Intelligent agent task execution methods, devices, equipment, media and products

Abstract

Embodiments of the present application provide an agent task execution method, which can be applied to the field of artificial intelligence technology. The agent task execution method comprises: generating dictionary programming data corresponding to user task information through a preset logical form analysis rule; analyzing the dictionary programming data to generate a logical form operator graph according to a preset logical form analysis operator library; and generating an AND / OR matrix sequence according to the logical form operator graph, the AND / OR matrix sequence being used to implement a task execution process corresponding to the user task information. Embodiments of the present application also provide an agent task execution device, equipment, a storage medium and a program product.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to the field of image processing technology, and more specifically to a method, apparatus, device, medium, and product for the task execution of an intelligent agent. Background Technology

[0002] Artificial intelligence (AI) is a key driving force of the new round of technological revolution and industrial transformation. It is a new and crucial technological science that studies and develops theories, methods, technologies, and application systems to simulate, extend, and expand human intelligence. As an important component of intelligent science, AI attempts to understand the essence of intelligence and produce a new type of intelligent machine (i.e., intelligent agent) that can react in a way similar to human intelligence.

[0003] To perform corresponding actions and tasks, intelligent agents need to conduct various complex task planning processes. Currently, intelligent agents based on generative models (such as Large Language Models, or LLM language models) need to map their output unstructured text to instructions (such as robotic arm actions, API calls, and HTTP access) when planning tasks. In this case, the generative model is usually required to output structured parameters so that the processor can make correct calls and calculations.

[0004] On the one hand, while existing intelligent agents provide their own action space, enabling them to execute specific atomic actions and advanced skills, these actions and skills also require invocation with structured parameters. However, the language understanding modules currently provided by intelligent agents cannot correctly generate structured invocation parameters. On the other hand, the task planning and decomposition process relying on the aforementioned generative models typically exhibits good common sense and generalization capabilities. However, as the length of the output characters increases, the training and inference costs continuously rise, which places higher demands on engineering.

[0005] Therefore, existing intelligent agents cannot perform efficient action planning for large batches and at low cost based on existing generative models. Moreover, existing technologies cannot provide detailed information feedback related to task planning and execution processes, resulting in users being unable to accurately understand the task execution process of the intelligent agent, leading to a poor interactive experience. At the same time, the intelligent agent is also unable to correct its own execution state based on task execution. Summary of the Invention

[0006] In view of at least one of the technical problems existing in the prior art, the embodiments of the present invention aim to achieve an effective connection between flexible expression of natural language and precise task execution, thereby achieving efficient task operator invocation, and realizing a task execution method, device, equipment, medium and product for intelligent agents with low cost and high efficiency in task planning and execution. This allows for the use of low-cost, language-friendly dedicated programming rules, eliminating the need for natural language processing to generate long structured parameters, thus ensuring the accuracy and inference speed of task planning and execution. Furthermore, through syntax checking and parsing execution, it provides the ability to map invocation parameters to intelligent agent action skills, not only deconstructing the agent's dependence on natural language parsing, but also providing additional interpretability information feedback to the generative model, enabling it to correct its execution state before task execution, thereby greatly improving the user interaction experience and achieving a higher level of intelligence.

[0007] One aspect of the present invention provides a task execution method for an intelligent agent, comprising: generating dictionary programming data corresponding to user task information through a preset logical form parsing rule; parsing the dictionary programming data to generate a logical form operator graph according to a preset logical form parsing operator library; and generating an AND-OR matrix sequence according to the logical form operator graph, wherein the AND-OR matrix sequence is used to implement the task execution process corresponding to the user task information.

[0008] According to an embodiment of the present invention, before generating dictionary programming data corresponding to user task information by parsing rules through a preset logical form, the method further includes: generating user task information based on the interaction application scenario between the agent and the user; and generating class extensible tag data corresponding to the user task information through a preset natural language processing model.

[0009] According to an embodiment of the present invention, the process of generating dictionary programming data corresponding to user task information by parsing rules in a preset logical form includes: parsing class extensible tag data by parsing rules in a preset logical form to generate dictionary programming data.

[0010] According to an embodiment of the present invention, parsing dictionary programming data to generate a logical form operator graph based on a preset logical form parsing operator library includes: sequentially calling at least one execution operator required for task execution through a preset logical form parsing operator library based on the parsing information of the dictionary programming data; and generating a logical form operator graph according to the simulated execution results of each execution operator in the at least one execution operator.

[0011] According to an embodiment of the present invention, the task execution method of the intelligent agent further includes: obtaining the task atomic skills and / or task atomic actions required for task execution based on the AND-OR matrix sequence and the preset skill action mapping relationship; and performing the task execution of the intelligent agent based on the task atomic skills and / or task atomic actions.

[0012] According to an embodiment of the present invention, in obtaining the task atomic skills and / or task atomic actions required for task execution based on the AND-OR matrix sequence and the preset skill action mapping relationship, the method further includes: calling a preset natural language generation model to convert the exception description information corresponding to the logical form of the AND-OR matrix sequence and the user task information, and generating interpretable information required for task execution.

[0013] Another aspect of the present invention provides a task execution device for an intelligent agent, comprising a logic parsing module, an operator graph generation module, and a sequence generation module. The logic parsing module generates dictionary programming data corresponding to user task information through preset logic form parsing rules; the operator graph generation module parses the dictionary programming data to generate a logic form operator graph based on a preset logic form parsing operator library; and the sequence generation module generates an AND-OR matrix sequence based on the logic form operator graph, the AND-OR matrix sequence being used to implement the task execution process corresponding to the user task information.

[0014] Another aspect of the present invention provides an electronic device including one or more processors and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the task execution method of the intelligent agent described above.

[0015] Another aspect of the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the task execution method of the intelligent agent described above.

[0016] Another aspect of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the task execution method of the intelligent agent described above.

[0017] The task execution method for intelligent agents provided in this invention can at least partially solve the problem of low intelligence level in the task execution process of intelligent agents in related technologies, and therefore can achieve at least one of the following technical effects:

[0018] Based on low-cost (simple and concise) language processing-friendly dedicated programming rules, natural language processing does not need to generate long structured parameters, thus ensuring the accuracy of task planning and execution and the speed of inference. In addition, through syntax checking and parsing execution, it can provide the ability to map to the calling parameters of the agent's action skills. This not only deconstructs the agent's dependence on natural language parsing, but also provides additional interpretable information feedback to the generative model, so that it can correct its own execution state before executing the task, thereby greatly improving the user interaction experience and achieving a higher level of intelligence.

[0019] Therefore, the task execution method of the intelligent agent described in the embodiments of the present invention can effectively connect the flexible expression of natural language and the precise task execution, thereby achieving efficient task operator invocation, realizing low-cost and efficient task planning and execution, and providing corresponding interpretable information as a basis for correcting the state of the intelligent agent itself, thus achieving a good interactive experience with the user.

[0020] It should be understood that the above general description and the following specific embodiments are merely exemplary and illustrative, and do not limit the scope of the invention. Attached Figure Description

[0021] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0022] Figure 1 The illustration schematically depicts application scenarios of a task execution method, apparatus, device, medium, and program product of an intelligent agent according to embodiments of the present invention.

[0023] Figure 2 A flowchart illustrating a task execution method of an intelligent agent according to an embodiment of the present invention is shown schematically;

[0024] Figure 3A The illustration schematically shows a graphical representation of a logical form of an operator diagram of a task execution method for an intelligent agent according to an embodiment of the present invention;

[0025] Figure 3B This schematically illustrates an AND-OR matrix sequence data architecture diagram of a task execution method for an intelligent agent according to an embodiment of the present invention;

[0026] Figure 4 This schematically illustrates a flowchart of another application scenario of the task execution method of an intelligent agent according to an embodiment of the present invention;

[0027] Figure 5 A schematic diagram illustrating the structure of a task execution device for an intelligent agent according to an embodiment of the present invention is shown; and

[0028] Figure 6 A block diagram of an electronic device suitable for implementing a task execution method for an intelligent agent according to an embodiment of the present invention is shown schematically.

[0029] The accompanying drawings mentioned above are part of the specification of embodiments of the present invention, illustrating exemplary embodiments of the invention. The drawings, together with the description in the specification, serve to illustrate the principles of the embodiments of the present invention. It should be understood that the above general description with reference to the drawings and the following detailed description are merely exemplary and illustrative, and do not limit the scope of the invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the spirit of the contents disclosed in the present invention will be clearly explained below with reference to the accompanying drawings and detailed description. After understanding the embodiments of the present invention, any person skilled in the art can make changes and modifications based on the technology taught in the present invention without departing from the spirit and scope of the present invention.

[0031] The illustrative embodiments and descriptions of the present invention are used to explain the invention, but are not intended to limit the invention. Furthermore, elements / components using the same or similar reference numerals in the drawings and embodiments are used to represent the same or similar parts.

[0032] The terms "first," "second," etc., used in this invention do not specifically refer to any order or sequence, nor are they intended to limit the invention; they are merely used to distinguish elements or operations described using the same technical terms.

[0033] The directional terms used in this invention, such as up, down, left, right, front, or back, are merely for reference to the accompanying drawings. Therefore, the directional terms used are for illustrative purposes and not for limiting the scope of this invention.

[0034] The terms “comprising,” “including,” “having,” “containing,” etc., used in this invention are all open-ended terms, meaning that they include but are not limited to.

[0035] The term "and / or" as used in this invention includes any or all combinations of the things mentioned.

[0036] In this invention, "multiple" includes "two" and "more than two"; in this invention, "multiple groups" includes "two groups" and "more than two groups".

[0037] The terms "approximately," "about," etc., used in this invention are intended to modify any quantity or error that may vary slightly, but these slight variations or errors do not change the essence of the quantity or error. Generally, the range of slight variations or errors modified by such terms may be 20% in some embodiments, 10% in some embodiments, 5% in some embodiments, or other values. Those skilled in the art should understand that the aforementioned values ​​can be adjusted according to actual needs and are not limited thereto.

[0038] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0039] When expressions such as "at least one of A, B, and C" are used, they should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). When expressions such as "at least one of A, B, or C" are used, they should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, or C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). A person skilled in the art should also understand that any conjunction and / or phrase that substantially arbitrarily indicates two or more optional items, whether in the specification, claims, or drawings, should be understood to indicate the possibility of including one of these items, either of these items, or both items. For example, the phrase “A or B” should be understood as including the possibility of “A” or “B”, or “A and B”.

[0040] In view of at least one of the technical problems existing in the prior art, the embodiments of the present invention aim to achieve an effective connection between flexible expression of natural language and precise task execution, thereby achieving efficient task operator invocation, and realizing a task execution method, device, device, medium and product for intelligent agents with low cost and high efficiency in task planning and execution. This allows for the use of low-cost, language-friendly, dedicated programming rules, eliminating the need for natural language processing to generate long structured parameters, thus ensuring the accuracy and inference speed of task planning and execution. Furthermore, through syntax checking and parsing execution, it provides the ability to map invocation parameters to intelligent agent action skills, not only deconstructing the agent's dependence on natural language parsing, but also providing additional interpretability information feedback to the generative model, enabling it to correct its execution state before task execution, thereby greatly improving the user interaction experience and achieving a higher level of intelligence.

[0041] One aspect of the present invention provides a task execution method for an intelligent agent, comprising: generating dictionary programming data corresponding to user task information through a preset logical form parsing rule; parsing the dictionary programming data to generate a logical form operator graph according to a preset logical form parsing operator library; and generating an AND-OR matrix sequence according to the logical form operator graph, wherein the AND-OR matrix sequence is used to implement the task execution process corresponding to the user task information.

[0042] Figure 1 The illustration schematically depicts an application scenario of a task execution method, apparatus, device, medium, and program product of an intelligent agent according to embodiments of the present invention.

[0043] like Figure 1 As shown, application scenario 100 according to this embodiment may include terminal devices 101, 102, and 103, network 104, and server 105. Network 104 is used as a medium to provide a communication link between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0044] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0045] Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0046] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using terminal devices 101, 102, and 103 (for example only). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0047] It should be noted that the task execution method of the intelligent agent provided in the embodiments of the present invention can generally be executed by server 105. Correspondingly, the task execution device of the intelligent agent provided in the embodiments of the present invention can generally be located in server 105. The task execution method of the intelligent agent provided in the embodiments of the present invention can also be executed by a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Correspondingly, the task execution device of the intelligent agent provided in the embodiments of the present invention can also be located in a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105.

[0048] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0049] The following will be based on Figure 1 The described scene, through Figures 2-4 The task execution method of the intelligent agent in the disclosed embodiments will be described in detail.

[0050] like Figure 2 As shown, one aspect of an embodiment of the present invention provides a task execution method for an intelligent agent, which includes operations S201 to S203.

[0051] In operation S201, dictionary programming data corresponding to user task information is generated by parsing rules in a preset logical form.

[0052] In operation S202, dictionary programming data is parsed to generate a logical form operator graph from the operator library according to a preset logical form; and

[0053] In operation S203, an AND-OR matrix sequence is generated based on the logical form operator graph. The AND-OR matrix sequence is used to implement the task execution process corresponding to the user task information.

[0054] The intelligent agent can be the executing subject of the task execution method described in the embodiments of the present invention, or it can be an executor controlled by the task execution method. Specifically, it can be a humanoid intelligent robot or other AI device, which usually has its own actuators capable of performing specific action tasks. For example, a humanoid robot can use a robotic hand to perform the action task of picking up objects, or even perform action tasks such as packaging products on an industrial production line. These action tasks can serve as the target execution tasks of the intelligent agent. To complete these target execution tasks, the intelligent agent needs to perform a series of execution actions in sequence.

[0055] Preset logical form parsing rules can be understood as preset rules for a dedicated programming language based on logical forms, used to realize the mapping relationship between natural language and executable task sequences of intelligent agents. Specifically, these preset logical form parsing rules can be implemented based on the Preset Logical Form Parser (LFP) module to parse and execute data conforming to logical form programming languages. By customizing operators and syntax, a small computation graph is constructed to realize the mapping from natural language to action sequences.

[0056] LFP can be developed based on the Python language. By providing an abstract layer syntax parser and parser for generative models oriented towards Natural Language Processing (NLP), it decouples action planning at the natural language level from action execution at the agent execution level, thereby achieving flexibility and independence on the NLP side. Furthermore, by providing a lexical checker and a syntax checker, it ensures the robustness of the agent's execution of instructions.

[0057] User task information can be task information generated by an intelligent agent in an application scenario involving interaction with a user. This task information is usually presented in natural language and can include the target task that the intelligent agent believes the user needs and the related task execution data, state data, etc. For example, when a user points to an apple on the table to the left and asks the intelligent agent, "How many apples are on the table?", the corresponding user task information for the intelligent agent could be simply "who the user is," "what the user is pointing to," and "the user's question, 'How many apples are on the table?'" It could also include the intelligent agent's own state execution information (such as the action it is currently performing). Ultimately, it can be used to instruct the intelligent agent to observe the table and objects on it following the user's pointing finger, to distinguish the apples on the table and confirm their quantity, and to convert the confirmation result into natural language and play it to the user through a speaker, thus providing an interactive response to the user. Therefore, user task information can serve as the basis for the intelligent agent to respond to user needs. Furthermore, user task information can be represented as natural language instructions or statements that can be recognized and processed by the NLP side.

[0058] It is important to note that the agent can detect its environment and state in real time to confirm the progress of each task. Specifically, it can assess its spatial environment and state during each iteration. The current execution state can be at least one of the agent's current actions or the state of its environment. The state of its actions can include the name of the action and corresponding execution information (such as wrist torque, angle, estimated execution time, and spatial position of key points of the fingers). The state of its environment can include information about people and objects in the agent's current location, such as the size, position (distance), shape, and color of obstacles along the path, or dynamic changes in people in the surrounding environment, such as their speed and facial expressions.

[0059] Based on pre-defined logical parsing rules, user task information is initially processed to achieve data transformation, thereby generating corresponding dictionary programming data. Dictionary programming data can be based on the Python dictionary (Dic) structure, typically consisting of key-value pairs, with a format similar to dic = {key1 :value1, key2 : value2}. Therefore, dictionary programming data can effectively bridge the gap between natural language expression and precise action execution.

[0060] Furthermore, the preset logical form parsing operator library can be a database of custom execution operators designed based on preset logical form parsing rules. This includes numerous action-type and skill-type execution operators that can be executed or invoked by the agent execution module. Each execution operator includes basic execution information such as execution definition, category, execution description, and supported LFP version. For example, for the "calculate quantity" execution operator, its execution definition is "count(*individuals)", its category is "operator", its execution description is "calculate quantity", and its supported LFP version is 0.6.14. Each execution operator can be used to define corresponding execution actions or execution skills. When invoked by the execution module or executed automatically, it can provide the executor with the corresponding execution basis during the task execution phase. Since the preset logical form parsing operator library is a dedicated operator library for this embodiment of the invention, it can flexibly add, delete, and modify operations in a custom form, enabling it to support the addition of new agent actions. It can also provide protocol layers, encapsulation details, and sample data to meet needs such as language model training and generation.

[0061] During the parsing of dictionary programming data, matching execution operators from a pre-defined logical form parsing operator library can be continuously invoked. The execution results of these operators are predicted by simulating their execution in a simulated interactive environment. Based on these execution results, the execution operators are ordered to form the required logical form operator graph. For details on parsing dictionary programming data, please refer to the process of processing Python dictionary format data; it will not be elaborated upon here.

[0062] Logical form operator graphs (LFPs) are computational graphs formed by multiple execution operators. They are not truly graphical representations; their primary purpose is to define the execution order or relationship of the various execution operators invoked during the dictionary-based programming data parsing process. Because LFPs offer a more direct computational graph representation than agent execution modules, they enable the simplified conversion of complex natural language-expressed user task information into programming based on logical form. This lays the groundwork for efficient and correct task execution by the agent, effectively eliminating the need for structured parameters used in traditional techniques.

[0063] An AND-OR matrix sequence (OAMatrix) can be a unified structured data used for data transfer between execution operators. Specifically, it can be an execution sequence that includes AND (AND) relational data, OR (OR) relational data, and corresponding atomic action or skill information. See below for details. Figure 3BThe content shown is as follows. AND-OR matrix sequences can be generated based on the execution operators and their corresponding execution relationships in the logical operator graph. These sequences can then serve as input data for intelligent agent execution modules (such as Pibaby), enabling action skill parsing for task execution, and thus allowing the execution module to directly execute and invoke them.

[0064] In this context, AND-OR matrix sequences can often be used as data transfer between execution operators. If the computation process between each execution operator is required to remain random, then the output of the previous execution operator and the input of the next execution operator must maintain the OMAtrix data format of the AND-OR matrix sequence. Therefore, based on this flexible data structure of AND-OR matrix sequences, different possibilities in natural language expressions can be represented as AND-OR graphs, allowing for the free selection of the optimal skill action sequence during the subsequent planning and execution phase.

[0065] Therefore, this AND-OR matrix sequence enables the intelligent agent's task execution to support the ambiguity of natural language. For example, "Go to the table and get a banana," but there are 3 bananas, so taking any one is fine—this is an OR relationship. However, "Get a banana + wipe the table," if there are 3 bananas and 2 tables, then "get a banana + wipe the table" is an AND relationship, with 6 possibilities. This AND-OR matrix sequence effectively supports the execution ambiguity generated by such natural language instructions, helping the intelligent agent to make execution responses that are most similar to those of real humans during task execution.

[0066] In summary, it is evident that pre-defined logical form parsing rules can bridge the gap between unstructured natural language and structured parameter calls, thereby providing execution sequence data (AND / OR matrix sequences) of an AND / OR matrix architecture. Therefore, by leveraging low-cost, language-processing-friendly dedicated programming languages ​​(LFPs), natural language processing can avoid generating long structured parameters, ensuring the accuracy and speed of task planning and execution. Furthermore, through syntax checking and parsing, the ability to map to call parameters representing agent action skills can be provided. This not only deconstructs the agent's dependence on natural language parsing but also provides additional interpretability feedback to the generative model, enabling it to correct its execution state before task execution. This significantly enhances the user experience and achieves a higher level of intelligence.

[0067] As can be seen, the task execution method of the intelligent agent described in the embodiments of the present invention can effectively connect the flexible expression of natural language and the precise task execution, thereby achieving efficient task execution operator invocation, realizing low-cost and efficient task planning and execution, and achieving a good interactive experience with users.

[0068] To enable those skilled in the art to have a clearer understanding of the task execution method of the intelligent agent described in the embodiments of the present invention, the following further provisions are provided. Figures 3A-4 Explanation.

[0069] like Figures 2-4 As shown, according to an embodiment of the present invention, before operation 201 generates dictionary programming data corresponding to user task information by parsing rules through a preset logical form, the method further includes:

[0070] Generate user task information based on the interaction scenarios between the intelligent agent and the user;

[0071] The system generates class-based extensible tagging data corresponding to user task information using a pre-defined natural language processing model.

[0072] Interactive application scenarios can be scenarios where there is interaction between the intelligent agent and the user, such as scenarios where the user and the intelligent agent have a conversation, scenarios where the intelligent agent shakes hands or makes eye contact with the user, or scenarios where the agent provides corresponding actions and facial expressions based on the user's color and expression. Specifically, these can represent application scenarios generated by the interaction process between the two parties.

[0073] like Figure 4 As shown, user task information 401 is the task requirement information generated by the agent based on the interactive application scenario, used to define the target task for the agent's task execution. For example, in a dialogue scenario, the user points to an apple on the table and asks the agent, "How many apples are on the table?" At this time, the agent can generate the user's natural language data "How many apples are on the table?" based on the interactive application scenario. This data serves as the required instructions or statements for subsequently defining multiple target tasks such as "turning face, looking at the table, distinguishing apples from other objects, and confirming the number of apples," constituting a component of the user task information. This user task information can be input into the preset natural language processing model in the form of a string (str).

[0074] like Figure 4As shown, the preset natural language processing model 402 is a preset generation model based on natural language processing (NLP) technology to process the user task information 401 in string form mentioned above and generate corresponding extensible markup data. This language model can be trained on data in advance to acquire the corresponding information processing capabilities. Specifically, the preset natural language processing model 402 can understand and analyze the instructions or statements in the user task information 401 and then output the aforementioned extensible markup data. The extensible markup data is typically markup language data in XML (Extensible Markup Language) format. This type of data can achieve format consistency during data transmission, and can also be effectively structured for storage, facilitating subsequent parsing and processing. Furthermore, it provides readability, is easy to understand and edit, and has high flexibility and extensibility.

[0075] For example, regarding the user task information corresponding to the question "How many apples are on the table?" mentioned above, the pre-defined output class of the natural language processing model 402 can be: 'describe(<arg>>question = "number of apples on the table", arg> individuals = count( <arg> individuals< / arg> =grounding ( <arg> text< / arg> = “Apple on the table”)))'.

[0076] Furthermore, such as Figure 4 As shown, the definition of the Logic Form parsing intermediate layer 403 can be achieved based on preset logical form parsing rules, and parsing of natural language instructions can be realized with the help of LFP. This Logic Form parsing intermediate layer 403 is based on a unified and independent U abstraction layer protocol, entirely based on Python class functions and parameter interface definitions, and provides data conversion such as JSON and XML data formats on top of pydantic, while also providing lexical and syntactic analysis.

[0077] like Figures 2-4 As shown, according to an embodiment of the present invention, in operation 201, the dictionary programming data corresponding to user task information is generated by parsing rules in a preset logical form, including:

[0078] The class-based extensible tag data is parsed using preset logical parsing rules to generate dictionary-based programming data.

[0079] like Figure 4As shown, according to the logic parsing module 431 (lf_dict) based on preset logical form parsing rules, the class extensible tag data can be parsed, and the corresponding dictionary programming data can be generated by parsing this data. For example, for the final output class extensible tag data of the preset natural language processing model 402, its Python dictionary format dictionary programming data can be:

[0080] { 'u_name':'describe' ,

[0081] 'u_args':{

[0082] 'question': 'The number of apples on the table'

[0083] 'individuals': {

[0084] 'u_name': 'count',

[0085] 'u_args': {

[0086] 'individuals':{

[0087] 'u_name': "grounding',

[0088] 'u_args': {'text':'Apple on the table'}

[0089] }}}}}

[0090] Specifically, the logic parsing module 431 can provide logic form conversion tools (such as lf_converter) to parse class extensible tag data into dictionary programming data in formats such as dictionaries or JSON.

[0091] Therefore, based on the unified and independent U abstraction layer protocol, data conversion for JSON and YAML can be provided entirely based on Python class function and parameter interface definitions. This allows for effective bridging of the gap between flexible natural language expression and precise action execution through the custom logic form parsing middleware 403. Furthermore, as described later, the logic form parsing middleware 403 allows for the implementation of lexical analyzers, syntax analyzers, and operator systems specifically for logic form programming languages, better meeting the demands for flexibility and scalability.

[0092] like Figures 2-4 As shown, according to an embodiment of the present invention, in operation 202, parsing dictionary programming data to generate a logical form operator graph from the parsing operator library according to a preset logical form includes:

[0093] Based on the parsing information of dictionary programming data, the operator library is parsed in a preset logical form and at least one execution operator required for task execution is called in sequence.

[0094] Generate a logical operator graph based on the simulated execution results of each execution operator in at least one execution operator.

[0095] like Figure 4 As shown, the operator graph generation module 432 can receive the dictionary programming data output by the logic parsing module 431, and perform parsing on the dictionary programming data to extract the information related to operator calls as the parsing information of the dictionary programming data. The information related to operator calls may include the name of the specific execution action or execution skill.

[0096] According to the parsing information of the dictionary programming data, the corresponding execution operators are sequentially called in the preset logical form parsing operator library. The preset logical form parsing operator library can be an extensible operator library based on a unified interface and data structure, specifically including numerous execution operators. Each execution operator can be categorized into multiple types such as motion and operator according to the LFP supported version. For example, the execution operator for looking at an object can be defined as "look at" and belongs to the motion category, while the execution operator for calculating quantity can be defined as "count" and belongs to the operator category. Accordingly, in this embodiment of the invention, some execution operators of the preset logical form parsing operator library and their related information are shown in Table 1 below:

[0097] Operator definition category describe LFP minimum supported version count(*individuals) operator Calculate the quantity 0.6.14 calc_volume(*individuals) operator Calculate the volume and size of an object. 0.6.14 intersect(*individuals) operator Calculate the intersection of multiple object lists 0.6.14 look_at(individuals) motion Looking at an object 0.6.14 stand_up() motion Stand up 0.6.14 point_at(individuals) motion Pointing to an object 0.6.14 wave_hand() motion wave 0.6.14 shake_head() motion Shaking head 0.6.14 nod() motion nod 0.6.14 turn_on(individuals) motion Turn on the appliance 0.6.14 look_in_mirror() motion Looking in the mirror 0.6.14 turn(direction=""after") motion Turn around (left, right, back) 0.6.14 seek(goal) motion Search 0.6.14 track_grab(individuals) motion snatch 0.6.15

[0098] Table 1

[0099] like Figure 4 As shown, the operator graph generation module 432 can call at least three execution operators of describe, count and grounding in the operator library in a preset logical form based on the parsing information of dictionary programming data, such as the skill parsing content 'u_name':'describe', 'u_name': 'count' and 'u_name': "grounding'.

[0100] For each invoked execution operator, the operator graph generation module 432 can perform virtual execution based on the agent's virtual execution environment to obtain the corresponding simulated execution result. Therefore, the simulated execution result can be the virtual execution result of each execution operator within the virtual execution environment. The virtual execution environment is typically a simulation environment of the interaction scenario between the agent and the user. For example, scanning and modeling the living room environment where the agent and user are located forms a virtual complete living room environment. By executing the corresponding execution operator in this complete living room environment, the corresponding simulated execution result is generated based on the execution of the execution operator in this virtual environment. This simulated execution result can correspond to the execution order of each execution operator, thereby establishing execution relationships between multiple invoked execution operators and forming a computation graph, i.e., a logical form of operator graph.

[0101] like Figure 3A As shown, the logical form of the operator graph can be a first execution operator 301, a second execution operator 302, and a third execution operator 303. When the simulated execution result of the first execution operator 301 can serve as a condition required for the execution of the second execution operator 302, the execution of the first execution operator 301 is defined as preceding the execution of the second execution operator 302. For example, in response to the user's question "How many apples are on the table?", the agent can call the grounding execution operator to perform an observation task. Specifically, this involves environmental detection using the agent's visual sensors (such as a camera) (e.g., looking at the table as the user points and analyzing all images generated during this "looking" process to obtain all observed content). Further, the count execution operator is called to perform a counting task, specifically identifying and distinguishing objects from the observations generated during the grounding process, differentiating between the table and apples, and counting the number of apples. Finally, the describe execution operator is called to perform a description task, specifically providing a natural language description of the observed content from the grounding process and the objects and their quantities distinguished by the count process. At this point, the first execution operator 301 can be grounding, the second execution operator 302 can be count, and the third execution operator 303 can be describe. The three can form the above-mentioned logical form of the operator graph.

[0102] Specifically, the operator graph generation module 432 can parse the aforementioned dictionary programming data and execute operator calls using a parsing tool such as lf_parser, ultimately generating the required logical form operator graph. Therefore, the logical form operator graph can be an AND-OR computation graph of skill action sequences.

[0103] As can be seen, the operator graph generation module 432 can provide corresponding command-line and GUI tools to parse and process dictionary programming data based on logical form programming languages, generating corresponding computation graphs, thereby converting natural language instructions into easily executable computation graphs. Furthermore, by defining a dedicated operator library as a preset logical form parsing operator library, flexible addition and deletion of operators can be implemented to accommodate new actions added by the agent. Simultaneously, a protocol layer is provided to encapsulate details and provide sample data to support the training and generation of language models (such as LLM language models). Finally, by providing an implementation of the computation graph based on Logic Form and the operator library, complex instructions can be expressed as AND / OR computation graphs, significantly improving the efficiency and accuracy of task planning and execution. Additionally, this allows for the return of semantically interpretable information when errors occur during the computation graph parsing and execution phases.

[0104] like Figure 4 As shown, the sequence generation module 433 can further generate an AND-OR matrix sequence based on the above logical operator graph. This AND-OR matrix sequence can be the data that the agent's execution module (such as Pibaby) can use to perform the task. Figure 3B As shown, the AND-OR matrix sequence can include AND relational data and OR relational data, where the AND relational data can include corresponding atomic actions. Through the AND-OR matrix array, a unified data structure for data transfer between operators can be provided, avoiding the need for pre-defined natural language processing models 402 to focus on parameter adaptation details.

[0105] like Figures 2-4 As shown, according to an embodiment of the present invention, the task execution method of the intelligent agent further includes:

[0106] Based on the AND-OR matrix sequence and the preset skill action mapping relationship, obtain the task atomic skills and / or task atomic actions required for task execution;

[0107] The agent performs tasks based on task atomic skills and / or task atomic actions.

[0108] like Figure 4 As shown, the execution sequence interface 404 can serve as a data interface for the execution module of the intelligent agent, processing the AND-OR matrix sequence output by the received sequence generation module 433 through a preset skill action mapping relationship. The AND-OR matrix sequence is an executable sequence of the aforementioned execution sequence interface 404 (such as ObeyU).

[0109] The preset skill action mapping relationships can include the mapping relationship between atomic skills and AND / OR matrix sequences, as well as the mapping relationship between atomic actions and AND / OR matrix sequences. An atomic skill can be understood as the smallest skill unit U that cannot be further subdivided, and an atomic action can be understood as the smallest action unit Action that cannot be further subdivided. Through these preset skill action mapping relationships, atomic actions and atomic skills corresponding to the AND and OR relational data of the AND / OR matrix sequences can be extracted to generate corresponding task atomic skills and / or task atomic actions, i.e., U and / or Action.

[0110] A task atomic skill can be the smallest unit skill required for task execution in an embodiment of the present invention. Correspondingly, a task atomic action can be the smallest action required for task execution in an embodiment of the present invention. Task atomic skills and / or task atomic actions can be directly invoked by the execution module of the intelligent agent to achieve sequential execution of tasks.

[0111] Therefore, by implementing the mapping between the U abstraction layer (UAbs) and uvlib in the execution module (such as Pibaby), such as one-to-one or one-to-many, an adaptation scheme for coordinating natural language and the specific execution skills of the agent based on the pre-defined logical form parsing rules is provided. Detailed error reporting and execution description information can also be provided.

[0112] like Figures 2-4 As shown, according to an embodiment of the present invention, in obtaining the task atomic skills and / or task atomic actions required for task execution based on the AND-OR matrix sequence and the preset skill action mapping relationship, the method further includes:

[0113] The system calls a pre-defined natural language generation model to parse and transform the exception description information corresponding to the AND-OR matrix sequence and user task information in logical form, generating interpretable information required for task execution.

[0114] like Figure 4 As shown, the preset natural language generation model 405 can be a preset model based on natural language generation (NLG) technology. It can be trained on language data in advance and has the ability to generate natural language text. It is used to process the input data in natural language and generate natural language text information that conforms to human expression habits.

[0115] The exception description information in the logical form parsing can be the exception information generated during the generation of the AND-OR matrix sequence of the intermediate layer 403 in the above logical form parsing. Specifically, it can be information such as execution failure information and its cause, which are generated when the corresponding execution process fails to achieve the expected goal.

[0116] like Figure 4As shown, in the logic parsing module 431, during the process of parsing class extension marker data to generate dictionary programming data, if the corresponding dictionary programming data cannot be generated, a corresponding logic parsing exception information E1 is generated as an exception information for the dictionary programming data generation process, i.e., a token exception. Correspondingly, in the operator graph generation module 432, if the corresponding execution operator cannot be effectively called, a corresponding operator graph exception information E2 is generated as an exception information for the process failing to achieve a logical form operator graph, i.e., a parse exception. Similarly, in the sequence generation module 433, if a valid AND-OR matrix sequence cannot be generated, a corresponding sequence generation exception information E3 is generated as an exception information for the AND-OR matrix sequence failing to be achieved. These logic parsing exception information E1, operator graph exception information E2, and sequence generation exception information E3 are used to describe the adverse results and corresponding adverse reasons generated during their respective execution processes, and can finally be uniformly described to generate a corresponding logical form parsing exception description information E4.

[0117] Furthermore, by invoking the preset natural language generation model 405, the logical form parsing exception description information E4 can be refined using NLG-based natural language processing. This allows the final generated natural language text to be input as a meaningful message to the execution sequence interface 404, thereby completing the conversion of the logical form parsing exception description information corresponding to the user task information. Furthermore, during the process of the execution sequence interface 404 receiving the corresponding AND-OR matrix sequence and generating task atomic skills and / or task atomic actions, the preset natural language generation model 405 can also be invoked to describe the corresponding information generated in this process using natural language, generating the corresponding natural language text, thereby completing the conversion of the corresponding AND-OR matrix sequence.

[0118] The natural language text converted by the aforementioned pre-defined natural language generation model 405 is used as interpretable information. It can retain detailed natural language description text, and describe in detail the problems and other contents generated during the task execution method of the above embodiment of the present invention. In this way, the intelligent agent can realize self-awareness, explain its own problems, explain its own execution status, explain its own execution process, provide detailed explanations and specific natural language responses that conform to human expression habits, so that the intelligent agent can demonstrate a higher level of intelligence.

[0119] In the operator graph generation process of the operator graph generation module 432, a natural language description of the calculation result of each operator is also provided, forming a chronological description of the calculation graph, saving all calculation details and results, and providing them to the downstream preset natural language generation model for the generation of natural language text.

[0120] Interpretable information can be natural language text that describes the execution process of the agent's task execution method in detail after natural language generation (NLG) processing, including execution exception information, etc. Therefore, the task execution method of this embodiment can provide interpretability functionality for language model-oriented operator runtime, enabling the agent to have a higher level of intelligence and more in line with real human reactions.

[0121] In summary, the task execution method for intelligent agents provided in the embodiments of the present invention can achieve at least one of the following technical effects:

[0122] (1) It provides better encapsulation of the action details of the agent, which facilitates the generation of action planning by the language model.

[0123] (2) The LFP-based preset logic form parsing operator library is an easily expandable operator library that can support agents to easily add various actions, and the operator adjustment is more flexible and can adapt to various task executions.

[0124] (3) Provides the ability to align natural language with objects in the scene, and can map the concepts in the instructions to objects in the scene.

[0125] (4) It can check the correctness of the logic-form format and syntax of the language model output and use it as a pre- and post-processing tool for the language model.

[0126] (5) It can provide semantic natural language information and support the interpretability of intelligent agents.

[0127] Based on the above-described task execution method for intelligent agents, the present invention also provides a task execution device 500 for intelligent agents. The following will be combined with... Figure 5 The device 500 is described in detail.

[0128] Figure 5 A schematic block diagram of a task execution device 500 for an intelligent agent according to an embodiment of the present invention is shown.

[0129] like Figure 5 As shown, the task execution device 500 of the intelligent agent in this embodiment includes a logic parsing module 510, an operator graph generation module 520, and a sequence generation module 530.

[0130] The logic parsing module 510 is used to generate dictionary programming data corresponding to user task information through preset logic parsing rules. In one embodiment, the logic parsing module 510 can be used to perform the operation S201 described above, which will not be repeated here.

[0131] The operator graph generation module 520 is used to parse dictionary programming data to generate a logical form operator graph from the operator library according to a preset logical form. In one embodiment, the operator graph generation module 520 can be used to perform the operation S202 described above, which will not be repeated here.

[0132] The sequence generation module 530 is used to generate an AND-OR matrix sequence based on the logical operator graph. The AND-OR matrix sequence is used to implement the task execution process corresponding to the user task information. In one embodiment, the sequence generation module 530 can be used to perform the operation S203 described above, which will not be repeated here.

[0133] According to embodiments of the present invention, any plurality of modules in the logic parsing module 510, operator graph generation module 520, and sequence generation module 530 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the logic parsing module 510, operator graph generation module 520, and sequence generation module 530 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or other hardware or firmware implementations such as integrating or packaging circuits, or implemented in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the logic parsing module 510, operator graph generation module 520, and sequence generation module 530 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0134] Figure 6 A block diagram of an electronic device suitable for implementing a task execution method for an intelligent agent according to an embodiment of the present invention is shown schematically.

[0135] The electronic device provided in the embodiments of the present invention includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the task execution method of the intelligent agent.

[0136] like Figure 6As shown, an electronic device 600 according to an embodiment of the present invention includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0137] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0138] According to an embodiment of the present invention, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0139] The present invention also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the task execution method of the aforementioned intelligent agent.

[0140] The computer-readable storage medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0141] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, a computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0142] Embodiments of the present invention also include a computer program product comprising a computer program that, when executed by a processor, implements the task execution method of the intelligent agent described above.

[0143] The computer program includes program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the methods provided in the embodiments of the present invention.

[0144] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this invention. According to embodiments of the invention, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0145] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0146] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this embodiment of the invention. According to embodiments of the invention, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0147] According to embodiments of the present invention, program code for executing the computer programs provided in the embodiments of the present invention can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0148] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0149] Furthermore, all actions involving the acquisition of information, signals, or data in this invention are carried out in compliance with the relevant data protection laws, regulations, and policies of the country where the invention is located, and with the authorization granted by the owner of the corresponding device.

[0150] Those skilled in the art will understand that the features described in the various embodiments and / or claims of the present invention can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments and / or claims of the present invention can be combined or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0151] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of the invention is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A method for task execution of an agent, characterized in that, The method comprises the following steps: generating dictionary programming data corresponding to user task information through preset logical form analysis rules; wherein, the preset logical form analysis rules are preset rules of a special programming language based on logical form, used to realize the mapping relationship between natural language and executable task sequence of an agent; according to the analysis information of the dictionary programming data, at least one execution operator required for task execution is called in sequence through a preset logical form analysis operator library; wherein, the preset logical form analysis operator library is a database of self-defined execution operators designed based on the preset logical form analysis rules, which includes action class execution operators and skill class execution operators executed or called by the agent execution module; generating a logical form operator graph according to the simulation execution result of each execution operator in the at least one execution operator; wherein, the simulation execution result is a virtual execution result of each corresponding execution operator in a virtual execution environment, and the execution relationship between the called execution operators is established according to the execution order of each execution operator to form a logical form operator graph; the logical form operator graph is specifically an operator calculation graph formed based on multiple execution operators, mainly used to define the execution order or execution relationship of each execution operator called according to the dictionary programming data analysis process; and generating an and / or matrix sequence according to the logical form operator graph, which is used to realize the task execution process corresponding to the user task information; wherein, the and / or matrix sequence is a unified structure data used for data transmission between execution operators, including and relationship data, or relationship data, and execution sequence of corresponding atomic actions or skills; obtaining task atomic skills and / or task atomic actions required for task execution according to the and / or matrix sequence and a preset skill action mapping relationship; performing task execution of the agent according to the task atomic skills and / or task atomic actions.

2. The method of claim 1, wherein, Before the dictionary programming data corresponding to the user task information is generated through the preset logical form analysis rules, the method further comprises the following steps: generating user task information according to the interaction application scenario between the agent and the user; generating class extensible mark data corresponding to the user task information through a preset natural language processing model.

3. The method of claim 2, wherein, In the step of generating dictionary programming data corresponding to user task information through preset logical form analysis rules, the method comprises the following steps: analyzing the class extensible mark data through preset logical form analysis rules to generate the dictionary programming data.

4. The method of claim 1, wherein, In the step of obtaining task atomic skills and / or task atomic actions required for task execution according to the and / or matrix sequence and a preset skill action mapping relationship, the method further comprises the following steps: calling a preset natural language generation model to convert the logical form analysis exception description information corresponding to the and / or matrix sequence and the user task information to generate the interpretability information required for task execution.

5. An apparatus for task execution of an agent, characterized by, The method comprises the following steps: The logic analysis module is configured to analyze the dictionary programming data corresponding to the user task information by using a preset logic form analysis rule. The preset logic form analysis rule is a preset rule of a special programming language based on a logic form, and is used to realize a mapping relationship between a natural language and an executable task sequence of an agent. The operator graph generation module is configured to analyze the dictionary programming data, and generate a logic form operator graph from a preset logic form analysis operator library according to the dictionary programming data. The generation includes: calling at least one execution operator required for the task execution from the preset logic form analysis operator library according to analysis information of the dictionary programming data in sequence. The preset logic form analysis operator library is a database of custom execution operators designed based on the preset logic form analysis rule, and includes action class execution operators and skill class execution operators executed or called by the agent execution module. The logic form operator graph is generated according to a simulation execution result of each execution operator in the at least one execution operator. The simulation execution result is a virtual execution result of each corresponding execution operator in a virtual execution environment. An execution relationship is established between the called execution operators according to an execution order of each execution operator, and the logic form operator graph is formed. The logic form operator graph is specifically an operator calculation graph formed based on the execution operators, and is mainly used to define an execution order or an execution relationship of each execution operator called in the dictionary programming data analysis process. The sequence generation module is configured to generate an and / or matrix sequence from the logic form operator graph. The and / or matrix sequence is used to realize a task execution process corresponding to the user task information. The and / or matrix sequence is a unified structure data used for data transmission between execution operators, and includes and relationship data, or relationship data, and an execution sequence of corresponding atomic actions or skills. The and / or matrix sequence is used to obtain task atomic skills and / or task atomic actions required for the task execution in combination with a preset skill action mapping relationship. The task atomic skills and / or task atomic actions are used to perform the task execution of the agent. 6.An electronic device, comprising: one or more processors; memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the method of any one of claims 1-4. 7.A computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to perform the method of any one of claims 1-4. 8.A computer program product comprising a computer program which, when executed by a processor, implements the method of any one of claims 1-4. ​

Images