Robot programming data analyzing and processing platform

Abstract

The invention relates to a robot programming data analysis and processing platform, which integrates the functions of multi-modal data input and preprocessing, motion trail code generation based on a Transform model, code post-processing and verification and virtual simulation feedback. The platform can receive and process voice, character and picture input of a user and uniformly convert the voice, character and picture input into a standardized token sequence, so that the consistency and accuracy of data processing are improved, a Transform model efficiently extracts global and local features of the input sequence by using a multi-head self-attention mechanism, a motion track code matched with the user input is generated, and the user experience is improved. According to the platform, the code generation efficiency and accuracy are ensured, a user can visually observe the motion trail of the robot through the virtual simulation and feedback module, and the platform has adaptive learning and continuous optimization capabilities, can keep high performance in complex and changeable application scenes, and is widely applicable to various robot programming tasks.

Description

Technical Field

[0001] The present invention relates to a robot programming data analysis and processing platform, and in particular to a robot programming data analysis and processing platform applied in the field of robot programming technology. Background Art

[0002] With the rapid development of artificial intelligence and robotics, robot programming technology has also been widely used. However, most existing robot programming methods rely on programmers to manually write codes, which is a complex and error-prone process. At the same time, traditional robot programming tools lack a unified processing method when processing multimodal input data (such as voice, text, and pictures), resulting in low data conversion and processing efficiency. In addition, the generated code often lacks an effective automatic verification mechanism, which may cause logical errors or instability in the actual execution of the code.

[0003] Chinese invention patent CN110032363B discloses a data structure construction method and system for robot programming. The method obtains the abstract syntax tree generated by the original code and performs secondary packaging to clean up useless data to simplify the data structure. This method reduces the difficulty of implementing robot programming and shortens the programming time. However, the patent mainly focuses on the construction and optimization of the syntax tree, and lacks in-depth discussion on how to process multimodal input data and generate high-quality code. In addition, the automatic verification and optimization mechanism of the code has not been fully considered.

[0004] Chinese invention patent CN111061465B discloses a reverse mapping method and system for visual robot programming. By collecting path data during the robot's movement and generating action block sequences in the programming interface, intuitive interaction of robot programming is achieved. Although this method simplifies the programming process and improves the user experience, it is mainly used for path planning and action sequence generation, and has limited processing capabilities for complex logical control and multimodal data. In addition, this method has deficiencies in the verification and optimization links after code generation, and it is difficult to ensure the stability and reliability of the generated code in complex application scenarios.

[0005] The above design improves the ease of use and efficiency of robot programming to a certain extent by optimizing the data structure and providing a visual programming interface, but there are still some limitations. For example, there is a lack of unified processing of multimodal data, and it is unable to effectively deal with data fusion problems in complex tasks. At the same time, the automatic verification and optimization mechanism of the code is not perfect, which may lead to errors or instability in the actual execution of the generated code. In addition, the existing programming tools often require a lot of manual adjustments when dealing with complex application scenarios and lack adaptive optimization capabilities. Summary of the invention

[0006] In view of the above-mentioned prior art, the technical problem to be solved by the present invention is how to uniformly process multimodal data, efficiently generate motion trajectory codes, enhance code verification and automatic repair, provide intuitive simulation feedback and adaptive learning and continuous optimization.

[0007] To solve the above problems, the present invention provides a robot programming data analysis and processing platform, including a multimodal data input module, a data preprocessing module, a serialization and encoding module, a Transformer model, a code post-processing and verification module, and a virtual demonstration and feedback module;

[0008] Multimodal data input module, used to receive user's voice, text and picture input;

[0009] A data preprocessing module, used to preprocess multimodal input data;

[0010] The data preprocessing module includes converting speech input into text, performing word segmentation, part-of-speech tagging, and syntactic analysis on text input, and feature extraction on image input;

[0011] The serialization and encoding module is used to convert the preprocessed multimodal data into a token sequence and add position information;

[0012] Transformer model, used to generate motion trajectory codes corresponding to user input based on token sequences;

[0013] Code post-processing and verification module, used to perform format and logic verification on the generated motion trajectory code;

[0014] The virtual demonstration and feedback module is used to display the generated motion trajectory in a simulation environment and perform iterative optimization based on user feedback.

[0015] In the above-mentioned robot programming data analysis and processing platform, a multimodal data input and preprocessing module is used to realize unified processing and conversion of different forms of input such as voice, text and pictures, thereby improving the consistency and accuracy of data processing. By using the Transformer model, the present invention can efficiently extract features from multimodal input data and generate motion trajectory codes that match user input, thereby improving the efficiency and accuracy of code generation. A code post-processing and verification module is integrated to ensure that the generated code is correct in syntax and logic through syntax parsing, logic verification and automatic repair mechanisms, thereby significantly improving the correctness and execution reliability of the code. Through the virtual simulation and feedback module, users can observe the execution effect of the generated code in real time, and optimize the code generation process through feedback iteration, thereby improving the intelligence level of the system and user experience. Through online learning and iterative optimization mechanisms, the code generation capability can be continuously optimized according to user feedback and simulation results, adapting to complex and changeable task scenarios, thereby ensuring that the generated code has high stability and performance in various application environments.

[0016] As a further improvement of the present application, the multimodal data input module includes a voice input unit, a text input unit and a picture input unit;

[0017] The voice input unit is used to receive the user's voice input; the text input unit is used to receive the user's text input; the picture input unit is used to receive the user's picture input.

[0018] As a further improvement of the present application, the data preprocessing module further includes a speech recognition unit, a natural language processing unit and a computer vision processing unit;

[0019] The speech recognition unit is used to convert speech input into text and perform cleaning and normalization processing; the natural language processing unit is used to perform word segmentation, part-of-speech tagging, syntactic parsing and semantic understanding on the text input; the computer vision processing unit is used to perform feature extraction and target detection on the image input.

[0020] As a further improvement of the present application, the serialization and encoding module is used to convert the preprocessed text, voice and image data into a token sequence, and add sequence order information through a position encoding mechanism.

[0021] As another improvement of the present application, the Transformer model includes an encoder and a decoder;

[0022] The encoder is used to extract global and local features of the input sequence through a multi-head self-attention mechanism; the decoder is used to gradually generate a sequence of target motion trajectory codes based on the feature representation generated by the encoder through multi-head self-attention and encoder-decoder attention mechanisms.

[0023] As another improved supplement of the present application, the output layer of the Transformer model is used to map the sequence generated by the decoder into specific motion trajectory code instructions, and calculate the probability distribution through the Softmax layer to select the optimal token.

[0024] As another improved supplement of the present application, the code post-processing and verification module includes a syntax parsing unit, a logic verification unit and an automatic repair unit;

[0025] The syntax parsing unit is used to perform syntax checking on the generated motion trajectory code; the logic verification unit is used to check the correctness of the control flow and operation logic in the code; the automatic repair unit is used to identify and correct common errors in the code.

[0026] As another improvement of the present application, the virtual demonstration and feedback module includes a simulation environment configuration unit, a motion trajectory simulation unit and a user feedback collection unit;

[0027] The simulation environment configuration unit is used to create and configure the virtual simulation environment; the motion trajectory simulation unit is used to execute the generated motion trajectory code in real time and display the motion process; the user feedback collection unit is used to collect user feedback on the simulation results and perform iterative optimization.

[0028] The data preprocessing module, Transformer model, code post-processing and verification module, and virtual demonstration and feedback module work together to continuously improve the accuracy and reliability of the generated motion trajectory code through online learning and iterative optimization mechanisms.

[0029] A robot programming data analysis and processing platform, the working method of which comprises the following steps;

[0030] S1: Receive and preprocess multimodal data, use the multimodal data input module and data preprocessing module to process the voice, text and image data input by the user, and convert them into token sequences;

[0031] S2: Add position information to the token sequence through the serialization and encoding module to generate serialized data suitable for Transformer model processing;

[0032] S3: Input the serialized data into the Transformer model to generate the motion trajectory code corresponding to the user input;

[0033] S4: Verify the syntax and logic of the generated motion trajectory code to ensure its executable;

[0034] S5: Execute the motion trajectory code in the simulation environment, and iteratively optimize the code and simulation process based on user feedback until user needs are met.

[0035] In summary, this application has the following beneficial effects:

[0036] 1. Improve the consistency and accuracy of data processing. The multimodal data input and preprocessing module integrates three data input forms: voice input, text input, and image input. It also standardizes different forms of data through voice recognition, natural language processing, and computer vision technology. Through the serialization and encoding module, these data are uniformly converted into token sequences and the location information is retained, so that subsequent processing modules can consistently and accurately process data from multiple sources. This unified preprocessing method significantly improves the consistency of data processing and ensures the accuracy of input data in subsequent analysis and processing.

[0037] 2. Efficient feature extraction and code generation. The Transformer model can efficiently extract global and local features from the input token sequence through the multi-head self-attention mechanism, and generate motion trajectory codes that match the user input. The collaborative work of the encoder and decoder enables the generated code to accurately reflect the complex relationship of the input data. In addition, the Softmax layer of the output layer ensures the generation of optimal code by calculating the probability distribution of the generated sequence. This code generation method based on the Transformer model not only improves the efficiency of motion trajectory code generation, but also ensures the accuracy and logic of the code, and is suitable for complex robot programming tasks.

[0038] 3. Enhance the correctness and execution reliability of the code. The code post-processing and verification module provides a comprehensive code inspection and correction mechanism through syntax parsing, logic verification and automatic repair unit. This module ensures that the generated motion trajectory code is syntactically correct and can be executed correctly logically, thereby greatly reducing the risk of errors that may occur during the execution process. The automatic repair unit can identify and correct common code errors, making the code generation process more intelligent and efficient, further enhancing the reliability of the code.

[0039] 4. Provide intuitive simulation feedback and user participation. The virtual demonstration and feedback module provides an intuitive simulation environment through simulation environment configuration, motion trajectory simulation and user feedback collection unit. Users can observe the motion trajectory of the robot after executing the code in real time. Through simulation feedback, users can promptly discover potential problems in the code and evaluate the execution effect of the code. User feedback is effectively integrated into the iterative optimization process of code generation to ensure that the final generated code can meet actual needs and expected effects. Through this feedback mechanism, users can actively participate in the optimization process of the code, which improves the system's ease of use and user satisfaction.

[0040] 5. Adaptive learning and continuous optimization. Through online learning and iterative optimization mechanisms, this platform continuously uses user feedback and simulation results to optimize the code generation capabilities of the Transformer model. The platform can adapt to changing task scenarios and user needs, and gradually improve the quality and execution effect of code generation. This adaptive learning capability enables the system to maintain high working performance in various complex environments, ensuring that the generated code always meets the requirements of actual application scenarios.

[0041] 6. Expand the flexibility and breadth of applications. The modular design and functional integration of the present invention make the platform highly flexible and able to adapt to different types of robot programming tasks. Whether in industrial automation, service robots or education, the platform can provide efficient programming support. In addition, the system has strong scalability and can be customized and adjusted according to specific application requirements, further expanding its wide application in various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is the overall interactive diagram of this application;

[0043] Figure 2 Local interaction for this application Figure 1 ;

[0044] Figure 3 Local interaction for this application Figure 2 ;

[0045] Figure 4 Local interaction for this application Figure 3 . DETAILED DESCRIPTION

[0046] Three embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0047] Example 1

[0048] like Figures 1 to 4 As shown, this embodiment relates to a multimodal data input and preprocessing part of a robot programming data analysis and processing platform, which aims to receive and process various input forms of users (voice, text, pictures) and convert them into standardized data sequences that can be used by subsequent data processing modules.

[0049] The platform includes a multimodal data input module for receiving voice, text and image data input by users, and specifically includes the following units:

[0050] Voice input unit: used to receive user voice input, and can collect voice data through a microphone or other audio input devices.

[0051] Text input unit: used to receive text input from the user, such as input through the keyboard or reading text from a file.

[0052] Image input unit: used to receive user image input and supports obtaining image data from cameras, image files or scanning devices.

[0053] The data preprocessing module is used to perform preliminary processing on the received multimodal data to ensure the uniformity and standardization of the data. The module includes the following subunits:

[0054] Speech recognition unit: used to convert the received speech data into text, using advanced speech recognition technologies such as a combination of convolutional neural networks (CNN) and long short-term memory networks (LSTM). The converted text will be cleaned to remove noise and non-essential information to obtain pure text data.

[0055] Natural language processing unit: used to perform further semantic processing on the text data received or generated by the speech recognition unit. Specific operations include word segmentation, part-of-speech tagging, syntactic parsing and semantic understanding to ensure the structuring and semantic correctness of the text data.

[0056] Computer vision processing unit: used to extract features from received image data, use convolutional neural network (CNN) model for image processing and target detection, identify key objects in the image and their location information, and convert this information into structured data for subsequent processing.

[0057] The serialization and encoding module converts the multimodal data generated by the data preprocessing module into a standardized token sequence and adds location information to each token. This process includes:

[0058] Serialization processing: Convert the results of text, speech-to-text, and image feature extraction into token sequences. These tokens represent the basic units of data.

[0059] Position encoding mechanism: Add position information to each token to ensure that the order and contextual dependencies of the data in the sequence are maintained. This mechanism is usually implemented by adding sine and cosine function values.

[0060] Working process

[0061] Data input: Users provide data in various forms through voice input units, text input units or image input units. Voice is input through a microphone, text is input through a keyboard or read from a file, and images are taken through a camera or imported from a file.

[0062] Speech recognition and conversion: The speech data received by the speech input unit is processed by the speech recognition unit and converted into corresponding text. The recognition process includes noise reduction, feature extraction, model matching and other steps, and finally generates clear text data.

[0063] Text preprocessing: Whether it is directly input text or text generated by speech recognition, it is segmented, POS tagged and parsed through the natural language processing unit to ensure semantic integrity and consistency.

[0064] Image feature extraction: The image data received by the image input unit is processed by the computer vision processing unit, which uses a deep learning model to identify the target object in the image and extract its location information.

[0065] Data serialization and encoding: All preprocessed data (text, speech-to-text, image features) are converted into a unified token sequence through the serialization and encoding module. Each token represents a data unit and is accompanied by its position information in the sequence.

[0066] Data transmission: The processed token sequence will be transmitted to the Transformer model module for subsequent motion trajectory code generation.

[0067] Improve the consistency of data processing: Through a unified preprocessing process, data in various input forms are standardized into token sequences, ensuring that subsequent data processing modules can consistently process data from different sources.

[0068] Enhance data understanding capabilities: Through advanced speech recognition, natural language processing, and computer vision technologies, deep analysis of input data is performed to improve the semantic understanding of data and provide a more accurate basis for subsequent code generation.

[0069] Optimize user experience: Users can input commands through voice, text or pictures, and the system can process this data intelligently, improving the platform's ease of use and flexibility.

[0070] Ensure data order and context preservation: The position information added by the serialization and encoding modules ensures that the order and context information during data processing are preserved, allowing the subsequent Transformer model to generate motion trajectory codes more accurately.

[0071] Example 2

[0072] like Figures 1 to 4 As shown, this embodiment relates to a robot programming data analysis and processing platform based on the Transformer model, which focuses on generating motion trajectory codes from preprocessed data and continuously improving the accuracy and reliability of the codes through an iterative optimization mechanism.

[0073] The core of the platform is the Transformer model, which includes the following components:

[0074] Encoder: It is used to extract global and local features from the input token sequence through a multi-head self-attention mechanism.

[0075] Multi-head self-attention mechanism: The key part of the encoder, which uses multiple parallel attention heads to focus on different parts of the input sequence and extract important features in the sequence.

[0076] Feedforward neural network: After the multi-head self-attention mechanism in each layer, the extracted features are transformed nonlinearly through the feedforward neural network to further enhance the expressive power of feature representation.

[0077] Decoder: It is used to gradually generate a sequence of target motion trajectory codes based on the feature representation generated by the encoder through multi-head self-attention and encoder-decoder attention mechanism.

[0078] Masked Multi-Head Self-Attention Mechanism: During the decoding process, the masking mechanism ensures that the decoder can only access part of the generated sequence to avoid information leakage.

[0079] Encoder-Decoder Attention Mechanism: The decoder not only pays attention to the sequence it generates, but also pays attention to the feature representation generated by the encoder through the attention mechanism, thereby generating a code sequence corresponding to the input data.

[0080] Output layer: used to map the sequence generated by the decoder into specific motion trajectory code instructions.

[0081] Softmax layer: The output layer calculates the probability distribution of the generated sequence through the Softmax layer, selects the optimal token as the output, and generates the final motion trajectory code.

[0082] The platform also integrates online learning and iterative optimization mechanisms to improve the quality of motion trajectory code generation and the overall performance of the system.

[0083] The data preprocessing module, Transformer model, code post-processing and verification module, and virtual demonstration and feedback module work together: the modules work together to achieve continuous learning of data and iterative optimization of the model.

[0084] Online learning: The platform continuously updates and optimizes the Transformer model based on user input and feedback to improve the quality of motion trajectory code generation.

[0085] Iterative optimization: The accuracy and execution effect of the generated code are gradually improved through multiple cycles of code generation, verification, simulation demonstration, and user feedback collection.

[0086] Working process

[0087] Data input and preprocessing: From Example 1, the data processed by the data preprocessing module (already converted into a token sequence with position information attached) is passed to the Transformer model.

[0088] The encoder receives the serialized and encoded token sequence and uses the multi-head self-attention mechanism to extract global and local features in the sequence.

[0089] The multi-head self-attention mechanism captures important dependencies in the sequence by focusing on different parts of the input sequence and further processes the features through a feed-forward neural network.

[0090] The encoder generates high-level feature representations and passes them to the decoder for further processing.

[0091] Based on the feature representation generated by the encoder, the decoder uses a masked multi-head self-attention mechanism to generate a target motion trajectory code sequence.

[0092] The decoder also focuses on the features of the encoder output through the encoder-decoder attention mechanism to ensure that the generated code sequence is consistent with the contextual information of the input sequence.

[0093] The decoder gradually generates the motion trajectory code sequence, calculates the probability distribution through the Softmax layer of the output layer, and selects the optimal token as the final code output.

[0094] The generated motion trajectory code is transmitted to the code post-processing and verification module, and the syntax is checked by the syntax parsing unit to ensure that the code format is correct.

[0095] The logic verification unit checks the correctness of the code's control flow and operation logic, identifies and fixes common errors, and ensures code executableness.

[0096] The code is simulated and run in a simulation environment through a virtual demonstration and feedback module to observe the robot's motion trajectory and evaluate the accuracy of code generation.

[0097] Based on the simulation results, user feedback is collected and used in the online learning module to adjust the Transformer model for the next round of optimization.

[0098] Based on user feedback and simulation results, the platform continuously optimizes the Transformer model and generates more accurate motion trajectory code.

[0099] This process is iterated until the generated code achieves the desired performance and accuracy.

[0100] Efficient feature extraction and code generation: Through the multi-head self-attention mechanism, the Transformer model can efficiently extract key features in the input sequence and generate high-quality motion trajectory codes that match the user input.

[0101] Adaptive optimization: Through online learning and iterative optimization mechanisms, the platform can continuously optimize its code generation capabilities based on user feedback, ensuring that the generated code has high accuracy and execution effect in various scenarios.

[0102] Improve execution reliability: After verification by the code post-processing and verification module, the generated code is strictly checked and corrected in terms of format and logic, reducing potential risks during code execution.

[0103] User participation in the optimization process: User feedback is effectively utilized and directly involved in the system optimization process, making the final code generation results closer to user needs and actual application scenarios.

[0104] Example 3

[0105] like Figures 1 to 4 As shown, this embodiment focuses on describing a system and method for code post-processing, verification and virtual simulation feedback for a robot programming data analysis and processing platform. The system implements syntax and logic verification of generated code through a series of modules, and collects user feedback through simulation demonstration, thereby optimizing the accuracy of code generation and execution effect.

[0106] The code post-processing and verification module is responsible for comprehensive inspection and correction of the motion trajectory code generated by the Transformer model, including the following sub-units:

[0107] Syntax parsing unit: This unit performs syntax checking on the generated code to ensure that the code meets the specification requirements of the programming language. During the syntax checking process, keywords, grammatical structures, punctuation marks, and variable declarations are analyzed and verified in detail.

[0108] Logic verification unit: After the syntax check is passed, the logic verification unit further checks the control flow and operation logic of the code. This unit ensures that the logical relationships such as conditional statements, loop structures, function calls and variable assignments in the code are correct to avoid potential execution errors.

[0109] Automatic repair unit: During the syntax parsing and logic verification process, if common errors or unreasonable logical structures are detected, the automatic repair unit will attempt to make automatic corrections. The unit uses predefined repair rules or machine learning models to automatically adjust errors in the code and provide correction suggestions when necessary.

[0110] The virtual demonstration and feedback module creates a virtual simulation environment to execute and display the generated motion trajectory code and collects user feedback for iterative optimization. It specifically includes the following sub-units:

[0111] Simulation environment configuration unit: responsible for creating and configuring a virtual simulation environment. The simulation environment includes a virtual reproduction of the robot's working scene, such as simulation of terrain, obstacles, target objects, and environmental conditions. The unit supports custom environment configuration to test the execution effect of the code in different scenarios.

[0112] Motion trajectory simulation unit: In the configured simulation environment, the motion trajectory simulation unit executes the generated motion trajectory code in real time to display the robot's motion process in the virtual environment. The unit provides a visual motion trajectory demonstration to help users intuitively observe the code execution effect.

[0113] User feedback collection unit: After the motion trajectory simulation is completed, the user feedback collection unit allows the user to evaluate the simulation results and provide feedback. The feedback information may include satisfaction with the motion trajectory, potential problems found, and optimization suggestions. The feedback data collected by this unit will be used in the subsequent code generation optimization process.

[0114] Working process

[0115] The generated motion trajectory code is firstly checked by the syntax parsing unit to ensure that the code format is correct and meets the specification requirements of the programming language.

[0116] After the syntax check passes, the code enters the logic verification unit for detailed logic checking to verify the control flow and operation logic of the code to ensure its correctness.

[0117] If errors or inconsistencies are found during syntax or logic checking, the auto-repair unit will try to automatically correct the code.

[0118] Automatic repair includes correcting syntax errors, adjusting unreasonable logical structures, and ensuring that the corrected code still meets the expected functional requirements.

[0119] The fixed code will be verified again in terms of syntax and logic to ensure the correctness of the correction.

[0120] The verified and repaired code is transmitted to the simulation environment configuration unit, which creates and configures a virtual simulation environment, including simulated terrain, obstacles, target objects, etc., to ensure that the test environment is similar to the actual application scenario.

[0121] The motion trajectory simulation unit executes the code in a simulation environment and displays the robot's motion trajectory in real time. This process visualizes the robot's behavior path and action sequence after executing the code.

[0122] After the simulation is completed, the user feedback collection unit collects user feedback on the simulation results, including satisfaction with the robot's motion, potential problems found, and suggestions for improving the motion trajectory.

[0123] The collected user feedback will be used to guide the further optimization of the Transformer model and adjust the code generation strategy to generate motion trajectory codes that better meet user expectations.

[0124] This process can be iterated multiple times until the generated code meets user requirements and performs well in the simulation environment.

[0125] Improve the correctness and reliability of the code: Through syntax parsing, logic verification and automatic repair, ensure that the generated motion trajectory code is syntactically and logically correct, reducing the risk of errors that may occur during code execution.

[0126] Enhanced code optimization capabilities: The automatic repair unit can automatically identify and correct common errors, making the code generation process more intelligent and efficient.

[0127] Provide intuitive simulation feedback: Through virtual simulation demonstration, users can intuitively observe the behavior of the robot after executing the code, and discover and correct potential problems in time.

[0128] Improve user engagement and code quality: User feedback is effectively integrated into the iterative optimization process of code generation to ensure that the final generated code meets actual needs and expected results.

[0129] Adaptive optimization: Through continuous user feedback and code optimization cycles, the platform gradually improves the quality of code generation and execution results to adapt to complex and changing task scenarios.

[0130] In view of current practical needs, the above-mentioned implementation mode adopted in this application is not limited to the scope of protection. Various changes made within the knowledge scope of technical personnel in this field without departing from the concept of this application still fall within the scope of protection of the present invention.

Claims

1. A robot programming data analysis and processing platform, characterized by: It includes multimodal data input module, data preprocessing module, serialization and encoding module, Transformer model, code post-processing and verification module and virtual demonstration and feedback module; The multimodal data input module is used to receive voice, text and picture input from the user; The data preprocessing module is used to preprocess the multimodal input data; The data preprocessing module includes converting speech input into text, performing word segmentation, part-of-speech tagging, and syntactic analysis on the text input, and performing feature extraction on the image input; The serialization and encoding module is used to convert the preprocessed multimodal data into a token sequence and add position information; The Transformer model is used to generate a motion trajectory code corresponding to the user input based on the token sequence; The code post-processing and verification module is used to perform format and logic verification on the generated motion trajectory code; The virtual demonstration and feedback module is used to display the generated motion trajectory in a simulation environment and perform iterative optimization based on user feedback.

2. A robot programming data analysis and processing platform according to claim 1, characterized in that: The multimodal data input module includes a voice input unit, a text input unit and a picture input unit; The voice input unit is used to receive the user's voice input; the text input unit is used to receive the user's text input; The picture input unit is used to receive picture input from the user.

3. A robot programming data analysis and processing platform according to claim 1, characterized in that: The data preprocessing module further includes a speech recognition unit, a natural language processing unit and a computer vision processing unit; The speech recognition unit is used to convert speech input into text and perform cleaning and normalization processing; The natural language processing unit is used to perform word segmentation, part-of-speech tagging, syntactic parsing and semantic understanding on text input; the computer vision processing unit is used to perform feature extraction and target detection on image input.

4. A robot programming data analysis and processing platform according to claim 1, characterized in that: The serialization and encoding module is used to convert the preprocessed text, voice and image data into a token sequence, and add sequence order information through a position encoding mechanism.

5. The robot programming data analysis and processing platform according to claim 1, characterized in that: The Transformer model includes an encoder and a decoder; The encoder is used to extract global and local features of the input sequence through a multi-head self-attention mechanism; The decoder is used to gradually generate a sequence of target motion trajectory codes based on the feature representation generated by the encoder through multi-head self-attention and encoder-decoder attention mechanisms.

6. A robot programming data analysis and processing platform according to claim 1, characterized in that: The output layer of the Transformer model is used to map the sequence generated by the decoder into specific motion trajectory code instructions, and calculate the probability distribution through the Softmax layer to select the optimal token.

7. The robot programming data analysis and processing platform according to claim 1, characterized in that: The code post-processing and verification module includes a syntax parsing unit, a logic verification unit and an automatic repair unit; The syntax parsing unit is used to perform syntax checking on the generated motion trajectory code; the logic verification unit is used to check the correctness of the control flow and operation logic in the code; The automatic repair unit is used to identify and correct common errors in the code.

8. The robot programming data analysis and processing platform according to claim 1, characterized in that: The virtual demonstration and feedback module includes a simulation environment configuration unit, a motion trajectory simulation unit and a user feedback collection unit; The simulation environment configuration unit is used to create and configure a virtual simulation environment; the motion trajectory simulation unit is used to execute the generated motion trajectory code in real time and display the motion process; The user feedback collection unit is used to collect user feedback on simulation results and perform iterative optimization.

9. The robot programming data analysis and processing platform according to claim 1, characterized in that: The data preprocessing module, Transformer model, code post-processing and verification module and virtual demonstration and feedback module work together to continuously improve the accuracy and reliability of the generated motion trajectory code through online learning and iterative optimization mechanisms.

10. A robot programming data analysis and processing platform according to any one of claims 1 to 9, characterized in that: Its working method The steps include: S1: Receive and preprocess multimodal data, use the multimodal data input module and data preprocessing module to process the voice, text and image data input by the user, and convert them into token sequences; S2: Add position information to the token sequence through the serialization and encoding module to generate serialized data suitable for Transformer model processing; S3: Input the serialized data into the Transformer model to generate the motion trajectory code corresponding to the user input; S4: Verify the syntax and logic of the generated motion trajectory code to ensure its executable; S5: Execute the motion trajectory code in the simulation environment, and iteratively optimize the code and simulation process based on user feedback until user needs are met.

Images