A multi-granularity path planning method based on environmental complexity self-adaption
By constructing multi-resolution maps and using adaptive path planning driven by environmental complexity factors, the problems of lagging planning granularity switching and poor path quality in existing technologies are solved, achieving efficient and accurate path planning in complex environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG UNIV OF TECH
- Filing Date
- 2026-04-30
- Publication Date
- 2026-06-05
AI Technical Summary
Existing multi-granularity path planning methods lack quantitative assessment of environmental complexity, resulting in lagging or excessively frequent switching of planning granularity. Traditional algorithms are inefficient in large environments or have slow convergence speed in narrow channels, and heuristic weights need to be manually set and are difficult to adaptively adjust, leading to poor path quality and curvature discontinuity.
A multi-resolution occupancy grid map is constructed, and the environmental complexity factor is calculated by obstacle density and information entropy. The planning granularity level is dynamically selected and the heuristic weights are adjusted. Curvature penalty terms and weighted B-spline curves are introduced to smooth the path, forming an adaptive path planning closed loop.
It achieves efficient and accurate path planning in complex environments, balancing planning efficiency and path quality, ensuring path continuity and feasibility, and adapting to environmental changes.
Smart Images
Figure CN122149494A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of robot autonomous navigation and path planning technology, and in particular relates to a multi-granularity path planning method based on environmental complexity adaptation. Background Technology
[0002] Path planning is one of the core technologies for autonomous navigation of mobile robots. Its goal is to plan a collision-free feasible path from the starting point to the target point within the workspace. Based on the level of environmental information available, path planning can be divided into two categories: global planning based on a known static map and local planning based on real-time sensor perception. With the continuous expansion of applications such as autonomous driving, warehousing and logistics, and service robots, the working environments faced by robots are becoming increasingly complex, often containing multiple structures such as open areas, areas with dense obstacles, and narrow passages. This places higher demands on path planning technology: it needs to quickly reach the target in simple, open scenarios, and accurately plan feasible paths in complex, narrow scenarios. To address this, researchers have begun to explore multi-granularity path planning methods that combine global coarse planning with local fine planning, aiming to balance planning efficiency and accuracy under varying environmental complexities.
[0003] However, existing multi-granularity path planning methods still have the following shortcomings: First, they lack quantitative assessment methods for environmental complexity and cannot automatically select the appropriate map resolution level based on the accessibility of local areas, resulting in delayed or excessively frequent switching of planning granularity. Second, traditional graph search algorithms (such as Dijkstra's algorithm) are inefficient and computationally complex in large environments, while random sampling-based methods (such as RRT and PRM) have slow convergence speed and poor path quality in narrow passage environments and are insensitive to environment type. Third, local planning algorithms (such as DWA) are prone to getting trapped in local optima when lacking global guidance, resulting in poor path smoothness. Fourth, heuristic weights in existing methods usually need to be set manually, making it difficult to automatically adjust the search strategy according to environmental changes. They cannot achieve an adaptive balance between accelerating the search and path optimality, and when switching planning granularity, there are often curvature discontinuities at the connection points of segmented paths, affecting the executability of the trajectory.
[0004] Therefore, this invention proposes a multi-granularity path planning method based on adaptive environmental complexity. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention proposes a multi-granularity path planning method based on adaptive environmental complexity, thereby resolving the issues present in the prior art.
[0006] To achieve the above objectives, this invention provides a multi-granularity path planning method based on adaptive environmental complexity, comprising: Construct at least two layers of occupied raster maps with different resolutions; On the lowest resolution map, obstacle density and information entropy are calculated based on the obstacle occupancy ratio within the sliding window, and the environmental complexity factor is obtained through weighted fusion. The granularity level of the current plan is determined based on the environmental complexity factor, and the heuristic weights of the path search algorithm are adjusted based on the granularity level. Based on the selected granularity level, a global path search is performed using an evaluation function containing curvature penalty terms and heuristic weights to obtain the planned path; When the change in environmental complexity triggers a map level switch based on the planned path, the fusion anchor point is determined by using the end point of the previous level path and the starting point of the next level path. A weighted B-spline curve is then used to smoothly fuse the segmented paths based on the fusion anchor point to obtain a continuous path.
[0007] Optionally, the process of constructing at least two layers of occupied raster maps with different resolutions includes: constructing three layers of occupied raster maps with fine, medium and coarse resolutions, wherein the raster side lengths of adjacent layers increase in an integer multiple ratio; The grid side length ratio of the fine, medium, and coarse map layers is 1:2:4.
[0008] Optionally, on the lowest resolution map, the process of calculating obstacle density and information entropy based on the obstacle occupancy ratio within the sliding window, and then obtaining the environmental complexity factor through weighted fusion, includes: A square sliding window is set with the current position as the center, and the obstacle density is obtained based on the ratio of the number of occupied grids in the square sliding window to the total number of grids; Calculate information entropy based on the obstacle density; The obstacle density and the information entropy are fused using a weighted summation method to obtain the environment complexity factor.
[0009] Optionally, the expression for calculating the information entropy is: ; In the formula, For information entropy, The density is the obstacle density.
[0010] Optionally, the process of determining the granularity level of the current plan based on the environmental complexity factor includes: The environmental complexity factor is compared with the preset lower and upper thresholds of complexity to obtain the comparison results; Based on the comparison results, a low-resolution map is selected when the environmental complexity factor is less than the lower threshold, a medium-resolution map is selected when it is between the lower threshold and the upper threshold, and a high-resolution map is selected when it is greater than the upper threshold.
[0011] Optionally, the expression for the heuristic weights based on the granularity-level adjustment path search algorithm is: ; In the formula, Scaling factor For heuristic weights, This represents the environmental complexity factor.
[0012] Optionally, the expression for the valuation function is: ; In the formula, From the starting point to the current node The actual cumulative path cost, For the current node Heuristic cost estimation to the target point For heuristic weights, For the path in the node Curvature penalty term at the location, For curvature penalty weighting coefficients, This is the evaluation function.
[0013] Optionally, the process of smoothly fusing the segmented paths using weighted B-spline curves based on the fusion anchor points to obtain a continuous path includes: Calculate the Euclidean distance between the end point of the previous level path and the start point of the next level path, and mark the corresponding point pairs whose Euclidean distance is less than a set threshold as fusion anchor points; Using the fusion anchor point as the control vertex, a cubic uniform B-spline curve is constructed for interpolation smoothing to obtain a continuous path.
[0014] Optionally, the expression for calculating a cubic uniform B-spline curve is: ; In the formula, For the coordinates of the control points, For cubic B-spline basis functions, For control point weighting coefficients, It is represented by a curve.
[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention constructs a multi-resolution occupancy grid map and calculates obstacle density and information entropy on the lowest resolution map using a sliding window. Weighted fusion yields a quantitative environmental complexity factor characterizing local passage difficulty, thus achieving an objective assessment of environmental complexity. This complexity factor simultaneously drives the dynamic selection of planning granularity levels and the continuous adjustment of heuristic weights, enabling the algorithm to automatically accelerate search in simple, open environments and prioritize path optimality in complex, narrow environments. This overcomes the shortcomings of traditional methods, such as lag in granularity switching and the need for manual setting of heuristic weights. A curvature penalty term is introduced into the path search to effectively suppress sharp path transitions and generate smoother trajectories. When environmental changes trigger granularity switching, weighted B-spline curves are used to smoothly fuse segmented paths, eliminating curvature discontinuities at path connections. The overall method forms a complete closed loop from environmental assessment, granularity decision-making, adaptive search to path smoothing, balancing planning efficiency, accuracy, and trajectory executability in complex dynamic environments. Attached Figure Description
[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the multi-scale raster map pyramid structure according to an embodiment of the present invention. Detailed Implementation
[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0018] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0019] Example 1 The overall framework of this plan is as follows: Figure 1As shown, to address the challenge of balancing efficiency and accuracy in complex environments using traditional path planning, this scheme constructs a three-layer multi-scale raster map pyramid (fine, medium, and coarse). Obstacle density and information entropy are statistically analyzed using a sliding window on the coarse-grained map, and then fused to generate an environmental complexity factor. This complexity factor drives the granularity decision controller to dynamically select the map resolution level and continuously adjusts the heuristic weights of the improved A* algorithm, achieving an adaptive balance between search speed and path quality. When granularity switching results in segmented paths, weighted B-spline curves are used to smoothly fuse anchor points, outputting a continuous and executable trajectory. The system also includes a dynamic environmental feedback loop, supporting real-time replanning. The specific steps are as follows: The first step involves the mobile robot acquiring real-time 3D point cloud data of its surrounding environment using its onboard LiDAR, depth camera, or vision sensor. The raw point cloud is then subjected to distortion correction, statistical filtering for noise reduction, and coordinate system normalization transformation to obtain structured environmental information represented in the robot's body coordinate system.
[0020] The second step involves simultaneously constructing at least two layers of occupancy raster maps with different resolutions based on the preprocessed environmental data. Ideally, three layers should be constructed: a fine-resolution map, a medium-resolution map, and a coarse-resolution map. Figure 2 As shown, maps of different granularities are distinguished by setting different grid side lengths. In the specific parameter settings, the grid side lengths of adjacent map layers are set to be integer multiples of each other. Preferably, the resolution scaling factor is set to 2, meaning the grid side lengths of the fine, medium, and coarse maps are constructed in a 1:2:4 ratio to ensure accurate alignment of different granularity levels in spatial projection. The fine-resolution map is used to accurately describe local obstacle boundaries and narrow passages, while the coarse-resolution map is used for rapid global overview searches. Spatial correspondence between different levels is maintained through grid aggregation.
[0021] The third step is to set a square sliding window with a size of W×W grid cells on the coarse-grained map, centered on the robot's current position, where W is an integer between 3 and 7. Then, iterate through all the cells within the window and determine the number of cells that are considered occupied. Total number of grid cells within the window .
[0022] The fourth step is to calculate the obstacle density within the window area. : ; This density value quantitatively reflects the proportion of the area occupied by obstacles in a local space.
[0023] Fifth, treat the occupancy status of the grid within the window as a binary random variable, and approximate the occupancy probability as... Then the free probability is Calculate the information entropy of this local region. To depict the degree of disorder in the distribution of obstacles: ; when When approaching 0 or 1, The smallest value indicates a simple environmental structure. When... When it approaches 0.5, The largest value indicates the most complex interplay between obstacles and free space.
[0024] Step 6: Adjust obstacle density With information entropy Perform weighted fusion and define the environment complexity factor. as follows: ; in and satisfy Typical value , In a preferred embodiment, take , This complexity factor comprehensively reflects the impact of obstacle density and spatial disorder on the difficulty of path planning.
[0025] Step 7, based on complexity factor Dynamically determine the current planning granularity level. Set a threshold for pre-defined complexity. With upper threshold The result calculated in step six The value is compared with two thresholds respectively, and the following granularity selection logic is executed: when When a coarse-grained map is used, subsequent planning should be carried out; when When a medium-granularity map is used, subsequent planning should be carried out; when At this time, a fine-grained map is selected for subsequent planning. To avoid frequent granularity fluctuations, a hysteresis anti-jitter mechanism is introduced, and the granularity level switch is only actually executed when the C-value judgment results of three consecutive control cycles are consistent.
[0026] Step 8: Based on the granularity level selected in Step 7, calculate the improvement... Heuristic weights of the algorithm : ; in, This is the scaling factor, typically ranging from 1.0 to 3.0. When... When it is small, that is, in a simple and open environment, Automatically increase to speed up the search; when When it is large, that is, in a complex and narrow environment, It automatically approaches 1, making the search more focused on path optimality. The value is determined by Continuous drive eliminates the need for manual setting of segment thresholds, enabling adaptive adjustment.
[0027] Step 9: During the node expansion process, for the two path directions formed by three consecutive path nodes, calculate the change in the included angle. Then the curvature penalty term of the current node Defined as This penalty term is used to suppress sharp turns in the path, prompting the planner to tend to generate paths with gentler turns.
[0028] Step 10: Based on the granularity level determined in Step 7, call the map layer with the corresponding resolution and apply the improved... The algorithm performs a global path search. An evaluation function for path node expansion is defined. for ; in, From the starting point to the current node The actual cumulative path cost, For the current node The heuristic cost estimate to the target point can be calculated using Euclidean distance. These are heuristic weighting coefficients; For the path in the node Curvature penalty term at the location, This is the curvature penalty weighting coefficient.
[0029] Step 11: When the robot's planning granularity changes due to variations in environmental complexity during its movement, record the last M waypoints of the previous granularity path segment and the first M waypoints of the next granularity path segment, where M ranges from 5 to 20. For these two sets of waypoints, calculate the Euclidean distance in the global coordinate system, and select the waypoints whose distances are less than a set threshold. The corresponding point pairs are marked as fusion anchor points.
[0030] Step 12: Using the fusion anchor points determined in Step 11 as control vertices, construct a cubic uniform B-spline curve to interpolate and smooth the segmented path. The curve expression is: ; in For the coordinates of the control points, For cubic B-spline basis functions, These are the control point weighting coefficients. During the fusion process, path points from the fine-grained planning results are assigned higher weights to reflect the reliability advantages of fine-grained planning paths.
[0031] Step 13 involves performing velocity planning and time parameterization on the smooth, continuous geometric path obtained in Step 12, applying constraints on maximum linear velocity, maximum angular velocity, and maximum acceleration to generate a time-domain trajectory sequence that conforms to the robot's kinematic constraints. This sequence is then sent to the underlying motion controller for execution. During the journey, new sensor data is continuously received, and the environmental complexity is updated. If a change in complexity triggers a granularity switch, the process returns to Step 7 to re-execute the planning, continuously adapting to the dynamic environment.
[0032] This invention constructs a three-layer, multi-scale grid map pyramid structure (fine, medium, and coarse), and implements a dynamic feedback closed loop based on environmental complexity. The scheme sets the side lengths of adjacent grid layers to an integer multiple of 1:2:4 to provide a unified spatial reference. The system continuously updates environmental information during its movement, adjusting the planning granularity in real time according to changes in environmental complexity. A coarse-grained map is used for global evaluation and rapid preliminary search, while a fine-grained map enables precise obstacle avoidance and passage through narrow passages. When complexity triggers a switching signal, the system re-executes the planning, ensuring that the robot can balance the efficiency and accuracy of path planning with the reliability of real-time obstacle avoidance in complex dynamic environments.
[0033] This invention defines a sliding window on a coarse-grained map and integrates obstacle density and information entropy to generate an environmental complexity factor. This factor quantitatively assesses the traversal difficulty of local areas, comprehensively reflecting the impact of obstacle density and spatial disorder on path planning. The environmental complexity factor drives the selection of planning granularity and the adaptive adjustment of heuristic weights, thereby achieving a balance between accuracy and efficiency and adapting to dynamically changing environments.
[0034] In traditional path planning algorithms, heuristic weights typically require manual setting. This invention proposes a mechanism that dynamically adjusts heuristic weights based on environmental complexity factors. When the environment is simple and open, the heuristic weights automatically increase to accelerate the search; when the environment is complex and confined, the heuristic weights approach 1, prioritizing path optimality. This allows path planning to achieve efficient and accurate searches in different environments.
[0035] This invention introduces a curvature penalty term into path planning. By limiting the change in the directional angle of continuous path segments, it suppresses sharp turns in the path, resulting in a smoother path. Furthermore, when granularity switching causes path breakage, a cubic uniform B-spline curve is used to perform weighted smoothing and merging of the segmented paths. This gives fine-grained path points higher weight during the merging process, ensuring the reliability and executability of the smooth path.
[0036] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A multi-granularity path planning method based on adaptive environmental complexity, characterized in that, Includes the following steps: Construct a raster map with at least two layers of different resolutions; On the lowest resolution map, obstacle density and information entropy are calculated based on the obstacle occupancy ratio within the sliding window, and the environmental complexity factor is obtained through weighted fusion. The granularity level of the current plan is determined based on the environmental complexity factor, and the heuristic weights of the path search algorithm are adjusted based on the granularity level. Based on the selected granularity level, a global path search is performed using an evaluation function containing curvature penalty terms and heuristic weights to obtain the planned path; When the change in environmental complexity triggers a map level switch based on the planned path, the fusion anchor point is determined by using the end point of the previous level path and the starting point of the next level path. A weighted B-spline curve is then used to smoothly fuse the segmented paths based on the fusion anchor point to obtain a continuous path.
2. The multi-granularity path planning method based on adaptive environmental complexity according to claim 1, characterized in that, The process of constructing at least two layers of occupied raster maps with different resolutions includes: constructing three layers of occupied raster maps with fine, medium and coarse resolutions, wherein the raster side lengths of adjacent layers increase in an integer multiple ratio; The grid side length ratio of the fine, medium, and coarse map layers is 1:2:
4.
3. The multi-granularity path planning method based on adaptive environmental complexity according to claim 1, characterized in that, On the lowest resolution map, the process of calculating obstacle density and information entropy based on the obstacle occupancy ratio within a sliding window, and then obtaining the environmental complexity factor through weighted fusion, includes: A square sliding window is set with the current position as the center, and the obstacle density is obtained based on the ratio of the number of occupied grids in the square sliding window to the total number of grids; Calculate information entropy based on the obstacle density; The obstacle density and the information entropy are fused using a weighted summation method to obtain the environment complexity factor.
4. The multi-granularity path planning method based on adaptive environmental complexity according to claim 3, characterized in that, The expression for calculating the information entropy is: ； In the formula, For information entropy, The density is the obstacle density.
5. The multi-granularity path planning method based on adaptive environmental complexity according to claim 1, characterized in that, The process of determining the granularity level of the current plan based on the aforementioned environmental complexity factors includes: The environmental complexity factor is compared with the preset lower and upper thresholds of complexity to obtain the comparison results; Based on the comparison results, a low-resolution map is selected when the environmental complexity factor is less than the lower threshold, a medium-resolution map is selected when it is between the lower threshold and the upper threshold, and a high-resolution map is selected when it is greater than the upper threshold.
6. The multi-granularity path planning method based on adaptive environmental complexity according to claim 1, characterized in that, The expression for the heuristic weights based on the aforementioned granularity-level adjustment path search algorithm is as follows: ； In the formula, Scaling factor For heuristic weights, This represents the environmental complexity factor.
7. The multi-granularity path planning method based on adaptive environmental complexity according to claim 1, characterized in that, The expression for the valuation function is: ； In the formula, From the starting point to the current node The actual cumulative path cost, For the current node Heuristic cost estimation to the target point For heuristic weights, For the path in the node Curvature penalty term at the location, For curvature penalty weighting coefficients, This is the evaluation function.
8. The multi-granularity path planning method based on adaptive environmental complexity according to claim 1, characterized in that, The process of smoothly fusing segmented paths using weighted B-spline curves based on the fusion anchor points to obtain continuous paths includes: Calculate the Euclidean distance between the end point of the previous level path and the start point of the next level path, and mark the corresponding point pairs whose Euclidean distance is less than a set threshold as fusion anchor points; Using the fusion anchor point as the control vertex, a cubic uniform B-spline curve is constructed for interpolation smoothing to obtain a continuous path.
9. The multi-granularity path planning method based on adaptive environmental complexity according to claim 8, characterized in that, The expression for calculating a cubic uniform B-spline curve is: ； In the formula, For the coordinates of the control points, For cubic B-spline basis functions, For control point weighting coefficients, It is represented by a curve.