Large model scene drilling system based on AI-VR

Abstract

The invention relates to the technical field of artificial intelligence and virtual reality, in particular to a large model scene drilling system based on AI-VR. The system comprises a student portrait construction module, a dynamic scene generation module, a multi-source data evaluation module and a holographic guide interaction module. According to the invention, the student portrait construction module collects multi-dimensional information and constructs a detailed portrait through clustering analysis, the dynamic scene generation module generates an initial scene model, and the multi-source data evaluation module constructs an operation step dependence graph, so that the operation condition in the learning process of the student is comprehensively and accurately evaluated; the holographic guidance interaction module uses a holographic projection technology to accurately guide students to operate correctly, and meanwhile, the AI large model dynamically adjusts scene difficulty and complexity according to real-time performance of the students, so that individual requirements of the students in different learning stages are met, error sources and influences are deeply analyzed, and a quantitative and scientific basis is provided for teaching evaluation and improved learning of the students.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and virtual reality technology, and more specifically, to a large-scale model scene simulation system based on AI-VR. Background Technology

[0002] With the rapid development of technology, scenario simulation systems based on AI-VR large models have emerged, bringing revolutionary changes to training and emergency simulation in various fields. This system integrates technologies such as intelligent decision-making, natural language processing, computer vision, and machine learning from artificial intelligence (AI) with the immersive environment and interactive design advantages of virtual reality (VR).

[0003] In existing AI-VR scenario training systems, scenario generation often relies on fixed templates or simple parameter adjustments, making it difficult to dynamically generate training environments tailored to the trainees' actual abilities. For example, in medical simulation training, trainees of different levels facing surgical scenarios of the same difficulty may experience frustration for beginners due to excessive difficulty, while advanced trainees may fail to improve their skills due to a lack of challenge. Furthermore, teaching assessments often rely on single-dimensional data, such as operation completion rate and answer accuracy, neglecting the trainees' behavioral details, physiological reactions, and interactions with the scenario during the operation. This makes it impossible to comprehensively and accurately assess the trainees' learning outcomes and potential problems. Therefore, we propose a large-scale AI-VR scenario training system. Summary of the Invention

[0004] The purpose of this invention is to address the problem that in existing AI-VR scene training systems, scene generation often uses fixed templates or simple parameter adjustments, making it difficult to dynamically generate a suitable training environment based on the actual ability of the learners. Teaching evaluation often relies on single-dimensional data, such as operation completion rate and answer accuracy, ignoring the behavioral details, physiological reactions, and interactions with the scene during the operation process, and failing to comprehensively and accurately evaluate the learners' learning effects and potential problems.

[0005] To achieve the above objectives, the present invention provides a large-scale model scene training system based on AI-VR, including a student profile construction module, a dynamic scene generation module, a multi-source data evaluation module, and a holographic guided interaction module;

[0006] The student profile building module collects students' historical learning data, test scores, and operation records, and performs cluster analysis on students' knowledge reserves, skill levels, and learning styles to build a detailed student ability profile.

[0007] The dynamic scene generation module receives natural language instructions input by the trainee, parses the semantics of the instructions through natural language processing technology, extracts key scene elements, combines the trainee profile, retrieves an initial scene model that meets the trainee's requirements from the scene resource library, and optimizes the physical rules of the generated scene.

[0008] The multi-source data evaluation module defines each operation step as a node in the graph and defines node attributes. Based on logical relationships and dependency order, directed edges are established between related nodes to represent the order of operation steps. An operation step dependency graph is established. Based on the operation step dependency graph, the original error point is determined. Combining the error type, the stage of occurrence, and the degree of impact on subsequent operations, the influence weight of each error point is calculated.

[0009] The holographic guided interaction module generates a minimized remedial path based on the original error points and their impact weights, provides students with targeted improvement suggestions, generates holographic projection images of operation guidance instructions and error markers to guide students to perform correct operations, and uses an AI big model to dynamically adjust the difficulty and complexity of the scene based on the students' real-time learning performance.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0011] 1. This AI-VR-based large-scale model scenario training system has a student profile building module that collects multi-dimensional information and uses cluster analysis to build detailed profiles. It can also generate initial profiles and update them dynamically, providing a precise basis for the system's subsequent personalized services, meeting the personalized needs of students at different learning stages, and making learning more targeted.

[0012] 2. The dynamic scene generation module combines natural language processing, part-of-speech and syntactic analysis and named entity recognition technologies to accurately parse instructions, extract elements, generate an initial scene model, and optimize physical rules from aspects such as physical model construction, interactive simulation optimization, and effect visualization enhancement. The generated scene not only meets the needs of students, but also highly restores the real physical environment, providing students with an immersive and personalized learning scenario.

[0013] 3. The multi-source data evaluation module constructs an operation step dependency graph, determines the impact weight of error points by combining error type, occurrence stage and impact degree, comprehensively and accurately evaluates the operation of students in the learning process, deeply analyzes the root causes and impact of errors, and provides quantitative and scientific basis for teaching evaluation and students to improve their learning;

[0014] 4. The holographic guided interaction module generates remedial paths based on error analysis and uses holographic projection technology to accurately guide students to operate correctly. At the same time, the AI ​​big model dynamically adjusts the difficulty and complexity of the scene according to the student's real-time performance to ensure that the scene is adapted to the student's learning progress, improve learning efficiency, and ensure scene coherence and logical rationality during the adjustment process to maintain a good learning experience.

[0015] As a further improvement to this technical solution, the student profile construction module designs diversified basic test questions for different disciplines and training fields. When students use the system for the first time, they generate an initial ability profile by completing the basic test and filling in the learning information form.

[0016] As a further improvement to this technical solution, the student profile construction module establishes a profile update mechanism, which triggers updates at key points such as when students complete important learning tasks or when test scores fluctuate significantly, and collects students' learning data in real time to update the student profile.

[0017] The beneficial effect of adopting the above-mentioned further improvements is that each learner is a unique individual with different learning characteristics. By constructing a learner competency profile, the system can gain a deep understanding of the learner's knowledge reserves, identify the knowledge the learner has mastered and the knowledge gaps in various subjects or fields, and tailor exclusive learning content, learning paths, and learning scenarios for the learner, thereby improving the relevance and effectiveness of learning and stimulating the learner's interest and enthusiasm.

[0018] As a further improvement to this technical solution, the dynamic scene generation module combines part-of-speech tagging and syntactic analysis results to determine the relationship between scene elements, clarifies the grammatical relationship between words through dependency parsing, and identifies key entities in the semantically parsed text using named entity recognition technology.

[0019] As a further improvement to this technical solution, the dynamic scene generation module optimizes the physical rules of the generated scene, including refined construction of the physical model, optimization of real-time physical interaction simulation, and enhanced visualization of physical effects.

[0020] The beneficial effect of adopting the above-mentioned further improvements is that by parsing the semantics of student instructions and extracting key scene elements through natural language processing technology, we can accurately understand student needs. Combined with student profiles, we can retrieve the initial scene model that best matches the student's requirements from the scene resource library based on the student's knowledge reserves, skill level and learning style.

[0021] The generated scenes undergo physical rule optimization, encompassing refined physical model construction, real-time physical interaction simulation optimization, and enhanced visualization of physical effects. In physical model construction, precise physical properties are set for scene objects, and complex structural mechanics analysis is performed to ensure that object motion and forces conform to realistic laws. Real-time physical interaction simulation optimization ensures realistic and natural collision detection and force transmission. Enhanced visualization of physical effects uses realistic lighting and dynamic physical phenomena, such as simulated water flow and flames, to make learners feel as if they are in a real environment, comprehensively improving the immersive experience of the VR scene and providing learners with a more realistic experience during practice, which helps deepen their understanding and mastery of knowledge and skills.

[0022] As a further improvement to this technical solution, the multi-source data evaluation module defines node attributes that include the name, type, and description of the operation steps, and edge attributes that include the logical relationship type and dependency strength.

[0023] As a further improvement to this technical solution, the multi-source data evaluation module uses a topological sorting algorithm to sort the operation steps to ensure that the order of nodes in the graph conforms to the dependency relationship of the operation steps. When constructing the graph, operation steps without prerequisite dependencies are first placed into the graph as starting nodes, and then subsequent operation steps are added in sequence, and nodes are connected according to logical relationships.

[0024] As a further improvement to this technical solution, the multi-source data evaluation module determines the impact weight of each error point by establishing an error type evaluation system, determining the evaluation criteria for the error occurrence stage, and evaluating the degree of impact of the error on subsequent operations.

[0025] The beneficial effect of adopting the above-mentioned further improvements is that the operation step dependency graph clearly presents the logical relationship and sequence between each operation step. When a student makes a mistake, the system can quickly trace back to the original operation step that caused the error based on the graph, accurately locating the root cause of the error;

[0026] By combining the error type, the stage of occurrence, and the degree of impact on subsequent operations, the impact weight of the error point can be calculated, which can comprehensively and scientifically assess the severity of each error.

[0027] Once the original error point and its impact weight are identified, the system can generate a minimum remedial path and targeted improvement suggestions for trainees. For errors with high impact weight, the system will provide detailed and comprehensive solutions and reinforcement exercises, while for errors with low impact weight, it will also provide concise and effective prompts to help trainees correct errors efficiently.

[0028] As a further improvement to this technical solution, the holographic guidance and interaction module obtains the position and posture information of the learner in the virtual space in real time through the spatial positioning sensor built into the VR device. Based on the preset logical rules and the learner's current operation status, it determines the specific position and display angle of the holographic image in the VR scene to guide the learner.

[0029] As a further improvement to this technical solution, the holographic guided interaction module optimizes and reorganizes scene elements and task processes through an AI big model, ensuring the continuity and logic of the scene while adjusting the difficulty and complexity of the scene.

[0030] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0032] Figure 2 This is a schematic diagram of the student ability profiling process of the present invention;

[0033] Figure 3 This is a schematic diagram of the dynamic scene generation process of the present invention;

[0034] Figure 4 This is a schematic diagram of the multi-source data evaluation process of the present invention.

[0035] The meanings of the labels in the diagram are as follows:

[0036] 100. Student profile building module; 200. Dynamic scene generation module; 300. Multi-source data evaluation module; 400. Holographic guided interaction module. Detailed Implementation

[0037] The technical solutions in 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.

[0038] Currently, existing AI-VR scenario training systems often use fixed templates or simple parameter adjustments to generate scenarios, making it difficult to dynamically generate suitable training environments based on the actual abilities of trainees. Teaching assessments often rely on single-dimensional data, such as operation completion rate and answer accuracy, ignoring the behavioral details, physiological reactions, and interactions with the scenario during the operation process. This makes it impossible to comprehensively and accurately assess the learning outcomes and potential problems of trainees.

[0039] Therefore, this invention proposes a learning profile construction module that collects multi-dimensional information, uses cluster analysis to construct a detailed profile, generates an initial profile, and dynamically updates it. A dynamic scene generation module combines natural language processing, part-of-speech and syntactic analysis, and named entity recognition technologies to accurately parse instructions, extract elements, and generate an initial scene model. A multi-source data evaluation module constructs an operation step dependency graph, determines the impact weight of error points based on error type, stage of occurrence, and degree of impact, and comprehensively and accurately evaluates the learner's operational performance during the learning process. A holographic guidance and interaction module generates remedial paths based on error analysis and uses holographic projection technology to accurately guide learners to operate correctly. Simultaneously, an AI big data model dynamically adjusts the difficulty and complexity of the scene based on the learner's real-time performance.

[0040] Specifically as follows:

[0041] Please see Figure 1 As shown, the present invention provides a large-scale model scene training system based on AI-VR, including a student profile construction module 100, a dynamic scene generation module 200, a multi-source data evaluation module 300, and a holographic guided interaction module 400;

[0042] The student profile building module 100 collects students' historical learning data, test scores, and operation records, and performs cluster analysis on students' knowledge reserves, skill levels, and learning styles to build a detailed student ability profile.

[0043] In order to better construct the initial competency profile of trainees, the trainee profile construction module 100 designs diversified basic test questions for different disciplines and training fields. When trainees use the system for the first time, they generate an initial competency profile by completing the basic test and filling in the learning information form.

[0044] For example, in the field of education, for mathematics, the test covers basic knowledge points such as algebra, geometry, and probability, and includes various question types such as multiple choice, fill-in-the-blank, and short answer questions to comprehensively assess students' understanding of mathematical concepts, application of formulas, and problem-solving abilities. For language subjects, the test includes questions on vocabulary, grammar, reading comprehension, and writing to assess students' language knowledge and application abilities. In the field of vocational training, such as mechanical repair training, the test includes Q&A on basic mechanical principles, multiple choice questions on simple fault identification, and simulation questions on the operation of basic repair tools to test students' mastery of mechanical knowledge and basic operational skills.

[0045] The test employs a VR immersive testing format, where trainees wear VR devices to enter a virtual testing environment. The system presents questions using a combination of voice prompts and text displays, and trainees answer questions through the interactive functions of the VR device (such as gestures and controller operations). The system records the trainee's answering time, actions during the answering process (such as whether they modified their answers multiple times or skipped questions), and the final answer in real time, ensuring complete test data is obtained.

[0046] When students enter the information form filling interface, the system uses voice prompts and animations to explain in detail the requirements and purpose of each piece of information. For some complex options, such as learning style preferences, the system provides example explanations to help students understand and fill in the form accurately. The system integrates the student's basic test score data with the content filled in the learning information form to generate an initial ability profile of the student.

[0047] In order to better update student profiles, the student profile building module 100 establishes a profile update mechanism, which triggers updates at key points such as when students complete important learning tasks or when test scores fluctuate significantly. It collects students' learning data in real time and updates student profiles accordingly.

[0048] The system presets key node trigger conditions. Important learning task completion nodes include completing a subject-specific study, passing a phased skills assessment, and submitting a large project assignment, which are significant learning outcomes. Significant fluctuations in test scores are set as a single subject score increase or decrease of more than 15%, or a change in overall score ranking of more than 20% of the class size. The system continuously monitors students' learning progress and performance data, and once the trigger conditions are met, the profile update program is immediately activated.

[0049] The system updates learning style profiles based on changes in learners' learning behaviors before and after key milestones. If a learner demonstrates strong collaboration and communication skills in group projects, while the previous profile indicated an independent learner, the system will recalculate the weight of collaborative learning in the learning style dimension, adjust the learning style category, and simultaneously update the learner's preference for learning resources.

[0050] like Figure 2 As shown, by inputting data and performing cluster analysis, and by completing the first test when the student uses the system for the first time to generate an initial profile, subsequent real-time updates are triggered at key nodes (such as completing important learning tasks, significant fluctuations in test scores, etc.), ultimately jointly constructing and improving the student's ability profile.

[0051] In addition, the dynamic scene generation module 200 receives natural language instructions input by the trainee, parses the semantics of the instructions through natural language processing technology, extracts key scene elements, combines the trainee profile, retrieves an initial scene model that meets the trainee's requirements from the scene resource library, and optimizes the physical rules of the generated scene.

[0052] In order to better extract key scene elements, the dynamic scene generation module 200 combines part-of-speech tagging and syntactic analysis results to determine the relationship between scene elements, clarifies the grammatical relationship between words through dependency parsing, and uses named entity recognition technology to identify key entities in the semantically parsed text.

[0053] First, Named Entity Recognition (NER) technology is used to process the semantically parsed text and identify key entities such as people, places, objects, times, and events. For the instruction "A chemical reaction experiment will be conducted in the lab next Monday," NER technology can identify "next Monday" (time), "laboratory" (place), and "chemical reaction experiment" (event). These key entities are the basic elements for constructing the scene, providing clear objects for subsequent analysis.

[0054] Part-of-speech tagging is performed simultaneously during word segmentation, marking each word with a noun, verb, adjective, or other part-of-speech tagging. Part-of-speech tagging provides basic information for understanding the function and role of words in a sentence and helps to initially determine the relationship between words. For example, verbs usually indicate actions, while nouns often act as the executor, receiver, or location of the action.

[0055] Through dependency parsing, the grammatical relationships between words in a sentence are analyzed in depth to determine the core and subordinate relationships of each word. By combining the results of part-of-speech tagging and dependency parsing, the relationships between scene elements are further determined. Combining part-of-speech information and dependency relationships, it is clarified that "next Monday" is the time when the event "conducting a chemical reaction experiment" occurs, and "laboratory" is the location where the event occurs. Together with "chemical reaction experiment," they constitute a complete scene description.

[0056] In order to better optimize the physical rules of the scene, the dynamic scene generation module 200 optimizes the physical rules of the generated scene, including fine-tuning the physical model, optimizing the real-time physical interaction simulation, and enhancing the visualization of physical effects.

[0057] Refined physical model construction: setting precise physical properties for objects in the scene, such as the mass, density, and friction of mechanical parts, and the mechanical properties of building materials; for complex structures, using the finite element analysis method to calculate unit stress and strain, and optimize structural design; at the same time, clarifying the constraints such as fixation and hinge between objects, and regulating the motion of objects.

[0058] Real-time physics interaction simulation optimization: Collision detection is optimized using a hierarchical bounding box algorithm to quickly and accurately determine collisions and calculate the physical changes after a collision based on object properties; the transmission of forces is simulated based on Newton's laws of motion to realize the interaction between objects; in the face of multi-object interaction scenarios, parallel computing and priority scheduling are used to ensure the smoothness and realism of key interactions.

[0059] Enhanced visualization of physical effects: Using physically based rendering technology, combined with object materials and light sources, realistic light and shadow are simulated; using particle systems and fluid simulation algorithms, physical phenomena such as water flow, smoke, and flames are dynamically presented; the system records the physical changes of objects and supports playback for students to help them understand the physical principles.

[0060] like Figure 3 As shown, the student inputs a natural language command, which is then parsed and processed by NLP to extract key scene elements. Subsequently, the scene is generated by combining the student's profile with the scene, and the generated scene is then physically optimized before finally outputting a VR scene.

[0061] In addition, the multi-source data evaluation module 300 defines each operation step as a node in the graph and defines node attributes. Based on logical relationships and dependency order, it establishes directed edges between related nodes to represent the order of operation steps, establishes an operation step dependency graph, determines the original error point based on the operation step dependency graph, and calculates the impact weight of each error point by combining the error type, the stage of occurrence, and the degree of impact on subsequent operations.

[0062] In order to better define the attributes of nodes and edges, the multi-source data evaluation module 300 defines node attributes as including the name, type, and description of the operation steps, and edge attributes as including the logical relationship type and dependency strength.

[0063] When constructing the operation step dependency graph, clearly defined node and edge attributes provide crucial support for system operation. Node attributes encompass the operation step name, type, and description. The name uniquely identifies the step, the type distinguishes the nature of the step, and the description details the step's content and requirements. Edge attributes include the logical relationship type and dependency strength. Logical relationship types have a sequential relationship, such as "installation base" and "installation bracket" having a sequential relationship.

[0064] By combining key scene elements extracted through natural language processing, dependency graphs can be efficiently constructed. When a student inputs the instruction "First place the beaker on the lab bench, then add the reagent and perform the heating reaction," natural language processing parses out key scene elements such as "place the beaker," "add the reagent," and "heat the reaction," which are then used as nodes. Based on part-of-speech tagging and syntactic analysis, the logical relationship between "place the beaker," "add the reagent," and then "heat the reaction" is clarified, and directed edges are constructed. Through the accurate setting and application of node and edge attributes, the logic and dependencies between operation steps can be accurately presented, providing a solid data foundation and logical framework for subsequent error analysis, scene optimization, and student guidance.

[0065] In order to better construct the operation step dependency graph, the multi-source data evaluation module 300 uses a topological sorting algorithm to sort the operation steps to ensure that the order of nodes in the graph conforms to the dependency relationship of the operation steps. When constructing the graph, the operation steps without prerequisite dependencies are first placed into the graph as the starting node, and then the subsequent operation steps are added in sequence, and the nodes are connected according to the logical relationship.

[0066] Using a topological sorting algorithm ensures that the order of nodes in the graph matches the dependencies between operation steps. Specifically, based on the dependencies between operation steps in the node attributes, operation steps without prerequisite dependencies are identified and placed into the graph as the starting node.

[0067] Continuously search for steps in the remaining operations that have all their preceding dependent nodes in the graph, and add them to the graph sequentially. During the addition process, based on the logical relationship type (such as sequential relationship, conditional relationship) and dependency strength in the edge attributes, establish directed edges between the newly added nodes and existing nodes;

[0068] By repeatedly adding nodes and connecting edges, until all operation steps are incorporated into the graph, a complete graph is formed with nodes arranged in a manner consistent with the dependencies between operation steps. This graph not only clearly presents the logical structure between each operation step but also provides accurate and organized data support and a logical framework based on the attribute information of nodes and edges. This helps students understand the operation process more clearly and improves their learning and practice effectiveness.

[0069] In order to better determine the impact weight, the multi-source data evaluation module 300 determines the impact weight of each error point by establishing an error type evaluation system, determining the evaluation criteria for the error occurrence stage, and evaluating the degree of impact of the error on subsequent operations.

[0070] The drills are meticulously categorized according to their characteristics. For example, mechanical assembly drills are divided into parts installation errors, tool usage errors, and sequence errors. A five-level scoring system is adopted, and each type of error is rated based on its impact on the task results and the potential risk, thus determining the initial severity score.

[0071] The exercise tasks were divided into different stages according to the operation procedures, and the start and end marks of each stage were clearly defined. The importance of each stage in the task and its impact on subsequent operations were considered and assigned an importance score.

[0072] Determine the scope of the impact of the error on subsequent operation steps. If it affects 3 or more operation steps, the impact is considered large, corresponding to 8-10 points; if it affects 1-2 operation steps, the impact is considered medium, corresponding to 4-7 points; if it has no direct impact, the impact is considered small, corresponding to 1-3 points.

[0073] like Figure 4 As shown, the operation steps are clearly defined. First, a dependency graph is constructed according to the operation steps. Then, a topological sorting algorithm is used on the graph to ensure that the node order conforms to the dependency relationship. After that, error location is performed based on the sorted graph to find the original error point in the operation. Then, the influence weight of each error point is calculated by combining error type and other factors. Finally, a minimum remedial path is generated based on the error point and its weight, providing students with targeted improvement guidance.

[0074] In addition, the holographic guided interaction module 400 generates a minimized remedial path based on the original error points and their impact weights, providing students with targeted improvement suggestions, generating holographic projection images of operation guidance instructions and error markers to guide students to perform correct operations. At the same time, it uses an AI big model to dynamically adjust the difficulty and complexity of the scene based on the students' real-time learning performance.

[0075] In order to better guide learners, the holographic guidance and interaction module 400 uses the spatial positioning sensor built into the VR device to obtain the learner's position and posture information in the virtual space in real time. Based on the preset logical rules and the learner's current operation status, it determines the specific position and display angle of the holographic image in the VR scene to guide the learner.

[0076] By utilizing the IMU inertial measurement unit, camera and other spatial positioning sensors in the VR device, the three-dimensional coordinates and posture data of key parts such as the head and hands of the trainee in the virtual space are tracked in real time. The data collection frequency can reach hundreds of times per second to ensure the real-time performance and accuracy of the data.

[0077] Next, the collected position and posture information is transmitted to the system backend. The system follows preset logic rules, such as when a student approaches a certain operating device, the holographic guidance image automatically appears at a suitable distance in front of the device; or according to the student's current operating status, if the student makes an operating error, the error marker holographic image immediately floats above the position of the error.

[0078] Then, through complex spatial calculation algorithms, combined with the spatial coordinate system of the virtual scene, the precise position and display angle of the holographic image in the VR scene are determined, so that it is presented to the students from the best perspective and avoids obstruction or poor viewing angle.

[0079] In order to better adjust the difficulty and complexity of the scene, the holographic guided interaction module 400 optimizes and reorganizes the scene elements and task flow through the AI ​​big model, ensuring the continuity and logic of the scene while adjusting the difficulty and complexity of the scene.

[0080] The AI ​​big data model first integrates multi-source information, including student learning data, operation step dependency graph data, and error point impact weight data. By analyzing students' historical learning data and real-time operation data, it understands students' knowledge mastery, skill level, and learning habits; combined with the operation step dependency graph, it grasps the logical relationships of the task flow; and based on the error point impact weight, it identifies key links and error-prone areas in the task.

[0081] The physical and functional properties of scene elements are dynamically adjusted. In a chemical experiment scenario, the reaction rate and conditions of reagents are adjusted based on the trainees' understanding of the experimental principles to match the difficulty of the experiment with the trainees' abilities.

[0082] During the adjustment of scene elements and task flow, the AI ​​large model designs reasonable transition links, using animation demonstrations, prompts and other methods to allow trainees to clearly understand scene changes and task transitions, avoiding any abruptness.

[0083] In summary, the working principle of this solution is as follows:

[0084] This AI-VR-based large-scale model scenario training system has a student profile building module 100 that collects multi-dimensional information and uses cluster analysis to build detailed profiles. It can also generate initial profiles and update them dynamically, providing a precise basis for the system's subsequent personalized services, meeting the personalized needs of students at different learning stages, and making learning more targeted.

[0085] The Dynamic Scene Generation Module 200 combines natural language processing, part-of-speech and syntactic analysis, and named entity recognition technologies to accurately parse instructions, extract elements, generate an initial scene model, and optimize physical rules from aspects such as physical model construction, interactive simulation optimization, and effect visualization enhancement. The generated scene not only meets the needs of students but also highly restores the real physical environment, providing students with an immersive and personalized learning scenario.

[0086] The multi-source data evaluation module 300 constructs an operation step dependency graph, determines the impact weight of error points by combining error type, occurrence stage and impact degree, comprehensively and accurately evaluates the operation of students in the learning process, deeply analyzes the root causes and impact of errors, and provides quantitative and scientific basis for teaching evaluation and students to improve their learning.

[0087] The holographic guided interaction module 400 generates remedial paths based on error analysis and uses holographic projection technology to accurately guide students to operate correctly. At the same time, the AI ​​big model dynamically adjusts the difficulty and complexity of the scene according to the student's real-time performance to ensure that the scene is adapted to the student's learning progress, improve learning efficiency, and ensure scene coherence and logical rationality during the adjustment process to maintain a good learning experience.

[0088] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A large-scale model scene simulation system based on AI-VR, characterized in that: It includes a student profile construction module (100), a dynamic scene generation module (200), a multi-source data evaluation module (300), and a holographic guided interaction module (400); The student profile building module (100) collects students' historical learning data, test scores and operation records, performs cluster analysis on students' knowledge reserves, skill levels and learning styles, and builds a detailed student ability profile. The dynamic scene generation module (200) receives natural language instructions input by the student, parses the semantics of the instructions through natural language processing technology, extracts key scene elements, combines the student profile, retrieves an initial scene model that meets the student's requirements from the scene resource library, and optimizes the physical rules of the generated scene. The multi-source data evaluation module (300) defines each operation step as a node in the graph and defines node attributes. Based on logical relationships and dependency order, it establishes directed edges between related nodes to represent the order of operation steps, establishes an operation step dependency graph, determines the original error point based on the operation step dependency graph, and calculates the impact weight of each error point by combining the error type, the stage of occurrence, and the degree of impact on subsequent operations. The holographic guidance and interaction module (400) generates a minimized remedial path based on the original error points and influence weights, provides students with targeted improvement suggestions, generates holographic projection images of operation guidance instructions and error markers to guide students to perform correct operations, and uses an AI big model to dynamically adjust the difficulty and complexity of the scene based on the students' real-time learning performance.

2. The AI-VR-based large-scale model scene training system according to claim 1, characterized in that: The student profile building module (100) designs diverse basic test questions for different disciplines and training fields. When students use the system for the first time, they generate an initial ability profile by completing the basic test and filling in the learning information form.

3. The AI-VR-based large-scale model scene training system according to claim 2, characterized in that: The student profile construction module (100) establishes a profile update mechanism, which triggers updates at key points when students complete important learning tasks or when test scores fluctuate significantly. It collects students' learning data in real time and updates the student profiles accordingly.

4. The AI-VR-based large-scale model scene training system according to claim 1, characterized in that: The dynamic scene generation module (200) combines part-of-speech tagging and syntactic analysis results to determine the relationship between scene elements, clarifies the grammatical relationship between words through dependency parsing, and identifies key entities in the semantically parsed text using named entity recognition technology.

5. The AI-VR-based large-scale model scene training system according to claim 1, characterized in that: The dynamic scene generation module (200) optimizes the physical rules of the generated scene, including refined physical model construction, real-time physical interaction simulation optimization, and enhanced visualization of physical effects.

6. The AI-VR-based large-scale model scene training system according to claim 1, characterized in that: The multi-source data evaluation module (300) defines node attributes including the name, type, and description of the operation steps, and edge attributes including the logical relationship type and dependency strength.

7. The AI-VR-based large-scale model scene training system according to claim 1, characterized in that: The multi-source data evaluation module (300) uses a topological sorting algorithm to sort the operation steps to ensure that the order of nodes in the graph conforms to the dependency relationship of the operation steps. When constructing the graph, the operation steps without prerequisite dependencies are first placed into the graph as the starting nodes. Then, subsequent operation steps are added in sequence, and the nodes are connected according to the logical relationship.

8. The AI-VR-based large-scale model scene training system according to claim 1, characterized in that: The multi-source data evaluation module (300) determines the impact weight of each error point by establishing an error type evaluation system, determining the evaluation criteria for the error occurrence stage, and evaluating the degree of impact of the error on subsequent operations.

9. The AI-VR-based large-scale model scene training system according to claim 1, characterized in that: The holographic guidance and interaction module (400) obtains the position and posture information of the student in the virtual space in real time through the spatial positioning sensor built into the VR device. Based on the preset logical rules and the student's current operation status, it determines the specific position and display angle of the holographic image in the VR scene and guides the student.

10. The AI-VR-based large-scale model scene training system according to claim 1, characterized in that: The holographic guided interaction module (400) optimizes and reorganizes scene elements and task processes through an AI big model, ensuring the continuity and logic of the scene while adjusting the difficulty and complexity of the scene.

Images