Machine session language construction method for task collaboration in complex scene

Abstract

The invention relates to a machine session language construction method for task collaboration in a complex scene, which is used for information exchange of multiple unmanned devices and supports cooperative work of the unmanned devices. The method is mainly used for solving the problem of scientific construction of an unmanned equipment collaborative information interaction machine session language system, and unmanned equipment can efficiently interact states, observation and demands by means of a machine session language and negotiate to solve accidental problems. According to the method, the natural language and artificial intelligence technology are utilized, the machine session language is constructed for the collaborative task scene, the method is used for describing and expressing complex unforeseen circumstances possibly occurring in the collaborative task scene, unmanned equipment is supported to collaboratively solve accidental problems, and the task collaboration efficiency and capacity of the unmanned equipment are improved.

Description

Technical Field

[0001] This invention relates to a method for constructing a machine conversational language for collaborative tasks in complex scenarios, used for information exchange among multiple unmanned devices, and supporting collaborative work among unmanned devices. Background Technology

[0002] Unmanned vehicles, unmanned ships, and drones are widely used in daily life and industrial production, and their application is evolving towards swarm collaboration, enabling more complex tasks to be completed through the cooperation of unmanned devices. During collaborative tasks, unmanned devices need to exchange information to achieve mutual understanding and interoperability. This requires the support of machine conversational languages ​​to enhance the mutual understanding capabilities between unmanned devices, just as natural language plays a role in information exchange during human collaborative work.

[0003] Current technologies primarily employ formatted message passing, which involves pre-planning and designing standardized message formats for the interactive information required for collaborative tasks involving unmanned equipment. During task collaboration, the unmanned equipment transmits various types of information according to these message standards. However, these pre-defined message standards are fixed and cannot account for situations beyond what is specified in the standards.

[0004] In actual task collaboration, unexpected environmental changes and unforeseen malfunctions of unmanned equipment can significantly impact the collaboration process. If the unmanned equipment cannot effectively negotiate and resolve these impacts, task execution efficiency may decrease, or even the collaborative task may fail. Therefore, a more flexible messaging mechanism is urgently needed to support unmanned equipment in describing and negotiating solutions to complex unexpected situations. This allows the current task to continue execution after the problem is resolved, until the entire task is completed, thereby improving the success rate of collaborative task execution by unmanned equipment clusters under complex conditions. Summary of the Invention

[0005] This invention provides a method that starts with documents describing task collaboration processes within a domain, and uses techniques such as large language models and natural language semantic analysis to scientifically construct a machine conversational language that meets the requirements of task collaboration within the domain.

[0006] The technical solution adopted by this invention to solve its technical problem is: A method for constructing a machine conversational language for collaborative tasks in complex scenarios includes the following steps: Step 1: Using workflow description documents of various collaborative work scenarios in the business domain as input, a semantic analysis method based on a large language model is used to extract information interaction statements between various unmanned devices from the documents. Step 2: Encode all extracted information interaction statements into semantic vectors, then cluster them into multiple clusters, find the common sentence pattern of each cluster, establish the mapping relationship between each sentence component and associated vocabulary within the cluster, and the sentence pattern within the cluster and its corresponding associated vocabulary together constitute the machine conversation language. Step 3: Build a simulation environment, design and generate various test cases, including test cases for normal execution process and test cases for abnormal execution problems caused by various accidental factors; use various test cases to test the machine conversation language, determine whether the sentence patterns and related words can meet the requirements of information interaction and problem negotiation in the application scenario, and supplement and improve the machine conversation language according to the judgment results until it meets the information expression requirements of various normal execution process test cases and abnormal execution problem test cases. Step 4: Construct a machine conversation language ontology using ontology methods. Organize machine conversation languages ​​applicable to different business domains and collaborative work scenarios to form a hierarchical machine conversation language expression system. The first layer is the business domain, the second layer is the different collaborative work scenarios under each business domain, and the third layer is the definition and description of each collaborative work scenario, as well as all sentence patterns and related vocabulary used in the corresponding scenario.

[0007] Furthermore, step 2 specifically includes the following processes: Step 201: Use the SBERT model to encode all extracted information interaction statements into semantic vectors; Step 202: Use the DBSCAN clustering algorithm to perform cluster analysis on the vectorized sentences. After clustering, traverse each cluster, segment the sentences within the cluster, and then perform dependency parsing. By finding the longest common subsequence, find the common sentence pattern of each cluster. Step 203: Based on the dependency parsing results, establish the mapping relationship between each sentence structure component and its associated vocabulary; the intra-cluster sentence patterns and their corresponding associated vocabulary together constitute the machine conversational language.

[0008] Furthermore, in step 3, the accidental factors include environmental accidental factors and accidental factors of the unmanned equipment itself; environmental accidental factors include weather changes, topographical changes, changes in lighting conditions and electromagnetic interference, and accidental factors of the unmanned equipment itself include sensor failure, mechanical failure, communication failure, software failure and power failure.

[0009] This invention utilizes natural language and artificial intelligence technologies to construct a machine conversational language for collaborative task scenarios. This language is used to describe and express complex and unexpected situations that may occur in collaborative task scenarios, supporting unmanned equipment in collaboratively solving occasional problems and improving the task collaboration capabilities of unmanned equipment. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the machine conversation language construction method of the present invention.

[0011] Figure 2 This is a schematic diagram of the organizational structure of the machine conversational language system of this invention. Detailed Implementation

[0012] The present invention will now be further described with reference to the accompanying drawings.

[0013] This invention is mainly used to solve the problem of scientifically constructing a machine conversation language system for collaborative information interaction of unmanned devices. With the help of machine conversation language, unmanned devices can efficiently interact with each other's status, observations and needs, negotiate and solve unexpected problems, and improve the efficiency and capability of unmanned device task collaboration.

[0014] In daily life, language communication in specific scenarios only requires a limited set of sentence structures to meet the needs of information exchange. Inspired by this, this invention uses technical means to automatically extract and condense the most essential syntax and vocabulary from technical documents on collaborative work of unmanned equipment, forming a machine conversational language system to meet the conversational needs between unmanned devices.

[0015] The working principle of this invention is as follows: Figure 1 As shown, taking a workflow description document for a collaborative task scenario within the domain as input, the document filters out statements describing the exchange of information between unmanned devices through task collaboration information interaction. The filtered statements are then vectorized, analyzed for sentence patterns, and their constituent words are analyzed to form a machine conversation language. A simulation environment is constructed to verify and refine the machine conversation language. Finally, the syntax and sentence patterns formed in the above steps are organized and structured according to the domain and scenario to create a clear and logical machine conversation language system that is easy to maintain and use.

[0016] A method for constructing a machine conversational language for collaborative tasks in complex scenarios includes the following steps: Step 1: Identify the information interaction requirements for typical collaborative work scenarios of unmanned systems within the domain: Using workflow description documents for various collaborative task scenarios within different business areas as input, a semantic analysis method based on a large language model is employed to extract statements about information exchange between various unmanned devices from the documents.

[0017] Taking the collaboration between drones and unmanned vehicles in the field of agricultural plant protection as an example, in order to implement on-demand precision fertilization based on the plant growth status of different plots, it is necessary to transmit the plant growth status observed by the drone to the unmanned vehicle. Therefore, the following description of information transmission can be extracted from the system user manual: "...The drone sends the plant growth status within the current travel strip, marked by granularity per square meter, to the unmanned vehicle, and at the same time transmits the wind direction and wind speed to the unmanned vehicle every second to ensure that the unmanned vehicle can carry out precise spraying..." These statements represent the information interaction needs of typical collaborative work scenarios of unmanned systems in the field of agricultural plant protection.

[0018] Step 2: Build the machine conversational language: All extracted information interaction statements are encoded into semantic vectors, then clustered into multiple clusters. A common sentence pattern is found for each cluster, and a mapping relationship is established between each sentence component and its associated vocabulary within the cluster. The sentence patterns within each cluster and their corresponding associated vocabulary together constitute the machine conversational language. The specific process is as follows: Step 201, Vectorization of statements: The Sentence-BERT (SBERT) model is used to encode all extracted information interaction statements into semantic vectors.

[0019] Step 202, Sentence Pattern Analysis: The DBSCAN clustering algorithm is used to perform cluster analysis on vectorized sentences. After clustering, each cluster is traversed, and the sentences within the cluster are segmented. Then, dependency parsing is performed, and the common sentence pattern of the cluster is discovered by finding the longest common subsequence.

[0020] Assume the cluster contains two sentences: "The drone sends information about plant growth to the unmanned vehicle" and "The drone sends information about wind direction and speed to the unmanned vehicle." The statement "The drone sends plant growth data to the unmanned vehicle" was analyzed using dependency parsing. The analysis results are as follows: send ├--Drone (nsubj) ├--Plant growth (dobj) └--Give (cmp) └--Autonomous vehicles (POBJ) The analysis results indicate that the "drone" is the executor of the action, the "plant growth" is the recipient of the action, and the "unmanned vehicle" is the target of the action.

[0021] Similarly, dependency parsing was used to analyze the statement "The drone sends wind direction and wind speed information to the unmanned vehicle". The analysis results are as follows: send ├--Drone (nsubj) ├--Wind direction and speed (dobj) └--Give (cmp) └--Autonomous vehicles (POBJ) The analysis results indicate that the "drone" is the executor of the action, "wind direction and wind speed" are the recipients of the action, and the "unmanned vehicle" is the target of the action.

[0022] By finding the longest common subsequence, the general sentence pattern of this cluster can be expressed as: root ├-- nsubj ├-- dobj └-- cmp └--pobj In this context, root represents the predicate, nsubj represents the noun subject, dobj represents the direct object, cmp represents the complement, and pobj represents the prepositional object.

[0023] Step 203, Analysis of related vocabulary of sentence components: Based on the results of dependency parsing, a mapping relationship is established between each sentence component and related words; The related words for root are "send", the related words for nsubj are "drone", the related words for dobj are "plant growth", "wind direction" and "wind speed", and the related word for cmp is "to unmanned vehicle". The related words are the legal words used in each sentence component. Intra-cluster sentence patterns and their corresponding conjunctions together constitute machine conversational language.

[0024] Step 3: Build a simulation environment to verify and improve the machine conversation language: A simulation environment was constructed, and various test cases were designed and generated to verify whether the sentence structure and vocabulary met the actual information interaction requirements of task collaboration in the application scenario. To closely resemble real-world application scenarios, the test cases included normal execution process test cases as well as test cases for abnormal execution problems caused by various unforeseen factors, including environmental factors and unforeseen factors inherent to the unmanned equipment itself. The machine conversation language was tested using various test cases to see if it could meet the requirements of information interaction and problem negotiation in the application scenario. The machine conversation language was supplemented and improved based on deficiencies until it met the information expression requirements of various normal and abnormal test cases. Unforeseen factors included weather changes, terrain changes, lighting conditions changes, and electromagnetic interference; unforeseen factors inherent to the unmanned equipment included sensor failure, mechanical failure, communication failure, software failure, and power failure.

[0025] The aforementioned internal and external factors, acting alone or in combination, produce two types of consequences: one where the problem cannot be resolved, and the unmanned equipment (UAV) cannot continue its mission; and another where the problem can be resolved independently or with the help of neighboring UAVs, allowing the UAV to continue participating in the collaborative mission. In the first case, it is necessary to use conversational statements to inform the neighboring UAVs that the mission cannot continue. In the second case, it is necessary to use conversational statements to explain the reason for the problem and the type of assistance needed. In the first case, after receiving notification, the neighboring UAV can replace it with another UAV as a teammate to continue the collaborative mission. In the second case, after receiving notification, the neighboring UAV can assist the problematic UAV in resolving the problem and then continue cooperating with it to perform the collaborative mission.

[0026] The conversational communication achieved through the aforementioned conversational statements can salvage some unmanned equipment collaborative tasks that would otherwise have failed.

[0027] Step 4: Construct the machine conversational language ontology: like Figure 2 As shown, an ontology approach is used to construct a machine conversational language ontology. This ontology organizes machine conversational languages ​​applicable to different task domains and collaborative work scenarios, forming a machine conversational language expression system. This system is a hierarchical structure: the first layer represents the business domain; the second layer represents different collaborative work scenarios within each business domain; and the third layer contains the definition and description of each collaborative work scenario, along with all sentence patterns and related vocabulary used in that scenario. This conversational language ontology facilitates the maintenance and use of the machine conversational language.

Claims

1. A method for constructing a machine conversational language for collaborative tasks in complex scenarios, characterized in that, Includes the following steps: Step 1: Using workflow description documents of various collaborative work scenarios in the business domain as input, a semantic analysis method based on a large language model is used to extract information interaction statements between various unmanned devices from the documents. Step 2: Encode all extracted information interaction statements into semantic vectors, then cluster them into multiple clusters, find the common sentence pattern of each cluster, establish the mapping relationship between each sentence component and associated vocabulary within the cluster, and the sentence pattern within the cluster and its corresponding associated vocabulary together constitute the machine conversation language. Step 3: Build a simulation environment, design and generate various test cases, including test cases for normal execution and test cases for abnormal execution problems caused by various accidental factors; The machine conversation language is tested using various test cases to determine whether the sentence patterns and related vocabulary can meet the requirements of information interaction and problem negotiation in the application scenario. Based on the judgment results, the machine conversation language is supplemented and improved until it meets the information expression requirements of various normal execution process test cases and abnormal execution problem test cases. Step 4: Construct a machine conversation language ontology using ontology methods. Organize machine conversation languages ​​applicable to different business domains and collaborative work scenarios to form a hierarchical machine conversation language expression system. The first layer is the business domain, the second layer is the different collaborative work scenarios under each business domain, and the third layer is the definition and description of each collaborative work scenario, as well as all sentence patterns and related vocabulary used in the corresponding scenario.

2. The method for constructing a machine conversational language for collaborative tasks in complex scenarios according to claim 1, characterized in that, Step 2 includes the following specific steps: Step 201: Use the SBERT model to encode all extracted information interaction statements into semantic vectors; Step 202: Use the DBSCAN clustering algorithm to perform cluster analysis on the vectorized sentences. After clustering, traverse each cluster, segment the sentences within the cluster, and then perform dependency parsing. By finding the longest common subsequence, find the common sentence pattern of each cluster. Step 203: Based on the dependency parsing results, establish the mapping relationship between each sentence structure component and its associated vocabulary; the intra-cluster sentence patterns and their corresponding associated vocabulary together constitute the machine conversational language.

3. The method for constructing a machine conversational language for collaborative tasks in complex scenarios according to claim 1, characterized in that, In step 3, accidental factors include environmental accidental factors and accidental factors of the unmanned equipment itself; environmental accidental factors include weather changes, topographical changes, changes in lighting conditions and electromagnetic interference; accidental factors of the unmanned equipment itself include sensor failure, mechanical failure, communication failure, software failure and power failure.

Images