An ai generation editing and quality closed loop method for high-precision map collaborative production

Abstract

This invention provides an AI-generated, edited, and quality-closed-loop method for collaborative production of high-precision maps. The method includes the following steps: S1, Project Import and Task Splitting: S1.1, Project Import: The system imports high-precision map production project data for a newly collected area in a city, including information such as the work area, road structure, and production schedule. The AI-generated, edited, and quality-closed-loop method for collaborative production of high-precision maps provided by this invention organizes AI generation, AI editing interaction, manual editing confirmation, task phase transition, and project-level quality feedback into a unified production framework. This improves high-precision map production efficiency, reduces repetitive work, enhances quality closed-loop capabilities, and ensures the stability of the final deliverables. Furthermore, through a layered carrying mechanism of candidate layer, task working copy layer, and project deliverable layer, it reduces the risk of unstable results directly entering the final deliverables.

Description

Technical Field

[0001] This invention relates to the field of high-precision map production, and in particular to an AI-generated editing and quality closed-loop method for collaborative production of high-precision maps. Background Technology

[0002] High-precision map production differs from ordinary mapmaking or general graphic editing. It not only requires high geometric accuracy but also complete object attributes, correct topological relationships, consistent feature associations, and meets requirements for cross-regional continuity and project-level consistency. Typical production processes in the industry usually include project import, task breakdown, partial editing, review and confirmation, edge matching, post-processing, and overall quality inspection.

[0003] Existing high-precision map production platforms typically possess basic capabilities such as task management, map editing, rule quality inspection, and results aggregation. However, AI-related capabilities are mostly limited to local recognition, candidate recommendation, or independent auxiliary tools, and have not yet been deeply integrated with the formal production chain. The existing technologies mainly suffer from the following shortcomings: The AI-generated results lack a stable binding relationship with the task object, stage status, and role permissions, making it difficult to directly incorporate them into the formal production process. The lack of hierarchical isolation and step-by-step release mechanisms between AI results, manually edited results, and official results can easily lead to unclear boundaries between results. There is a lack of a unified feedback loop between local editing, edge processing, post-processing and project-level quality inspection. After quality problems are discovered, there is still a heavy reliance on manual positioning and manual order assignment. The large amount of confirmation, modification, rejection, and rework feedback generated in multi-person collaborative production has not been structured and accumulated, making it difficult to continuously feed back into AI-assisted capabilities and production strategies.

[0004] Therefore, there is an urgent need for an AI-generated editing and quality closed-loop method for collaborative production of high-precision maps to solve the above-mentioned technical problems. Summary of the Invention

[0005] This invention provides an AI-generated editing and quality closed-loop method for collaborative production of high-precision maps, which solves the problem that in the production of high-precision maps, there is a lack of stable binding relationship between AI-generated results and task objects, stage status and role permissions, making it difficult to directly incorporate them into the formal production process.

[0006] To address the aforementioned technical problems, this invention provides an AI-generated editing and quality closed-loop method for collaborative production of high-precision maps, comprising the following steps: S1. Project Import and Task Splitting: S1.1 Project Import: The system imports project data including the work area, road structure, and production requirements from the high-precision map. S1.2 Task Breakdown: Based on project requirements, the system breaks down the project into multiple local task segments; S2. Task Context Construction: After an editor enters a local task, the system automatically constructs the current task context for that task. S3, AI-generated editing: S3.1 Editing Request: The editor initiates an AI-generated editing request based on the task context, which can be done through natural language, semi-structured operations, and interaction with selected objects; S3.2 Structured Editing Objects: The system combines editing requests with task context to form structured editing objects that describe the actions of adding, modifying, and deleting. S3.3 Candidate Result Generation: The system generates candidate geometry, candidate attributes, candidate relationships, and candidate repair results based on the structured editing object, and writes them into the candidate layer; S4. Manual Confirmation and Rule Validation: S4.1 Viewing and Fine-tuning: Editors can view candidate results in the interface and make necessary fine-tuning and confirmation; S4.2 Rule Validation: The system performs rule validation on candidate results; S4.3 Result Confirmation: After the verification is passed, the candidate result enters the task working copy layer from the candidate layer and becomes the valid processing result of the task at the current stage; S5. Task flow and review: S5.1 Task Flow: After a partial task is completed, the system flows it to the review stage; S5.2 Review Processing: Reviewers view the previous editing results, candidate records, and current rule check prompts, and perform confirmation, modification, rejection, and supplementation operations. If the flow line is found to be incomplete, the reviewers trigger AI to generate and edit again, forming new candidate results and confirming them. S6. Joint treatment: S6.1 Edge Connection Task Creation: When adjacent local tasks have completed their phase processing, the system automatically generates an edge connection task. S6.2 Boundary Processing: The edge connection task reads the valid results of adjacent tasks in the task working copy layer, as well as the boundary region context information. The system generates edge connection candidate results, proposes connection point alignment schemes and boundary attribute unification suggestions, and writes them into the candidate layer. S6.3 Result Confirmation: The edge processing personnel confirm and revise the candidate results. After approval, the results enter the working copy layer of the edge processing task, forming a valid result indicating that the boundary consistency has been processed. S7. Post-processing: S7.1 Post-processing task: After the edge connection task is completed, the system will transfer the processing result to the post-processing stage; S7.2, Unified Relationship and Format Check: The post-processing task checks and corrects issues related to the global unified relationship and format specifications, generates candidate correction results, and writes them into the candidate layer; S7.3 Result Confirmation: The post-processing personnel confirm the candidate results, and after they pass, they enter the post-processing task working copy layer; S8. Project-level quality inspection and return to work: S8.1 Project-level quality inspection: The system performs project-level quality inspection to comprehensively test project deliverables; S8.2 Problem Discovery and Backtracking: If a problem is discovered, the system organizes the problem into problem objects and determines their affiliation based on the problem location, related elements, stage, and historical processing information, and generates a backtracking task. S8.3 Problem Handling: The problem object re-enters the processing chain. The repair personnel trigger AI generation and editing in the corresponding task to generate new candidate repair results and confirm them. After the repair is completed, the results are transferred to the review, necessary connection and post-processing checks again. After the release conditions are met, they are synchronized to the project deliverables layer. S9. Feedback Accumulation and Continuous Optimization: S9.1 Feedback Record: Throughout the process, the system continuously records processing information; S9.2, Structured Deposition: The above records are structured and deposited into feedback objects. Feedback objects are used to trace the production process, optimize subsequent candidate sorting, problem prompts and handling strategies, and drive the co-evolution of AI-assisted capabilities and production processes. S10. Results Synchronization and Project Completion: When the task phase and its related edge processing, post-processing and backflow repair results meet the project-level release conditions, the system will synchronize the corresponding phase's valid results to the project deliverables layer to form a formal high-precision map result.

[0007] Preferably, in S1.2, each task is associated with its own spatial range, task stage, layer template, field rules, existing results, and reference data.

[0008] Preferably, in S2, the task context includes the task scope, current stage, role permissions, layer template, field rules, existing results, reference data, and information on the current view scope and selected object.

[0009] Preferably, in S3.3, the structured editing object includes a description of the target object type, target layer, spatial range, completion action, attribute inheritance relationship, and applicable rule constraints.

[0010] Preferably, in step S3.3, the candidate results include a suggested new lane boundary geometry, a set of candidate attribute values, and a description of the connection relationship with existing road objects.

[0011] Preferably, in S4.2, the rule verification includes checking whether the candidate result falls within the task space, whether it meets the lane line layer template requirements, and whether it conflicts with the topological relationship of adjacent roads.

[0012] Preferably, in S7.2, the system checks whether the lane object codes are consistent, whether the relationship between adjacent road objects is continuous, and whether the attribute naming meets the project template specifications.

[0013] Preferably, in S8.3, the repair personnel can directly view the problem location, related elements, and historical processing records.

[0014] Preferably, in S8.3, when AI-generated editing is triggered, the system generates new candidate repair results and proposes adjustments to abnormal connection relationships.

[0015] Preferably, in S9.1, the recorded processing information includes the editor's confirmation and fine-tuning record of the initial candidate results, the reviewer's confirmation record of the supplementary candidate results, the boundary revision record of the edge task, the unified rule correction record of the post-processing task, the problem objects generated by the project-level quality inspection task and their return destination, and the final closed-loop result of the local repair task.

[0016] Compared with related technologies, the AI-generated editing and quality closed-loop method for collaborative production of high-precision maps provided by this invention has the following beneficial effects: This invention provides an AI-generated editing and quality closed-loop method for collaborative production of high-precision maps. By organizing AI generation, AI editing interaction, manual editing confirmation, task phase flow, and project-level quality feedback into a unified production framework, it improves the production efficiency of high-precision maps, reduces repetitive work, enhances the quality closed-loop capability, and ensures the stability of the final deliverables. It elevates AI generation and editing capabilities from an independent auxiliary tool to an integral part of the formal production chain of high-precision maps. Through task-based organization and contextual constraints, it enables multi-person, multi-role, and multi-stage collaborative processing to revolve around a unified production object. Furthermore, through a layered carrying mechanism of candidate layer, task working copy layer, and project deliverable layer, it reduces the risk of unstable results directly entering the final deliverables. Finally, through problem objectification and a task feedback mechanism, project-level problems can be decomposed into locally executable processing units, shortening the rework chain, accumulating feedback objects, and enhancing the platform's continuous efficiency improvement and quality consistency in long-term production processes. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a preferred embodiment of the AI-generated editing and quality closed-loop method for collaborative production of high-precision maps provided by the present invention. Figure 2 for Figure 1 The diagram shows a task context constraint model. Figure 3 for Figure 1 The flowchart shown is a transformation process from AI editing interaction to structured editing objects. Figure 4 for Figure 1 The diagram shows the hierarchical flow of the candidate layer, task working copy layer, and project deliverable layer. Figure 5 for Figure 1 The diagram shows the loopback between local editing, edge stitching, post-processing, and project-level quality inspection. Figure 6 for Figure 1 The diagram shows a multi-person collaborative confirmation and feedback accumulation process. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Please refer to the following: Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 ,in, Figure 1 This is a schematic diagram of a preferred embodiment of the AI-generated editing and quality closed-loop method for collaborative production of high-precision maps provided by the present invention. Figure 2 for Figure 1 The diagram shows a task context constraint model. Figure 3 for Figure 1 The flowchart shown is a transformation process from AI editing interaction to structured editing objects. Figure 4 for Figure 1 The diagram shows the hierarchical flow of the candidate layer, task working copy layer, and project deliverable layer. Figure 5 for Figure 1 The diagram shows the loopback between local editing, edge stitching, post-processing, and project-level quality inspection. Figure 6 for Figure 1 The diagram illustrates multi-person collaborative confirmation and feedback accumulation. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps includes the following steps: S1. Project Import and Task Splitting: S1.1 Project Import: The system imports high-precision map production project data for a newly added collection area in a city, including information such as work area, road structure, and production schedule; S1.2 Task Breakdown: Based on the work area, road structure and production schedule, the system breaks down the project into multiple local work tasks. The main line of the expressway and its ramp area are broken down into local work task T1, and the adjacent road extension area is broken down into local work task T2. T1 and T2 belong to the same project. Each task is associated with its own spatial range, task stage, layer template, field rules, existing results and reference data. S2. Task Context Construction: After editor A opens local job task T1, the system automatically constructs the current task context of the task. The task context includes the spatial range of T1, the current stage, editor A's role and permissions, lane line layer template, road object field rules, existing base map results, collected point cloud and image reference data, as well as the current view range and selected object information. S3, AI-generated editing: S3.1 Edit Request: Editor A discovers that a section of lane boundary line is missing at the ramp entrance. He inputs an edit request via natural language: "Complete the missing right lane boundary line at the ramp entrance and inherit the attributes of the adjacent lanes." S3.2, Structured Editing Object: The system combines the editing request with the task context of T1, identifies the operation type as local completion and attribute inheritance, forms a structured editing object, and describes the target object type, target layer, spatial range, completion action, attribute inheritance relationship and applicable rule constraints; In this embodiment, after the editor enters the task and before the AI-generated editing is triggered, the task context construction unit completes a context reading and packaging to form a structured context object. Subsequently, each time the AI-generated editing is triggered, this object is passed to the AI-generated editing unit as an execution parameter.

[0020] The constraints on the transmitted content are detailed in Table 1: Table 1. Effects of Task Context Constraints The constraint transmission method is not limited to a specific technical interface form (such as prompt injection, parameter serialization or context injection). This invention protects the method step relationship of "first constructing the task context, and then using the task context to constrain the AI ​​generation and execution", rather than the specific implementation mechanism.

[0021] S3.3 Candidate Result Generation: The system generates candidate results based on structured editing objects. The candidate results are written to the candidate layer and not directly to the final result. The structured editing objects include descriptions of the target object type, target layer, spatial range, completion actions, attribute inheritance relationships, and applicable rule constraints. The candidate results include a suggested new lane boundary line geometry, a set of candidate attribute values, and a description of the connection relationship with existing road objects. S4. Manual Confirmation and Rule Validation: S4.1 Review and fine-tune: Editor A reviews the candidate results, fine-tunes the endpoint positions, and confirms that the attribute values ​​are correct; S4.2 Rule Validation: The system performs rule validation on candidate results. Rule validation includes checking whether they fall within the T1 task space, whether they meet the lane line layer template requirements, and whether they conflict with the topology of adjacent roads. S4.3 Result Confirmation: After the verification is passed, the candidate result enters the task working copy layer of T1 from the candidate layer and becomes the valid processing result of the current stage; In this embodiment, the isolation between the candidate layer and the task working copy layer is not simply a matter of different data storage locations, but rather a controlled boundary formed by "different trust levels, different usage permissions, and different flow conditions".

[0022] First layer of isolation: State isolation, see Table 2 for details: Table 2 Isolation Table between Candidate Layer and Task Working Copy Layer In other words, the data in the candidate layer only represents "suggested results that can be processed by humans" and does not necessarily have the validity of the current task. Only after manual confirmation, rule verification, or stage review is it transferred to the task working copy layer and becomes the valid processing result of the current stage of the task.

[0023] Second layer of isolation: Write path isolation: The sources of input for the candidate layer include: candidate results generated by the AI-generated editing unit, intermediate results that have been manually supplemented and modified but not yet confirmed, and suggestions for repair formed after the issue is returned.

[0024] Writing to the task working copy layer is not done through direct editing, but rather through a "confirm and commit" process: edit request / repair request, generate candidate results, write to the candidate layer, manual confirmation / modification / rejection, rule validation or stage review, and finally, write to the task working copy layer after approval. Therefore, the two layers cannot communicate freely; instead, they must pass through a "confirmation or verification" node to allow passage.

[0025] Third layer of isolation: Isolation between visibility and reference relationships: The results in the candidate layer are only for the current editor, reviewer, or stage reviewer to view and process; subsequent follow-up tasks, post-processing tasks, and project-level quality inspection tasks read the valid results in the task working copy layer, not the unconfirmed results in the candidate layer. The candidate layer is used to "see suggestions and make judgments"; the task working copy layer is used to "continue to flow as the formal processing baseline of the current task".

[0026] This can prevent unconfirmed AI results from directly entering the edge computing, post-processing, or project-level quality inspection process.

[0027] Fourth layer of isolation: Change control isolation: Results in the candidate layer can be regenerated multiple times, modified multiple times, or rejected entirely; however, once the results in the task working copy layer are formed, if adjustments are needed later, they are replaced by new candidate results, patched revisions, or reprocessed by a reflow task, rather than directly overwriting the candidate layer with the official results.

[0028] This means that the candidate layer undertakes the function of "trial and error and negotiation", while the task working copy layer undertakes the function of "effective version within the stage".

[0029] The above isolation is not limited to a specific technical means. For example, it can be achieved through independent layers, independent data tables, independent version spaces, status identifiers, or access control. The hierarchical control relationship that "candidate results shall not be used as valid results of the current task without confirmation" is not implemented through a specific storage.

[0030] S5. Task flow and review: S5.1 Task Flow: After editor A completes the initial processing of T1, the system will transfer T1 to the review stage; S5.2 Review Processing: Reviewer B checks the previous editing results, candidate records, manual confirmation records and current rule check prompts, finds that the flow line expression is incomplete, and triggers AI again to generate an editing request to complete the flow line and unify the attribute expression; S5.3 New Candidate Result Generation and Confirmation: The system generates new candidate results and writes them into the candidate layer. After reviewer B confirms them, the system writes them into the task working copy layer of T1. The editing stage and the review stage form a continuous processing chain around the same task object. S6. Joint treatment: S6.1 Edge Connection Task Creation: After both T1 and the adjacent task T2 have completed their phase processing, the system identifies the edge connection requirement and automatically generates the edge connection task E1. S6.2 Boundary Processing: The edge connection task E1 reads the valid results and boundary region context information of T1 and T2 in the task working copy layer. After comparison, the system finds that the newly added lane boundary line in T1 and the existing road boundary in T2 have a slight offset at the connection point, and the expression of road attributes on both sides is inconsistent. When the edge connection processing personnel C triggers AI-assisted repair in E1, the system generates edge connection candidate results and proposes connection point alignment schemes and boundary attribute unification suggestions. The candidate results still enter the candidate layer first. S6.3 Result Confirmation: Edge processing personnel C confirms and revises the candidate results. After approval, the results enter the task working copy layer of E1, forming a valid result indicating that boundary consistency has been processed. S7. Post-processing: S7.1 Post-processing task: After the edge-connecting task is completed, the system will transfer the processing results to the post-processing stage and generate post-processing task P1. S7.2, Unified Relationship and Format Check: Post-processing task P1 checks and corrects the global unified relationship and format specifications, finds inconsistencies in coding methods, generates candidate correction results and writes them into the candidate layer. The system checks whether the lane object coding is unified, whether the relationship between adjacent road objects is continuous, and whether the attribute naming meets the project template specifications. S7.3 Result Confirmation: Post-processing personnel D confirm the correction result, and the system writes it into the task working copy layer of P1; S8. Project-level quality inspection and return to work: S8.1 Project-level quality inspection: The system executes project-level quality inspection task Q1 to comprehensively inspect the project deliverables from a global perspective; S8.2 Problem Discovery and Backflow: The system discovers a local anomaly in the lane connection relationship within the T1 area. The system organizes the anomaly into a problem object, associating the problem location, type, related elements, stage, source task, and historical processing information. Based on the problem location and related elements, the system determines that it is a local repair problem and generates a local repair task R1. In this embodiment, the method for determining the affiliation of a problem object can be summarized as follows: the system reads multiple determination fields in the problem object and sequentially performs spatial affiliation judgment, semantic type judgment, stage affiliation judgment, and historical backflow correction judgment to determine which type of processing task the problem should be backflowed to.

[0031] The inputs for the decision include, as detailed in Table 3: Table 3 Input Table for Problem Attribution Determination The basic steps for attribution determination are as follows: create a problem object; read the problem location, related elements, problem type, stage, source task, and historical processing information; compare it with the task space range and boundary range; determine the scope of influence: single task / cross-boundary / multi-task or global; determine the nature of the problem: local expression / edge consistency / post-processing uniformity / review and confirmation; determine the return unit by combining the stage and historical processing information; generate the corresponding processing task.

[0032] Therefore, the "attribution determination method" protects a structured routing method based on the problem object field, rather than a fixed algorithm model.

[0033] In this implementation, the problem object carries the source task identifier when it is created. Regardless of whether the problem originates from rule verification in the local editing stage, manual rejection in the review stage, or batch testing in the project-level quality inspection stage, the system records the "current task identifier that generated the problem" when recording the problem object.

[0034] The function of the associated fields is detailed in Table 4: Table 4. Problem Object Related Fields Table Once a reflow task is created, the corresponding issue object is included in the task context of the reflow task. When an editor or AI-generated editing unit processes the reflow task, it can directly obtain the issue location, issue type, and related elements as the starting point for processing, without the need for manual relocation.

[0035] S8.3 Problem Handling: The problem object re-enters the processing chain as part of the task context. The repair personnel trigger AI generation and editing in R1 to generate new candidate repair results and confirm them. After the repair is completed, R1 flows again to review, necessary connection and post-processing checks. After the release conditions are met, it is synchronized to the project deliverables layer to replace the original abnormal expression. The repair personnel can directly view the problem location, related elements and historical processing records in R1. When triggering AI generation and editing, the system generates new candidate repair results based on this and proposes adjustment suggestions for abnormal connection relationships. In this implementation, the rules for determining the attribution of a problem can be executed according to the following four categories: Rule 1: Spatial Scope Rule - First determine the scope of the problem's impact; Rule content: If the problem location and related elements fall within the scope of only a single task space, it is determined to be a single-task problem; If the problem location is in the boundary area of ​​adjacent tasks, or the related elements belong to two adjacent tasks, it is determined to be a boundary problem; If the problem involves multiple task scopes, a unified relationship network, or a project-level unified format, it is determined to be a global problem.

[0036] Function: Spatial scope rules are used to first define whether the problem is "locally processed" or "cross-locally processed".

[0037] Rule 2: Problem type rule, based on the spatial scope, then determine the nature of the problem; Rule content: If the problem manifests as a geometric error, missing attribute, missing object, or local relation error within a single task, it is classified as a local repair task; If the problem manifests as discontinuous boundary connections, repetitive expressions, inconsistent attributes on both sides of the boundary, or inconsistent relationships across the boundary, it is classified as an edge repair task. If the problem manifests as a uniform relationship error, uniform format error, uniform encoding error, or batch standardization problem across the entire project, it is classified as a post-processing repair task. If the issue is essentially about confirming, rejecting, supplementing, or re-examining the results of the current stage, and does not yet involve cross-boundary or globally unified processing, it falls under the category of review processing tasks.

[0038] Purpose: Problem type rules are used to determine the type of functional unit to be returned.

[0039] Rule 3: Stage-specific rules: When the same problem type is exposed in different stages, the backflow entry point can be different; Rule content: If the problem is discovered during the review stage and is still an expression problem within the current task, it will be returned to the review processing task or the partial repair task; If the problem is discovered during the edge connection phase, and the core issue is boundary consistency, it is classified as an edge connection repair task. If the problem is discovered during the post-processing or project-level quality inspection stage, and it involves a unified relationship or a unified format, it is classified as a post-processing repair task. If a problem is not exposed in the upstream stage, but is discovered during project-level quality inspection and is determined to affect only a single task, it can still be reverted to a local repair task.

[0040] Function: The stage rule is used to prevent mechanical backtracking based solely on "at which stage the problem was discovered", and instead combines the "discovery stage" and "the nature of the problem" for judgment.

[0041] Rule 4: Historical processing information rule, historical processing information is used to correct the initial judgment result; Rule content: If the same problem has been backflowed to the local repair task for processing, but reappears and is exposed as a boundary consistency problem, it will be upgraded to the edge repair task; If the same issue has been addressed in the edge task but still recurs and is displayed as a unified standard issue, it will be escalated to a post-processing repair task. If a problem has been successfully closed in a certain processing unit, then subsequent similar problems can use the same backflow path; If the same issue is rejected multiple times, repeatedly returned, or remains unresolved for a long period, its return priority can be increased or the processing unit level can be adjusted based on historical records.

[0042] The "synchronization" from the task working copy layer to the project output layer is not a real-time mirror synchronization, but a "controlled improvement after edge integration, post-processing, and project-level release".

[0043] The overall synchronization logic can be expanded as follows: Local task candidate layer, local task working copy layer, edge connection task reads adjacent task working copy, edge connection candidate result, edge connection task working copy layer, post-processing task reads the valid result of completed edge connection, post-processing candidate result, post-processing task working copy layer, project-level quality inspection / release judgment, project deliverable layer.

[0044] The key points are as follows: First, the edge-connection phase does not directly rewrite the project deliverables; instead, it first generates valid results for the edge-connection tasks. Once adjacent local tasks have completed their current stage of processing, the system creates a connecting task. The input basis for the connecting task should be the valid results of the adjacent tasks in the task working copy layer, rather than the unconfirmed results in their respective candidate layers.

[0045] After the edge-connection task handles issues such as discontinuous connections, duplicate representations, inconsistent attributes, or inconsistent relationships within the boundary area, its results still first undergo the following processes: generating edge-connection candidate results; confirmation or revision by the edge-connection processing personnel; and finally, writing them into the task working copy layer corresponding to the edge-connection task.

[0046] At this point, what has been achieved is a "valid result of the phase where boundary consistency has been handled," but it is still not the official outcome of the project.

[0047] Second, the post-processing stage reads the valid results from the stage that has already completed the preceding release: The post-processing task takes the working copy results of the task released in the previous stage as input and continues to process issues such as global consistency, formatting standards, and expression consistency. New adjustments generated during post-processing also first enter the candidate layer, and then, after confirmation, enter the working copy layer of the post-processing task.

[0048] Therefore, the project deliverables layer is not generated directly from local task working copies, but rather from "a set of valid stage results that have been connected, post-processed, and passed project-level checks".

[0049] Third, the actions synchronized to the project deliverables level are merging or replacing after the phase release: Once the edge detection, post-processing, and project-level quality checks all meet the release criteria, the system executes a deliverable enhancement action, synchronizing the valid results from the corresponding task working copy to the project deliverable layer. This synchronization action can manifest in one or more of the following ways: writing the released results within the corresponding spatial range into the project deliverable layer; replacing the old version results in the project deliverable layer on a per-boundary-region or object-set basis; merging new, revised, and deleted results into the project deliverable layer using incremental updates; and recording a stage release snapshot to ensure the project deliverable layer corresponds to a traceable official version.

[0050] Regardless of the method used, the core principle is: only valid results that have passed the initial release stage can enter the project deliverables layer.

[0051] IV. The relationship between the project deliverables layer and the task working copy layer is one-way promotion, not two-way free write-back: To ensure the stability of the final deliverables, the project deliverable layer is used as the official release layer and will not fluctuate in real time due to candidate modifications or temporary local edits. If project-level quality checks discover problems before or after the project deliverables are formed, the system does not arbitrarily modify the project deliverable layer directly. Instead, it proceeds as follows: discovers the problem, identifies the problem object, determines its attribution, generates a local repair task / edge repair task / post-processing repair task, reprocesses the problem in the corresponding task chain, and updates the project deliverable layer only after the release conditions are met again. Therefore, the project deliverables layer always remains the "official version," while the fixes are still completed in a closed loop within the task chain.

[0052] Fifth, “synchronization” can be understood as controlled release rather than ordinary data copying. More accurately, it means: elevating the effective results of a stage to the formal deliverables of the project; incorporating the results of task-level processing into the unified deliverables of the project; and releasing local corrections in a controlled manner after edge processing and post-processing.

[0053] Therefore, the emphasis is on the process relationship of "layered support + phased release + controlled improvement".

[0054] It does not limit which technical means, such as database submission, version merging, object replacement, or message distribution, must be used for synchronization of project deliverables. What it protects is the methodological relationship that "the result of the task working copy must be connected, post-processed, and released at the project level before it can be promoted to the formal project deliverable."

[0055] S9. Feedback Accumulation and Continuous Optimization: S9.1 Feedback Record: Throughout the process, the system continuously records processing information; S9.2, Structured Deposition: The above records are structured and deposited as feedback objects, used to trace back the complete production process of T1, optimize subsequent candidate sorting, problem prompts and handling strategies. The processing information includes the confirmation and fine-tuning records of editor A for the initial candidate results, the confirmation records of reviewer B for the supplementary candidate results, the boundary revision records of the edge task E1, the unified rule correction records of the post-processing task P1, the problem objects generated by the project-level quality inspection task Q1 and their return destination, and the final closed-loop results of the local repair task R1. In this implementation, feedback objects are automatically created by the system when the following actions occur: candidate result is accepted, candidate result is modified (including the modified content), candidate result is rejected (including the rejection reason if any), the return destination of quality issues, and the final processing result of the return task.

[0056] The information contained in the feedback recipients is detailed in Table 5: Table 5 Standard Table of Feedback Object Fields How feedback affects the generation rules (three levels): First layer: Candidate ranking optimization. In the context of similar tasks, candidate result types with higher acceptance rates are given priority in subsequent generation; candidate result types with low acceptance rates or those frequently modified have their recommendation weight reduced. This layer does not modify the generation model, but only adjusts the ranking strategy of candidate results.

[0057] The second layer: Problem prompts and pre-verification rule adjustments. If a certain type of problem repeatedly appears in a specific task stage or a specific area, the system can include it in the pre-verification scope of AI generation and editing in that scenario, so that AI can actively avoid known high-frequency problem types when generating candidate results. This layer corresponds to the adjustment of rule constraints before generation.

[0058] The third layer: AI-assisted capability optimization (long-term). After accumulating a sufficient amount of three-element records of "original candidate results + manually modified and accepted results + corresponding task context", these can be used as training or fine-tuning data to improve the industry adaptability of AI-generated results. This layer is an optimization of the AI ​​generation model itself and belongs to the long-term operation scenario. It is summarized as "continuous optimization" in the disclosure document.

[0059] The specific implementation of feedback on the generation rules is not limited; what is protected is the methodological relationship of "structuring and accumulating the conclusions of manual processing in the production process into feedback objects, and using the feedback objects to drive candidate ranking, pre-validation rules, and AI capability adjustment".

[0060] S10. Results Synchronization and Project Completion: When T1 and its related edge-connection, post-processing and backflow repair results meet the project-level release conditions, the system will synchronize the corresponding stage's valid results to the project results layer to form a formal high-precision map result.

[0061] Thus, an editing request that originally started by filling in missing elements has completed a full closed loop from a local task to the official project outcome, through candidate generation, manual confirmation, review and supplementation, edge processing, post-processing unification, project-level quality inspection, problem return and repair, and feedback accumulation.

[0062] In this implementation, the process is organized under a unified task object and a unified result flow mechanism, so that the high-precision map production process forms a traceable, reversible, and continuously optimized collaborative production link.

[0063] The system establishes a task object system around the project. The task object system includes one or more of the following: local operation tasks, review tasks, connection tasks, post-processing tasks, and project-level quality inspection tasks. Each task object is associated with at least one or more of the following information: task scope, current stage, role permissions, data template or layer specifications, current existing results, upstream or downstream task relationships, and current problem objects or objects to be processed. Through the above organizational method, different roles can carry out continuous processing around the same task object at different stages, so that the production of high-precision maps can form a collaborative operation mechanism with tasks as boundaries, stages as sequence, and result status as the basis for release.

[0064] Based on task-oriented collaborative production organization, the system introduces AI capabilities into the formal editing process. AI capabilities can generate candidate results based on the current task context, and can also generate editing suggestions based on the natural language input by the editor, semi-structured instructions, selected object operations, or system trigger conditions. The system first obtains the current task context, then combines the interactive input with the current task context to form a structured editing object. The structured editing object can describe one or more actions such as adding, modifying, deleting, completing, connecting, revising attributes, adding or repairing relationships. Subsequently, the system generates candidate geometry, candidate attributes, candidate relationships and candidate repair results based on the structured editing object and provides them to the editor or reviewer for processing. This approach moves AI beyond simply providing suggestions; it integrates with human editing processes to form a part of the formal production process.

[0065] To prevent AI output from deviating from production constraints, the system constructs contextual constraints corresponding to the current task before executing AI-generated editing. These contextual constraints include: the spatial scope of the current task, the stage to which the current task belongs, the current operator's role and permissions, the layer templates and field rules applicable to the current task, the existing results of the current task, nearby features or reference data, the rule set and quality requirements associated with the current task, the current view, and the currently selected object or the current problem object. Based on the above contextual constraints, the system restricts AI-generated editing behavior to be executed within the current production unit, thereby improving the consistency between the results and industry production requirements.

[0066] In this implementation, to ensure the quality of the final results, results with different levels of credibility are carried out in a layered manner, including: a candidate layer, which carries newly generated AI results, results to be confirmed and modified, and results to be repaired; a task working copy layer, which carries data that can be used as the current task's valid results after manual confirmation, rule verification, or stage review; and a project result layer, which carries the final results formed after edge matching, post-processing, and project-level quality checks.

[0067] AI-generated results and manually modified results are first written into the candidate layer; after manual confirmation, rule verification, or current stage review, they enter the task working copy layer; and after meeting the release conditions of the upper stage, they enter the project output layer. Through the above method, a clear boundary is formed between candidate results, task results, and formal results, which is conducive to result control in the formal production process of high-precision maps.

[0068] The quality handling process is expanded from a simple error reporting mechanism to a problem objectification and feedback loop mechanism. Problems discovered during local editing, review and confirmation, edge processing, post-processing, or project-level quality inspection are organized into problem objects. Each problem object is associated with at least one or more of the following: problem location, problem type, associated elements, stage, source task, and historical processing information. The system determines the attribution of a problem based on the problem object and then routes the problem back to the corresponding processing unit. The processing units include: local repair tasks, edge repair tasks, post-processing repair tasks, and review processing tasks. After the reflow task is generated, the relevant problem objects re-enter the task context, and are processed again by combining AI generation and editing, manual processing and stage submission, thus forming a quality closed loop between the local and the global.

[0069] In this embodiment, during the long-term production of high-precision maps, editors, reviewers, quality inspectors, and the system will continuously generate a large amount of feedback. In order to improve subsequent production efficiency, the following information can be structured and accumulated as feedback objects: records of accepted candidate results, records of modified candidate results, records of rejected candidate results, types of quality problems and their return destinations, and the final processing results of return tasks. The feedback can be used for production process traceability, as well as for subsequent candidate ranking, problem recommendation, processing priority adjustment, and AI-assisted capability optimization.

[0070] The implementation system may include a project management unit, a task management unit, a role and permission management unit, a task context construction unit, an AI generation and editing unit, a manual editing and confirmation unit, a rule verification and quality check unit, an issue feedback and task splitting unit, as well as a candidate layer, a task working copy layer and a project output layer for carrying the results; Each of the above units can be implemented by one or more software modules, intelligent processing programs, rule engine programs, editing service programs, or workflow orchestration programs.

[0071] Each unit collaborates around a unified task object, so that AI-generated editing, manual confirmation, stage submission, and quality feedback are integrated into the same production chain.

[0072] The system forms a task context around the current task. The task context includes one or more of the following: task scope, task stage, role permissions, template rules, existing results, reference data, current problem object, and current view. Both AI-generated editing and manual editing are performed under the constraints of the task context.

[0073] The system first writes the AI-generated results and manually modified results into the candidate layer, then confirms, verifies or reviews them before they enter the task working copy layer, and finally enters the project deliverable layer after meeting the release conditions of the upper stage. This achieves the hierarchical carrying and step-by-step release of candidate results, task results and formal deliverables.

[0074] In this implementation, the system organizes the problems found in the local editing stage, review stage, edge matching stage, post-processing stage, or project-level quality inspection stage into problem objects, and generates corresponding backflow tasks according to the problem's affiliation. The backflow tasks re-enter the task context construction, AI generation and editing, manual processing, and stage submission process, thereby forming a quality closed loop in the collaborative production of high-precision maps.

[0075] The method for determining the attribution of a problem object can be summarized as follows: the system reads multiple determination fields in the problem object and sequentially performs spatial attribution judgment, semantic type judgment, stage attribution judgment, and historical backflow correction judgment to determine which type of processing task the problem should be backflowed to.

[0076] Historical processing rules are used to prevent errors from repeatedly circulating in inappropriate processing units for extended periods.

[0077] Mapping relationship after attribution determination The rules for attribution are summarized in Table 6: Table 6 Problem Attribution Determination Mapping Table When multiple rules are applied simultaneously, in actual production, a problem object may have multiple characteristics. For example, it may originate from a single task and be discovered during project-level quality inspection; or it may exhibit both local errors and affect boundary consistency. In response, the judgment principle of "from smallest to largest scope of impact, and prioritizing the essence of the problem" is adopted.

[0078] Priority is as follows: if the problem can be resolved within a single task, it belongs to the local repair task; If addressing the issue would affect the consistency of adjacent task boundaries, then edge repair tasks take priority over local repair tasks. If the problem requires unified rules, unified format, or batch relationship adjustment, then post-processing repair tasks take precedence over local repair tasks and edge repair tasks. If the core issue is not data repair, but rather stage confirmation, re-examination, or return, then the review processing task should be given priority as the entry point.

[0079] The purpose of this priority is to ensure that problems are prioritized for processing within the smallest possible closed loop and that the processing unit best reflects the essence of the problem.

[0080] The feedback object is not an arbitrary operation log or ordinary process record, but a structured record object that can reflect "how the system output is evaluated, handled and produces results by humans or subsequent processes" in the production chain; A record is considered a feedback object only when it demonstrates that "a clear conclusion has been reached after processing of the candidate result or the problem object", and that the conclusion can be subsequently traced, statistically analyzed, or optimized.

[0081] In this embodiment, the criteria for determining the feedback object can be grasped from the following five aspects.

[0082] Standard 1: Originating from the formal production chain Feedback should originate from the formal production chain of this invention, rather than from irrelevant behavior outside the chain.

[0083] Criteria for recognition: The record was generated during partial editing, review, edge stitching, post-processing, project-level quality inspection, or reflow processing; This record corresponds to candidate results, problem objects, reflow tasks, or stage processing results; This record is generated by editors, reviewers, quality inspectors, or the system in the processing chain, and its source must be directly related to the "task object-driven collaborative production process".

[0084] Standard Two: A clear conclusion exists regarding the handling of the case. The feedback should reflect how a candidate result or problem was handled.

[0085] The criteria for recognition are as follows: For candidate results, there are clear conclusions such as acceptance, modification, rejection, or supplementary confirmation; for problem objects, there are clear conclusions such as attribution determination, return destination, and whether the loop is closed; for return tasks, there is a final processing result, such as repaired, transferred to other processing units, continued return, or failed.

[0086] If a record only indicates that it has been "viewed" or "opened" but has not resulted in a processing conclusion, it is generally not appropriate to identify it as a feedback object on its own.

[0087] Standard 3: Able to be linked to the evaluated object Feedback objects must be traceable to their corresponding candidate results, problem objects, or task results.

[0088] Identification criteria: It can be associated with one or more of the following: candidate result identifier, problem object identifier, or task object identifier; it can locate the corresponding task context, stage, handler, or processing time; it can explain "which output result this feedback was generated for".

[0089] If there are no related objects, and only isolated subjective opinions or scattered notes, it is difficult to serve as a reusable feedback object.

[0090] Standard 4: Possesses structured utilization value The feedback should not be purely natural language miscellaneous notes, but rather records that can be further statistically analyzed, sorted, recommended, or optimized by the system.

[0091] Criteria for recognition: Recorded content can be structured and represented as fields such as action type, difference content, problem type, backflow path, and final status; recorded content can support subsequent candidate sorting, problem prompts, priority adjustment, or AI capability optimization; similar records can be summarized and compared across tasks, personnel, and stages.

[0092] For example, simple chat-style comments may not naturally be considered feedback objects; however, if they are categorized by the system into structured fields such as "reason for rejection," "modification type," and "issue type," they can be transformed into feedback objects.

[0093] Criterion 5: It has a feedback effect on subsequent decision-making. The core value of feedback lies in its ability to contribute to subsequent production, rather than merely leaving a record.

[0094] Criteria for recognition: This record can be used for subsequent candidate result ranking adjustments; it can be used for high-frequency problem identification and problem recommendation; it can be used for adjustments to backflow strategies, processing priorities, or pre-verification rules; and it can serve as a data source for optimizing AI-assisted capabilities after long-term accumulation.

[0095] In this implementation, if a record is generated in the production chain but has no practical value for traceability, statistics, and optimization, it does not need to be forcibly included in the feedback scope.

[0096] Typical cases that can be identified as feedback recipients The circumstances under which identification can be made are summarized in Table 7: Table 7. Identification of Feedback Recipients Field standards for feedback objects From an implementation perspective, feedback recipients must meet at least three requirements: "identifiable, traceable, and usable." Therefore, their content must be associated with at least one or more of the following fields, as detailed in Table 8: Table 8. Meaning of Field Judgment for Feedback Objects If a record meets the requirements of the core fields mentioned above and can be included in the subsequent statistical and optimization process, it can be identified as a feedback object.

[0097] AI-generated editing in localized tasks: In high-precision map production projects, the system first breaks down the work area, road structure or production schedule into multiple local work tasks. After the editor opens one of the local work tasks, the system loads the scope, stage, template, existing results and reference data corresponding to the task to form the current task context. Editors can initiate editing requests through natural language, semi-structured operations, or by selecting objects. The system combines the request with the current task context to form a structured editing object and generates candidate results accordingly. Candidate results can be presented as suggestions for adding features, geometric completion, revising attributes, or establishing relationships. Candidate results first enter the candidate layer, and after being confirmed or modified by the editor, they enter the task working copy layer as the effective processing result of the current local task.

[0098] In this implementation, AI-generated editing capabilities are directly embedded into local job tasks, rather than serving as an auxiliary tool independent of the formal production process.

[0099] Continuous collaborative processing during the review phase: After the initial processing of a local task is completed, the system will transfer the corresponding results to the review stage. Reviewers can view the previous editing results, relevant candidate records, and current quality inspection prompts, and continue to perform confirmation, modification, rejection, and supplementation operations around the same task object. When necessary, reviewers can also trigger AI to generate editing again based on the current task context to form new candidate results. Thus, the editing stage and the review stage are no longer separated from each other, but form a continuous processing chain around the unified task object.

[0100] Boundary consistency handling in edge connection tasks: After adjacent local tasks have completed their phase processing, the system creates a boundary task to handle issues such as discontinuous connections, repetitive expressions, or inconsistent attributes within the boundary area. The system loads the results of adjacent tasks and the boundary area context into the boundary task and generates candidate repair results based on this context. The boundary processing personnel can confirm or revise the candidate repair results to ensure that objects within the boundary area maintain consistency in connection relationships, expression methods, and attribute relationships. After processing is completed, the corresponding results continue to flow to the subsequent processing or project-level quality inspection phase.

[0101] It can cover representative edge-connecting scenarios in the high-precision map industry and enable AI-assisted capabilities to serve consistent processing across task boundaries.

[0102] Handling of project-level quality issues: During post-processing or project-level quality checks, the system can identify issues that can only be exposed from a global perspective. The system organizes these issues into issue objects and determines the processing unit to which they should be returned based on the issue's location, related elements, stage, and historical processing information. If the issue is a local expression problem, it is returned as a local repair task; if the issue involves the boundary of adjacent tasks, it is returned as an edge repair task; if the issue is a uniform relationship or uniform format problem, it is returned as a post-processing repair task. After the corresponding return task is generated, the issue object re-enters the processing chain as a new task context and is repaired again by combining AI-generated editing, manual processing, and stage submission.

[0103] Continuous optimization driven by accumulated feedback: During long-term production, the system continuously records the acceptance, modification, rejection of candidate results, as well as the return and final handling of quality issues. These records can be used as feedback objects to trace the production process and can also be used to optimize subsequent candidate sorting, problem prompts, and handling strategies. This allows AI-assisted capabilities to evolve together with the high-precision map production process, making it suitable not only for one-time production processing but also for continuous optimization scenarios for long-term production operations.

[0104] The working principle of the AI-generated editing and quality closed-loop method for collaborative production of high-precision maps provided by this invention is as follows: The system first breaks down the project into multiple local tasks and associates spatial range, task stage, template rules, existing results, and reference data with each task. After the editor enters a local task, the system constructs the current task context. Based on the current task context, the editor initiates an AI-generated editing request. The system forms a structured editing object and generates candidate results. The candidate results first enter the candidate layer, and after manual confirmation and rule verification, they enter the task working copy layer. After the local task is completed, the system continues to transfer it to the review stage. In the review stage, AI-generated editing can be triggered again around the same task object to confirm the results. After adjacent tasks are completed... After completion, the system generates a connection task to handle inconsistencies within the boundary area. The connection result continues to flow to the post-processing stage to correct issues related to uniform relationships and format. Subsequently, the system performs a project-level quality check. If problems are found, they are organized into problem objects, and local repair tasks, connection repair tasks, or post-processing repair tasks are generated based on the problem's attribution. This allows the problem object to re-enter the processing chain. Once the relevant repair results meet the release conditions, they are synchronized to the project deliverables layer. At the same time, the system continuously records the acceptance, modification, rejection, problem return destination, and final processing results of candidate results, and accumulates them in a structured manner as feedback objects for traceability and subsequent optimization. This forms a complete closed loop from local editing to project deliverables.

[0105] Compared with related technologies, the AI-generated editing and quality closed-loop method for collaborative production of high-precision maps provided by this invention has the following beneficial effects: By organizing AI generation, AI editing interaction, manual editing confirmation, task phase flow, and project-level quality feedback into a unified production framework, the efficiency of high-precision map production is improved, repetitive work is reduced, the quality closed-loop capability is enhanced, and the stability of the final deliverables is ensured. The AI ​​generation and editing capabilities are elevated from an independent auxiliary tool to an integral part of the formal production chain of high-precision maps. Through task-based organization and contextual constraints, multi-person, multi-role, and multi-stage collaborative processing revolves around a unified production object. Furthermore, through a layered carrying mechanism of candidate layer, task working copy layer, and project deliverable layer, the risk of unstable results directly entering the final deliverables is reduced. Through problem objectification and a task feedback mechanism, project-level problems can be decomposed into locally executable processing units, shortening the rework chain, accumulating feedback objects, and enhancing the platform's continuous efficiency improvement and quality consistency in the long-term production process.

[0106] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An AI-generated editing and quality closed-loop method for collaborative production of high-precision maps, characterized in that, The following steps are included: S1. Project Import and Task Splitting: S1.1 Project Import: The system imports project data including the work area, road structure, and production requirements from the high-precision map. S1.2 Task Breakdown: Based on project requirements, the system breaks down the project into multiple local task segments; S2. Task Context Construction: After an editor enters a local task, the system automatically constructs the current task context for that task. S3, AI-generated editing: S3.1 Editing Request: The editor initiates an AI-generated editing request based on the task context, which can be done through natural language, semi-structured operations, and interaction with selected objects; S3.2 Structured Editing Objects: The system combines editing requests with task context to form structured editing objects that describe the actions of adding, modifying, and deleting. S3.3 Candidate Result Generation: The system generates candidate geometry, candidate attributes, candidate relationships, and candidate repair results based on the structured editing object, and writes them into the candidate layer; S4. Manual Confirmation and Rule Validation: S4.1 Viewing and Fine-tuning: Editors can view candidate results in the interface and make necessary fine-tuning and confirmation; S4.2 Rule Validation: The system performs rule validation on candidate results; S4.3 Result Confirmation: After the verification is passed, the candidate result enters the task working copy layer from the candidate layer and becomes the valid processing result of the task at the current stage; S5. Task flow and review: S5.1 Task Flow: After a partial task is completed, the system flows it to the review stage; S5.2 Review Processing: Reviewers check the previous editing results, candidate records, and current rule check prompts, and perform confirmation, modification, rejection, and supplementation operations. If the flow line is found to be incomplete, the reviewers trigger AI to generate and edit again, forming new candidate results and confirming them. S6. Joint treatment: S6.1, Edge Connection Task Creation: When adjacent local tasks have completed their phase processing, the system automatically generates an edge connection task. S6.2 Boundary Processing: The edge connection task reads the valid results of adjacent tasks in the task working copy layer, as well as the boundary region context information. The system generates edge connection candidate results, proposes connection point alignment schemes and boundary attribute unification suggestions, and writes them into the candidate layer. S6.3 Result Confirmation: The edge processing personnel confirm and revise the candidate results. After approval, the results enter the working copy layer of the edge processing task, forming a valid result indicating that the boundary consistency has been processed. S7. Post-processing: S7.1 Post-processing task: After the edge connection task is completed, the system will transfer the processing result to the post-processing stage; S7.2, Unified Relationship and Format Check: The post-processing task checks and corrects issues related to the global unified relationship and format specifications, generates candidate correction results, and writes them into the candidate layer; S7.3 Result Confirmation: The post-processing personnel confirm the candidate results, and after they pass, they enter the post-processing task working copy layer; S8. Project-level quality inspection and return to work: S8.1 Project-level quality inspection: The system performs project-level quality inspection to comprehensively test project deliverables; S8.2 Problem Discovery and Backtracking: If a problem is discovered, the system organizes the problem into problem objects and determines their affiliation based on the problem location, related elements, stage, and historical processing information, and generates a backtracking task. S8.3 Problem Handling: The problem object re-enters the processing chain. The repair personnel trigger AI generation and editing in the corresponding task to generate new candidate repair results and confirm them. After the repair is completed, the results are transferred to the review, necessary connection and post-processing checks again. After the release conditions are met, they are synchronized to the project deliverables layer. S9. Feedback Accumulation and Continuous Optimization: S9.1 Feedback Record: Throughout the process, the system continuously records processing information; S9.2, Structured Deposition: The above records are structured and deposited into feedback objects. Feedback objects are used to trace the production process, optimize subsequent candidate sorting, problem prompts and handling strategies, and drive the co-evolution of AI-assisted capabilities and production processes. S10. Results Synchronization and Project Completion: When the task phase and its related edge processing, post-processing and backflow repair results meet the project-level release conditions, the system will synchronize the corresponding phase's valid results to the project deliverables layer to form a formal high-precision map result.

2. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S1.2, each task is associated with its own spatial range, task stage, layer template, field rules, existing results, and reference data.

3. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S2, the task context includes the task scope, current stage, role permissions, layer template, field rules, existing results, reference data, current view scope, and selected object information.

4. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S3.3, the structured editing object includes a description of the target object type, target layer, spatial range, completion action, attribute inheritance relationship, and applicable rule constraints.

5. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S3.3, the candidate results include a proposed new lane boundary geometry, a set of candidate attribute values, and a description of the connection relationship with existing road objects.

6. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S4.2, rule verification includes checking whether the candidate result falls within the task space, whether it meets the lane line layer template requirements, and whether it conflicts with the topological relationship of adjacent roads.

7. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S7.2, the system checks whether the lane object codes are consistent, whether the relationship between adjacent road objects is continuous, and whether the attribute naming meets the project template specifications.

8. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S8.3, repair personnel can directly view the problem location, related elements, and historical processing records.

9. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S8.3, when AI-generated editing is triggered, the system generates new candidate repair results and proposes adjustments to abnormal connection relationships.

10. The AI-generated editing and quality closed-loop method for collaborative production of high-precision maps according to claim 1, characterized in that, In S9.1, the recorded processing information includes the editor's confirmation and fine-tuning records of the initial candidate results, the reviewer's confirmation records of the supplementary candidate results, the boundary revision records of the edge-connecting tasks, the unified rule correction records of the post-processing tasks, the problem objects generated by the project-level quality inspection tasks and their return destinations, and the final closed-loop results of the local repair tasks.

Images