used for acquiring an initial sentence, extracting the initial sentence by utilizing the large-scale pre-training language model to generate a task sequence, wherein the task sequence comprises a high-level interface and a low-level interface; the simulator module is used for constructing a simulation environment from top to bottom according to the high-level interface and the low-level interface, abstracting the real operation of the robot in the real environment into the simulation operation in the simulation environment, and verifying whether the simulation operation meets the expected target; the data generation module is used for determining a task target in the simulation environment, searching a task path according to the task target, generating an initial text according to the task path, and performing text enhancement on the initial text to obtain an enhanced text. The method realizes the use of a simple data structure in a high-level interface through a top-down simulator design and data generation method, reduces the construction cost of a new scene, and reduces the calculation force requirement in the verification and generation process.

Description

Robot mission planning model construction system, method, equipment and storage medium

Technical field

The invention relates to the field of robot control technology, and in particular to a robot task planning model construction system, method, equipment and storage medium.

Background technique

Robot task planning is one of the pain points in the current application of the robot industry. It is mainly reflected in the fact that robots often need to operate in a variety of scenarios. Therefore, the capabilities of most robots are basic atomic capabilities, such as "navigation", "grasping", "Broadcast", etc., if it needs to be combined into a complete function through business arrangement in actual scenarios, such as the function of getting water, it needs to be arranged as: "Navigation → Recognition → Move → Grab → Return → Broadcast → Submit" process, the parameters in the process also need to be reset, and different scenarios require different functions, and the same function requires different instruction sequences. This results in the need to redesign and implement the functions in each new scenario. This increases the difficulty and cost of implementing robots in various scenarios. The purpose of the mission planning model is to break down fuzzy functions into capability sequences that can be directly executed.

The input of the above-mentioned task planning model needs to include the user's statements and information used to assist the model in making judgments. The main form is {"user": "Get me a bottle of water", "item_in_view": ["Cola", " Dining table"], "current_place": "Restaurant"}, and the output is the currently planned operation path, which may not be the complete task process, such as [{"action": "Navigation", "action_object": "Kitchen"}, {"action": "Object recognition"}] This represents the first two steps of "navigation" and "recognition" of the above-mentioned water-receiving function. Feedback from the external system is required to determine what objects have been recognized in order to continue planning subsequent actions. One problem with the data in the above method is that it can only be collected within a specific scene, and the cost of collection is often high, the collection efficiency is slow, and data from different scenes cannot be reused, which often results in work only in closed scenes. And cannot quickly adapt to new open scenarios.

Existing technologies often directly use the simulation environment of the physics engine, adopting a bottom-up simulation environment construction method. For advanced interfaces, data generation is slow, judgment conditions are difficult, and verification requires large computing power. The underlying physics must be implemented. It is difficult to directly adapt to the needs of various levels of functions (for example, the higher-level interface is "navigate to the kitchen", and the lower-level interface is "search for the kitchen on the map → get the current location → use the path planning algorithm → call the movement command move along the planned path") and other issues.

Contents of the invention

In view of this, the purpose of the present invention is to overcome the deficiencies in the existing technology and provide a robot mission planning model construction system, method, equipment and storage medium.

The present invention provides the following technical solutions:

In a first aspect, embodiments of the present disclosure provide a robot mission planning model construction system, which includes:

The model module is used to obtain the initial statement input by the user, extract the initial statement using a large-scale pre-trained language model, and generate a task sequence, where the task sequence includes a high-level interface and a low-level interface;

The simulator module is used to construct a simulation environment in the form of graph-structured data from top to bottom according to the high-level interface and the low-level interface, and abstract the real operations of the robot in the real environment into simulated operations in the simulation environment. , and verify whether the simulation operation meets the expected goals;

A data generation module is used to determine a task goal in the simulation environment, search for a corresponding task path according to the task goal, generate initial text according to the task path, and perform text enhancement on the initial text to obtain enhanced text.

Further, the model module includes:

The model main unit is used to obtain batch data, pre-train the batch data, generate the large-scale pre-trained language model, and determine the interface type;

A pre- and post-processing unit is used to obtain the initial statement, and use the large-scale pre-trained language model to perform regularized extraction on the initial statement to generate a planned task sequence.

Further, the simulator module includes:

A simulation environment unit, configured to construct the simulation environment in the form of graph structure data from top to bottom according to the high-level interface and the low-level interface, wherein the nodes of the data graph in the graph structure data are in the simulation environment specific objects, the edges of the data graph in the graph structure data are the relationships between the specific objects in the simulation environment, and the information amount of the high-level interface in the simulation environment is less than the information amount of the low-level interface;

A request execution unit configured to abstract the robot's own state and the robot's real operations in the real environment into its own state and simulated operations in the simulated environment;

A verification unit configured to verify whether the simulation operation meets the expected goal by retrieving the attribute information and side information of the specific object in the simulation environment.

Furthermore, the data generation module includes:

A data sampling unit, configured to sample and obtain the task target by searching for attribute information of the specific object in the simulation environment;

A path search unit, configured to obtain the task path by searching the data graph according to the task target;

A data enhancement unit is configured to generate the initial text using a preset template according to the task path, and perform text enhancement on the initial text through a preset text enhancement tool to obtain the enhanced text.

In the second aspect, embodiments of the present disclosure provide a method for constructing a robot mission planning model, which is applied to the robot mission planning model construction system described in the first aspect. The system includes a model module, a simulator module and a data generation module, The methods include:

Obtain the initial statement input by the user through the model module, extract the initial statement using a large-scale pre-trained language model, and generate a task sequence, where the task sequence includes a high-level interface and a low-level interface;

The simulator module constructs a simulation environment in the form of graph structure data from top to bottom according to the high-level interface and the low-level interface, and abstracts the real operations of the robot in the real environment into simulated operations in the simulation environment. , and verify whether the simulation operation meets the expected goals;

The data generation module determines a task goal in the simulation environment, searches for a corresponding task path according to the task goal, generates initial text according to the task path, and performs text enhancement on the initial text to obtain enhanced text.

Further, the initial statement input by the user is obtained through the model module, the initial statement is extracted using a large-scale pre-trained language model, and the task sequence is generated, including:

Obtain batch data, perform pre-training on the batch data, generate the large-scale pre-trained language model, and determine the interface type;

The initial statement is obtained, and the large-scale pre-trained language model is used to perform regularized extraction on the initial statement to generate a planned task sequence.

Further, the top-down construction of a simulation environment in the form of graph-structured data based on the high-level interface and the low-level interface abstracts the real operations of the robot in the real environment into simulated operations in the simulation environment, And verify whether the simulation operation meets the expected goals, including:

The simulation environment is constructed in the form of graph structure data from top to bottom according to the high-level interface and the low-level interface, wherein the nodes of the data graph in the graph structure data are specific objects in the simulation environment, and the The edges of the data graph in the graph structure data are the relationships between the specific objects in the simulation environment, and the information amount of the high-level interface in the simulation environment is less than the information amount of the low-level interface;

Abstract the robot's own state and the robot's real operations in the real environment into its own state and simulated operations in the simulated environment;

By retrieving the attribute information and side information of the specific object in the simulation environment, it is verified whether the simulation operation meets the expected goal.

Further, determining a task goal in the simulation environment, searching for a corresponding task path according to the task goal, generating an initial text according to the task path, and performing text enhancement on the initial text to obtain the enhanced text, including :

By searching the attribute information of the specific object in the simulation environment, the task target is sampled;

Obtain the task path by searching the data graph according to the task goal;

The initial text is generated using a preset template according to the task path, and the initial text is text enhanced using a preset text enhancement tool to obtain the enhanced text.

In a third aspect, an embodiment of the present disclosure provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the computer program in the second aspect is implemented. The steps of the robot mission planning model construction method described above.

In a fourth aspect, embodiments of the present disclosure provide a computer-readable storage medium that stores a computer program. When the computer program is executed by a processor, the robot task described in the second aspect is implemented. Steps in planning a model-building approach.

The embodiments of the present application have the following advantages:

The robot task planning model construction system provided by the embodiment of the present application includes: a model module, used to obtain the initial statement input by the user, extract the initial statement using a large-scale pre-trained language model, and generate a task sequence, wherein: The above task sequence includes a high-level interface and a low-level interface; a simulator module is used to construct a simulation environment in the form of graph-structured data from top to bottom according to the high-level interface and the low-level interface, and abstract the actual operation of the robot in the real environment. To simulate operations in the simulation environment and verify whether the simulation operations meet expected goals; a data generation module for determining task goals in the simulation environment, searching for corresponding task paths according to the task goals, and The task path generates initial text, and text enhancement is performed on the initial text to obtain enhanced text. This application uses top-down simulator design and corresponding data generation methods to implement high-level scenarios, using simple data structures to reduce the construction costs required for new scenarios, reduce simulation, and reduce the need for verification and generation. Computing power requirements during the process.

In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, preferred embodiments are given below and described in detail with reference to the accompanying drawings.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and therefore do not It should be regarded as a limitation of the scope. For those of ordinary skill in the art, other relevant drawings can also be obtained based on these drawings without exerting creative efforts. In the various drawings, similar components are numbered similarly.

Figure 1 shows a flow chart of a robot mission planning model construction system provided by an embodiment of the present application;

Figure 2 shows a schematic diagram of a simulation environment provided by an embodiment of the present application;

Figure 3 shows a schematic structural diagram of a method for constructing a robot mission planning model provided by an embodiment of the present application.

Description of main component symbols:

100-Robot mission planning model construction system; 110-Model module; 111-Model main unit; 112-Pre- and post-processing unit; 120-Simulator module; 121-Simulation environment unit; 122-Request execution unit; 123-Verification unit; 130 -Data generation module; 131-data sampling unit; 132-path search unit; 133-data enhancement unit.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and cannot be understood as limiting the present invention.

It should be noted that when an element is referred to as being "fixed" to another element, it can be directly on the other element or intervening elements may also be present. When an element is said to be "connected" to another element, it can be directly connected to the other element or there may also be intervening elements present. In contrast, when an element is referred to as being "directly on" another element, there are no intervening elements present. The terms "vertical," "horizontal," "left," "right" and similar expressions are used herein for illustrative purposes only.

In the present invention, unless otherwise clearly stated and limited, the terms "installation", "connection", "connection", "fixing" and other terms should be understood in a broad sense. For example, it can be a fixed connection or a detachable connection. , or integrated; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be an internal connection between two elements or an interaction between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

In addition, the terms “first” and “second” are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of indicated technical features. Therefore, features defined as "first" and "second" may explicitly or implicitly include one or more of these features. In the description of the present invention, "plurality" means two or more than two, unless otherwise explicitly and specifically limited.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs. The terminology used herein in the description of the template is for the purpose of describing specific embodiments only and is not intended to limit the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

Example 1

The robot mission planning model system proposed in this application mainly solves two problems. One is the problem of information migration, and the other is that interfaces at different levels of the scene require different levels of functions. If you use the physics engine to build it directly, you first need to Implementing the underlying physical implementation increases the complexity of the simulation work and makes it difficult to migrate scenarios in different scenarios. Thirdly, for most simulators, it is difficult to generate inputs and labels through sampling, which reduces data generation. s efficiency. As shown in Figure 1, it is a schematic structural diagram of a robot mission planning model construction system 100 in an embodiment of the present application. The system includes:

The model module 110 is used to obtain the initial statement input by the user, extract the initial statement using a large-scale pre-trained language model, and generate a task sequence, where the task sequence includes a high-level interface and a low-level interface.

Specifically, the model module 110 also includes: a model main unit 111, used to obtain batch data, pre-train the batch data, generate a large-scale pre-trained language model, and determine the interface type; a pre- and post-processing unit 112, used to obtain the initial Statements, and use a large-scale pre-trained language model to regularize the initial statements to generate a planned task sequence.

It can be understood that the main goal of the robot task planning model construction system 100 of the present application is to train a set of models that decompose the fuzzy instructions of the robot during the interaction with humans into a model that the robot can directly execute the instructions, and realize its task decomposition. solution function. In the present invention, the system includes several parts: model module 110, simulator module 120 and data generation module 130.

Among them, the model module 110 is used to implement model training and inference, including a model main unit 111 and a pre- and post-processing unit 112. The model main unit 111 in the model module 110 mainly obtains batch data and pre-trains the batch data to obtain large-scale data. Pre-training language model, the input and output of this model are text, and because it is pre-trained on large-scale batch data, using the knowledge learned in model pre-training, the knowledge of a certain scene can be transferred to functions in other scenarios, thereby reducing the training costs required for new scenarios.

In addition, the main job of the pre- and post-processing unit 112 in the model module 110 is to construct the input and output into a form corresponding to the output. The input of the pre- and post-processing unit 112 is the initial sentence input by the user or other external input, and the output of the pre- and post-processing unit 112 is The planned task sequence has been extracted through regularization of a large-scale pre-trained language model. The elements in the task sequence (ie, high-level interfaces and low-level interfaces) are determined at the beginning of pre-training of the model main unit 111.

It should be noted that in this embodiment, regularization extraction mainly includes prompt engineering and some information extraction methods. The specific extraction method can be determined according to the actual situation, and this embodiment does not limit this.

For the model part, the present invention chooses a large-scale pre-trained language model as a solution. This solution can use the knowledge learned by the language model on other texts to allow the model to understand what the specific content is currently doing, and increases model generalization. The ability to reach new scenarios.

The simulator module 120 is used to construct a simulation environment in the form of graph-structured data from top to bottom based on high-level interfaces and low-level interfaces, abstract the real operations of the robot in the real environment into simulated operations in the simulated environment, and verify the simulated operations. Whether it meets the expected goals.

Specifically, the simulator module 120 includes: a simulation environment unit 121, which is used to construct a simulation environment in the form of graph structure data from top to bottom according to high-level interfaces and low-level interfaces, wherein the nodes of the data graph in the graph structure data are specific objects in the simulation environment, the edges of the data graph in the graph structure data are the relationships between specific objects in the simulation environment, and the amount of information of the high-level interfaces in the simulation environment is less than that of the low-level interfaces; a request execution unit 122, which is used to abstract the robot's own state and the robot's real operation in the real environment into its own state and simulated operation in the simulation environment; a verification unit 123, which is used to verify whether the simulation operation meets the expected goal by retrieving the attribute information and edge information of the specific objects in the simulation environment.

It can be understood that the main function of the simulator module 120 is to provide scenario basis for data sampling in the data generation module 130 and to provide verification for the generated task plan, which is mainly implemented by tree-structured scenario data and execution methods.

Specifically, the simulation environment unit 121 constructs a simulation environment in the form of graph-structured data from top to bottom according to the high-level interface and the low-level interface. In the present invention, the simulation environment is constructed in the form of graph-structured data. In this graph-structured data The nodes of the graph are specific objects in the simulation environment, and the edges of the graph are the relationships between specific objects. The robot is abstracted as a collection of states. Each operation of the robot is considered to be a modification of the robot state or a modification of the graph structure data. With additions, deletions, modifications and searches, as the interface gradually goes deeper into the bottom layer, the node information in the graph structure gradually becomes more detailed. As shown in Figure 2, the simulation environment in the simulator module 120 is designed as a multi-level graph data structure.

In Figure 2, the left side is the high-level interface, and the right side is the low-level interface. There is a scene root node under all scenes. Except for the root node, the scene is divided into two layers, namely the location layer (the scene root node in the picture) and the item layer (the kitchen in the picture). ), the location layer elements will store information about specific objects in the room (such as bedrooms, bathrooms, kitchens, etc.) and a separate item layer (such as stoves, iron pots, trash cans, etc.), in the high-level interface on the left , since there is no need to care about the specific movement path of the room, the relevant information of specific objects in the room (such as "room interconnection" in the low-level interface on the right) can be omitted; while in the high-level interface on the left, the elements of the item layer will store a single specific Item-related attributes and information about other items.

Therefore, there is less information in the higher-level interfaces, because the high-level interfaces do not need to consider how the low-level interfaces are implemented. As the interfaces gradually penetrate into the bottom layer, the node information in the graph structure gradually becomes more detailed. In the scene, through attributes and relationships, Supplementing additional information allows the simulator module 120 to properly represent the required functions. Such a design does not require additional computing power to calculate the logic implementation of the low-level interface when only using high-level interfaces, reduces the complexity and computing power requirements of high-level interface simulation implementation, and facilitates verification of subsequent task planning effects.

Furthermore, the request execution unit 122 of the simulator module 120 mainly represents the state of the robot itself and the impact of the actual operation of the robot in the real environment. In the present invention, the state of the robot itself and the impact of the actual operation in the real environment It is abstracted as the influence of its own state or simulated operations in the simulation environment. For example, when a robot performs a "navigation to the kitchen" behavior, it modifies the robot's own state and the location is the kitchen, and performs an "open trash can" operation. The attributes of the trash can in the simulation environment can be opened or closed. For interfaces at different levels, the behaviors that can be executed are different, and the abstraction level needs to be determined.

The verification unit 123 of the simulator module 120 is mainly used to evaluate whether the robot's simulation operation achieves the expected goal by retrieving attribute information and side information of specific objects in the simulation environment.

By constructing the simulation environment from top to bottom, we avoid sampling difficulties caused by the complexity of low-level implementation. At the same time, through high-level feature sampling, we can directly obtain the basic tasks that may be performed in the simulation environment, avoiding the problem of requiring a large number of manual annotations. , and at the same time, because the scene construction is simple and easy to migrate to other scenes, it reduces the consumption of landing in different scenes.

The data generation module 130 is used to determine the task goal in the simulation environment, search for the corresponding task path according to the task goal, generate the initial text according to the task path, and perform text enhancement on the initial text to obtain the enhanced text.

Specifically, the data generation module 130 includes: a data sampling unit 131, used to sample and obtain the task target by searching the attribute information of specific objects in the simulation environment; a path search unit 132, used to obtain the task path by searching the data graph according to the task target. ; Data enhancement unit 133, used to generate initial text using a preset template according to the task path, and perform text enhancement on the initial text through a preset text enhancement tool to obtain the enhanced text.

It can be understood that the data generation module 130 is mainly used for the acquisition, generation and enhancement of training data. The main job is to provide an executable goal, the task path input by the user before executing the goal, and the task path in a given simulation environment. Steps to execute the goal.

Specifically, the data sampling unit 131 in the data generation module 130 selects tasks in the form of "put {item1} into {item2}", "open {item}" and other operations by sampling the attribute information of specific objects in the simulation environment. Targets, such as: mobile, wok, stove. The path search unit 132 in the data generation module 130 searches the data graph in the simulator module 120 according to the task target and obtains the corresponding task path, for example: searching for an iron pot and picking up an iron pot. The data enhancement unit 133 in the data generation module 130 obtains the initial text through a preset template, for example: take the iron pot to the stove and put the iron pot on the stove. Then use preset text enhancement tools (transcription, rewriting, etc.) to obtain enhanced text, for example: take the pot to the stove and put the pot back on the stove.

The method of the present invention is mainly aimed at the current situation that problems such as lack of data and difficulty in constructing a simulation environment in existing robot mission planning scenarios lead to difficulties in model implementation. A set of robot mission planning model construction systems 100 are proposed to complete data acquisition, data enhancement, and model building. Processes such as verification reduce the complexity of model implementation and scene migration. Compared with the existing bottom-up simulation environment construction method, the present invention uses a top-down construction method. There is no need to implement low-level logic for high-level interfaces, which reduces the overall complexity of the system and reduces the time required for simulation. Computing power requirements during the device verification process. Determined by the characteristics of this simulator, the system can easily obtain the data inside the scene to avoid the cost of manual annotation, and at the same time ensure the diversity of languages through text enhancement.

The robot mission planning model construction system 100 provided by the embodiment of the present application implements high-level scenarios through top-down simulator design and corresponding data generation methods, using simple data structures to reduce the investment required for new scenarios. Construction costs are reduced for simulations while reducing computational power requirements during verification and generation.

Example 2

As shown in Figure 3, it is a flow chart of a robot mission planning model construction method in an embodiment of the present application. The robot mission planning model construction method provided by the embodiment of the present application is applied to the robot mission planning model as described in Embodiment 1. Construct the system 100, which includes a model module 110, a simulator module 120 and a data generation module 130, specifically including the following steps:

Step S310: Obtain the initial statement input by the user through the model module 110, extract the initial statement using a large-scale pre-trained language model, and generate a task sequence, where the task sequence includes a high-level interface and a low-level interface.

In an optional implementation, the initial statement input by the user is obtained through the model module 110, the initial statement is extracted using a large-scale pre-trained language model, and a task sequence is generated, including:

Obtain batch data, perform pre-training on the batch data, generate the large-scale pre-trained language model, and determine the interface type;

The initial statement is obtained, and the large-scale pre-trained language model is used to perform regularized extraction on the initial statement to generate a planned task sequence.

The specific steps have been described in Embodiment 1 and will not be repeated here.

Step S320: The simulator module 120 constructs a simulation environment in the form of graph structure data from top to bottom according to the high-level interface and the low-level interface, and abstracts the actual operation of the robot in the real environment into the simulation environment. Simulation operations in and verify whether the simulation operations meet the expected goals.

In an optional implementation, the top-down construction of a simulation environment in the form of graph-structured data is based on the high-level interface and the low-level interface, and the real operations of the robot in the real environment are abstracted into the Simulate operations in a simulated environment and verify that said simulated operations meet expected goals, including:

The simulation environment is constructed in the form of graph structure data from top to bottom according to the high-level interface and the low-level interface, wherein the nodes of the data graph in the graph structure data are specific objects in the simulation environment, and the The edges of the data graph in the graph structure data are the relationships between the specific objects in the simulation environment, and the information amount of the high-level interface in the simulation environment is less than the information amount of the low-level interface;

Abstract the robot's own state and the robot's real operations in the real environment into its own state and simulated operations in the simulated environment;

By retrieving the attribute information and side information of the specific object in the simulation environment, it is verified whether the simulation operation meets the expected goal.

The specific steps have been described in Embodiment 1 and will not be repeated here.

Step S330: Determine a task goal in the simulation environment through the data generation module 130, search for a corresponding task path according to the task goal, generate initial text according to the task path, and perform text enhancement on the initial text. Get enhanced text.

In an optional implementation, the task goal is determined in the simulation environment, the corresponding task path is searched according to the task goal, the initial text is generated according to the task path, and the initial text is text-based. Enhance, get enhanced text, including:

By searching the attribute information of the specific object in the simulation environment, the task target is sampled;

Obtain the task path by searching the data graph according to the task goal;

The initial text is generated using a preset template according to the task path, and the initial text is text enhanced using a preset text enhancement tool to obtain the enhanced text.

The specific steps have been described in Embodiment 1 and will not be repeated here.

The robot mission planning model construction method provided by the embodiment of the present application is applied to the robot mission planning model construction system 100 as described in Embodiment 1. The system includes a model module 110, a simulator module 120 and a data generation module 130. Through the The model module 110 obtains the initial statement input by the user, uses a large-scale pre-trained language model to extract the initial statement, and generates a task sequence, where the task sequence includes a high-level interface and a low-level interface; through the simulator module 120 From top to bottom, a simulation environment is constructed in the form of graph-structured data according to the high-level interface and the low-level interface, the real-life operations of the robot in the real environment are abstracted into simulated operations in the simulation environment, and the simulation is verified Whether the operation meets the expected goal; determine the task goal in the simulation environment through the data generation module 130, search for the corresponding task path according to the task goal, generate initial text according to the task path, and conduct the initial text Text enhancement, get enhanced text. The above method is implemented through top-down simulator design and corresponding data generation methods. For high-level scenarios, simple data structures are used to reduce the construction costs required for new scenarios, reduce simulation, and reduce the need for verification and generation. Computing power requirements during the process.

An embodiment of the present disclosure also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the robot described in Embodiment 2 is implemented. The steps of the task planning model construction method will not be described again here.

Embodiments of the present disclosure also provide a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the robot mission planning model construction described in Embodiment 2 is implemented. The steps of the method will not be repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are only illustrative. For example, the flow charts and structural diagrams in the accompanying drawings show the possible implementation architecture and functions of the devices, methods and computer program products according to multiple embodiments of the present invention. and operations. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more components for implementing the specified logical function(s). Executable instructions. It should also be noted that, in alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two consecutive blocks may actually execute substantially in parallel, or they may sometimes execute in the reverse order, depending on the functionality involved. It will also be noted that each block in the structure diagrams and/or flowchart illustrations, and combinations of blocks in the structure diagrams and/or flowchart illustrations, can be configured with specialized hardware-based systems that perform the specified functions or actions. to be implemented, or may be implemented using a combination of dedicated hardware and computer instructions.

In addition, each functional module or unit in various embodiments of the present invention can be integrated together to form an independent part, each module can exist alone, or two or more modules can be integrated to form an independent part.

If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which can be a smart phone, a personal computer, a server, a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. .

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention. should be covered by the protection scope of the present invention.

Claims (10)

1. A robot mission planning model construction system, characterized in that the system includes: The model module is used to obtain the initial statement input by the user, extract the initial statement using a large-scale pre-trained language model, and generate a task sequence, where the task sequence includes a high-level interface and a low-level interface; The simulator module is used to construct a simulation environment in the form of graph-structured data from top to bottom according to the high-level interface and the low-level interface, and abstract the real operations of the robot in the real environment into simulated operations in the simulation environment. , and verify whether the simulation operation meets the expected goals; A data generation module is used to determine a task goal in the simulation environment, search for a corresponding task path according to the task goal, generate initial text according to the task path, and perform text enhancement on the initial text to obtain enhanced text. 2. The robot mission planning model construction system according to claim 1, characterized in that the model module includes: The model main unit is used to obtain batch data, pre-train the batch data, generate the large-scale pre-trained language model, and determine the interface type; A pre- and post-processing unit is used to obtain the initial statement, and use the large-scale pre-trained language model to perform regularized extraction on the initial statement to generate a planned task sequence. 3. The robot mission planning model construction system according to claim 1, characterized in that the simulator module includes: A simulation environment unit, configured to construct the simulation environment in the form of graph structure data from top to bottom according to the high-level interface and the low-level interface, wherein the nodes of the data graph in the graph structure data are in the simulation environment specific objects, the edges of the data graph in the graph structure data are the relationships between the specific objects in the simulation environment, and the information amount of the high-level interface in the simulation environment is less than the information amount of the low-level interface; A request execution unit configured to abstract the robot's own state and the robot's real operations in the real environment into its own state and simulated operations in the simulated environment; A verification unit configured to verify whether the simulation operation meets the expected goal by retrieving the attribute information and side information of the specific object in the simulation environment. 4. The robot mission planning model construction system according to claim 3, characterized in that the data generation module includes: A data sampling unit, configured to sample and obtain the task target by searching for attribute information of the specific object in the simulation environment; A path search unit, configured to obtain the task path by searching the data graph according to the task target; A data enhancement unit is configured to generate the initial text using a preset template according to the task path, and perform text enhancement on the initial text through a preset text enhancement tool to obtain the enhanced text. 5. A robot mission planning model construction method, characterized in that it is applied to the robot mission planning model construction system according to any one of claims 1 to 4, the system includes a model module, a simulator module and a data generation module , the method includes: Obtain the initial statement input by the user through the model module, extract the initial statement using a large-scale pre-trained language model, and generate a task sequence, where the task sequence includes a high-level interface and a low-level interface; The simulator module constructs a simulation environment in the form of graph structure data from top to bottom according to the high-level interface and the low-level interface, and abstracts the real operations of the robot in the real environment into simulated operations in the simulation environment. , and verify whether the simulation operation meets the expected goals; The data generation module determines a task goal in the simulation environment, searches for a corresponding task path according to the task goal, generates initial text according to the task path, and performs text enhancement on the initial text to obtain enhanced text. 6. The robot mission planning model construction method according to claim 5, characterized in that the initial statement input by the user is obtained through the model module, and the initial statement is extracted using a large-scale pre-trained language model to generate Task sequence, including: Obtain batch data, perform pre-training on the batch data, generate the large-scale pre-trained language model, and determine the interface type; The initial statement is obtained, and the large-scale pre-trained language model is used to perform regularized extraction on the initial statement to generate a planned task sequence. 7. The robot mission planning model construction method according to claim 5, characterized in that the top-down construction of the simulation environment in the form of graph structure data according to the high-level interface and the low-level interface, and the robot in reality Real-life operations in the environment are abstracted into simulated operations in the simulated environment, and verification is made whether the simulated operations meet the expected goals, including: The simulation environment is constructed in the form of graph structure data from top to bottom according to the high-level interface and the low-level interface, wherein the nodes of the data graph in the graph structure data are specific objects in the simulation environment, and the The edges of the data graph in the graph structure data are the relationships between the specific objects in the simulation environment, and the information amount of the high-level interface in the simulation environment is less than the information amount of the low-level interface; Abstract the robot's own state and the robot's real operations in the real environment into its own state and simulated operations in the simulated environment; By retrieving the attribute information and side information of the specific object in the simulation environment, it is verified whether the simulation operation meets the expected goal. 8. The method for constructing a robot mission planning model according to claim 7, characterized in that: determining a mission goal in the simulation environment, searching for a corresponding mission path according to the mission goal, and generating an initial initialization mission according to the task path. text, and perform text enhancement on the initial text to obtain enhanced text, including: By searching the attribute information of the specific object in the simulation environment, the task target is sampled; Obtain the task path by searching the data graph according to the task goal; The initial text is generated using a preset template according to the task path, and the initial text is text enhanced using a preset text enhancement tool to obtain the enhanced text. 9. A computer device, characterized in that it includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the robot task according to any one of claims 5-8. Steps in planning a model-building approach. 10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the robot task of any one of claims 5-8 is implemented. Steps in planning a model-building approach.