A method for robot intelligent interaction based on natural language understanding

Abstract

This invention proposes a method for intelligent robot interaction based on natural language understanding. The method includes the following steps: S1: The natural language interaction module receives the initial requirements input by the user in the form of voice or text, performs noise reduction and speech recognition processing on the voice signal to convert it into text data, and transmits the text data to the large model understanding and multi-turn dialogue module; S2: The large model understanding and multi-turn dialogue module uses a large model specifically trained in the field of industrial robots to perform semantic parsing and intent recognition on the text data. If the initial requirement information is incomplete or the parameters are ambiguous, it automatically generates targeted follow-up questions and completes the requirement information through multi-turn natural language dialogue interaction until the complete workstation function type, physical parameters of the processed object, working environment constraints, and performance index requirements are obtained.

Description

Technical Field

[0001] This invention belongs to the field of language intelligent interaction, and specifically relates to a method for robot intelligent interaction based on natural language understanding. Background Technology

[0002] Currently, industrial robots are increasingly widely used in industrial production. However, the configuration and operation of industrial robots require a high level of expertise from users, which is a significant factor restricting their widespread adoption. Traditionally, the configuration process for industrial robots requires specialized robot engineers who must master specific robot programming languages, be familiar with complex parameter setting interfaces, and possess knowledge of robot kinematics and other related fields.

[0003] Ordinary electrical engineers, lacking relevant robotics training, find it difficult to independently configure industrial robot system workstations. Currently, industrial robot interaction primarily relies on fixed instruction sets or complex graphical programming interfaces. Users need to input instructions according to strict formats or perform tedious drag-and-drop operations, resulting in high learning costs and time-consuming configuration processes. When production demands change, reconfiguring the robot system often requires repeating these complex steps, severely impacting production efficiency.

[0004] Furthermore, existing interaction methods cannot effectively understand users' natural language expressions, making it difficult for users to describe their needs in everyday conversational language, resulting in low interaction efficiency and further raising the barrier to entry for using industrial robots. Summary of the Invention

[0005] This invention proposes a method for intelligent robot interaction based on natural language understanding. This invention solves the problems of high technical barriers to configuration of existing industrial robot workstations, reliance on manual programming by professional personnel, unnatural interaction methods, and low efficiency. It enables the automated deployment and management of robot workstations through natural language dialogue without the need for complex training.

[0006] The technical solution of this invention is implemented as follows: a method for intelligent robot interaction based on natural language understanding, the method comprising the following steps: S1: The natural language interaction module receives the user's initial request in the form of voice or text, performs noise reduction and speech recognition processing on the voice signal to convert it into text data, and transmits the text data to the large model understanding and multi-turn dialogue module. S2: The large model understanding and multi-turn dialogue module uses a large model specifically trained in the field of industrial robots to perform semantic parsing and intent recognition on text data. If the initial requirement information is incomplete or the parameters are ambiguous, it automatically generates targeted follow-up questions and completes the requirement information through multi-turn natural language dialogue interaction until it obtains the complete workstation function type, physical parameters of the processed object, working environment constraints and performance index requirements. S3: The industrial robot workstation automatic configuration module calls the industrial robot model and parameter library, end effector type library, workstation layout rule library and process parameter library in the knowledge base module according to the obtained requirements, automatically matches the robot model and end effector, generates workstation layout scheme, motion trajectory planning and process parameter settings, and automatically generates robot control program and workstation configuration file. S4: Convert the configuration scheme into a natural language description and feed it back to the user. If the user suggests modifications, receive and parse the user feedback through natural language interaction, and control the automatic configuration module to regenerate the configuration scheme until the user confirms it. S5: The execution and feedback module loads the user-confirmed control program and configuration file into the industrial robot control system, drives the industrial robot and peripheral equipment to perform tasks, and collects operating status data in real time, including work progress and fault information. S6: Converts runtime status data into natural language descriptions and feeds them back to the user. During system operation, it continuously collects user needs, configuration schemes, and runtime data, dynamically updates the knowledge base module, and achieves self-optimization of the knowledge base and continuous improvement of configuration accuracy.

[0007] The configuration and deployment of existing industrial robot workstations heavily rely on experienced engineers, resulting in numerous limitations: First, their interaction methods primarily rely on specialized programming languages, complex teach pendant operations, or graphical configuration interfaces, requiring users to possess deep knowledge of robotics, programming languages, and processes. This prevents ordinary operators or electrical engineers from directly participating in the configuration process, leading to rigidity and bottlenecks in human resource allocation. Second, existing processes are typically unidirectional and static. Once the program is generated and put into operation, modifications and adjustments become extremely cumbersome, requiring reprogramming, simulation, and debugging. This fails to adapt to flexible and ever-changing production demands, resulting in low efficiency and high error-tolerance costs. Furthermore, traditional systems have weak knowledge accumulation and reuse capabilities, making it difficult to systematically accumulate and pass on expert experience. Each new application scenario requires a large amount of repetitive work starting from scratch.

[0008] The core technical challenge that this invention aims to overcome lies in how to break through this technical barrier. The specific breakthrough is reflected in: realizing the intelligent conversion and mapping between highly uncertain natural language and precise structured industrial parameters. By adopting a large language model finely tuned with industrial domain knowledge, combined with intent recognition and slot filling technology, it accurately understands the user's vague and incomplete colloquial descriptions, and can actively ask for key parameters through generative multi-turn dialogue to ensure the completeness and accuracy of the demand information. This is the difficulty of understanding unstructured language into structured parameters. This invention solves the problem of generating automated solutions based on multiple constraints. The system needs to comprehensively consider many complex factors such as robot load, working radius, accuracy, end effector adaptability, layout rationality, collision-free motion trajectory, and process requirements. This invention automatically matches the optimal equipment resources from the knowledge base and generates reliable control programs and configuration files by constructing an industrial knowledge graph and using a reasoning optimization algorithm based on constraint satisfaction problem (CSP), replacing the traditional process that relies on manual decision-making and programming. A closed-loop self-optimizing system was constructed. The technical challenge lies in how to analyze the robot's operation data in real time and convert it into natural language feedback that users can intuitively understand. At the same time, it continuously learns from successful configuration schemes and operation data, dynamically updates and expands the knowledge base, so that the system has continuous performance improvement and adaptive capabilities, thus ultimately forming a virtuous cycle of becoming smarter the more it is used.

[0009] As a preferred implementation, the large-scale model understanding and multi-turn dialogue module adopts a large language model fine-tuned with professional knowledge in the field of industrial robots. The large language model is optimized through domain-adaptive pre-training and human feedback reinforcement learning, and has the ability to perform semantic parsing and multi-turn dialogue management in industrial scenarios. The multi-turn dialogue adopts a dynamic reasoning mechanism based on intent recognition and slot filling, and generates accurate clarifying questions through dialogue state tracking and entity relationship extraction technology to ensure the complete extraction of all key process parameters and operational constraints.

[0010] As a preferred implementation, the domain-adaptive pre-training uses industrial robot technical manuals, process specification documents, and fault diagnosis reports to construct a pre-training corpus of related domain texts. Domain vocabulary expansion and semantic enhancement techniques are used to improve the model's understanding accuracy of technical terms and process parameters. The human feedback reinforcement learning design a multi-dimensional reward function for robot programming scenarios, comprehensively considering grammatical correctness, technical specification compliance, and operational safety. A training dataset is constructed by ranking the generated results according to preferences. The multi-turn dialogue management adopts a knowledge graph-based dialogue state tracking mechanism, establishing a slot-filling framework that includes parameters such as robot working range, load capacity, and accuracy indicators.

[0011] In a preferred embodiment, the knowledge base module in step S3 uses ontology modeling to construct a knowledge graph for the industrial robot domain. The industrial robot domain knowledge graph includes a robot ontology model library, an end effector function library, a process rule library, and a layout constraint library. The automatic configuration module uses a reasoning mechanism based on constraint satisfaction problems. By traversing the equipment parameters and process rules in the knowledge base, it performs matching calculations of the robot's working radius, load capacity, repeatability accuracy, and task requirements. Based on collision detection and reachability analysis, it generates an interference-free motion trajectory and finally synthesizes a robot control code and workstation configuration parameter set that can be directly deployed.

[0012] In a preferred embodiment, the execution and feedback module is integrated with the control system through an industrial standard communication protocol to collect equipment operating status data in real time; it processes the operating data using time-series pattern recognition and anomaly detection algorithms, and converts equipment status, task progress, and fault diagnosis results into interpretable multimodal feedback information through natural language generation technology, providing users with real-time monitoring and decision support.

[0013] As a preferred implementation, the knowledge base update mechanism adopts incremental learning and knowledge graph embedding technology. The system automatically collects operation data and user interaction records, continuously optimizes the domain knowledge representation through feature extraction and knowledge fusion, and dynamically updates the equipment attribute association and process constraint relationship to achieve adaptive improvement of configuration strategy and continuous improvement of system performance.

[0014] After adopting the above technical solution, the beneficial effects of the present invention are: the core value lies in greatly reducing the technical threshold and usage cost of industrial robot applications, enabling ordinary electrical engineers or even technicians without in-depth programming or robotics backgrounds to describe their task requirements through the most intuitive natural language, thereby directly driving the automated configuration and deployment of the entire robot workstation, greatly liberating the productivity of professional engineers, shortening the personnel training cycle, and broadening the application audience of robots. This method completely changes the traditional rigid and one-way configuration process. Through multi-round dialogue interaction, it realizes dynamic, collaborative and refined requirement definition, ensuring that the final configuration solution can accurately match the user's true intention. At the same time, any modification suggestions can be put forward and responded to quickly through the same convenient natural language. The system can quickly iterate and generate new solutions. This high degree of flexibility and interactivity significantly improves the efficiency of the entire configuration process and user experience, adapting to the rapid response needs of modern manufacturing for small-batch, multi-variety production. The system minimizes human intervention through an automated solution generation mechanism, avoiding configuration errors or performance differences caused by differences in human experience. The solutions generated by the built-in knowledge base and optimization algorithms are often more consistent and optimal in terms of reliability, security and performance indicators, thereby improving the overall operational quality and productivity of the workstation. This system constructs an intelligent agent with self-learning and continuous evolution capabilities. Its knowledge base is not static but can continuously absorb new knowledge, optimize existing parameters, and learn new process strategies with each deployment and operation of the system and each human-computer interaction. This dynamic update mechanism enables the system's decision-making ability to continuously improve over time, and the configuration scheme becomes more accurate and efficient, forming a virtuous cycle of becoming smarter the more it is used, providing enterprises with a continuous core competitiveness for digital and intelligent transformation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: like Figure 1 As shown, a method for intelligent robot interaction based on natural language understanding is provided, the method comprising the following steps: S1: The natural language interaction module receives the user's initial request in the form of voice or text, performs noise reduction and speech recognition processing on the voice signal to convert it into text data, and transmits the text data to the large model understanding and multi-turn dialogue module. S2: The large model understanding and multi-turn dialogue module uses a large model specifically trained in the field of industrial robots to perform semantic parsing and intent recognition on text data. If the initial requirement information is incomplete or the parameters are ambiguous, it automatically generates targeted follow-up questions and completes the requirement information through multi-turn natural language dialogue interaction until it obtains the complete workstation function type, physical parameters of the processed object, working environment constraints and performance index requirements. S3: The industrial robot workstation automatic configuration module calls the industrial robot model and parameter library, end effector type library, workstation layout rule library and process parameter library in the knowledge base module according to the obtained requirements, automatically matches the robot model and end effector, generates workstation layout scheme, motion trajectory planning and process parameter settings, and automatically generates robot control program and workstation configuration file. S4: Convert the configuration scheme into a natural language description and feed it back to the user. If the user suggests modifications, receive and parse the user feedback through natural language interaction, and control the automatic configuration module to regenerate the configuration scheme until the user confirms it. S5: The execution and feedback module loads the user-confirmed control program and configuration file into the industrial robot control system, drives the industrial robot and peripheral equipment to perform tasks, and collects operating status data in real time, including work progress and fault information. S6: Converts runtime status data into natural language descriptions and feeds them back to the user. During system operation, it continuously collects user needs, configuration schemes, and runtime data, dynamically updates the knowledge base module, and achieves self-optimization of the knowledge base and continuous improvement of configuration accuracy.

[0019] This application document describes a robot intelligent interaction method based on natural language understanding, which is specifically implemented in the configuration scenario of an automotive parts handling workstation. In this application scenario, an electrical engineer from a manufacturing company needs to deploy a robot workstation for handling engine blocks. The user inputs the initial requirement via voice: "I need a robot workstation that can handle engine blocks." The natural language interaction module first preprocesses the voice signal, including noise reduction and speech recognition, converting it into structured text data and transmitting it to the large-scale model understanding and multi-turn dialogue module. This module uses a pre-trained large-scale model trained with expertise in the field of industrial robotics. Through semantic parsing and intent recognition of the text, it identifies the missing key parameter information in the current requirement. The system generates targeted follow-up questions based on a dialogue state tracking mechanism: "Please specify the weight range, handling distance, and cycle time requirements for the engine block," and presents this to the user through the natural language interaction module. The user supplements the response: "Weight approximately 50 kg, handling distance 5 meters, and at least 60 handling operations per hour." The large model continuously extracts key parameters through multi-turn dialogue management technology until it fully obtains the workstation's function type, including material handling, object weight parameters, workspace constraints, and production cycle performance indicators.

[0020] Subsequently, the industrial robot workstation automatic configuration module initiates a knowledge base query process, calling upon the industrial robot model parameter library, fixture type library, workstation layout rule library, and handling process parameter library stored in the knowledge base module. Based on constraint satisfaction problem-solving methods, the system first filters robot models with a rated load greater than 50 kg based on load requirements, and determines the applicable model by combining the working radius parameter. In the end-effector matching stage, a case-based reasoning mechanism is used to select a suitable dedicated fixture for engine block handling from vacuum suction cups and mechanical grippers based on workpiece characteristics. In the workstation layout planning stage, collision detection algorithms and reachability analysis are used to generate the optimal spatial relationship between the robot base and the loading / unloading points. The motion trajectory generation module calculates an interference-free optimized trajectory based on path planning algorithms, while the process parameter setting module calculates the motion speed and acceleration parameters of each axis based on cycle time requirements. Finally, the robot control program and workstation configuration file, including logic control instructions and safety interlock settings, are automatically generated.

[0021] The configuration plan was converted into an easily understandable description using natural language generation technology: "A six-axis industrial robot with a load capacity of 80 kg is recommended, equipped with an adaptive vacuum suction cup gripper, a working radius of up to 6 meters, and a maximum handling capacity of 72 times per hour." After review, the user suggested modifications: "A visual inspection function needs to be added." The system used semantic understanding to parse the new requirement, restarted the configuration process, matched the vision sensor model in the knowledge base and integrated it into the configuration plan, generating a new robot program containing visual positioning instructions. After user confirmation, the execution and feedback module downloaded the control program to the robot controller via the OPC UA protocol, driving the robot body, vision system, and conveyor belt equipment to work collaboratively. The system collects real-time operational data such as robot joint torque, motor current, and vision processing results, and monitors the operational status through anomaly detection algorithms. When the vision system detects a workpiece position deviation, the execution and feedback module immediately adjusts the robot trajectory and generates a natural language prompt: "Workpiece position deviation detected; gripping position automatically adjusted." All operational data, including the number of tasks completed, equipment operating status, and abnormal events, are converted into natural language reports and fed back to the user. Simultaneously, the system continuously collects actual operating parameters and user operation records, updating the process parameters in the knowledge base through an incremental learning mechanism to optimize the accuracy and adaptability of subsequent configuration schemes, forming a closed-loop optimization system. The entire implementation process demonstrates how natural language interaction can lower the professional threshold for configuring robot workstations, achieving an intelligent transformation from requirements description to system deployment.

[0022] In a real-world application scenario, an electrical engineer in a factory needs to configure an industrial robot assembly workstation for assembling small electronic components.

[0023] 1. Initial user input: The user says to the system via voice, "I need to configure a robot workstation for assembling small electronic components." The natural language interaction module converts the speech into text and transmits it to the large model understanding and multi-turn dialogue module.

[0024] 2. Obtain complete requirements through multi-turn dialogue: After the large model understanding and multi-turn dialogue module analyzes the data, the natural language interaction module asks follow-up questions: "What is the maximum size of the electronic component that needs to be assembled? What is its approximate weight?" Answer: "The largest part is 5 centimeters in size and weighs about 20 grams." The large model understanding and multi-turn dialogue module continues to ask: "Where should the two parts be placed during assembly? Where should the finished product be placed after assembly?" Answer: "Part A is placed on the left tray, part B is placed on the right tray, and the assembled finished product is placed on the middle conveyor belt." The large model understanding and multi-turn dialogue module confirms the completeness of the requirements: Assemble small electronic parts weighing less than 5 cm and about 20 grams. Part A is on the left tray, part B is on the right tray, and the finished product is placed on the middle conveyor belt.

[0025] Generate and feedback configuration scheme: The industrial robot workstation automatic configuration module calls the knowledge base, selects a small collaborative robot of model XX with a load capacity of 500 grams and a working radius of 800 mm, equipped with a precision pneumatic gripper as the end effector. Plan workstation layout: The robot is positioned between the material tray and the conveyor belt, generating the motion trajectory of picking up part A from the left material tray, picking up part B from the right material tray, assembling them, and placing them on the middle conveyor belt. The assembly speed is set to 1 time per second, with an accuracy of ±0.1 mm, and a control program is generated.

[0026] The large model understanding and multi-turn dialogue module converts the configuration scheme into natural language feedback: "The configuration scheme is for a collaborative robot of model XX, equipped with a precision pneumatic gripper, which can assemble electronic parts with a maximum size of 5 cm and a weight of 20 grams. Take part A from the left tray and part B from the right tray, assemble them, and place them on the middle conveyor belt. It can complete one assembly per second with an accuracy of ±0.1 mm. Are you satisfied?" Solution Confirmation and Execution: The system says, "Can the assembly speed be increased to 1.5 times per second?" The system readjusts its configuration, setting the assembly speed to 1.5 times per second, and sends feedback for confirmation. After confirmation, the execution and feedback module loads the control program, and the robot begins work. The execution and feedback module provides real-time feedback: "Robot is running; 50 assembly operations have been completed so far; operation is normal." As can be seen from this example, an electrical engineer, without any professional robotics training, completed the configuration of the industrial robot assembly workstation solely through natural language interaction. The entire process was simple and efficient, fully demonstrating the advantages of this invention.

[0027] In the precision assembly scenario of automotive parts, the large-scale model understanding and multi-turn dialogue module demonstrates its core value. When an assembly line engineer raises the initial requirement of "configuring a robotic workstation for assembling gearbox gear sets," the large-scale language model, fine-tuned with expertise in the industrial robotics field, immediately initiates the semantic parsing process. Through domain-adaptive pre-training, this model has deeply integrated professional terminology and operational specifications of assembly processes, enabling it to accurately identify the key intent of "gearbox gear set assembly." The model employs a deep architecture based on transformers, encoding the input text through a self-attention mechanism to capture the semantic features and interrelationships of key concepts such as "precision assembly" and "gear set." In the intent recognition stage, the model invokes a pre-trained industrial scenario classifier to accurately categorize the requirement as "precision assembly" and activates the corresponding parameter extraction template.

[0028] Since the initial requirements lacked key process parameters, a multi-turn dialogue management mechanism was activated using a dynamic reasoning framework based on intent-slot. The dialogue state tracker maintained the current dialogue context, identifying missing essential parameters including assembly precision requirements, gear set size range, and production cycle time. Entity relation extraction technology extracted "gearbox" as the core entity from existing statements and generated precise clarifying questions: "Please specify the assembly precision requirements, maximum gear set size, and hourly production capacity requirements." When the user responded "micron-level assembly precision is required," the model, through semantic parsing, recognized this as an ultra-high precision requirement, immediately triggering a dedicated follow-up questioning process for precision assembly, further inquiring about professional parameters such as "whether force-controlled assembly functionality is needed" and "visual guidance requirements."

[0029] Throughout the multi-round dialogue, the model continuously employs an industry knowledge-enhanced dialogue management strategy. Each follow-up question is generated based on a process knowledge graph, ensuring logical coherence and technical relevance between questions. When the user provides information that "the gear set contains multiple gears of different sizes," the model identifies this as a complex assembly task through relation extraction and automatically adds inquiries about the assembly sequence of each gear and specific process requirements. Through this hierarchical dialogue management, the system ultimately obtains all key parameters, including assembly accuracy level, component size range, force control requirements, vision system requirements, and production cycle time, laying a solid foundation for subsequent automated configuration.

[0030] In the configuration scenario of an aerospace composite material laying workstation, the knowledge base module, constructed using an ontology modeling approach, plays a crucial role in building a knowledge graph for the industrial robot domain. This knowledge graph, built using the OWL ontology language, comprises four core sub-libraries: a robot ontology model library, an end effector function library, a process rule library, and a layout constraint library. The robot ontology model library employs a unified ontology-based modeling method to formally represent the structural parameters, motion performance, and working range of various industrial robots. Each robot model is described as an instance with clearly defined attribute constraints. The end effector function library uses functional ontology to categorize various actuators, establishing semantic associations such as "vacuum suction cup - suitable for flat surfaces" and "flexible clamp - suitable for curved workpieces."

[0031] When the system receives a request for composite material wing skin installation, the automatic configuration module initiates a constraint-based reasoning mechanism. The reasoning engine first extracts key constraints from the requirements, including material properties, surface curvature requirements, and installation accuracy standards. Then, it iterates through the knowledge base, examining equipment parameters and process rules to perform multi-level matching calculations. During robot selection, the system comprehensively considers workspace constraints and accuracy requirements, eliminating models with insufficient working radius, and then filters out candidate models that meet the accuracy criteria. In the end effector matching stage, a non-metallic vacuum suction cup is selected based on the composite material characteristics, and an actuator with compliant adjustment capabilities is determined based on the surface installation requirements.

[0032] The workstation layout planning employs a knowledge-based reasoning optimization algorithm to determine the relative positional relationship between the robot and the tiling mold according to the tiling process requirements. The motion trajectory generation module, considering the process characteristics of composite material tiling, generates interference-free tiling paths based on collision detection and accessibility analysis. The system specifically considers the unique requirements of composite material tiling, incorporating constant pressure control points and material tension control parameters into the trajectory planning. The final generated robot control code includes a complete tiling process: from material pickup and positioning to compaction and leveling, each action is equipped with corresponding force control parameters and visual correction instructions. The workstation configuration parameter set details the collaborative working parameters and safety interlock logic of each device, ensuring the entire workstation operates efficiently and reliably.

[0033] In the application scenario of precision testing workstations for electronic products, the execution and feedback module is deeply integrated with the control system through industrial standard communication protocols. This module uses the OPC UA unified architecture to establish real-time data connections with the robot controller, vision system, and conveyor belt controller, building a complete equipment communication network. The data acquisition system obtains the robot's joint positions, motor torque, vision system detection results, and conveyor belt operating status in real time through a subscription mechanism, forming a comprehensive view of the equipment's operating status.

[0034] The temporal pattern recognition algorithm performs in-depth analysis of the collected operational data to establish a multivariate temporal model under normal production conditions. When performing mobile phone motherboard inspection tasks, the system monitors in real time the temporal characteristics of the robot's pick-and-place actions, the processing time of visual inspection, and the conveyor belt's operating rhythm. The anomaly detection algorithm employs a deep learning-based multimodal anomaly detection model, simultaneously analyzing both numerical and image data to promptly detect subtle operational deviations. For example, when the visual inspection time abnormally increases, the system immediately analyzes possible causes: these could be changes in lighting conditions, lens contamination, or product positioning deviation.

[0035] Natural language generation technology transforms complex operational states into interpretable multimodal feedback information. When an extended robot pick-and-place cycle is detected, the system generates a clear prompt: "The robot arm's pick-and-place cycle is longer than the standard value; it is recommended to check the vacuum generator performance." When the vision system detects multiple consecutive products with the same type of defect, it generates a diagnostic report: "Multiple consecutive products have been found to have solder bridging defects; it is recommended to check the soldering process parameters." This feedback information is presented to the operator in a multimodal manner, combining a graphical interface and voice prompts, providing both detailed numerical information and clear handling suggestions.

[0036] The system also establishes a real-time monitoring system based on digital twins, displaying the operational status of the entire workstation through 3D visualization. When equipment malfunctions, the system not only highlights the malfunctioning device on the interface but also describes the scope and urgency of the malfunction in natural language, providing operators with comprehensive decision support. This in-depth status monitoring and intelligent feedback mechanism significantly improves equipment utilization and production quality.

[0037] In the optimization scenario of a white goods painting production line, the knowledge base update mechanism demonstrates its continuous learning capability. The system automatically collects the operating data of the painting robot and the configuration parameters of the workstation through incremental learning technology, including key process data such as painting path, paint film thickness detection results, and paint consumption. After each production shift, the data preprocessing module automatically cleans and extracts features from the collected data, removing outliers and noisy data, and retaining a representative set of process parameters.

[0038] Knowledge graph embedding technology employs a representation learning method based on graph neural networks to map equipment attributes, process parameters, and performance indicators into a low-dimensional vector space. This embedding representation can capture complex relationships between entities, such as "the optimal matching parameters between robot model-A and paint-B" or "the nonlinear relationship between spraying distance and paint film thickness." When new spraying process data arrives, the system discovers new patterns or optimization spaces not recorded in the existing knowledge graph through similarity calculation of the embedded vectors.

[0039] The feature extraction module employs an autoencoder architecture to learn latent feature representations from operational data. For example, it extracts the optimal combination of multiple parameters such as spray gun movement speed, atomization pressure, and paint viscosity from successful spraying cases. The knowledge fusion mechanism integrates these newly discovered knowledge with the existing knowledge graph, enriching and refining the knowledge base through entity alignment and relational reasoning. When the system discovers that a new type of paint achieves better spraying results when using specific parameters, it automatically updates the relevant parameter recommendations in the process rule base.

[0040] The dynamic update mechanism employs a confidence-based weighted update strategy, assigning higher weight to reliable knowledge that has been verified multiple times, while adopting a cautious adoption strategy for single instances of anomalous data. The equipment attribute association update module continuously optimizes equipment selection recommendation rules by analyzing a large amount of operational data. For example, if long-term operational data reveals that a certain model of robot performs more stably in continuous spraying operations, this model will be prioritized for equipment recommendations. Process constraints are also continuously refined with the accumulation of practical experience, adding new constraints or relaxing unnecessary restrictions. This continuous self-optimization mechanism enables the system to constantly adapt to new production demands and technological developments, always maintaining the optimal performance of the configuration scheme.

[0041] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for natural language understanding based robot intelligent interaction, characterized in that, The method includes the following steps: S1: The natural language interaction module receives the user's initial request in the form of voice or text, performs noise reduction and speech recognition processing on the voice signal to convert it into text data, and transmits the text data to the large model understanding and multi-turn dialogue module. S2: The large model understanding and multi-turn dialogue module uses a large model specifically trained in the field of industrial robots to perform semantic parsing and intent recognition on text data. If the initial requirement information is incomplete or the parameters are ambiguous, it automatically generates targeted follow-up questions and completes the requirement information through multi-turn natural language dialogue interaction until it obtains the complete workstation function type, physical parameters of the processed object, working environment constraints and performance index requirements. S3: The industrial robot workstation automatic configuration module calls the industrial robot model and parameter library, end effector type library, workstation layout rule library and process parameter library in the knowledge base module according to the obtained requirements, automatically matches the robot model and end effector, generates workstation layout scheme, motion trajectory planning and process parameter settings, and automatically generates robot control program and workstation configuration file. S4: Convert the configuration scheme into a natural language description and feed it back to the user. If the user suggests modifications, receive and parse the user feedback through natural language interaction, and control the automatic configuration module to regenerate the configuration scheme until the user confirms it. S5: The execution and feedback module loads the user-confirmed control program and configuration file into the industrial robot control system, drives the industrial robot and peripheral equipment to perform tasks, and collects operating status data in real time, including work progress and fault information. S6: Converts runtime status data into natural language descriptions and feeds them back to the user. During system operation, it continuously collects user needs, configuration schemes, and runtime data, dynamically updates the knowledge base module, and achieves self-optimization of the knowledge base and continuous improvement of configuration accuracy.

2. The method for robot intelligent interaction based on natural language understanding according to claim 1, characterized in that: The large-scale model understanding and multi-turn dialogue module adopts a large language model that has been fine-tuned with professional knowledge in the field of industrial robots. The large language model is optimized through domain-adaptive pre-training and human feedback reinforcement learning, and has the ability to perform semantic parsing and multi-turn dialogue management in industrial scenarios. The multi-turn dialogue adopts a dynamic reasoning mechanism based on intent recognition and slot filling. It generates accurate clarifying questions through dialogue state tracking and entity relationship extraction technology to ensure the complete extraction of all key process parameters and operational constraints. 3.The method of claim 1, wherein: The domain-adaptive pre-training uses industrial robot technical manuals, process specification documents, and fault diagnosis reports to build a pre-training corpus of related domain texts. Domain vocabulary expansion and semantic enhancement techniques are used to improve the model's understanding accuracy of professional terms and process parameters. The human feedback reinforcement learning is designed with a multi-dimensional reward function for robot programming scenarios, taking into account grammatical correctness, technical specification compliance, and operational safety. The training dataset is constructed by ranking the generated results by preference. The multi-turn dialogue management adopts a knowledge graph-based dialogue state tracking mechanism to establish a slot filling framework that includes parameters such as robot working range, load capacity, and accuracy indicators.

4. The method for robot intelligent interaction based on natural language understanding according to claim 1, characterized in that: In step S3, the knowledge base module uses ontology modeling to construct a knowledge graph for the industrial robot domain. The industrial robot domain knowledge graph includes a robot ontology model library, an end effector function library, a process rule library, and a layout constraint library. The automatic configuration module uses a reasoning mechanism based on constraint satisfaction problems. By traversing the equipment parameters and process rules in the knowledge base, it performs matching calculations of the robot's working radius, load capacity, repeatability accuracy, and task requirements. Based on collision detection and reachability analysis, it generates interference-free motion trajectories and finally synthesizes directly deployable robot control code and workstation configuration parameter sets.

5. The method for robot intelligent interaction based on natural language understanding according to claim 1, characterized in that: The execution and feedback module is integrated with the control system through an industrial standard communication protocol to collect equipment operating status data in real time. It processes the operating data using time-series pattern recognition and anomaly detection algorithms, and converts equipment status, task progress, and fault diagnosis results into interpretable multimodal feedback information through natural language generation technology, providing users with real-time monitoring and decision support.

6. The method for intelligent robot interaction based on natural language understanding as described in claim 1, characterized in that: The knowledge base update mechanism adopts incremental learning and knowledge graph embedding technology. The system automatically collects operation data and user interaction records, continuously optimizes domain knowledge representation through feature extraction and knowledge fusion, and dynamically updates equipment attribute associations and process constraint relationships to achieve adaptive improvement of configuration strategies and continuous improvement of system performance.

Images