A dynamic updating path planning method and system based on voronoi error metric

Abstract

This invention relates to a dynamic path planning method and system based on Voronoi error metric. The method includes: acquiring and preprocessing input data; generating a seed set and constructing a Voronoi diagram based on the preprocessed data; extracting seed points from the Voronoi diagram and generating a candidate point set within the neighborhood of each seed point; calculating the comprehensive error of the candidate points in the candidate point set and updating the seed point positions based on the comprehensive error; locally reconstructing the Voronoi diagram when the seed point positions are updated or the local environment changes; constructing a navigation map based on the reconstructed Voronoi diagram; obtaining the path with the minimum comprehensive cost based on the navigation map; generating a new locally optimal path segment; and obtaining the path planning result. This invention updates only the affected area when the environment changes locally, thereby achieving real-time, low-overhead path reconstruction and replanning.

Description

Technical Field

[0001] This invention relates to the field of intelligent navigation and path planning technology, and in particular to a dynamic update path planning method and system based on Voronoi error metric. It is applicable to various autonomous mobile platforms and their control systems, including but not limited to mobile robots, warehouse handling robots, inspection robots, unmanned ground vehicles (UGVs), indoor service robots, and drones equipped with onboard computing units. It is especially suitable for application scenarios that require real-time replanning and high safety in complex or dynamic environments. Background Technology

[0002] With the rapid development of intelligent robots and autonomous driving technologies, achieving autonomous obstacle avoidance and efficient path planning in complex environments has become a key issue for mobile platforms. Existing commonly used path planning methods can be categorized into three types: graph / grid-based search methods (e.g., A*, D*), random sampling-based methods (e.g., RRT, PRM), and geometric skeleton-based planning methods (e.g., Voronoi diagrams or central axes). Graph-based search methods have well-defined optimality definitions and mature engineering implementations; sampling-based methods excel in high-dimensional spaces and complex constraints; geometric skeleton-based methods obtain safe paths "away from obstacles" by generating skeletons equidistant from obstacles, exhibiting natural advantages in narrow passages and areas with dense obstacles.

[0003] Existing technologies disclose a fusion path planning method for mobile robots based on the Voronoi skeleton. This approach typically uses a statically or periodically reconstructed Voronoi skeleton as the global navigation skeleton and combines it with local motion control or local obstacle avoidance algorithms (such as Dynamic Window Method (DWA)) for short-range obstacle avoidance or trajectory tracking. This type of approach can generate reasonable paths in static or slowly changing environments, but it suffers from the following drawbacks: (1) Insufficient dynamic response capability. Existing Voronoi skeleton-based methods mostly adopt the strategy of full graph construction or periodic reconstruction. When the environment changes locally, the entire Voronoi structure needs to be recalculated, which leads to replanning delay and makes it difficult to meet the real-time requirements.

[0004] (2) High computational resource consumption. Global Voronoi construction or large-scale sampling reconstruction has high requirements for CPU and memory resources, which is not conducive to the deployment of embedded or airborne computing platforms.

[0005] (3) Insufficient use of semantic information. Traditional solutions mainly rely on geometric distance as the planning basis, without inputting semantic information such as terrain type, accessibility or task orientation, which affects the scene adaptability of the path.

[0006] (4) Lack of local incremental correction mechanism. Existing technologies usually lack a clear and efficient local incremental reconstruction process. When the local environment changes, they still mostly adopt a global or coarse reconstruction method, which is inefficient. Summary of the Invention

[0007] The purpose of this invention is to provide a dynamic path planning method and system based on Voronoi error metric, which enables real-time, low-overhead path reconstruction and replanning.

[0008] To achieve the above objectives, the present invention provides the following solution: A dynamic path planning method based on Voronoi error metric includes: Collect environmental data and robot data, acquire input data and preprocess it, wherein the input data includes environmental grid data, obstacle boundary information, start and end point positions, optional semantic label map and robot geometric and motion constraint parameters; Based on the preprocessed data, seed sets are generated at obstacle boundaries, free space, and semantic interest points to construct a Voronoi diagram; Seed points are extracted from the Voronoi diagram, and a set of candidate points is generated in the neighborhood of the seed points. The candidate points in the candidate point set are calculated based on the path planning cost, and the seed points are updated based on the comprehensive error. When the seed point position is updated or the local environment changes, the Voronoi diagram is locally reconstructed to obtain the reconstructed Voronoi diagram; A navigation graph is constructed based on the reconstructed Voronoi graph. The path with the minimum overall cost is obtained based on the navigation graph, and a new locally optimal path segment is generated. Based on the new locally optimal path segment and the original global path, the path planning result is obtained.

[0009] Optionally, preprocessing the input data includes: The input data is preprocessed by noise filtering, coordinate alignment and scale normalization, and a distance representation structure for security assessment is constructed to evaluate the accessibility and security margin of each point in the environment.

[0010] Optionally, seed sets are generated at obstacle boundaries, free space, and semantic interest points to construct the Voronoi diagram, including: The obstacle boundary is sampled at a preset interval to obtain boundary seed points. In the free space, a sampling strategy is used to generate uniformly distributed seed points, with the start point, end point and key semantic interest points as forced seeds. The boundary seed points, the uniformly distributed seed points, and the forced seeds are merged and deduplicated to obtain the seed set; The Voronoi diagram is constructed using geometric calculation methods based on the seed set.

[0011] Optionally, performing a comprehensive error calculation on the candidate points in the candidate point set based on the path planning cost, and updating the position of the seed point based on the comprehensive error includes: Based on the path planning cost, construct a comprehensive cost function; The comprehensive error of each candidate point is calculated based on the comprehensive cost function, and the candidate point with the smallest comprehensive error is selected as the update target. when The seed point is updated in position according to the update target, wherein, This represents the combined error of candidate points after sampling in the neighborhood. This represents the overall error of the current seed point. Reduce the threshold for error.

[0012] Optionally, the comprehensive cost function for: ; in, This is a security cost based on the distance field; This represents the optional semantic passage cost; Costs guided by objectives; weights , , Used to balance the costs across different tasks.

[0013] Optionally, local reconstruction of the Voronoi diagram includes: The topological relationships between seeds are recorded using a Voronoi adjacency list, and a fast neighborhood retrieval is achieved using a KD-tree or an equivalent spatial index structure to obtain the set of affected Voronoi units. The Voronoi diagram is locally reconstructed based on the affected Voronoi cell set.

[0014] Optionally, constructing a navigation graph based on the reconstructed Voronoi diagram includes: The nodes and edges of the reconstructed Voronoi graph are mapped to the navigation graph, where nodes represent position points on the Voronoi graph, edges represent lines connecting adjacent skeleton nodes in the Voronoi graph, and the weights of the edges are obtained by the composite cost function. The composite cost function for: ; in, For the edge geometric length, For spatial points The distance field value represents the point. The minimum Euclidean distance to the nearest obstacle. For the edge Statistics on semantic access costs For curvature penalty term, , , , These represent the length weight, safety distance weight, semantic cost weight, and curvature constraint weight, respectively.

[0015] Optionally, the path to obtain the minimum overall cost includes: ; in, The path with the lowest overall cost, In the navigation map The sequence of passable paths from the starting point to the destination.

[0016] This invention also provides a dynamically updated path planning system based on Voronoi error metric, comprising: The data processing module is used to collect environmental data and robot data, acquire input data and perform preprocessing. The input data includes environmental grid data, obstacle boundary information, start and end point positions, optional semantic label map and robot geometric and motion constraint parameters. The Voronoi graph construction module is used to generate seed sets at obstacle boundaries, free space, and semantic interest points based on preprocessed data, and construct Voronoi graphs. The Voronoi graph reconstruction module is used to extract seed points from the Voronoi graph, generate a set of candidate points in the neighborhood of the seed points, calculate the comprehensive error of the candidate points in the candidate point set according to the path planning cost, update the position of the seed points according to the comprehensive error, and perform local reconstruction of the Voronoi graph when the position of the seed points is updated or the local environment changes, so as to obtain the reconstructed Voronoi graph. The path search module is used to construct a navigation graph based on the reconstructed Voronoi graph, obtain the path with the minimum comprehensive cost based on the navigation graph, and generate a new locally optimal path segment. The path output module is used to obtain the path planning result based on the new local optimal path segment and the original global path.

[0017] The beneficial effects of the present invention are: (1) High efficiency of dynamic response capability: Through error-driven local optimization and incremental Voronoi reconstruction, updates are performed only in the affected area, realizing real-time local replanning and significantly shortening the response delay; (2) Lightweight and easy to deploy in embedded systems: Avoids frequent global reconstruction or large-scale sampling, reduces computing and storage overhead, and is suitable for running on embedded or airborne computing platforms; (3) Path quality balances safety and executability: The navigation graph edge weights are designed by combining cost and curvature penalty, and the post-processing module is combined to ensure that the output path meets the dynamic constraints, thereby reducing the burden on the online controller; (4) Optional semantic adaptation capability: The semantic passage cost is included as an option in the error metric or edge weight calculation, so that the system can achieve intelligent adaptation to different terrains or task areas when needed; (5) System-level innovation: The creativity of this invention is reflected in the systematic design of “graph structure generation - error-driven optimization - incremental reconstruction - local replanning”, rather than a simple modification of an existing search algorithm (such as A*), thus possessing a wider scope of protection and engineering value.

[0018] In summary, this invention outperforms existing technologies in terms of dynamic responsiveness, lightweight implementation, security, and intelligence, demonstrating significant innovation and broad engineering application value. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a dynamic path planning method based on Voronoi error metric according to an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, this embodiment provides a dynamic path planning update method based on Voronoi error metric, including: Step 1, Input and Preprocessing: Collect environmental data and robot data, obtain input data and perform preprocessing. The input data includes environmental grid data, obstacle boundary information, start and end point positions, optional semantic label map and robot geometric and motion constraint parameters. Step 2, Initial Seed Generation and Voronoi Construction: Based on the preprocessed data, seed sets are generated at obstacle boundaries, free space, and semantic interest points to construct a Voronoi graph; Step 3, Voronoi error-driven local optimization: Extract seed points from the Voronoi diagram, generate a set of candidate points in the neighborhood of the seed points, calculate the comprehensive error of the candidate points in the candidate point set based on the path planning cost, and update the position of the seed points based on the comprehensive error. Step 4, Incremental Voronoi Local Reconstruction: When the candidate point position is updated or the local environment changes, the Voronoi diagram is locally reconstructed to obtain the reconstructed Voronoi diagram; Step 5, Navigation Graph Construction and Path Search: Construct a navigation graph based on the reconstructed Voronoi diagram, obtain the path with the minimum overall cost based on the navigation graph, and generate a new locally optimal path segment; Step 6, Dynamic Detection and Local Replanning: Obtain the path planning result based on the new local optimal path segment and the original global path.

[0024] Step 7, Post-processing of the path and output of the trajectory: Smooth the path obtained by replanning (such as B-spline or polynomial fitting) and verify its dynamic feasibility (curvature, velocity, acceleration, etc.). If necessary, call the trajectory optimizer for further optimization to generate the final trajectory and reference velocity sent to the controller. The output trajectory and control module interface can be given in the form of time or path parameterization.

[0025] Further preprocessing of the input data includes: The input data is preprocessed by noise filtering, coordinate alignment and scale normalization, and a distance representation structure for security assessment is constructed to evaluate the accessibility and security margin of each point in the environment.

[0026] Specifically, the system inputs include an environmental occupancy grid map or 2D grid data generated from point cloud projection, obstacle boundary information, starting and target positions, and an optional semantic label map. In addition, the system also receives geometric and motion constraint information related to the robot, including but not limited to parameters such as turning radius, speed, and acceleration limits, for subsequent path feasibility and safe distance assessments.

[0027] To improve the robustness and geometric analyzability of subsequent planning, the system performs preprocessing operations such as noise filtering, coordinate alignment, and scale normalization on the input data. Based on this, a distance representation structure for safety assessment is constructed, which can be a Euclidean distance field (ESDF) or other equivalent distance descriptions, to evaluate the accessibility and safety margin of each point in the environment.

[0028] After the above processing, the system obtains a map representation and constraint parameters with a unified format and noise suppression, providing basic data support for initial seed generation and Voronoi skeleton construction.

[0029] Furthermore, seed sets are generated at obstacle boundaries, free space, and semantic interest points to construct the Voronoi diagram, including: Obstacle boundaries are sampled at preset intervals to obtain boundary seed points. In free space, a sampling strategy is used to generate uniformly distributed seed points, with the start point, end point, and key semantic interest points used as forced seeds. Merge and deduplicate the boundary seed points, evenly distributed seed points, and forced seeds to obtain the seed set; Based on the seed set, a Voronoi diagram is constructed using geometric computation methods.

[0030] Specifically, based on the preprocessing results, the system generates an initial seed point set P at obstacle boundaries, free space, and semantic interest points. Obstacle boundaries are sampled at preset intervals to obtain boundary seed points; free space uses Poisson disk sampling or other uniformly distributed sampling strategies to generate evenly distributed seed points; simultaneously, the start point, end point, and key semantic interest points are added to the set as forced seeds. The merged seed points are then deduplicated to obtain the final seed set P.

[0031] Based on this set P, a Voronoi graph is constructed using computational geometry methods (such as Qhull, CGAL, etc.), and initial skeleton nodes and edges are extracted, thus naturally forming a safe path framework, which serves as the basic structure for subsequent optimization and path planning.

[0032] Furthermore, updating the positions of candidate points based on path planning costs includes: Construct a comprehensive cost function based on the path planning cost; The comprehensive error of each candidate point is calculated based on the comprehensive cost function, and the candidate point with the smallest comprehensive error is selected as the update target. when The seed point is updated in position according to the update target, where, This represents the combined error of candidate points after sampling in the neighborhood. This represents the overall error of the current seed point. Reduce the threshold for error.

[0033] Specifically, this embodiment proposes a navigation error measurement mechanism based on the Voronoi structure. Unlike the error definition used for geometric contour fitting in the traditional Voronoi algorithm, this embodiment extends the error measurement to a comprehensive cost function oriented towards path planning: ; in: To reflect the security cost based on the distance field, the point Relative safety to obstacles; This is the optional semantic passage cost, used to reflect the passage cost for different terrains or regions; The target-oriented cost is used to encourage seed distribution to move closer to the target direction or key passage areas; it can be a function of the distance or directional deviation from the target. Weights , , This function is used to balance various costs across different tasks and can be configured according to the task or scenario. The error function is defined at the Voronoi seed point and serves as the optimization objective to guide the Voronoi skeleton to automatically converge to a new optimal structure, achieving a balance between safety, smoothness, and objective orientation in the path, thereby realizing adaptive dynamic optimization of the Voronoi skeleton.

[0034] For seed point In its neighborhood Internally generated candidate point set And calculate the error for each candidate point. Select the candidate point with the smallest error as the update target: ; To balance optimization speed and stability, a phased functional strategy is adopted: larger position updates are allowed during phases of significant error reduction for rapid convergence; smaller adjustments are made when approaching local optima to ensure smooth convergence. Update implementation is based on an error reduction threshold. :when Position updates are performed only when necessary to avoid frequent reconstructions caused by minor fluctuations. This strategy ensures a balance between real-time performance and stability in the local optimization process.

[0035] Furthermore, local reconstruction of the Voronoi diagram includes: The topological relationships between seeds are recorded using a Voronoi adjacency list, and a fast neighborhood retrieval is achieved using a KD-tree or an equivalent spatial index structure to obtain the set of affected Voronoi units. The Voronoi diagram is locally reconstructed based on the set of affected Voronoi units.

[0036] Specifically, when some seed point locations are updated or local environmental changes are detected, the system only performs Voronoi cell reconstruction and updates the topology within the affected local area.

[0037] By maintaining a Voronoi adjacency list to record the topological relationships between seeds and using a KD-tree or equivalent spatial index structure to achieve fast neighborhood retrieval, the system can quickly determine the set of affected Voronoi units. Refactoring is only performed within this scope.

[0038] This process modifies only the map geometry, enabling efficient incremental updates of the Voronoi diagram. It significantly reduces global computational complexity, making the update time linearly related to the scale of local changes, thus achieving real-time response.

[0039] Furthermore, constructing a navigation graph based on the reconstructed Voronoi diagram includes: The nodes and edges of the reconstructed Voronoi graph are mapped to a navigation graph, where nodes represent position points on the Voronoi graph, edges represent lines connecting adjacent skeleton nodes in the Voronoi graph, and the weights of the edges are obtained by the composite cost function.

[0040] Specifically, the Voronoi diagram, updated through local optimization and incremental reconstruction, is mapped as a navigation diagram. , where the set of nodes Represents key points on the Voronoi skeleton, including skeleton nodes such as intersections and endpoints; edge set. This represents the connection between adjacent skeleton nodes, forming a passable path.

[0041] For each edge Assign composite cost weights: ; in, For the edge The geometric length; To define at a point in space The distance field value on the point represents the distance field value on the point. The minimum Euclidean distance to the nearest obstacle; For the edge The statistics of semantic passage cost are used to reflect the semantic passage cost of the region where the edge is located; This is a curvature penalty term used to characterize the curvature of the path segment corresponding to the edge and to penalize high-curvature segments to improve trajectory smoothness. Parameters , , , These represent the length weight, safety distance weight, semantic cost weight, and curvature constraint weight, respectively, which are used to weigh path length, safety margin, semantic traversability, and path smoothness in specific tasks.

[0042] The system in the weighted navigation map The algorithm employs a heuristic search algorithm or an equivalent graph search method to find the path with the minimum overall cost. Let the path be... Let be an ordered set of nodes, and adjacent nodes By corresponding edge Connecting them leads to the optimal path. Defined as: ; in, In the navigation map The sequence of passable paths from the starting point to the destination.

[0043] Resulting path It simultaneously meets the requirements of safety, smoothness, and semantic rationality, providing reliable reference input for subsequent path post-processing and trajectory optimization modules.

[0044] Furthermore, step 6 includes: The system continuously monitors sensor or map incremental update data. When it detects the addition or movement of obstacles, or changes in the local environment, it automatically triggers a local replanning process. Based on the Voronoi skeleton updated after incremental reconstruction, the system performs path search in the affected submap region, generates new locally optimal path segments, and smoothly stitches them with the original global path. This mechanism forms a closed-loop process of environmental change detection, local structure update, and path replanning, enabling the system to achieve real-time path updates and continuous navigation in dynamic environments, maintaining the stability and feasibility of the overall planning.

[0045] Step 7 includes: The path obtained from replanning is smoothed (e.g., by B-spline or polynomial fitting) and its dynamic feasibility is verified (curvature, velocity, acceleration, etc.). If necessary, the trajectory optimizer is called for further optimization to generate the final trajectory and reference velocity sent to the controller. The output trajectory and control module interface can be given in time or path parameterized form.

[0046] The method in this embodiment has the following effects: (1) A Voronoi seed generation and skeleton initialization method for path planning tasks: Based on the traditional Voronoi graph generation which is only based on geometric boundaries, this invention divides seed points into three categories: obstacle boundary seeds, free space seeds, and semantic interest seeds, and introduces a forced seed mechanism to ensure that key regions (such as the starting point, target, or semantic access area) are covered. By combining Poisson disk sampling with boundary equidistant sampling, a uniform and task-related initial seed set is formed, realizing the task-adaptive initialization of the Voronoi skeleton.

[0047] (2) Redefinition and Application of Voronoi Error Metric: This invention transforms the optimization objective of Voronoi from traditional geometric fitting to navigation cost evaluation, simultaneously reflecting safety, semantic cost, and goal guidance. This error definition forms a bridge between Voronoi structure and path planning, enabling skeleton generation to no longer be limited to geometric features but directly respond to changes in navigation cost. This redefined error metric provides a unified theoretical basis for the dynamic optimization and adaptive adjustment of Voronoi skeletons.

[0048] (3) Error-driven local seed point optimization mechanism: Candidate sampling points are generated in the local neighborhood of each seed point. The navigation error of the candidate points is evaluated, and the optimal update direction and position are selected based on the error reduction trend. To balance the convergence speed and stability of the algorithm, a phased stepping mechanism is introduced. The optimization response is accelerated during the stage of significant error reduction, and the stepping magnitude is smoothly adjusted when approaching the optimal region, thereby achieving a fast and stable local convergence effect. This mechanism enables the Voronoi skeleton to achieve continuous adaptive optimization in dynamic environments with low computational cost.

[0049] (4) Error-driven incremental reconstruction and local replanning closed-loop mechanism: After detecting changes in the environment or seed point, the system determines the region to be updated based on the error signal, performs Voronoi reconstruction only within a local area, and triggers path replanning. An adaptive closed-loop update mechanism is formed through local reconstruction and smooth path stitching, realizing a dynamic response process from environmental change detection, skeleton update to path output. This structure significantly reduces the computational burden of global replanning and enables real-time continuous updating of path planning results.

[0050] This embodiment also provides a dynamically updated path planning system based on Voronoi error metric, including: The data processing module is used to collect environmental data and robot data, acquire input data and perform preprocessing. The input data includes environmental grid data, obstacle boundary information, start and end point positions, optional semantic label map and robot geometric and motion constraint parameters. The Voronoi graph construction module is used to generate seed sets at obstacle boundaries, free space, and semantic interest points based on preprocessed data, and construct Voronoi graphs. The Voronoi diagram reconstruction module is used to extract seed points from the Voronoi diagram and generate a set of candidate points in the neighborhood of the seed points; calculate the comprehensive error of the candidate points in the candidate point set according to the path planning cost, and update the position of the seed points according to the comprehensive error; when the position of the seed points is updated or the local environment changes, the Voronoi diagram is locally reconstructed to obtain the reconstructed Voronoi diagram. The path search module is used to construct a navigation graph based on the reconstructed Voronoi diagram, obtain the path with the minimum overall cost based on the navigation graph, and generate a new locally optimal path segment. The path output module is used to obtain the path planning results based on the new local optimal path fragment and the original global path.

[0051] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A dynamic path planning method based on Voronoi error metric, characterized in that, include: Collect environmental data and robot data, acquire input data and preprocess it, wherein the input data includes environmental grid data, obstacle boundary information, start and end point positions, optional semantic label map and robot geometric and motion constraint parameters; Based on the preprocessed data, seed sets are generated at obstacle boundaries, free space, and semantic interest points to construct a Voronoi diagram; Seed points are extracted from the Voronoi diagram, and a set of candidate points is generated in the neighborhood of the seed points. The candidate points in the candidate point set are subjected to a comprehensive error calculation based on the path planning cost, and the seed points are updated according to the comprehensive error, including: Based on the path planning cost, construct a comprehensive cost function; The comprehensive error of each candidate point is calculated based on the comprehensive cost function, and the candidate point with the smallest comprehensive error is selected as the update target. when The seed point is updated in position according to the update target, wherein, This represents the combined error of candidate points after sampling in the neighborhood. This represents the overall error of the current seed point. Reduce the threshold for error; The comprehensive cost function for: ； in, This is a security cost based on the distance field; This represents the optional semantic passage cost; Costs guided by objectives; weights , , Used to balance the costs across different tasks; When the seed point position is updated or the local environment changes, the Voronoi diagram is locally reconstructed to obtain the reconstructed Voronoi diagram; A navigation graph is constructed based on the reconstructed Voronoi graph. The path with the minimum overall cost is obtained based on the navigation graph, and a new locally optimal path segment is generated. Based on the new locally optimal path segment and the original global path, the path planning result is obtained.

2. The dynamic path planning method based on Voronoi error metric according to claim 1, characterized in that, Preprocessing the input data includes: The input data is preprocessed by noise filtering, coordinate alignment and scale normalization, and a distance representation structure for security assessment is constructed to evaluate the accessibility and security margin of each point in the environment.

3. The dynamic path planning method based on Voronoi error metric according to claim 2, characterized in that, Generate seed sets at obstacle boundaries, free space, and semantic interest points, and construct a Voronoi diagram including: The obstacle boundary is sampled at a preset interval to obtain boundary seed points. In the free space, a sampling strategy is used to generate uniformly distributed seed points, with the start point, end point and key semantic interest points as forced seeds. The boundary seed points, the uniformly distributed seed points, and the forced seeds are merged and deduplicated to obtain the seed set; The Voronoi diagram is constructed using geometric calculation methods based on the seed set.

4. The dynamic path planning method based on Voronoi error metric according to claim 1, characterized in that, Local reconstruction of the Voronoi diagram includes: The topological relationships between seeds are recorded using a Voronoi adjacency list, and a fast neighborhood retrieval is achieved using a KD-tree or an equivalent spatial index structure to obtain the set of affected Voronoi units. The Voronoi diagram is locally reconstructed based on the affected Voronoi cell set.

5. The dynamic path planning method based on Voronoi error metric according to claim 1, characterized in that, Constructing a navigation graph based on the reconstructed Voronoi diagram includes: The nodes and edges of the reconstructed Voronoi graph are mapped to the navigation graph, where nodes represent position points on the Voronoi graph, edges represent lines connecting adjacent skeleton nodes in the Voronoi graph, and the weights of the edges are obtained by the composite cost function. The composite cost function for: ； in, For the edge geometric length, For spatial points The distance field value represents the point. The minimum Euclidean distance to the nearest obstacle. For the edge Statistics on semantic access costs For curvature penalty term, , , , These represent the length weight, safety distance weight, semantic cost weight, and curvature constraint weight, respectively.

6. The dynamic path planning method based on Voronoi error metric according to claim 5, characterized in that, The paths to obtain the minimum overall cost include: ； in, The path with the lowest overall cost, In the navigation map The sequence of passable paths from the starting point to the destination.

7. A dynamic path planning system based on Voronoi error metric, used to implement the dynamic path planning method based on Voronoi error metric as described in any one of claims 1-6, characterized in that, include: The data processing module is used to collect environmental data and robot data, acquire input data and perform preprocessing. The input data includes environmental grid data, obstacle boundary information, start and end point positions, optional semantic label map and robot geometric and motion constraint parameters. The Voronoi graph construction module is used to generate seed sets at obstacle boundaries, free space, and semantic interest points based on preprocessed data, and construct Voronoi graphs. The Voronoi graph reconstruction module is used to extract seed points from the Voronoi graph, generate a set of candidate points in the neighborhood of the seed points, calculate the comprehensive error of the candidate points in the candidate point set according to the path planning cost, update the position of the seed points according to the comprehensive error, and perform local reconstruction of the Voronoi graph when the position of the seed points is updated or the local environment changes, so as to obtain the reconstructed Voronoi graph. The path search module is used to construct a navigation graph based on the reconstructed Voronoi graph, obtain the path with the minimum comprehensive cost based on the navigation graph, and generate a new locally optimal path segment. The path output module is used to obtain the path planning result based on the new local optimal path segment and the original global path.

Images