Robot task planning method, system and electronic device based on graph representation learning

Abstract

The present disclosure relates to the technical field of robot control, and proposes a robot task planning method and system based on graph representation learning and an electronic device, including the following steps: constructing an a priori knowledge reasoning model; based on the constructed a priori knowledge reasoning model, extracting robot planning elements to be processed as graph nodes according to obtained natural language information of robot service execution, taking the relationship between the robot planning elements as the edges of the graph, and constructing a directed graph; performing vectorization on the information of the constructed directed graph and the natural language information of robot service execution to obtain a vectorized representation of task planning; and based on a deep learning model, using a progressive method of template generation and template filling to generate a phased task planning of robot actions. The robot task planning is effectively graphically represented considering service execution requirements, and the service-related natural language is autonomously converted into a selective task planning under the guidance of reinforcement learning.

Description

Technical Field

[0001] This disclosure relates to the field of robot control technology, specifically to a robot task planning method, system, and electronic device based on graph representation learning. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] With the development of artificial intelligence technology, natural language-based knowledge learning has become an important way for humans to improve their skills. In the field of home services, the internet contains a large amount of natural language information related to home service execution, enabling home service robots to have natural language learning and understanding capabilities. This allows them to autonomously generate task plans based on natural language, greatly enhancing their service skills and improving their intelligence in task execution. Robots are widely used in home services, and robot actions can be divided into low-level actions and high-level actions. Low-level actions mainly involve basic robot operations, including specific motion parameters such as distance, joint angles, and grasping direction. High-level actions, on the other hand, are more semantic and abstract descriptions of actions, such as "grasp," "put down," and "deliver," which more intuitively express the robot's intention and goals. Task planning for home service robots mainly involves the recognition and execution of high-level actions to improve the robot's autonomous decision-making capabilities, enabling it to better adapt to the service needs of different home environments.

[0004] Currently, task planning for home service robots mostly employs rule-based design methods. These methods primarily revolve around the structural representation of natural language, designing corresponding planning templates and achieving task planning through template matching. For example, Misra et al. utilized Natural Language Processing (NLP) technology to construct information extraction templates using semantic tags (POS, Part of Speech), extracting information elements from natural language sentences and generating corresponding task plans based on the templates. However, this method imposes significant limitations on the structure of natural language, struggles to handle diverse and complex language expressions, and lacks flexibility.

[0005] With the development of deep learning technology, researchers have gradually attempted to endow robots with learning capabilities, enabling them to autonomously convert natural language into task plans, thus improving the robots' understanding of natural language and the intelligence level of task planning. However, the inventors have discovered that existing deep learning-based task planning methods still have many problems. First, these methods typically do not consider the actual execution environment of the robot in home service, and the generated task plans are often disconnected from the actual environment, lacking environmental adaptability, which may lead to significant uncertainty for the robot when performing specific tasks. Second, the task planning representation of existing methods is relatively simple, usually generating only a single execution plan and failing to provide multiple coping strategies, making it difficult for the robot to adjust the task plan in abnormal situations, affecting the successful execution of the task. In addition, most existing deep learning-based task planning methods limit the scale and vocabulary of natural language in order to reduce computational complexity and training difficulty, thus limiting the robot's ability to understand complex and diverse natural language. These problems make current task planning methods unable to meet the execution needs of actual home service robots, limiting the robot's adaptability and autonomous decision-making capabilities in complex environments. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes a robot task planning method, system, and electronic device based on graph representation learning. It effectively represents robot task planning graphically, considering service execution requirements, and autonomously transforms service-related natural language into selective task planning under reinforcement learning guidance. This enables the robot to cope with abnormal situations and complete unfamiliar service tasks, ultimately improving the robot's intelligence level and service skills.

[0007] To achieve the above objectives, the present disclosure adopts the following technical solution:

[0008] One or more embodiments provide a robot task planning method based on graph representation learning, including the following steps:

[0009] A prior knowledge reasoning model is constructed based on item characteristic reasoning and scene co-occurrence relationship modeling.

[0010] Based on the natural language information obtained from the robot's execution services, and based on the constructed prior knowledge reasoning model, the robot planning elements to be processed are extracted as graph nodes, and the relationships between the robot planning elements are used as the edges of the graph to construct a directed graph.

[0011] The nearest neighbor matrix of the directed graph is extracted, and the words and word positions in the natural language information related to robot services are vectorized and fused to obtain the vectorized representation of task planning.

[0012] The task planning model is trained by taking natural language information related to robot services as input and a vectorized representation of task planning as output.

[0013] The system acquires natural language instructions for the service robot to execute, and based on the trained task planning model, uses a progressive method of template generation and template filling to generate the task planning results for the robot's actions.

[0014] One or more embodiments provide a robot task planning system based on graph representation learning, including:

[0015] The prior construction module is configured to build a prior knowledge reasoning model based on item characteristic reasoning and scene co-occurrence relationship modeling.

[0016] The task planning graphical representation module is configured to extract the robot planning elements to be processed as graphical nodes based on the natural language information of the robot execution service obtained and the constructed prior knowledge reasoning model, and construct a directed graph with the relationship between the robot planning elements as the edges of the graph.

[0017] The vectorization representation module is configured to extract the nearest neighbor matrix of the directed graph and perform vectorization representation on the words and word positions in the natural language information related to robot services. After fusion, the vectorization representation of task planning is obtained.

[0018] The training module is configured to take robot service-related natural language information as input and a vectorized representation of task planning as output to train the constructed task planning model.

[0019] The task planning and generation module is configured to obtain the natural language information instructions to be executed by the service robot. Based on the trained task planning model, it uses a progressive method of template generation and template filling to generate the task planning results of the robot's actions.

[0020] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the steps in the above-described robot task planning method based on graph representation learning.

[0021] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0022] (1) This disclosure is based on graph learning, constructing a directed graph with items, actions, similar items, and existing scenes, and using a deep model for task planning generation. The task planning is represented by a graph structure, which makes the task planning not just a linear sequence, but a graph model containing items, actions, similar items, and scenes. The task planning is optimized based on the relational edges of the graph, which has stronger task expression ability and adaptability.

[0023] (2) This disclosure proposes a progressive task planning generation strategy of template generation and template filling. That is, first, a task planning template composed of part-of-speech tags (POS) is generated, and then the template is filled to form a task plan for robots. This reduces the complexity of autonomous task planning generation, does not require restrictions on the scale of natural language to meet the system requirements of task planning, and obtains accurate task planning based on reinforcement learning, ensuring that robot services can be successfully executed.

[0024] The advantages of this disclosure, as well as its additional advantages, will be described in detail in the following specific embodiments. Attached Figure Description

[0025] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute a limitation thereof.

[0026] Figure 1 This is a flowchart of the overall process of the robot task planning method based on graph representation learning according to Embodiment 1 of this disclosure;

[0027] Figure 2 This is a priori knowledge structure diagram of Embodiment 1 of this disclosure;

[0028] Figure 3 This is a schematic diagram of the task planning graphical representation of the robot according to Embodiment 1 of this disclosure;

[0029] Figure 4 This is a flowchart of the vectorization representation method in the robot task planning process of Embodiment 1 of this disclosure;

[0030] Figure 5 This is a flowchart of the task planning template generation process of Embodiment 1 of this disclosure;

[0031] Figure 6 This is a flowchart of the task planning template filling process of Embodiment 1 of this disclosure. Detailed Implementation

[0032] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0034] It should be noted that the terminology used herein is for descriptive purposes only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0035] Example 1

[0036] In one or more of the technical solutions disclosed in the embodiments, such as Figures 1 to 6 As shown, a robot task planning method based on graph representation learning includes the following steps:

[0037] Step 1: Based on item characteristic reasoning and scene co-occurrence relationship modeling, construct a prior knowledge reasoning model;

[0038] Step 2: Based on the natural language information of the robot's service, and based on the constructed prior knowledge reasoning model, extract the robot planning elements to be processed as graph nodes, and construct a directed graph with the relationships between the robot planning elements as the edges of the graph.

[0039] Step 3: Extract the nearest neighbor matrix of the directed graph, and perform vectorized representation of the words and word positions in the natural language information related to robot services. After fusion, the vectorized representation of task planning is obtained.

[0040] Step 4: Train the constructed task planning model by taking the robot service-related natural language information as input and the vectorized representation of task planning as output.

[0041] Step 5: Obtain the natural language information instructions to be executed by the service robot. Based on the trained task planning model, use a progressive method of template generation and template filling to generate the task planning results of the robot's actions.

[0042] This embodiment is based on graph learning, constructing a directed graph using items, actions, similar items, and existing scenes, and utilizing a deep model for task planning generation. Employing graph representation to represent task planning ensures that it is not merely a linear sequence, but a graph model encompassing items, actions, similar items, and scenes. Task planning is optimized based on graph relationships (order, function, similarity, and position), resulting in stronger task expressiveness and adaptability. A progressive task planning generation strategy involving template generation and template filling is proposed. First, a task planning template composed of part-of-speech tags (POS) is generated, and then the template is filled to form a robot-oriented task plan. This reduces the complexity of autonomous task planning generation, does not require limitations on the scale of natural language to meet system requirements, and obtains accurate task planning based on reinforcement learning, ensuring successful execution of robot services.

[0043] In this embodiment, the task planning model autonomously transforms service-related natural language into task plans for robot service execution, as shown in the overall block diagram below. Figure 1 As shown. Robot service-related natural language information refers to the service instructions issued by the user to the robot; service-related natural language is textual information used to describe the execution of household services, including the service execution actions required to perform the service, household items, and other elements.

[0044] like Figure 2 As shown, in step 1, the construction of the prior knowledge reasoning model includes item priors and scene priors. Item priors are the similar item reasoning model, and scene priors are the scene reasoning model. The specific construction method is as follows:

[0045] Step 11: Based on the prior knowledge of the items, the characteristics of the items are represented in a structured manner, and a reasoning mechanism is established based on the structured features to infer substitute items.

[0046] Step 11.1, Structured Representation of Item Characteristics: Ontology-based representation technology is used to represent the prior knowledge of items; the shape, size, and function of the items are selected as inference characteristics, and these characteristics are organized into a tree-structured knowledge base, thus obtaining the item prior information base.

[0047] An item can have multiple inference properties, such as inference property 1, inference property 2, and inference property 1, etc.

[0048] A specific example:

[0049] Parent node: Item name, such as "cup";

[0050] Child nodes: Reasoning characteristics of items, such as "cylindrical", "capacity", "300ml", "drinking function", etc.;

[0051] Step 11.2: Establish a reasoning mechanism based on the reasoning characteristics of items to identify substitute items, including:

[0052] An end-to-end reasoning model is constructed using a Long Short-Term Memory (LSTM) network. The model takes feature data from a tree-structured knowledge base of different items as input and identifies the similarity of reasoning characteristics among different items.

[0053] Set a similarity threshold for the inference characteristics of items, and filter out similar items with a similarity higher than the similarity threshold as substitute items;

[0054] This embodiment uses an end-to-end reasoning model to infer alternative items with similar functions or characteristics from a priori information database of items based on the characteristics of newly input items. This reasoning method enables the robot to infer suitable alternative items based on the characteristics of the target item when the target item is unavailable, thus enhancing the flexibility of task execution.

[0055] Step 12: For scenario priors, explore the co-occurrence relationship between items and scenarios, and recommend reasonable items under the given scenario: Use natural language processing to extract robot planning element words such as items and scenarios contained in the service robot's service execution requirements, and use statistical learning to statistically model the extracted robot planning elements to form a probabilistic item-scenario association model as scenario priors.

[0056] Step 12.1, the extraction of scene semantic elements, includes:

[0057] Natural Language Processing (NLP) technology is used to analyze the language information related to the service robot's service and extract key vocabulary such as items and scenarios involved.

[0058] Image recognition is used to label and classify items and scenes in image or video data;

[0059] Step 12.2, Item-Scene Association Model:

[0060] Statistical learning methods are used to model the co-occurrence relationship between items and scenes, forming a probabilistic item-scene association model to predict the most likely items to appear in different scenes.

[0061] For example, in an "office" scenario, the co-occurrence probability of documents, computers, and printers is relatively high;

[0062] In the "kitchen" scenario, cookware, ingredients, and seasonings have a high probability of co-occurrence;

[0063] The item-scene association model in this embodiment can output the probability of an item appearing in different scenes. For example, when the robot is looking for a remote control, it should prioritize searching the living room, then the bedroom, rather than the kitchen. This reasoning method enables the robot to efficiently locate target items using environmental information, optimize search strategies, and improve task execution efficiency.

[0064] In this embodiment, prior knowledge is constructed from two aspects: item prior and scene prior. Based on the item reasoning characteristics and the item-scene association, the substitutes for the item and the proposed existing scene can be inferred, enabling the robot to effectively deal with abnormal situations such as missing items and items that cannot be located, increasing the selectivity of the robot's task planning and execution, and improving environmental adaptability.

[0065] In step 2, the robot planning elements cover the core information that the robot needs to understand and process during the task planning process, and can provide the necessary contextual support for task generation. These elements may include items, actions, similar items, and existing scenarios.

[0066] Optionally, the relationships between robot planning elements may include sequential relationships, action relationships, similarity relationships, and positional relationships, etc.

[0067] like Figure 3 and Figure 4 As shown, in step 2, the constructed directed graph serves as the graphical representation of task planning. Specifically, based on the service-related natural language of the robot's service execution requirements, nodes are defined by four categories of elements: actions, items, similar items (substitutes), and existing scenarios, with sequential relationships R. 顺序 Relationship R 作用 Similarity relation R 相似 and positional relationship R 位置 Construct a directed graph with edges between nodes to represent the relationships between actions, items, and the environment during service execution;

[0068] In this embodiment, the order relationship, action relationship, similarity relationship and position relationship are used as the edges of the graph to represent the mutual connections between the above nodes, forming a directed graph;

[0069] Step 3 involves converting linguistic and textual information into digital data, extracting the nearest neighbor matrix vector of the directed graph, and vectorizing the natural language information related to robot services. These vectorized representations are then fused to obtain the vectorized representation of the task planning, such as... Figure 4 As shown, the specific steps are as follows;

[0070] Step 31: Extract the nearest neighbor matrix from the directed graph to obtain the vocabulary in the natural language information related to robot services;

[0071] Step 32: Represent the relationship of the nearest neighbor matrix, perform lexical representation on the word vectors, and perform positional encoding on the word positions to obtain the vectorized digital representation;

[0072] In this embodiment, the vectorized representation of task planning includes three parts: relation representation, lexical representation, and positional representation. Task planning is expressed in vector form, specifically:

[0073] Relational representation: For the nearest neighbor matrix of a directed graph, the graph nodes and edges that constitute the task planning are expressed in a matrix as a relational representation;

[0074] Lexical representation utilizes the word2vector method to represent the words in the service-related natural language information of the robot's service execution as word vectors. Word2Vec is a neural network-based word embedding technique used to convert words in text into fixed-dimensional continuous vectors.

[0075] Location representation is obtained by encoding the location of words in the natural language information of the robot performing services using one-hot encoding.

[0076] The robot task planning vector consists of: the nearest neighbor matrix vector of the directed graph, the word vectors corresponding to the semantic words in the task planning, and the one-hot vectors representing the positions of the semantic words in the task planning. These three types together constitute the task planning vector.

[0077] Step 32, Vector Fusion: The numerical vectors corresponding to relation representation, lexical representation and position representation are fused through attention mechanism to obtain the vectorized representation of task planning;

[0078] Specifically, relation vectors, word vectors, and position vectors are fused in a concatenated manner, and selective feature extraction is performed under the guidance of an attention mechanism to serve as the task planning representation result; remember the task planning sequence:

[0079] [Action 1, Item 1 (Similar Item 1, Scene 1)];

[0080] [Action 1, Item 1 (Similar Item 1, Scene 1)];

[0081] ...

[0082] [action n, item n (similar items n, scene n)];

[0083] Steps 1 and 3 above obtain task planning sequences from the natural language information (training data) of the robot's service execution. First, prior knowledge reasoning is performed to obtain similar items and multiple possible scenarios. Then, based on the fusion of directed graph and vectorized natural language data, multiple task planning sequences are obtained. This allows the robot to choose to execute the next alternative task plan after failing to execute one task plan. The above process involves prior knowledge reasoning (model), directed graph construction, and information fusion, which is complex. The results obtained at each step may be inaccurate, requiring manual correction during training and dataset construction. Furthermore, the numerous processes are not conducive to real-time generation of task plans. In this embodiment, the multiple task planning sequences obtained in step 3 are used as the model output to construct and train a task planning model. By directly using the robot service-related natural language information as input, a task planning sequence containing multiple task plans can be obtained. This improves the real-time performance and accuracy of robot task planning, reduces the failure rate of robot command execution, and enhances customer experience.

[0084] In step 4, a task planning model is constructed, including a template generation module and a template filling module. Two identical models can be built by combining the attention mechanism and the Transformer model, which serve as the template generation model and the template filling model, respectively. The task planning is generated in stages according to the order from template generation to template filling.

[0085] The task planning model described above comprises two modules: a template generation module and a template filling module. Each module is designed based on the Transformer model, and both modules have corresponding inputs and outputs. The template generation module takes service-related natural language information as input and outputs a task planning template. The template filling module takes template vectors and corresponding vectors of service-related natural language as input and outputs selective task plans.

[0086] In step 4, a progressive method of template generation and template filling based on a deep learning model is used. First, a task planning template is generated based on part-of-speech attributes (POS). Then, reinforcement learning is used to optimize and fill the template, progressively generating the task planning for robot actions. The method for training the constructed task planning model includes the following steps:

[0087] Step 41: Build a template generation model based on the Transformer model, and generate a task planning template represented by the part of speech attribute (POS) under the guidance of the attention mechanism;

[0088] Step 41-1: Obtain the input vector: Express the natural language information of the robot's service using vectors, obtain word vectors, sentence vectors and paragraph vectors, fuse them, and use them as the input vector for the template generation model;

[0089] Task planning templates are represented by word class attributes, and task planning template generation is completed with the assistance of an attention mechanism. The template generation process is as follows: Figure 5 As shown, service-related natural language information is represented by word vectors, and sentence vectors and paragraph vectors are obtained based on the word vectors. Weight coefficients are introduced to fuse the three types of vectors, and the fused vector is used as the input vector for the template generation model.

[0090] Step 41-2: For the obtained input vector, generate a task planning template represented by the part of speech attribute (POS) based on the template generation model built on the attention mechanism and Transformer.

[0091] In this embodiment, the template generation module is based on the Transformer network model. It takes a fused vector composed of word vectors, sentence vectors, and segment vectors as input and, guided by an attention mechanism, generates a task planning template characterized by part-of-speech (POS) attributes. The task planning template is represented as follows:

[0092] VB([NN,…],rel:[<VB,NN,lv-n> ])

[0093] In this context, VB represents the part of speech of a verb, referring to the action contained in the sentence, which here corresponds to the service execution action; NN represents the part of speech of a noun, referring to the item contained in the sentence, which can correspond to household items; and rel represents the relationship between the item and the action, corresponding to the relationship of action, sequence, similarity and position.

[0094] In this embodiment, template generation and template filling are implemented using Transformer. First, a task planning template is generated based on the grammatical structure, and then reinforcement learning is used to optimize the filling process, thereby improving the adaptability and generalization ability of the task planning. The template filling process is described below.

[0095] Step 42: Construct a template filling model based on the Transformer model. The template vector output by the template generation model and the input vector consisting of word vectors, sentence vectors, and paragraph vectors extracted from the natural language information of the robot's service execution are used together as input. A reinforcement learning algorithm is used to fill the generated task planning template. The model parameters are adjusted based on the set incentives to obtain the trained task planning model. The template filling process is as follows: Figure 6 As shown, it includes the following steps:

[0096] Step 42-1: Take the template vector and the input vector composed of word vectors, sentence vectors and paragraph vectors as input, and transmit them to the constructed template filling model. Perform a dot product on the template features and the input vector to fill the planning template and form the task planning result.

[0097] Step 42-2: Based on the task planning results obtained in Step 42-1, calculate semantic similarity incentives and action logic incentives. Using the stochastic gradient descent method, iteratively adjust the task planning generation strategy through the incentive values ​​corresponding to semantic similarity incentives and action logic incentives to obtain the trained task planning model.

[0098] In this embodiment, the incentives in the reinforcement learning process mainly include semantic similarity incentives and action logic incentives.

[0099] (1) Semantic similarity incentive: For the similarity comparison between input information and output action sequence at the sentence level and document level respectively, the similarity is used as semantic similarity incentive;

[0100] Specifically, the method for determining the semantic similarity incentive at the sentence level is as follows: at the sentence level, the semantic role labeling (SRL) method is used to extract the comparison elements between the input sentence (i.e., the natural language information of the service robot's service execution requirements) and the output action sequence. Based on the elements extracted by SRL, the semantic similarity between the elements and the corresponding action sequence is compared at the sentence level, and the comparison result is the semantic similarity incentive at the sentence level.

[0101] Optional, the extracted comparison elements can be represented as:

[0102] (f A0 ,f A0 ,f A0 ,V);

[0103] Among them, f A0 Indicates the agent performing the action; f A1 Indicates the direct recipient of the action, which can be a tool; f A2 It refers to the indirect recipient of an action, which can be an object affected by a tool.

[0104] For example, the robot wipes the table with a cloth, where f A0 f represents a robot A1 It means rag, and f A2 This represents a table. Finally, the elements extracted based on SRL are compared with their corresponding action sequences at the sentence level for semantic similarity, and the comparison result is used as a sentence-level semantic similarity incentive.

[0105] Specifically, the method for determining the semantic similarity incentive at the document level is as follows: the semantic similarity of the document vector composed of segment vectors is compared with the average vector composed of the planning vectors of each subtask, and the result is used as the semantic similarity incentive at the document level.

[0106] Specifically, task planning is generated based on service-related natural language information, which can be converted into word vectors, sentence vectors, and paragraph vectors; the average vector corresponding to each paragraph vector is used as the document vector.

[0107] A subtask planning vector represents a task planning vector consisting of subtasks. Task planning is a whole composed of a series of actions and items. Therefore, each group of actions and items is a subtask, which can be transformed into a subtask planning vector.

[0108] (2) Action logic stimulus: In this embodiment, actions in the natural language information of the robot's service and actions in the task planning are extracted on a document-by-document basis to form action lists L1 and L2; the sequence similarity of L1 and L2 is compared using the Dynamic Time Wrapping (DTW) method, and the comparison value is used as the action logic stimulus.

[0109] In this embodiment, different input vectors (word vectors, sentence vectors, and paragraph vectors) are fused together when generating task plans. The combination of reinforcement learning and attention mechanisms helps to generate selective and accurate task plans.

[0110] In this embodiment, the stochastic gradient descent method is used to guide the reinforcement learning process, ultimately realizing the transformation of service-related natural language into task planning based on feature label representation.

[0111] The task planning expression is: [Action, Item (Similar Items, Existing Scene)]. Here, "Similar Items" can represent a single item name or a list of item names, and "Existing Scene" refers to a home setting where the items might appear in the action sequence.

[0112] Step 5: Obtain the natural language instructions to be executed by the service robot. Based on the trained task planning model, use a progressive method of template generation and template filling to generate the task planning results for the robot's actions; specifically,

[0113] Step 51: Obtain the input vector: Express the natural language information instruction to be executed into a vector, obtain word vectors, sentence vectors and paragraph vectors, and fuse them as the input vector of the trained template generation model; the implementation process of this step is the same as step 41.

[0114] Step 52: For the obtained input vector, generate a task planning template represented by the part of speech attribute (POS) through the template generation model; the implementation process of this step is the same as step 41.

[0115] Step 53: The template vector output by the template generation model and the input vector composed of word vectors, sentence vectors, and paragraph vectors extracted from the natural language information instructions to be executed are used as input and transmitted to the trained template filling model. The template features and the input vector are multiplied together to fill the planning template and form the task planning result. The implementation process of this step is the same as step 42.

[0116] In this embodiment, the task planning representation of existing methods is relatively simple, usually generating only a single execution plan and failing to provide multiple coping strategies. This makes it difficult for robots to adjust task planning in abnormal situations, affecting the successful execution of tasks. To address this, a progressive task planning generation strategy of template generation and template filling is proposed. That is, a task planning template composed of part-of-speech tags (POS) is first generated, and then the template is filled to form a task plan for robots. This reduces the complexity of autonomous task planning generation, does not require restrictions on the scale of natural language to meet the system requirements of task planning, and obtains accurate task planning based on reinforcement learning, ensuring that robot services can be successfully executed.

[0117] Example 2

[0118] Based on Embodiment 1, this embodiment provides a robot task planning system based on graph representation learning, including:

[0119] The prior construction module is configured to build a prior knowledge reasoning model based on item characteristic reasoning and scene co-occurrence relationship modeling.

[0120] The task planning graphical representation module is configured to extract the robot planning elements to be processed as graphical nodes based on the natural language information of the robot execution service obtained and the constructed prior knowledge reasoning model, and construct a directed graph with the relationship between the robot planning elements as the edges of the graph.

[0121] The vectorization representation module is configured to extract the nearest neighbor matrix of the directed graph and perform vectorization representation on the words and word positions in the natural language information related to robot services. After fusion, the vectorization representation of task planning is obtained.

[0122] The training module is configured to take robot service-related natural language information as input and a vectorized representation of task planning as output to train the constructed task planning model.

[0123] The task planning and generation module is configured to obtain the natural language information instructions to be executed by the service robot. Based on the trained task planning model, it uses a progressive method of template generation and template filling to generate the task planning results of the robot's actions.

[0124] It should be noted that each module in this embodiment corresponds one-to-one with each step in embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0125] Example 3

[0126] Based on Embodiment 1, this embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps in the robot task planning method based on graph representation learning described in Embodiment 1.

[0127] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

[0128] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A robot task planning method based on graph representation learning, characterized in that, The method comprises the following steps: Based on the reasoning of the characteristics of the goods and the modeling of the co-occurrence relationship of the scene, a priori knowledge reasoning model is constructed; According to the obtained natural language information of the robot performing the service, based on the constructed priori knowledge reasoning model, the robot planning elements to be processed are extracted as graph nodes, and the relationship between the robot planning elements is taken as the edge of the graph to construct a directed graph; The adjacency matrix of the directed graph is extracted, and the words and word positions in the natural language information related to the robot service are vectorized, and the vectorized representation of the task planning is obtained after fusion; The robot service related natural language information is taken as the input, and the vectorized representation of the task planning is taken as the output, and the constructed task planning model is trained; The natural language information instruction of the service robot to be executed is obtained, and based on the trained task planning model, a progressive method of template generation and template filling is adopted to generate the task planning result of the robot action; The method for training the constructed task planning model comprises the following steps: A template generation model is constructed based on the Transformer model, and under the guidance of the attention mechanism, a task planning template characterized by word attributes is generated; A template filling model is constructed based on the Transformer model, the template vector output by the template generation model and the input vector composed of the word vector, sentence vector and segment vector extracted from the natural language information of the robot performing the service are taken as inputs, the generated task planning template is filled based on the reinforcement learning algorithm, the model parameters are adjusted based on the set incentive, and the trained task planning model is obtained; The method for training the constructed task planning model comprises the following steps: The template vector and the input vector composed of the word vector, sentence vector and segment vector are taken as inputs, transmitted to the constructed template filling model, and the template features and input vectors are multiplied to fill the planning template and form the task planning result; Based on the obtained task planning result, the semantic similarity incentive and the action logic incentive are calculated, the random gradient descent method is adopted, the generation strategy of the task planning is iteratively adjusted through the incentive values corresponding to the semantic similarity incentive and the action logic incentive, and the trained task planning model is obtained.

2. The robot task planning method based on graph representation learning according to claim 1, wherein: The constructed priori knowledge reasoning model comprises goods priori and scene priori; For the goods priori, the characteristics of the goods are structured and represented, and a reasoning mechanism is established based on the structured features to infer replaceable goods; For the scene priori, the robot planning element words contained in the service robot service execution demand are extracted using natural language processing, and the extracted robot planning elements are statistically modeled using statistical learning to form a probability form of goods-scene association model as the scene priori.

3. The robot task planning method based on graph representation learning according to claim 1, wherein: The robot planning elements comprise goods, actions, similar goods and existing scenes. The relationships between the robot planning elements include sequential relationships, action relationships, similar relationships, and location relationships.

4. The robot task planning method based on graph representation learning according to claim 1, wherein: According to the service-related natural language of the robot service execution demand, four types of elements, namely, actions, items, similar items, and existing scenes, are taken as nodes, and sequential relationships, action relationships, similar relationships, and location relationships are taken as edges between the nodes to construct a directed graph to represent the association between actions, items, and environments in the service execution process.

5. The robot task planning method based on graph representation learning according to claim 1, wherein: The neighbor matrix vector of the directed graph is extracted, and the robot service-related natural language information is vectorized to obtain a vectorized representation of the task planning, specifically: The neighbor matrix in the directed graph is extracted to obtain the vocabulary in the robot service-related natural language information; The relationship representation, the vocabulary representation, and the position encoding of the vocabulary are obtained through the attention mechanism to obtain the vectorized digital expression; The relationship representation, the vocabulary representation, and the position representation are fused through the attention mechanism to obtain the vectorized representation of the task planning.

6. The robot task planning method based on graph representation learning according to claim 5, wherein: The neighbor matrix of the directed graph is used to express the graph nodes and edges of the graph as a matrix representation, which is used as the relationship representation; The word2vector method is used to represent the vocabulary in the robot service-related natural language information as a word vector, which is used as the vocabulary vector; The one-hot encoding is used to encode the vocabulary in the robot service-related natural language information to obtain the position representation.

7. A robot task planning system based on graph representation learning, characterized in that, It includes: The prior construction module is configured to construct a prior knowledge reasoning model based on item characteristics reasoning and scene co-occurrence relationship modeling; The task planning graph representation module is configured to extract robot planning elements to be processed as graph nodes based on the constructed prior knowledge reasoning model according to the obtained natural language information of the robot service execution, and to construct a directed graph with the relationships between the robot planning elements as the edges of the graph; The vectorization representation module is configured to extract the neighbor matrix of the directed graph and vectorize the vocabulary and vocabulary position in the robot service-related natural language information, and to obtain the vectorized representation of the task planning after fusion; The training module is configured to train the constructed task planning model with the robot service-related natural language information as input and the vectorized representation of the task planning as output; The task planning generation module is configured to obtain the natural language information instruction to be executed by the service robot, generate the task planning result of the robot action based on the trained task planning model using a progressive method of template generation and template filling. The method for training the constructed task planning model includes the following steps: A template generation model is constructed based on the Transformer model to generate a task planning template characterized by word attributes under the guidance of the attention mechanism. The template filling model is constructed based on the Transformer model, the template vector output by the template generation model and the input vector composed of the word vector, the sentence vector and the segment vector extracted from the natural language information of the service performed by the robot are jointly used as the input, the generated task planning template is filled based on the reinforcement learning algorithm, the model parameters are adjusted based on the set incentive, and the trained task planning model is obtained; The method for filling the generated task planning template based on the reinforcement learning algorithm, adjusting the model parameters based on the set incentive, and obtaining the trained task planning model comprises the following steps: The template vector and the input vector composed of the word vector, the sentence vector and the segment vector are jointly used as the input, are transmitted to the constructed template filling model, the template features and the input vector are jointly multiplied, the planning template is filled, and the task planning result is formed; Based on the obtained task planning result, the semantic similarity incentive and the action logic incentive are calculated, the random gradient descent method is adopted, the generation strategy of the task planning is iteratively adjusted through the incentive values corresponding to the semantic similarity incentive and the action logic incentive, and the trained task planning model is obtained.

8. An electronic device, comprising: The computer instructions stored on the memory and running on the processor, when executed by the processor, complete the steps in the robot task planning method based on graph representation learning in any one of claims 1-6.

Images