A spatial intelligent scene self-reconstruction navigation system and method for dynamic obstacle intervention

Abstract

The application belongs to the technical field of space intelligence, and relates to a space intelligent scene self-reconstruction navigation system for dynamic obstacle intervention, comprising: a dynamic intervention perception module, which is used for identifying dynamic obstacle intervention factors existing in a space in real time and outputting a dynamic intervention event set; a space semantic topology reconstruction module, which is used for automatically updating a space semantic map and a topological connection relationship according to the dynamic intervention event set; a local topology rapid recombination module, which is used for rapidly recombining a local topology affected by intervention based on a topological incremental learning principle; and a perception and navigation bidirectional consistency optimization module, which is used for establishing a reverse feedback channel between perception and navigation. The application can realize real-time response and structure-level self-adaptation of a navigation system to environmental changes, so as to maintain stability and consistency of navigation decisions in a variable scene. The application also provides a space intelligent scene self-reconstruction navigation method for dynamic obstacle intervention.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of spatial intelligence, in particular to a spatial intelligence scene self-reconstruction navigation method for dynamic obstacle intervention, and belongs to the comprehensive application direction of artificial intelligence, spatial computing and autonomous navigation. The present application can be widely applied to indoor mobile robots, automated transportation equipment, intelligent security patrol, augmented reality navigation and intelligent building management and other intelligent scenes that require dynamic spatial understanding and adaptive path reasoning. BACKGROUND

[0002] At present, the dynamic navigation technology in the spatial intelligence scene mainly relies on static map modeling and local obstacle avoidance control. In the existing dynamic navigation system, a two-level architecture separating global path planning and local obstacle avoidance is generally adopted. Among them, the global path planning is usually based on static or low-frequency updated semantic map to generate macroscopic path, while the local obstacle avoidance module reacts to dynamic obstacles based on real-time sensor data for short-time reactive avoidance. The core feature of this kind of method is that dynamic obstacles are only handled at the path level, without triggering structural level update of semantic map or topology. When dynamic intervention (such as human flow, moving objects or temporary closed area) occurs in the environment, the system usually responds through short-time path correction, lacking systematic modeling and global consistency maintenance of spatial structure changes, resulting in insufficient stability of navigation in complex environment.

[0003] 1. Dynamic response is limited to path level, lacking structural level update. Traditional dynamic obstacle avoidance methods (such as velocity obstacle, dynamic window or reinforcement learning avoidance) only make local adjustments at the path level, without updating the spatial semantics or topology structure synchronously. When the environment changes continuously, the system repeatedly avoids obstacles or falls into local oscillation, failing to realize global spatial self-reconstruction and navigation consistency.

[0004] 2. Topology update relies on global reconstruction, with poor real-time performance. Existing semantic map and topology model mostly adopt global reconstruction strategy, which needs to recalculate the global connection relationship once the local structure changes, resulting in large amount of calculation and update lag, and being difficult to meet the real-time navigation demand in complex environment.

[0005] 3. Perception and navigation lack closed-loop feedback. In most systems, the perception module and the navigation module are in a one-way driving structure, and navigation anomalies (such as path failure, drift) cannot affect the perception model update in the reverse direction, resulting in lag of dynamic obstacle modeling and weak adaptability of the system to environmental changes.

[0006] 4. Incremental topology learning ability is insufficient. Existing topology learning mostly faces static scenes or small-scale node expansion, and lacks quick reorganization mechanism for local disturbance of node failure or edge reconnection, which cannot maintain structural coherence and path effectiveness under dynamic topology changes.

[0007] 5. Dynamic perception and semantic map update are disjointed. Existing dynamic detection (such as visual recognition or point cloud clustering) fails to cooperate with semantic map update, and the perceived dynamic intervention is difficult to reflect in the navigation map in real time, resulting in inconsistency between the spatial structure and the real environment.

[0008] 6. Lack of self-supervised and self-repair mechanisms. Most systems rely on manual annotation or predefined rules and cannot accumulate dynamic response experience through autonomous learning, lacking the ability to adapt to new intervention patterns in unknown environments.

[0009] 7. Navigation logic lacks consistency constraints. Existing dynamic planning algorithms focus on local path feasibility and do not establish global temporal and spatial logic constraints, making it easy to frequently change paths in complex dynamic scenarios, resulting in discontinuous or conflicting decisions.

[0010] Existing dynamic navigation systems have deficiencies in the coordination of dynamic perception-topology update-navigation feedback, making it difficult to cope with frequent structural changes in scenarios. There is an urgent need for a navigation method that can realize the linkage of dynamic obstacle intervention perception and spatial structure self-reconstruction to improve the real-time performance, stability and adaptive intelligence of the system. SUMMARY

[0011] The purpose of the present application is to provide a spatial intelligent scene self-reconstruction navigation method for dynamic obstacle intervention, to solve the problem that existing navigation technology cannot realize fast update of spatial structure and maintain path continuity in dynamic environments. By establishing a linkage mechanism of dynamic intervention perception and spatial topology self-reconstruction, the navigation system realizes real-time response to environmental changes and structural-level adaptation, thereby maintaining the stability and consistency of navigation decisions in variable scenarios.

[0012] The present application adopts the following technical solutions:

[0013] A spatial intelligent scene self-reconstruction navigation system for dynamic obstacle intervention, comprising:

[0014] A dynamic intervention perception module for real-time identification of dynamic obstacle intervention factors existing in the space, outputting a dynamic intervention event set;

[0015] A spatial semantic topology reconstruction module for automatically updating the spatial semantic map and the topological connection relationship according to the dynamic intervention event set;

[0016] A local topology rapid reorganization module for rapid reorganization of the local topology affected by the intervention based on the topology incremental learning principle;

[0017] A perception and navigation bidirectional consistency optimization module for establishing a reverse feedback path between perception and navigation.

[0018] Preferably, the system further comprises:

[0019] Self-supervised spatial disturbance learning module: for providing self-evolution and self-repair capabilities for the system, and realizing autonomous learning and parameter updating of the topological representation learning model by simulating disturbance intervention scenarios.

[0020] Preferably, the dynamic intervention event set is composed of a plurality of intervention event records, and each intervention event record contains:

[0021] Event identifier: for uniquely marking the intervention event;

[0022] Event type: for representing the intervention category, including but not limited to temporary obstacles, area closure, crowd gathering, temporarily opened channels, etc.;

[0023] Action range: for representing the influence area of the event in space, which can be represented by a region polygon, a grid set or a topological node set;

[0024] Occurrence time and duration: for describing the starting time and duration of the event;

[0025] Topological influence index: for pointing to the set of affected topological nodes and topological connection edges, so that the subsequent modules can be incrementally updated.

[0026] Preferably, for any connection edge, the system maintains its passage cost or passage weight, which is used to measure the availability and passage cost of the connection edge when planning a path; when the dynamic intervention event affects the connection edge, the system adjusts its passage weight based on the event type and action range:

[0027] If the event type is "area closure" or "persistent obstacle occupation", the related connection edge is marked as impassable, or its passage cost is set to a maximum value to avoid planning selection;

[0028] If the event type is "crowd gathering" or other situations that can be passed through but need to be cautious, the passage cost of the connection edge is increased or its selectable priority is reduced, so that the planning result tends to detour;

[0029] If the event represents a "temporarily opened channel", a corresponding connection edge is added and given a passage cost lower than that of the adjacent passable connection edge to encourage priority passage.

[0030] Preferably, the local topological quick reorganization module searches and updates only within the local subgraph range composed of the affected nodes and their neighbors when performing edge reconnection; the system adopts a comprehensive evaluation principle of "geometric rationality + semantic consistency" for candidate connection edges.

[0031] A spatial intelligent scene self-reconstruction navigation method for dynamic obstacle intervention, comprising:

[0032] Dynamic intervention perception and recognition: Real-time collection of environmental data through multi-modal sensors and recognition of dynamic obstacle types, generating a dynamic intervention event set;

[0033] Dynamic reconstruction of spatial semantic topology: The system automatically updates the semantic topology structure according to the dynamic intervention event set, dynamically marks the impassable nodes and incrementally adds new channels;

[0034] Local topology rapid reorganization: Only the area affected by intervention is executed for local topology reconstruction, and connectivity verification is performed after reconstruction and fused with the global topology;

[0035] Bidirectional consistency optimization of perception and navigation: Navigation feedback signals real-time correct perception model parameters, realizing dynamic closed loop of navigation-perception.

[0036] Preferably, the method further comprises:

[0037] Self-supervised learning by simulating topology disturbance scenarios to continuously optimize topology representation learning model parameters.

[0038] Preferably, dynamic intervention perception and recognition includes:

[0039] Multi-modal data synchronous collection: After the system is started, laser radar, depth camera and IMU module synchronously collect environmental information, and each sensor data is aligned through ROS time synchronization node;

[0040] Dynamic target detection and differential analysis: Moving target trajectory changes are identified through continuous frame point cloud difference, and target categories are determined in combination with image semantic segmentation model;

[0041] Intervention event generation and attribute labeling: Dynamic intervention event set is generated according to target position change and space occupancy rate;

[0042] Event unification and database storage: The system assigns a unique identifier to each dynamic intervention event, and dynamic intervention event information is written into shared database in real time;

[0043] System response and topology update trigger: When detecting dynamic intervention event, the system starts spatial semantic topology update process.

[0044] Preferably, local topology rapid reorganization includes:

[0045] Structure change detection and node positioning: The system detects spatial structure change and quickly locates the disturbed node area;

[0046] Local subgraph construction and edge adjustment: Extract the affected node and neighborhood subgraph, reset the edge direction and weight to form new topology structure;

[0047] Connectivity verification and fusion update: Perform connectivity check and incrementally fuse the local update result into global topology graph;

[0048] Synchronization and performance response: real-time synchronization of global models, topology repair and structural stability updates within hundreds of milliseconds.

[0049] Preferably, the bidirectional consistency optimization of perception and navigation includes:

[0050] Bidirectional feedback mechanism establishment: building a real-time closed-loop channel between navigation and perception;

[0051] Navigation deviation monitoring and feedback generation: deviation state is identified through multi-dimensional trajectory comparison algorithm;

[0052] Perception parameter adaptive adjustment: dynamically intervening perception module automatically corrects obstacle recognition weight and threshold according to navigation feedback;

[0053] Environmental semantic correction and topology update: dynamically updating scene semantics and topology relationship based on feedback;

[0054] Bidirectional closed-loop fusion and system optimization: perception and navigation form a bidirectional information flow mechanism, realizing the consistency optimization of the whole cognitive-decision-execution link.

[0055] The summary section is provided to introduce in simplified form a selection of concepts in order to provide a basic understanding of some aspects of the application. The summary section is not intended to identify key or essential features of the disclosure, nor is it intended to limit the scope of the disclosure. The summary section is not intended to be used to determine or predict the scope or the content of the disclosure.

[0056] Compared with existing navigation methods that only rely on static geometric information or local obstacle avoidance strategies, the present application has significant advantages in dynamic response capability, spatial structure self-reconstruction, path continuity, system robustness and universality, etc.

[0057] 1. Dynamic response capability is significantly improved.

[0058] The present application realizes the real-time update of semantic map and topology connection when dynamic obstacles appear by constructing an "intervention response type navigation structure". The system can quickly complete the structural level adjustment after environmental changes, maintain the real-time nature of navigation and the consistency of spatial cognition, and avoid the problems of response lag and path failure in traditional methods in dynamic scenes.

[0059] 2. Path continuity and stability are enhanced.

[0060] Using a topology incremental learning mechanism, the present application only reorganizes the affected local nodes, avoiding path interruption caused by global reconstruction. The system can continuously maintain the logical coherence and geometric reasonableness of the path during the spatial structure change, making the navigation process smoother and more stable, and significantly improving the continuity and reliability of task execution.

[0061] 3. Closed-loop self-optimization of perception and navigation.

[0062] The application proposes a "navigation feedback guidance perception" mechanism, establishing a two-way collaborative relationship between the perception layer and the navigation layer. Abnormal trajectory information in navigation execution can be used to correct the perception model in reverse, keeping dynamic obstacle detection and path planning consistent, thereby enhancing the system's adaptive ability to complex intervention scenarios.

[0063] 4. Self-supervised learning enhances the self-evolution ability of the system.

[0064] Through the self-supervised spatial disturbance learning mechanism, the system can autonomously learn spatial reconstruction and path recovery strategies during the simulation of obstacle intervention, without the need for human intervention to achieve self-repair and self-optimization, and has self-learning and adaptive ability under long-term operation.

[0065] 5. The system structure is lightweight and highly versatile.

[0066] The overall framework of the application adopts modular and parallel design, and the functional modules work cooperatively through standardized interfaces, with low computational load and good scalability. This method can adapt to various dynamic space application scenarios, including indoor mobile robots, intelligent building navigation, security patrol and augmented reality guidance, etc., and has good engineering deployment and promotion value.

[0067] The application realizes the structural self-reconstruction, path self-adaptation and system self-evolution of the spatial intelligent navigation system in a dynamic environment through the integration of dynamic intervention perception, topology incremental reconstruction, two-way feedback optimization and self-supervised learning, etc. The real-time performance, stability and intelligence level of the whole system are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0068] The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description of exemplary embodiments of the present disclosure taken in conjunction with the accompanying drawings in which like reference characters refer to the like parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the present disclosure.

[0069] Figure 1 A schematic diagram of a spatial intelligent scene self-reconstruction navigation system module in an embodiment of the application;

[0070] Figure 2 A schematic diagram of the overall flow of a spatial intelligent scene self-reconstruction navigation method in an embodiment of the application;

[0071] Figure 3 A schematic diagram of the steps of a spatial intelligent scene self-reconstruction navigation method in an embodiment of the application;

[0072] Figure 4 A dynamic intervention perception and identification flowchart provided in an embodiment of the application;

[0073] Figure 5 A schematic diagram of a local topology fast reconfiguration procedure is provided for an embodiment of the present application.

[0074] Figure 6 A schematic diagram of a perception and navigation bidirectional consistency optimization procedure is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0075] Embodiments of the present disclosure will be described in detail with reference to the drawings, wherein like reference numerals refer to like elements throughout. Although the embodiments of the present disclosure are shown and described in detail, it should be understood that the present disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0076] The term "comprising" and variations thereof as used herein are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted in an overly literal or overly formal sense unless expressly so defined herein.

[0077] For the convenience of understanding, the technical terms involved in the embodiments of the present application are briefly introduced as follows:

[0078] 1. Semantic map

[0079] The essence of a semantic map is different from a traditional navigation map in that it not only provides geometric and topological information of roads, but more importantly, it endows static and dynamic elements in the environment with machine-understandable semantic connotations, attribute states, and their internal logical relationships, thereby forming a precise world model that can be used for real-time reasoning.

[0080] From a technical composition perspective, a semantic map is a multi-layer fusion structured data model. Its foundation is a geometric layer composed of centimeter-precision three-dimensional point clouds and lane-level models, which defines the accurate framework of the physical space. The key to superimposing on it is the semantic layer, which finely classifies, identifies, and attribute-describes all entities in the road environment, covering lane lines, traffic signs, signal lights, curbs, guardrails, and other elements, and updates their states (such as signal light phases, construction area states) in real time or quasi-real time. The highest layer is the relationship layer, which defines the control, ownership, and constraint relationships between semantic elements in the form of digital rules, such as explicitly associating a specific signal light with the lanes it controls, or specifying the traffic rules of a lane at different times.

[0081] 2. Topology structure

[0082] Topology refers to the abstract logical model of digital road network used for path planning and navigation. It describes the connectivity and passability between locations by simplifying the complex physical road environment into a graph structure composed of nodes and connecting edges, and is the core data basis for route calculation of traditional navigation systems.

[0083] The core elements of topology include: nodes, usually representing road intersections, points of interest or shape points; connecting edges, representing road segments connecting two nodes with passability attributes, such as a road or a lane; weights, representing the cost of passing on a certain edge, such as length, expected travel time or toll amount. The system quickly searches for the optimal path from the starting point to the ending point on this structure through graph theory algorithms.

[0084] 3、Topology impact index

[0085] Topology impact index is a data structure and query mechanism designed for efficient and accurate incremental update in dynamic navigation semantic map systems. Its core function is: when the local topology elements of the map (such as lanes, connecting points) change, it can quickly and accurately locate and calculate all related topology elements, path planning results and downstream modules that depend on the map data that are directly or indirectly affected by the change.

[0086] The present application establishes a dynamic intervention response mechanism in the aspects of perception, spatial topology reconstruction, path generation and self-supervised learning, etc., to realize real-time self-reconstruction of spatial structure and maintenance of path continuity under environmental changes. When dynamic obstacles appear (such as crowd gathering, object movement, temporary closed areas, etc.), the system no longer only performs path obstacle avoidance, but triggers the synchronous update of semantic map and topology structure, thereby realizing self-reconstruction navigation at the spatial structure level.

[0087] The core innovation of the present application is that when the system perceives dynamic obstacles with persistent and structural impact, the response mechanism has undergone fundamental changes - the system is no longer limited to local path obstacle avoidance according to existing map data, but judges such events as substantial changes to the semantic layer of the map and the topological connectivity, and accordingly initiates real-time, incremental synchronous updating of the navigation semantic map and its topological structure. This process builds a dynamic closed loop of "perception-driven map evolution", enabling the navigation system to go beyond the traditional "detour in fixed map" mode and achieve navigation based on real-time reconstruction of the map. Specifically, the semantic information and topological network on which the system depends can be dynamically reconfigured according to real changes, and the global path is re-planned based on the updated map. This marks a shift in navigation paradigm from passive obstacle avoidance in static maps to active maintenance and use of a dynamically evolving high-fidelity map, thereby enabling the entire intelligent group to have collective awareness and forward-looking path planning capabilities for instantaneous road structure changes, fundamentally improving navigation safety, overall traffic efficiency and system intelligence in complex dynamic scenarios.

[0088] To achieve the above goal, the present application constructs a "intervention perception-topological reconstruction-path navigation-self-supervised optimization" spatial intelligent scene self-reconstruction navigation system for dynamic obstacle intervention. The system includes five core modules: dynamic intervention perception module, spatial semantic topological reconstruction module, local topological rapid reorganization module, perception and navigation bidirectional consistency optimization module, and self-supervised spatial disturbance learning module. These modules are coupled through data flow and feedback channels to form a complete technical link from environmental dynamic change perception, local topological structure reorganization to path logic reasoning and self-learning optimization.

[0089] Embodiment 1:

[0090] A spatial intelligent scene self-reconstruction navigation system for dynamic obstacle intervention, as shown in Figure 1 , includes:

[0091] Dynamic intervention perception module: used for real-time identification of dynamic obstacle intervention factors existing in space, is the input layer of the whole system.

[0092] Hardware composition and input conditions: the dynamic intervention perception module is composed of multi-modal sensing units (laser radar, RGB-D camera, inertial measurement unit IMU), data fusion controller and perception computing unit. The module collects environmental state data within 0.1-0.5 seconds.

[0093] Working process: (1) Obtain the position and velocity vector of each dynamic object in the scene through target detection and motion estimation algorithm; (2) Determine the intervention type (such as temporary obstacle, area closure, crowd gathering, etc.) by using time series analysis and spatial clustering method; (3) Generate the intervention event set and determine its influence range on the spatial topology.

[0094] Principle of action: The dynamic intervention perception module not only outputs the identification of the existence of obstacles, but also gives the topological coordinate index of the "influence area", so that the spatial semantic topology reconstruction module can directly locate the disturbed nodes at the topology level without the need to rebuild the whole graph. The output of the dynamic intervention perception module is the "dynamic intervention event set". The dynamic intervention event set is composed of a number of intervention event records, and the dynamic intervention event set E = {e1, e2, …, en} is composed of a number of intervention event records, and each intervention event e i} records contains: i

[0095] (1) Event identifier: used to uniquely mark the intervention event;

[0096] (2) Event type: used to represent the intervention category, including but not limited to temporary obstacle, area closure, crowd gathering, temporary opening of channel, etc.;

[0097] (3) Scope of action: used to represent the influence area of the event in space, which can be represented by area polygon, grid set or topology node set;

[0098] (4) Occurrence time and duration: used to describe the starting time and duration of the event;

[0099] (5) Topological influence index: used to point to the affected topology nodes and topology connection edge set, so as to realize incremental update in subsequent modules.

[0100] Spatial semantic topology reconstruction module: used to automatically update the spatial semantic map and topology connection relationship according to the dynamic intervention event set E output by the dynamic intervention perception module, and realize the real-time adjustment of the environmental semantic structure. The spatial semantic topology reconstruction module (structure level decision layer) solves the problem of "whether to change the spatial structure semantics and connectivity".

[0101] Data structure and input and output: input is the current semantic topology graph and the dynamic intervention event set E; the output is the updated semantic topology graph.

[0102] Technical scheme and working principle: The system adopts "intervention response type navigation structure", the core idea of which is that when the accessibility of space nodes or edges changes, instead of global reconstruction, the topology reconstruction is realized through incremental update operation.

[0103] ​For any connecting edge, the system maintains its passing cost or passing weight, which is used to measure the availability and passing cost of the edge when planning a path. When a dynamic intervention event affects the connecting edge, the system adjusts its passing weight based on the event type and the scope of action:

[0104] (1) If the event type is "area closure" or "persistent obstacle occupation", mark the relevant connecting edge as impassable, or set its passing cost to a maximum value to avoid planning selection;

[0105] (2) If the event type is "crowd gathering" or other passable but cautious passing situations, increase the passing cost of the connecting edge or reduce its selectable priority, so that the planning result tends to detour;

[0106] (3) If the event represents "temporary open passage", a new connecting edge is added and given a passing cost lower than that of the adjacent passable connecting edge to encourage priority passing.

[0107] The above adjustment rules can be implemented by a preset strategy table or a learnable weight mapping strategy, but their common goal is to convert dynamic intervention information into topology layer edge reachability and passing cost update.

[0108] New nodes (such as temporary open areas) are automatically added to the topology graph through semantic neighborhood search and logical edges are established with adjacent nodes. This module is the structural update core of the invention, ensuring that the navigation system remains semantically consistent and path connected after the spatial structure changes.

[0109] Local topology rapid reorganization module: used to rapidly reorganize the local topology affected by intervention based on the topology incremental learning principle, which is the key technical means for the invention to realize "real-time self-reconstruction". The local topology rapid reorganization module (algorithm level execution layer) solves the problem: "how to realize the above structural changes at the minimum calculation cost".

[0110] Working steps and working conditions:

[0111] (1) Identify the subgraph corresponding to the disturbed area ;

[0112] (2) Update node semantics and connection weights through local node embedding model;

[0113] (3) Perform local edge reconnection operation on the topology graph:

[0114] When performing edge reconnection, the local topology rapid reorganization module only searches and updates within the local subgraph range composed of the affected nodes and their neighbors. The system uses a comprehensive evaluation principle of "geometric rationality + semantic consistency" for candidate connecting edges:

[0115] Geometric rationality: candidate connections should satisfy the geometric constraints such as spatial distance, channel width, obstacle occupancy, turning radius, etc.

[0116] Semantic consistency: the functional semantics of the nodes at both ends of the candidate connection should be compatible, for example, "corridor-corridor", "room-corridor" can be connected, while "non-passable area- any area" cannot be connected.

[0117] Comprehensive selection: the system selects the connection edge with the minimum comprehensive cost from the candidate connections that satisfy geometric accessibility and semantic compatibility, and merges it back into the global topology structure.

[0118] (4) The reconstruction result is fused with the global topology to output the updated topology structure. The module completes the local topology repair within a few milliseconds without the need for full graph reconstruction, significantly improving the real-time response capability of the system.

[0119] Perception and navigation bidirectional consistency optimization module: used to form a dynamic consistency correction mechanism by establishing a reverse feedback channel between perception and navigation.

[0120] The perception and navigation bidirectional consistency optimization module includes a navigation control module, a feedback judgment unit, and a perception adjustment unit. The navigation control module, as a submodule of the perception and navigation bidirectional consistency optimization module, records the actual motion trajectory and planned trajectory of the agent in real time during navigation execution, and continuously monitors and analyzes the deviation between the two. The actual trajectory and the planned trajectory of the robot are recorded in real time.

[0121] Technical scheme and principle: when a path anomaly (such as trajectory deviation or continuous obstacle avoidance failure) is detected, a navigation deviation vector is generated; the deviation vector is converted into an obstacle model correction signal by the feedback judgment unit and is transmitted to the perception adjustment unit (perception layer); the perception adjustment unit updates the obstacle detection parameters and semantic classification threshold, realizing "navigation-perception" bidirectional self-regulation. Through bidirectional feedback, this module ensures that the spatial cognition and navigation execution of the system are consistent in time sequence and logic level, improving the navigation robustness and information stability.

[0122] Self-supervised spatial disturbance learning module: used to provide the system with self-evolution and self-repairing capabilities, and to realize autonomous learning and parameter updating of the topology representation learning model by simulating obstacle intervention scenarios. The topology representation learning model is a graph structure-based topology representation learning model used to embed the nodes and connection relationships in the spatial semantic topology graph. The topology representation learning model takes the current spatial semantic topology graph as input, encodes the node adjacency relationship, semantic attributes, and historical navigation results, and generates an intermediate representation for topology reconstruction and path reasoning.

[0123] Training and execution process: In the training phase, the self-supervised spatial disturbance learning module takes the system historical topology and navigation log as samples, periodically constructs "disturbed topology structure (such as deleting nodes, changing connection direction)", and requires the model to complete two types of recovery goals at the same time:

[0124] (1) Topology recovery goal: the recovered topology output by the model should be as close as possible to the connectivity and reachability annotation before disturbance, that is, to reduce the structural errors of "should be connected but not connected" and "should not be connected but connected";

[0125] (2) Path recovery goal: the planning path generated on the recovered topology should be as close as possible to the feasible reference path before disturbance, or at least maintain the reachability, traffic safety and path continuity.

[0126] The system takes the weighted combination of the above two types of target errors as the training target, wherein the weight coefficient is used to adjust the training emphasis of "more biased towards structure recovery" or "more biased towards path stability"; the weight coefficient can be preset or automatically selected through the validation set.

[0127] Automatic update of network weight to realize unsupervised learning; in the running phase, the system continuously performs online self-correction. Through the self-disturbance-self-recovery cycle, the model can also stably reconstruct the spatial structure and maintain the navigation continuity in the unknown dynamic environment.

[0128] The overall logic framework of the system is: the system takes dynamic intervention perception as the trigger source, realizes adaptive reconstruction of spatial semantic graph through topology incremental update and local recombination. Then under the bidirectional consistency optimization mechanism of perception and navigation, the system completes path update and execution feedback; at the same time, the self-supervised spatial disturbance learning module continuously optimizes the parameters in the background to realize structure self-evolution and self-recovery.

[0129] The running logic of the whole system can be summarized as the following process:

[0130] Environment perception stage: multi-source sensors collect and identify dynamic obstacle information to generate a set of dynamic intervention events;

[0131] Topology reconstruction stage: locally correct the semantic topology according to the set of dynamic intervention events;

[0132] Path reasoning stage: the navigation control module generates a dynamic feasible path on the reconstructed topology; execution feedback stage: deviation in navigation execution triggers reverse correction;

[0133] Self-supervised optimization stage: the model continuously improves the dynamic adaptability through disturbance learning.

[0134] Through the above structure and method, the application realizes the self-reconstruction navigation system at the space structure level, so that the navigation system can maintain the consistency and real-time performance of space understanding, topological logic and path execution in a high dynamic environment.

[0135] The operation process of the system is as follows:

[0136] Dynamic triggering stage: when a dynamic obstacle or a space intervention event occurs in the environment, the dynamic intervention perception module instantly identifies the intervention type and generates a dynamic intervention event set;

[0137] Structure response stage: the spatial semantic topology reconstruction module locally corrects the semantic topology graph according to the dynamic intervention event set, and triggers the topology update mechanism;

[0138] Incremental reorganization stage: the local topology rapid reorganization module rapidly updates the affected node and edge structure based on the incremental learning algorithm;

[0139] Path adaptation stage: the navigation control module of the perception and navigation bidirectional consistency optimization module generates a new optimal path according to the updated topology structure, and detects execution deviation through feedback signals;

[0140] Self-repairing stage: if continuous path instability or frequent environmental changes are detected, the self-supervised space disturbance learning module automatically adjusts the model parameters to complete strategy adaptation and structure self-evolution.

[0141] Through the closed-loop mechanism, the application realizes the whole-process self-reconstruction navigation of “dynamic intervention instant response, local topology rapid update, and path adaptive continuous planning”.

[0142] Information flow and working mechanism between modules:

[0143] The structure mapping relationship between perception and topology, the dynamic intervention event set output by the dynamic intervention perception module is input into the spatial semantic topology reconstruction module. The spatial semantic topology reconstruction module maps the physical obstacle information to the semantic topology layer through the intervention weight mapping function, realizes the constraint transmission from the geometric layer to the semantic layer. Dynamic constraint transmission from topology to path, after the topology graph is updated, the changed node weight and edge connection information are transmitted to the perception and navigation bidirectional consistency optimization module in real time. The navigation control module of the perception and navigation bidirectional consistency optimization module recalculates the accessibility matrix according to the updated topology relationship according to the navigation path generation algorithm, and instantly corrects the path planning. Feedback update from navigation to perception, if path jitter, detour or positioning failure occurs during navigation execution, the navigation control module will feedback the deviation signal to the perception adjustment unit to guide the perception system to optimize the obstacle recognition threshold and dynamic target clustering parameters, realizing information closed-loop optimization.

[0144] The long-term learning mechanism of self-supervised optimization, the self-supervised spatial disturbance learning module takes system operation logs as learning samples, periodically simulates a dynamic disturbance scene, and self-adjusts the topology representation learning model parameters. The topology representation learning model can gradually strengthen the prediction ability for environmental changes during long-term use, and realizes self-evolution and continuous stable operation.

[0145] The application realizes space structure level self-reconstruction navigation based on the following three key technical principles:

[0146] 1. Topology incremental learning principle: the change of the topology structure is usually limited to a local area, and the application realizes local reorganization instead of global reconstruction by incrementally updating the embedded features of the intervened nodes, thereby greatly reducing the computational overhead.

[0147] 2. Bidirectional consistency feedback principle: the perception and navigation modules share dynamic obstacle and trajectory deviation information through feedback signals, enabling the system to realize bidirectional correction at the semantic and geometric levels, and ensuring consistency between space cognition and behavior output.

[0148] 3. Self-supervised spatial disturbance principle: by simulating a random disturbance scene and self-recovering path logic, the system can autonomously accumulate reconstruction strategies in unknown environments, thereby having self-evolution and self-repairing capabilities.

[0149] Embodiment 2:

[0150] A space intelligent scene self-reconstruction navigation method for dynamic obstacle intervention, as shown in Figure 2 , comprising:

[0151] S100 collects environmental information in real time through a multi-modal sensor, identifies dynamic obstacles and intervention types, and outputs a dynamic intervention event set to provide positioning basis for subsequent structure reconstruction.

[0152] S200 dynamically adjusts the semantic topology relationship according to the dynamic intervention event set to realize adaptive updating of the local semantic structure. An intervention response weight function is used to adjust the node connection to maintain path connectivity and semantic consistency.

[0153] S300 uses topology incremental learning to perform local reconnection and semantic embedding optimization within the disturbed subgraph. The algorithm completes local topology repair within milliseconds without the need for global reconstruction.

[0154] S400 establishes a navigation-perception feedback channel to correct path deviation and obstacle recognition error in real time. Through a bidirectional self-adjusting mechanism, the consistency and robustness of space cognition and path execution are maintained.

[0155] S500 performs self-supervised learning by simulating a topology disturbance scene to continuously optimize the topology representation learning model parameters. The topology representation learning model realizes self-evolution and long-term stable operation in the disturbance and recovery cycle.

[0156] The following will make a more detailed description of a space intelligent scene self-reconstruction navigation method for dynamic obstacle intervention.

[0157] A space intelligent scene self-reconstruction navigation method for dynamic obstacle intervention, as shown in Figure 3 , comprises the following steps:

[0158] S101 Dynamic intervention perception and identification: real-time collection of environmental data by multi-modal sensors and identification of dynamic obstacle types to generate a dynamic intervention event set. As shown in Figure 4 , specifically, it comprises the following steps:

[0159] S1011 Multi-modal data synchronous collection: after the system is started, the laser radar, depth camera and IMU module synchronously collect environmental information. The data of each sensor is aligned by the ROS time synchronization node to ensure the consistency of multi-source input.

[0160] In the system startup phase, the dynamic intervention perception module synchronously collects environmental data by multi-modal sensors. The laser radar generates a two-dimensional point cloud map at a scanning frequency of 15 times per second, and the depth camera captures scene images and depth data at a rate of 30 frames per second. The IMU module provides attitude angle and velocity change, and all data are time-aligned by the ROS time synchronization node.

[0161] S1012 Dynamic target detection and differential analysis: mobile target trajectory changes are identified by continuous frame point cloud difference. The target category is determined by combining the image semantic segmentation model, such as personnel or obstacles.

[0162] The system performs dynamic target detection in the dynamic intervention perception module. The trajectory changes of moving objects are identified by continuous frame point cloud difference, and their types are identified by combining image semantic segmentation (e.g., personnel, mobile devices, obstacles).

[0163] S1013 Intervention event generation and attribute labeling: a dynamic intervention event set is generated according to target position change and space occupancy rate. Each event contains core attributes such as type, position and action range.

[0164] The dynamic intervention perception module automatically generates a dynamic intervention event set according to target position change and occupancy rate analysis, including intervention type, position coordinates, action range and duration.

[0165] S1014 Event unification and database storage: the system assigns a unique identifier to each intervention event. Event information is written into a shared database in real time to trigger subsequent updates.

[0166] Each event is assigned a unique identifier and stored in a shared database in real time as a trigger signal for subsequent topology updates.

[0167] S1015 System response and topology update trigger: When the intervention event is detected, the system starts the spatial topology update process. Dynamically adjust the path planning weight to realize the environment adaptive navigation.

[0168] Once the intervention event in the shared database is triggered, the system starts the spatial topology update process, adjusts the path planning weight, and realizes dynamic response.

[0169] Preferably, it can also include: S1016, rapid response to environmental changes: the system completes change identification and response within milliseconds. Ensure the continuity of the navigation process and the safe avoidance of dynamic obstacles.

[0170] This step ensures that the system can identify environmental changes and respond to dynamic obstacles within milliseconds.

[0171] S201 Dynamic reconstruction of spatial semantic topology: The system automatically updates the semantic topology structure according to the dynamic intervention event set, dynamically marks the impassable nodes and incrementally adds new channels. Keep the consistency of geometric and logical layers through asynchronous topology update without interrupting the navigation task. Specifically:

[0172] When the dynamic intervention event set is generated by the dynamic intervention perception module, the semantic topology reconstruction module automatically performs structure adjustment. The spatial semantic topology graph maintained by the system is composed of nodes (representing functional areas or channels) and connecting edges (representing the accessibility between areas). When a certain area is determined to be temporarily closed or there is a persistent intervention, the node is marked as "impassable"; at the same time, the connecting edge related to it is temporarily disabled in the system. If a new feasible channel (such as a door opening or a temporary path forming) is detected, the system automatically adds a node and a connection relationship, and completes incremental update in the topology structure. The entire topology update process is completed asynchronously in the background without interrupting the ongoing navigation task. Through dynamic marking and rapid edge relationship adjustment, the system ensures the real-time consistency of the semantic map in the geometric and logical layers.

[0173] S301 Local topology rapid reorganization: Only the local topology reconstruction of the area affected by the intervention is performed, which significantly reduces the computational burden. After reorganization, the connectivity is verified and fused with the global topology to realize rapid structure recovery. As shown in Figure 5 , specifically, it includes:

[0174] S3011 Structure change detection and node positioning: The system detects spatial structure changes and quickly locates the disturbed node area.

[0175] After the spatial structure changes, the system enters the local topology rapid reorganization stage. Unlike traditional full-map reconstruction methods, the present invention only updates the local area affected by the intervention, thereby significantly reducing the computational burden.

[0176] S3012 Local subgraph construction and edge adjustment: Extract the affected node set and the adjacent subgraph, reset the edge direction and weight to form a new topology.

[0177] First, according to the action range of the dynamic intervention event set, the affected node set is located, and then the local subgraph composed of these nodes and adjacent nodes is extracted. The system re-adjusts the direction and attribute of the connected edge according to the node reachability state, adjacency relationship and feasibility evaluation, and forms a new local topology structure.

[0178] S3013 Connectivity verification and fusion update: Perform connectivity check and incrementally fuse local update results into global topology graph.

[0179] After reorganization is completed, the module performs a global connectivity check to ensure that the topology structure has no isolated nodes or broken path. Finally, the updated local subgraph and global topology graph are merged to form a new spatial semantic network.

[0180] S3014 Synchronization and performance response: Real-time synchronization of global model, topology repair and structure stability update are completed within hundreds of milliseconds.

[0181] In this way, the system can complete local reorganization within hundreds of milliseconds, and maintain the integrity and continuity of the navigation structure.

[0182] S401 Perception and navigation bidirectional consistency optimization: Navigation feedback signal real-time correction of perception model parameters, realizing dynamic closed loop of navigation-perception. The system maintains the consistency and robustness of path planning and environment cognition through bidirectional information flow. Specifically, as shown in Figure 6 , including:

[0183] S4011 Bidirectional feedback mechanism establishment: The system builds a real-time closed loop between navigation and perception to realize synchronous transmission of task execution data and behavior coordination.

[0184] The present application introduces a bidirectional feedback mechanism during navigation execution, realizing the collaborative optimization of perception and navigation.

[0185] S4012 Navigation deviation monitoring and feedback generation: Deviation state is identified by multi-dimensional trajectory comparison algorithm, and feedback signal is generated in milliseconds to trigger dynamic correction.

[0186] The navigation control module records the actual trajectory and planned trajectory of the robot in real time, and when abnormal situations such as deviation, detour or path interruption occur, the system returns navigation feedback signal to the dynamic intervention perception module.

[0187] S4013 Perception parameter adaptive adjustment: The dynamic intervention perception module automatically corrects obstacle recognition weight and threshold according to navigation feedback, keeping the synchronization of perception output and execution state.

[0188] The perception module dynamically adjusts the obstacle recognition sensitivity and threshold according to the feedback.

[0189] S4014 Environment semantic correction and topology update: The system dynamically updates the scene semantics and topology relationship based on feedback, ensuring that the spatial model and traffic logic continue to match.

[0190] When the navigation avoids the same object multiple times, the system automatically judges that the object is a static passable obstacle and updates its state label; if the navigation path shakes in a certain area multiple times, the system will re-detect the traffic conditions in that area and update the topology connection.

[0191] S4015 Two-way closed loop fusion and system optimization: Perception and navigation form a two-way information flow mechanism to achieve consistent optimization of the cognitive-decision-execution link.

[0192] Through this two-way information flow, the system realizes a dynamic closed loop of "navigation influencing perception and perception guiding navigation", ensuring that the perception model and path planning results are always consistent.

[0193] S501 Self-supervised spatial disturbance learning mechanism: Through periodic simulation of environmental disturbances for self-supervised training, the system continuously optimizes system parameters. The topology representation learning model realizes self-evolution and long-term stable operation in the disturbance-recovery cycle.

[0194] To improve the system's adaptive ability in long-term operation, the embodiment sets up a self-supervised spatial disturbance learning module. The system periodically simulates environmental disturbances in the background, such as randomly setting some areas as impassable, adjusting obstacle positions or replacing channel directions, to test the topology reconstruction and navigation stability. In each simulation, the system will automatically perform topology reconstruction and path planning according to the disturbance conditions, and record the differences between the results and the previous state. If the path is successfully restored, the system will update the internal weight parameters of the topology representation learning model to strengthen the corresponding strategy; if the restoration fails, the parameters will be adjusted to reduce subsequent errors.

[0195] This learning mechanism does not require human annotation and achieves long-term stability and adaptability to unknown scenes through continuous self-evolution.

[0196] S601 Double-layer path reasoning and execution: The semantic layer is responsible for logical path planning, and the geometric layer is responsible for physical trajectory generation and obstacle avoidance execution. The double-layer linkage mechanism ensures continuous, reachable and smooth execution of the path.

[0197] After reconstruction, the system enters the path reasoning phase. The path reasoning of the present application is based on the updated spatial semantic topology structure, rather than directly searching on the underlying geometric point cloud or grid map. The present application adopts a semantic-geometric double-layer structure: the semantic layer determines the logical passing order, and the geometric layer generates physically executable paths. The semantic layer selects the starting and target nodes on the updated topology graph and reasons out the logical region sequence. The geometric layer calculates the specific trajectory according to the real-time environmental data, performs obstacle avoidance, speed adjustment, and path smoothing processing. When the geometric layer detects that the path is unreachable or the passing is limited, it immediately sends a feedback signal to the semantic layer, triggering a new semantic path generation, thereby ensuring path continuity and task uninterrupted. The navigation instructions are finally converted into motion control parameters (speed, angle, and attitude adjustment amount), which are executed through the chassis control unit to complete intelligent navigation in space.

[0198] Through the above steps, the navigation system of the present application can quickly complete structure updating and path reconstruction in a dynamic environment, maintaining stable navigation. Under multiple rounds of intervention conditions, it still maintains high success rate and adaptability, verifying the robustness and practicality of the method.

[0199] Running results and system performance: The navigation system of the present application shows good stability and real-time performance in various dynamic intervention scenarios. When there are moving crowds, closed passages, or newly added obstacles in the scene, the system can complete spatial structure updating and path reconstruction in a very short time, and the navigation path remains logically continuous and smooth in execution. In a multi-round continuous intervention environment, the system can still maintain a high path planning success rate and environmental adaptability, verifying the practicality and robustness of the method.

[0200] The above has described various embodiments of the present disclosure, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the described embodiments. The choice of terms used in the present application is intended to best explain the principles, practical applications, or technical improvements in the market of the embodiments, or to enable other ordinary skilled persons in the art to understand the disclosed embodiments.

Claims

1. A spatial intelligent scene self-reconstruction navigation system for dynamic obstacle intervention, characterized in that, include: Dynamic Intervention Perception Module: Used to identify dynamic obstacle intervention factors in the space in real time, and outputs a set of dynamic intervention events; each intervention event record in the set of dynamic intervention events includes an event identifier, event type, scope, occurrence time and duration, and a topological impact index pointing to the affected topological nodes and topological connection edges; Spatial semantic topology reconstruction module: used to automatically update the spatial semantic map and topological connection relationship based on the dynamic intervention event set; for the connection edge affected by the dynamic intervention event, the spatial semantic topology reconstruction module adjusts the passage cost of the connection edge based on the event type and duration; Local Topology Fast Reorganization Module: Based on the principle of incremental topology learning, this module rapidly reorganizes the local topology affected by intervention. When performing edge reconnection, the module searches and updates only within the local subgraph formed by the affected nodes and their neighborhoods, and adopts a comprehensive evaluation principle of "geometric rationality + semantic consistency" for candidate connection edges. Semantic consistency means that the functional semantics of the nodes at both ends of the candidate connection edge must be compatible. The perception and navigation bidirectional consistency optimization module is used to establish a reverse feedback path between perception and navigation. The perception and navigation bidirectional consistency optimization module includes a navigation control unit and a perception adjustment unit. The navigation control unit is used to record the actual movement trajectory and planned trajectory of the agent in real time, and generates a navigation deviation vector when path anomalies are detected. The perception adjustment unit is used to update obstacle detection parameters and semantic classification thresholds based on the navigation deviation vector. The self-supervised spatial perturbation learning module is used to provide the system with self-evolution and self-repair capabilities, and realizes the autonomous learning and parameter update of the topological representation learning model by simulating obstacle intervention scenarios.

2. The spatial intelligent scene self-reconfiguration navigation system according to claim 1, characterized in that, A dynamic intervention event set consists of several intervention event records, each of which contains: Event Identifier: Used to uniquely identify this intervention event; Event type: Used to characterize intervention categories, including temporary barriers, area closures, crowd gatherings, and temporary opening of passageways; Scope of influence: Used to characterize the area of ​​influence of the event in space, which can be represented by a region polygon, a grid set, or a set of topological nodes; Time of occurrence and duration: used to describe the start time and duration of the event; Topology Impact Index: Used to point to the set of affected topology nodes and topology connecting edges, so that subsequent modules can perform incremental updates.

3. The spatial intelligent scene self-reconfiguration navigation system according to claim 1, characterized in that, For any connecting edge, the system maintains its passage cost or passage weight, which is used to measure the availability and passage cost of the connecting edge during path planning; when a dynamic intervention event affects the connecting edge, the system adjusts its passage weight based on the event type and scope: If the event type is "area closure" or "persistent obstacle occupancy", then mark the relevant connecting edges as impassable, or set their passage cost to a maximum value to avoid planning selection; If the event type is "crowd gathering" which is passable but requires caution, then increase the passage cost of the connecting edge or reduce its optional priority, so that the planning result tends to detour; If the event characterizes "temporarily open passage", then a corresponding connecting edge is added and a passage cost lower than that of the adjacent passable connecting edge is assigned to encourage priority passage.

4. A spatial intelligent scene self-reconstruction navigation method for dynamic obstacle intervention, characterized in that, include: Dynamic intervention perception and recognition: Real-time environmental data is collected through multimodal sensors and dynamic obstacle types are identified to generate a dynamic intervention event set; each intervention event record in the dynamic intervention event set includes an event identifier, event type, scope of action, occurrence time and duration, and a topological impact index pointing to the affected topological nodes and topological connection edges; Spatial semantic topology dynamic reconstruction: The system automatically updates the semantic topology structure based on the dynamic intervention event set, dynamically marks impassable nodes and incrementally adds new channels; For connected edges affected by dynamic intervention events, the passage cost of the connected edge is adjusted based on the event type and duration; Local topology rapid reorganization: Local topology reconstruction is performed only on the regions affected by the intervention. After reconstruction, connectivity is verified and merged with the global topology. When reconnecting edges, the search and update are performed only within the local subgraph formed by the affected nodes and their neighbors. A comprehensive evaluation principle of "geometric rationality + semantic consistency" is adopted for candidate connection edges. Semantic consistency means that the functional semantics of the nodes at both ends of the candidate connection edge must be compatible. Perception and navigation bidirectional consistency optimization: navigation feedback signals correct perception model parameters in real time, realizing a dynamic closed loop of navigation and perception; The navigation control unit records the agent's actual movement trajectory and planned trajectory in real time, and generates a navigation deviation vector when path anomalies are detected; the perception adjustment unit updates obstacle detection parameters and semantic classification thresholds based on the navigation deviation vector. Self-supervised learning is performed by simulating topological perturbation scenarios to continuously optimize the parameters of the topological representation learning model.

5. The spatial intelligent scene self-reconstruction navigation method according to claim 4, characterized in that, Dynamic intervention perception and recognition include: Multimodal data synchronous acquisition: After the system starts up, the LiDAR, depth camera and IMU module synchronously acquire environmental information, and the data from each sensor are aligned by the ROS time synchronization node; Dynamic target detection and differential analysis: The trajectory changes of moving targets are identified by differential analysis of point clouds in consecutive frames, and the target category is determined by combining the image semantic segmentation model; Intervention event generation and attribute labeling: Generate a dynamic set of intervention events based on changes in target location and space occupancy rate; Event uniqueness and database storage: The system assigns a unique identifier to each dynamic intervention event, and the dynamic intervention event information is written to the shared database in real time; System response and topology update triggering: When a dynamic intervention event is detected, the system initiates the spatial semantic topology update process.

6. The spatial intelligent scene self-reconstruction navigation method according to claim 4, characterized in that, Local topology fast reorganization includes: Structural Change Detection and Node Localization: The system detects changes in spatial structure and quickly locates the affected node regions; Local subgraph construction and edge adjustment: Extract affected nodes and their neighborhood subgraphs, and reset edge directions and weights to form a new topology; Connectivity verification and merged update: Perform connectivity checks and merge the incremental results of local updates into the global topology graph; Synchronization and performance response: Real-time synchronization of the global model, completing topology repair and structural stability updates within hundreds of milliseconds.

7. The spatial intelligent scene self-reconstruction navigation method according to claim 4, characterized in that, Perception and navigation consistency optimization includes: Establishment of a two-way feedback mechanism: Constructing a real-time closed-loop pathway between navigation and perception; Navigation deviation monitoring and feedback generation: Identify deviation status through a multi-dimensional trajectory comparison algorithm; Adaptive adjustment of perception parameters: The dynamic intervention perception module automatically corrects the obstacle recognition weights and thresholds based on navigation feedback; Environmental semantic correction and topology update: Dynamically update scene semantics and topology relationships based on feedback; Two-way closed-loop fusion and system optimization: Perception and navigation form a two-way information flow mechanism to achieve consistent optimization of the entire cognition-decision-execution link.

Images