Robot path planning method based on K-value adaptive multi-algorithm cooperation

Abstract

The invention provides a robot intelligent path planning method based on K-value adaptive multi-algorithm cooperation, and belongs to the field of robot path planning. The robot path planning method mainly comprises the following steps: preprocessing an environment map, and constructing a three-layer data structure of a quadtree, a Voronoi graph and a connected graph; generating an initial path based on an improved fast extension random tree algorithm, and extracting key points; establishing a K value evaluation system including five dimensions of obstacle density, local complexity, channel characteristics, path curvature and topology complexity, and realizing accurate quantification of the environment; carrying out adaptive region division based on the K value, and dividing the environment into an open region, a narrow channel region and an obstacle dense region; aiming at the characteristics of different regions, an improved bidirectional A algorithm and an improved fast expansion random tree algorithm are respectively adopted to carry out collaborative planning; and finally, smoothing the path through a Bezier curve. According to the method, the quality and the planning efficiency of path planning in a complex environment can be improved, and higher robustness is achieved.

Description

Technical Field

[0001] The present invention relates to a robot path planning method in a complex environment and belongs to the field of robot path planning. Background Art

[0002] The robot path planning method refers to the technology of finding a feasible path from the starting point to the target point for a robot in a specific working environment. In the field of robot path planning, traditional single-algorithm methods have their own advantages and disadvantages: Search-based algorithms (such as A*, Dijkstra) are efficient in calculation and can ensure optimality in simple environments, but have high computational overhead in complex environments; Sampling-based algorithms (such as Rapidly-exploring Random Trees, PRM) have good exploration capabilities and can effectively handle high-dimensional spaces and complex constraints, but it is difficult to guarantee the path quality; Improved algorithms (such as Rapidly-exploring Random Trees, Informed-Rapidly-exploring Random Trees) have improved in path quality, but the calculation time has increased significantly. Single algorithms often cannot meet the requirements of both efficiency and quality in practical applications, especially in environments with large variations in complexity, and their performance is not ideal.

[0003] In recent years, hybrid algorithms and adaptive strategies have become research hotspots in the field of robot path planning. However, most of these methods are limited to improvements in specific scenarios or specific algorithms, lacking systematic quantitative analysis of environmental complexity and without establishing a complete multi-algorithm collaborative framework. Especially in complex environments, how to accurately evaluate local area characteristics and select the most suitable algorithm accordingly remains a key problem to be solved.

[0004] In actual application scenarios, the environment often exhibits obvious local regional characteristics, which can be divided into open areas, narrow channels, obstacle-dense areas, etc. The diversity of such characteristics requires the planning algorithm to have stronger adaptability and intelligence. Existing path planning technologies in complex environments mainly include the following two types. One is to improve traditional path planning methods for special environments, and the other is a path planning method based on adaptive parameter adjustment. However, these methods have deficiencies: The former fails to quantify the local environmental complexity and is difficult to balance efficiency and path quality; The latter relies heavily on empirical parameters and is prone to cause planning delays and instability. The present invention proposes a robot intelligent path planning method based on K-value adaptive multi-algorithm collaboration. By constructing a local feature evaluation model, it realizes real-time quantification of environmental complexity, thereby intelligently selecting the optimal planning algorithm, effectively improving the response speed, stability, and path optimization, and providing more reliable technical support for robot autonomous navigation in complex environments. Summary of the Invention

[0005] In view of the defects existing in the prior art, the present application provides an intelligent path planning method based on K - value adaptive multi - algorithm cooperation. This method uses a multi - parameter K - value evaluation system to accurately quantify environmental characteristics, and combines adaptive region division and a multi - algorithm cooperation framework. It can not only intelligently select the most suitable planning strategy to improve the adaptability and planning efficiency of the robot in complex environments, but also ensure the path quality through multi - level path optimization, and can be widely applied to various environments with significant regional characteristics such as complex obstacle distributions and narrow channels.

[0006] To achieve the above object, the technical solution of the present application is: A robot intelligent path planning method based on K - value adaptive multi - algorithm cooperation, specifically including the following steps:

[0007] Step 1: Pre - process the environmental map;

[0008] Model the obstacles in the environmental map and convert them into a grid map; combine the actual size of the robot and perform dilation processing on the obstacles; construct a three - layer data structure, including a quadtree for fast spatial retrieval, a Voronoi diagram for channel analysis, and a graph structure representing environmental connectivity.

[0009] Step 2: Generate an initial path based on the rapidly - exploring random tree path planning algorithm;

[0010] Step 3: Extract path key points;

[0011] Based on the planned initial path points, quadtree, Voronoi diagram, and connectivity graph, extract channel entrance points, exit points, corner points, and intersection points and merge them as path key points; on this basis, perform importance scoring on all identified key points and filter out redundant points.

[0012] Step 4: Establish an enhanced K - value evaluation system with multi - dimensional features;

[0013] Step 5: Adaptive region division;

[0014] Based on the path key points extracted in Step 3 and the K - value evaluation system in Step 4, divide the environment into open areas, narrow - channel areas, and obstacle - dense areas, and adjust the region radius size to adapt to different environmental characteristics.

[0015] Step 6: Multi - algorithm cooperative planning;

[0016] Calculate the K value of each sub-region after regional division. For the open area, select the bidirectional A* algorithm and quickly find the path through the optimized heuristic function and bidirectional search strategy; for the narrow channel area, on the basis of the original improved rapidly-exploring random tree algorithm, increase the probability of direction-guided sampling; for the area with dense obstacles, on the basis of the original improved rapidly-exploring random tree algorithm, add target-oriented sampling and dynamically adjust the step size according to the surrounding obstacle environment information;

[0017] Step 7: Perform path smoothing to obtain the final path.

[0018] Specifically, the three-layer data structure in Step 1 is established as follows: The quadtree is constructed by recursive splitting, dividing the non-homogeneous region into four child nodes; The Voronoi diagram is a network of boundary lines formed by connecting all points in the plane that are equidistant from the two nearest obstacles; The connectivity graph is established with Voronoi vertices as nodes, and the edge weights consider the path length and channel width.

[0019] Specifically, the specific method for generating the initial path in Step 2 using the rapidly-exploring random tree path planning algorithm is as follows:

[0020] Step 2.1: Initialize two rapidly-exploring random trees at the initial position X start and the target position X final respectively, denoted as T start and T goal ; Set the step size Δd and the maximum number of iterations N max ; Define the probability distribution q, p, w in the hybrid sampling strategy, such that q + p + w = 1, corresponding to target-oriented sampling, direction-guided sampling, and global uniform sampling respectively;

[0021] Step 2.2: Generate a random sampling point X rand in the configuration map through the hybrid sampling strategy. Select the target point X final as the sampling point with probability q, sample in the ±45° sector range in the current path direction with probability p, and perform uniform random sampling with probability w; Then perform a collision detection on the sampling point to ensure that it is in the free space;

[0022] Step 2.3: Find the node X rand with the closest Euclidean distance to X nearest in the existing rapidly-exploring random tree as the potential parent node, construct a path segment connecting X rand and X nearest , and perform a collision detection to verify the feasibility of the path segment;

[0023] Step 2.4: If the path satisfies the collision-free constraint, then add X rand as a new node to the rapidly-exploring random tree structure and establish a connection with Xnearest The parent - child relationship, calculate from X nearest to X rand the growth direction and step length;

[0024] Step 2.5: According to the step length, generate a new node X nearest from X rand in the direction of X new , check the collision situation of the path segment from X nearest to Xnew . If there is no collision, add X new to the tree structure and establish the parent - child relationship;

[0025] Step 2.6: Alternately expand the two trees T start and T goal . After each expansion, check the shortest distance between the two trees. When the distance is less than the set threshold, try to connect the two trees. If the connection is successful, a feasible path is found;

[0026] Step 2.7: Trace back from the connection point to the root nodes of the two trees, and merge the two path segments to obtain a complete path.

[0027] Specifically, the process of extracting path key points described in Step 3 is as follows: Identify the channel area based on the Voronoi diagram, and extract the entrance point and exit point of the channel as key nodes; Calculate the included angle θ formed by three adjacent points on the path. When θ is less than the angle threshold θ threshold , regard the middle point as a key node; Identify the node degree φ based on the connected graph. When φ is greater than the degree threshold φ threshold , mark it as a crossing point and regard it as a key node.

[0028] Specifically, the evaluation parameters in the enhanced K - value evaluation system described in Step 4 include: obstacle density W, local complexity γ, channel feature η, path curvature σ, and topological complexity The calculation methods are as follows:

[0029]

[0030] In the formula, S obs is the occupied area of obstacles in the region, S total is the total area of the region, d i is the distance from the obstacle to the center of the region, d avg is the average distance from all obstacles to the center of the region, n is the number of obstacles, Δθ is the channel turning change, w ref is twice the width of the robot, w min is the minimum width of the channel, k i is the curvature of path segment i, l i is the length of path segment i, L is the total length of the regional path, N cis the number of intersections, N max is the maximum number of intersections, D b is the degree of branching, D t is the total connectivity.

[0031] Specifically, the adaptive region division process described in step 5 is as follows:

[0032] Step 5.1: Based on the position of each key point extracted in step 3, set the basic radius R of the corresponding circular region base , and its calculation formula is:

[0033]

[0034] In the formula, L tol is the total path length planned by the improved rapidly-exploring random tree algorithm, n key is the number of key points extracted;

[0035] Step 5.2: Calculate the local obstacle density W and local space complexity γ of the region, and correct the basic radius R of the circular region base is R, and its calculation method is:

[0036] R = R base (1 + k w W + k y γ)

[0037] In the formula, k w is the density adjustment coefficient, and k y is the complexity adjustment coefficient;

[0038] Step 5.3: Calculate the K value of each region. According to the K value of each region, divide the regions into open regions, narrow channel regions, and obstacle-dense regions. The corresponding radii of the circular regions are denoted as R open , R nar , R den ;

[0039] Perform adaptive adjustment on the radius of each region, and the corresponding calculation method is as follows:

[0040]

[0041] In the formula, μ is the region expansion coefficient, and v1, v2 are the region contraction coefficients;

[0042] Step 5.4: Find the path segment not covered by the circular region, find the earliest generated parent node and the last generated child node in this segment, and use the Euclidean distance between these two characteristic nodes as the radius of the supplementary circular region;

[0043] Step 5.5: Calculate the overlap degree O and the K - value difference ΔK of adjacent regions, and merge the regions with large adjacent overlap or small K - value difference. Use the Euclidean distance between the earliest - generated parent node and the last - generated child node of the region as the radius of the merged circular region.

[0044] Specifically, in the bidirectional A* algorithm in step 6, the optimized heuristic function is: f(n) = g(n)+w1D(n)+w2C(n)

[0045] In the formula, g(n) is the current path cost, D(n) is the Euclidean distance from the node to the target, C(n) is the direction cost, and w1, w2 are dynamic weight coefficients;

[0046] Adopt a bidirectional search strategy with dynamic priority, and its specific implementation process is as follows: Calculate the expansion index to dynamically determine which search tree to expand preferentially, where d(n) is the shortest Euclidean distance from the current node to the opposite - side search tree; if E start <E goal , preferentially expand the start - point tree; if E start >E goal , preferentially expand the end - point tree;

[0047] If E start =E goal , then alternately expand the two trees.

[0048] Specifically, the method for dynamically adjusting the step size according to the surrounding obstacle environment information in step 6 is as follows:

[0049]

[0050] In the formula, base step is the base step size, d obs is the distance from the sampling point to the obstacle, d safe is the set safety distance, k is the direction influence factor, and θ goal is the angle between the current expansion direction and the target direction.

[0051] The present invention has at least the following beneficial effects on the prior art:

[0052] By constructing a quadtree, a Voronoi diagram, and a connected Figure 3 layer data structure, the efficiency of the algorithm in complex - environment space retrieval and channel analysis is improved. Combining the path key - point extraction and redundant - point filtering mechanism ensures the global rationality of path planning.

[0053] The algorithm quantifies environmental features in multiple dimensions through an enhanced K-value evaluation system, including obstacle density, local complexity, channel features, path curvature, and topological complexity. Using these feature quantification values and combining with a dynamically adjusted regional division method, the algorithm can adaptively adjust the planning strategy according to the characteristics of different regions. This regional division mechanism takes into account the adaptability in complex environments and the global planning efficiency.

[0054] For different regional characteristics, the algorithm flexibly selects a variety of improved path planning methods for collaborative planning: uses an optimized bidirectional A* algorithm for fast search in open areas, introduces an improved rapidly-exploring random tree algorithm in narrow channels and obstacle-dense areas, and combines direction-guided sampling, target-guided sampling, and dynamic step adjustment strategies to improve the path generation quality in complex environments. This multi-algorithm collaborative framework based on regional characteristics significantly enhances the adaptability and path planning efficiency of the robot in complex scenarios. Brief Description of the Drawings

[0055] Figure 1 It is a flowchart of the path planning method of the present invention.

[0056] Figure 2 It is a schematic diagram of the three-layer data structure for preprocessing in step 1; (a) is a quadtree of a complex environment, (b) is a Voronoi diagram of a complex environment, and (c) is a connectivity graph of a complex environment.

[0057] Figure 3 It is a schematic diagram of the adaptive regional division proposed by the present invention.

[0058] Figure 4 It is a schematic diagram of the final path planning result of the present invention.

[0059] Figure 5 It is a comparison chart of the performance of the path planning algorithm. Detailed Embodiment

[0060] The embodiments of the present invention are described in detail below, and the examples of the embodiments are shown in the drawings. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0061] The present invention provides a multi-algorithm collaborative path planning method based on K-value, adopting an adaptive regional division and a multi-algorithm collaborative parallel path planning mechanism to improve the path planning efficiency of the algorithm and reduce the computational complexity. The specific steps are as follows:

[0062] Step 1: Preprocessing of the environmental map. Model the obstacles in the environmental map and convert them into a grid map. Considering the safety of the robot during movement and combining the actual size of the robot, inflate the obstacles. To improve the efficiency of subsequent processing, the system constructs a three-layer data structure, as Figure 4 shown, including a quadtree for fast spatial retrieval, a Voronoi diagram for channel analysis, and a graph structure representing environmental connectivity;

[0063] The quadtree structure is constructed by recursive splitting. First, take the entire map as the root node and judge whether splitting is needed based on the homogeneity index of the node. When the homogeneity is insufficient, divide the node into four child nodes (upper left, upper right, lower left, lower right), and repeat this process for the child nodes until the node meets the homogeneity requirements; The construction of the Voronoi diagram first identifies all obstacle boundary points, and then generates Voronoi edges by calculating the set of points equidistant from the two nearest obstacles. When three or more Voronoi edges intersect, Voronoi vertices are formed; The construction of the connectivity graph uses Voronoi vertices as graph nodes and Voronoi edges as candidate connections. When the channel width is close to the size of the robot, the weight increases significantly to reduce the probability of this path being selected. At the same time, remove the edges with too large weights to ensure path safety.

[0064] Step 2: Generate an initial path based on an improved rapidly-exploring random tree path planning algorithm. The specific algorithm steps are as follows:

[0065] Step 2.1. Initialize two rapidly-exploring random trees at the initial position X start and the target position X final of the robot respectively, denoted as T start and T goal . The node storage structure of each tree includes information such as the current position, the parent node, and the path cost, which is used for subsequent path backtracking and optimization. Set the step size Δd and the maximum number of iterations N max . Define the probability distribution q, p, w in the hybrid sampling strategy so that q + p + w = 1, corresponding to goal-directed sampling, direction-guided sampling, and global uniform sampling respectively.

[0066] Step 2.2. Generate a random sampling point X rand in the configuration map by the hybrid sampling method. Select the target point X final as the random sampling point X rand with probability q, perform random sampling within the ±45° sector range of the current path direction with probability p, and perform uniform random sampling with probability w. Then perform collision detection on the sampling point to ensure that it is in the free space.

[0067] Step 2.3. Find the point in the existing rapidly-exploring random tree that is closest to Xrand The node X closest in distance nearest As a potential parent node, construct a connection to X rand and X nearest of the path segment, and perform collision detection to verify the feasibility of the path segment. The distance calculation usually uses the Euclidean distance:

[0068]

[0069] Step 2.4: If the path satisfies the collision-free constraint, then add X rand as a new node to the rapidly-exploring random tree structure, and establish a parent-child relationship with X nearest ;

[0070] Step 2.5: Generate a new node X nearest in the direction from X rand according to the step size: new :

[0071]

[0072] Step 2.6: Check the collision situation of the path segment from X neares t to X new . If there is no collision, add X new to the tree structure and establish a parent-child relationship.

[0073] Step 2.7: Alternately expand T start and T goal two trees. After each expansion, check the shortest distance d min between the two trees:

[0074]

[0075] When the distance is less than the set threshold, try to connect the two trees. If the connection is successful, a feasible path is found.

[0076] Step 2.8: Trace back from the connection point to the root nodes of the two trees, and merge the two path segments to obtain a complete path.

[0077] Step 3: Extract path key points. Based on the planned initial path points, quadtree, Voronoi, and connectivity graph, extract the channel entrance points, exit points, corner points, and intersection points and merge them as path key points. On this basis, perform importance scoring on all identified key points and filter out redundant points. The specific steps are as follows:

[0078] Step 3.1: Channel entrance point and exit point identification: First, search at the intersection of the Voronoi diagram and the quadtree boundary, and calculate the change in channel width within 5 step sizes before and after each candidate point. When the width change rate exceeds 30%, mark this point as the channel entrance point and exit point.

[0079] The channel width variation can be expressed as:

[0080] Δw(p i ) = |w(p i ) - w(p i+k )|

[0081] Wherein, w(p i ) represents the channel width at point p i , and k represents the step offset.

[0082] The channel width change rate can be expressed as:

[0083]

[0084] Step 3.2, corner point identification: Calculate the included angle θ formed by three adjacent points on the path. When θ is less than the angle threshold θ threshold , the middle point is taken as the key node;

[0085] The formula for calculating the included angle θ is:

[0086]

[0087] Wherein, and respectively represent the vectors of path segments P i-1 P i and P i P i+1 .

[0088] Step 3.3, intersection point identification: Based on the connected graph, identify the node degree φ. When φ is greater than the degree threshold φ threshold , it is marked as an intersection point and regarded as a key node;

[0089] Step 3.4, calculate the importance scores of intersection points, channel entrance points, and corner points in order of weight. Among them, intersection points are given the highest weight of 3.0 points because they connect multiple channels and affect path connectivity; channel entrance points and exit points represent important transition positions from a spacious area to a narrow channel, with a weight of 2.0 points; corner points are used to mark positions where the path needs to turn significantly, with a weight of 1.0 point.

[0090] Step 3.5, based on the calculated importance scores, sort the key points, and give priority to retaining the points with higher importance. On this basis, when the distance between two key points is less than the step length Δd, they are considered redundant, and only the one with higher importance is retained.

[0091] Step 4: Establish an enhanced K - value evaluation system for multi - dimensional features. The enhanced K - value evaluation system is constructed through five core dimensions: Obstacle density (W) reflects the space occupancy, local complexity (γ) characterizes the obstacle distribution characteristics, channel feature (η) quantifies the passage difficulty, path curvature (σ) evaluates the turning complexity, and topological complexity describes the environmental structure characteristics. Then, by optimizing the weights (α, β, δ, λ, ε), an evaluation formula is combined: It realizes the precise quantification of the environment and provides a decision - making basis for subsequent regional path planning.

[0092] Specifically, the obstacle density W, local complexity γ, channel feature η, path curvature σ, and topological complexity in the enhanced K - value evaluation system described in Step 4 are calculated as follows:

[0093]

[0094] In the formula, S obs is the occupied area of obstacles in the region, S total is the total area of the region, d i is the distance from the obstacle to the center of the region, d avg is the average distance from all obstacles to the center of the region, n is the number of obstacles, Δθ is the channel turning change, w ref is twice the width of the robot, w min is the minimum width of the channel, k i is the curvature of path segment i, l i is the length of path segment i, L is the total length of the regional path, N c is the number of intersection points, N max is the maximum number of intersection points, D b is the degree of branching, D t is the total connectivity.

[0095] Step 5: Adaptive regional division. Adaptive regional division adapts to different environmental characteristics by dynamically adjusting the size of the circular region. Based on the path key points extracted in Step 3, the basic radius of the circular region is calculated. Then, according to the K - value of each region, the region is divided into open regions, narrow - channel regions, and obstacle - dense regions. In narrow - channel regions and obstacle - dense regions, the system will automatically reduce the region radius to improve the planning accuracy; in open regions, the radius will be increased accordingly to improve the planning efficiency. At the same time, by analyzing the K - value difference and overlap degree of adjacent regions, it is judged whether regional merging is required. The specific steps are as follows:

[0096] Step 5.1: Based on the position of each key point extracted in Step 3, set and calculate its corresponding basic radius R base of the circular region, and its calculation formula is:

[0097]

[0098] Wherein, L tol is the total path length planned by the improved rapidly-exploring random tree algorithm, and n key is the number of key points extracted.

[0099] Step 5.2: Calculate the local obstacle density W and the local space complexity γ of the area, and correct the basic radius R of the circular area base is R, and its calculation method is:

[0100] R = R base (1 + k w W + k y γ)

[0101] Wherein, k w is the density adjustment coefficient, and k y is the complexity adjustment coefficient.

[0102] Step 5.3: Calculate the K value of each area. According to the magnitude of the K value of each area, divide the area into an open area, a narrow channel area, and an obstacle-dense area. The corresponding radii of the circular areas are denoted as R open 、R nar 、R den .

[0103] Perform adaptive adjustment on the radius of each area, and the corresponding calculation method is as follows:

[0104]

[0105] Wherein, μ is the area expansion coefficient, and v1, v2 are the area contraction coefficients.

[0106] Step 5.4: Find out the path segment not covered by the circular area, find the earliest generated parent node and the last generated child node in this segment, and use the Euclidean distance between these two characteristic nodes as the radius of the supplementary circular area.

[0107] Step 5.5: Calculate the coincidence degree O and the K value difference ΔK of adjacent areas, and merge the areas with large adjacent coincidence degrees or small K value differences. Use the Euclidean distance between the earliest generated parent node and the last generated child node of the area as the radius of the merged circular area.

[0108] Step 6: Multi-algorithm collaborative planning. Calculate the K value of each sub-region after regional division. For the open area, select the improved bidirectional A* algorithm to quickly find the path through the optimized heuristic function and bidirectional search strategy; for the narrow channel area, on the basis of the original improved bidirectional rapidly-exploring random tree algorithm, increase the probability of direction-guided sampling. For the area with dense obstacles, on the basis of the original improved bidirectional rapidly-exploring random tree algorithm, add goal-directed sampling and dynamically adjust the step size according to the surrounding obstacle environment information. The specific steps are as follows:

[0109] Step 6.1. Calculate the K value of each sub-region after regional division. Regard the earliest generated parent node of each region as the starting point of path planning and the last generated child node as the end point of path planning. If there is an overlap between two regions, set the starting point of the region with a larger K value as the end point of the region with a smaller K value.

[0110] Step 6.2. Use a parallel method to perform path planning for each region simultaneously. Among them, for the open area, select the improved bidirectional A* algorithm; for the narrow channel area, on the basis of the original improved bidirectional rapidly-exploring random tree algorithm, increase the probability of direction-guided sampling; for the area with dense obstacles, on the basis of the original improved bidirectional rapidly-exploring random tree algorithm, add goal-directed sampling and dynamically adjust the step size according to the surrounding obstacle environment information.

[0111] Specifically, the improved bidirectional A algorithm in Step 6 mainly realizes efficient path planning in the open area by optimizing the heuristic function and adopting a dynamic priority bidirectional search strategy.

[0112] The optimized heuristic function is:

[0113] In the formula, g(n) is the current path cost, D(n) is the Euclidean distance from the node to the target, where (x goal , y goal ) are the coordinates of the target node, and (x n , y n ) are the coordinates of the current node. C(n) is the direction cost, θ is the angle between the current path direction and the target direction, and w1, w2 are dynamic weight coefficients.

[0114] The specific implementation process of the dynamic priority bidirectional search strategy is as follows: Dynamically determine which search tree to expand preferentially by calculating the expansion index , where d(n) is the shortest Euclidean distance from the current node to the opposite search tree. Specifically, if E start < E goal , preferentially expand the starting point tree; if E start > E goal, preferentially expand the end tree; if E start = E goal , then alternately expand the two trees.

[0115] Specifically, the method for dynamically adjusting the step size according to the surrounding obstacle environment information in step 6 is as follows:

[0116]

[0117] In the formula, base step is the base step size, d obs is the distance from the sampling point to the obstacle, d safe is the set safety distance, k is the direction influence factor, θ goal is the angle between the current expansion direction and the target direction.

[0118] Step 7: Set the key points extracted in step 3 and the starting and ending points of each area as control points. Use the Bezier curve equation to smooth the path.

[0119] The Bezier curve equation is as follows:

[0120]

[0121] In the formula, P i is the i-th control point, is the Bernstein basis function.

[0122] After the above steps, a robot travel path with low path cost and smoothness is finally generated.

[0123] Table 1 Performance comparison of path planning algorithms

[0124] Algorithm type Planning time / s Total path length A* 5.62 845.6 Rapidly-exploring Random Tree 0.7 885.5 Rapidly-exploring Random Tree * 6.865 870.9 Multi-algorithm collaboration 0.5 841.2

[0125] Such as Figure 5As shown in Table 1, although the standard rapidly-exploring random tree algorithm (iterated 100 times) can quickly find a feasible solution, its path shows obvious zigzag characteristics and there are many redundant turns; the rapidly-exploring random tree* algorithm (iterated 20 times) optimizes the path through a rewiring mechanism, and the path is smoother than that of the standard rapidly-exploring random tree, but the optimization process increases the computational overhead; the A* algorithm is limited by the rasterization characteristics. Although it can find the shortest path, it shows obvious jagged characteristics and the path planning efficiency is not high. The multi-algorithm cooperation method combines the heuristic search of the A* algorithm and the fast random sampling characteristics of the rapidly-exploring random tree algorithm, generating a smooth and high-quality path, which well balances the path quality and computational efficiency. From the perspective of practical applications, the multi-algorithm cooperation method shows the best comprehensive performance: its path has high smoothness, good obstacle avoidance effect, and maintains good path continuity, which is more beneficial to the actual motion control of the robot.

Claims

1. A robot intelligent path planning method based on K - value adaptive multi - algorithm collaboration, specifically including the following steps: Step 1: Pre - process the environmental map; Model the obstacles in the environmental map and convert it into a grid map; combine the actual size of the robot and perform dilation processing on the obstacles; construct a three - layer data structure, including a quadtree for fast spatial retrieval, a Voronoi diagram for channel analysis, and a graph structure representing environmental connectivity; Step 2: Generate an initial path based on the rapidly - exploring random tree path planning algorithm; Step 3: Extract path key points; Based on the planned initial path points, quadtree, Voronoi diagram, and connectivity graph, extract channel entry points, exit points, corner points, and intersection points and merge them as path key points; on this basis, perform importance scoring on all identified key points and filter out redundant points; Step 4: Establish an enhanced K - value evaluation system with multi - dimensional features; Step 5: Adaptive region division; Based on the path key points extracted in Step 3 and the K - value evaluation system in Step 4, divide the environment into open areas, narrow - channel areas, and obstacle - dense areas, and adapt to different environmental characteristics by adjusting the size of the region radius; Step 6: Multi - algorithm collaborative planning; Calculate the K - value of each sub - region after region division. For the open area, select the bidirectional A* algorithm and quickly find the path through an optimized heuristic function and bidirectional search strategy; for the narrow - channel area, on the basis of the originally improved rapidly - exploring random tree algorithm, increase the direction - guiding sampling probability; for the obstacle - dense area, on the basis of the originally improved rapidly - exploring random tree algorithm, increase target - oriented sampling and dynamically adjust the step size according to the surrounding obstacle environment information; Step 7: Smooth the path to obtain the final path.

2. The robot intelligent path planning method based on K - value adaptive multi - algorithm cooperation as described in claim 1, wherein, The establishment of the three - layer data structure in Step 1 is as follows: The quadtree is constructed by recursive splitting, dividing the non - homogeneous region into four sub - nodes; the Voronoi diagram is a network of boundary lines formed by connecting all points in the plane that are equidistant from the two nearest obstacles; the connectivity graph is established by connecting Voronoi vertices as nodes, and the edge weights consider the path length and channel width.

3. The robot intelligent path planning method based on K - value adaptive multi - algorithm cooperation according to claim 1, characterized in that, The specific method for generating the initial path based on the rapidly - exploring random tree path planning algorithm in Step 2 is as follows: Step 2.

1. Initialize two rapidly-exploring random trees at the initial position X of the robot start and the target position X final respectively, denoted as T start and T goal ; set the step size Δd and the maximum number of iterations N max ; define the probability distribution q, p, w in the hybrid sampling strategy, such that q + p + w = 1, corresponding to goal-directed sampling, direction-guided sampling, and global uniform sampling respectively; Step 2.2: Generate a random sampling point X in the configuration map through a hybrid sampling strategy rand Select the target point X with probability q final as the sampling point, sample within the ±45° sector range in the current path direction with probability p, and perform uniform random sampling with probability w; then perform collision detection on the sampling point to ensure that it is in the free space Step 2.

3. Find the node X rand in the existing rapidly-exploring random tree that has the closest Euclidean distance to X rand as the potential parent node, and construct a path segment connecting X nearest nearest and X rand rand to perform collision detection to verify the feasibility of the path segment; nearest ​ Step 2.4: If the path satisfies the collision-free constraint, then add X rand as a new node to the rapidly-exploring random tree structure, establish the parent-child relationship with X nearest , and calculate the growth direction and step size from X nearest to X rand . Step 2.5: Generate a new node X in the X direction according to the step size nearest from X rand in the X new direction, and check the collision situation of the path segment from X nearest to Xnew . If there is no collision, add X new to the tree structure and establish a parent-child relationship; Step 2.

6. Alternately expand T start and T goal two trees. After each expansion, check the shortest distance between the two trees. When the distance is less than the set threshold, try to connect the two trees. If the connection is successful, a feasible path is found; Step 2.7: Trace back from the connection point to the root nodes of the two trees and merge the two path segments to obtain a complete path.

4. A robot intelligent path planning method based on K - value adaptive multi - algorithm collaboration as claimed in claim 1, wherein, Specifically, the process of extracting path key points described in step 3 is as follows: identify the channel area based on the Voronoi diagram, and extract the entrance point and exit point of the channel as key nodes; calculate the included angle θ formed by three adjacent points on the path, and when θ is less than the angle threshold θ threshold , take the middle point as a key node; Based on the connected graph, identify the node degree φ. When φ is greater than the degree threshold, φ threshold is marked as an intersection point and regarded as a key node.

5. A robot intelligent path planning method based on K - value adaptive multi - algorithm collaboration as claimed in claim 1, wherein The evaluation parameters in the enhanced K - value evaluation system described in step 4 include: obstacle density W, local complexity γ, channel feature η, path curvature σ, and topological complexity The calculation methods thereof are respectively: Where S obs is the occupied area of obstacles in the region, S total is the total area of the region, d i is the distance from the obstacle to the center of the region, d avg is the average distance from all obstacles to the center of the region, n is the number of obstacles, Δθ is the change in channel turning, w ref is twice the width of the robot, w min is the minimum width of the channel, k i is the curvature of path segment i, l i is the length of path segment i, L is the total length of the region path, N c is the number of intersections, N max is the maximum number of intersections, D b is the degree of branching, D t is the total connectivity.

6. The robot intelligent path planning method based on K - value adaptive multi - algorithm collaboration according to claim 1, wherein, The process of adaptive region division in Step 5 is as follows: Step 5.1: Based on each key point position extracted in Step 3, set the basic radius R of its corresponding circular region base , and its calculation formula is: Where L tol is the total path length planned by the improved rapidly-exploring random tree algorithm, and n key is the number of key points extracted; Step 5.2: Calculate the local obstacle density W and the local spatial complexity γ of the area, and correct the base radius R of the circular area base Take R, and its calculation method is as follows: R = R base (1 + k w W + k y γ) where k w is the density adjustment coefficient, and k y is the complexity adjustment coefficient; Step 5.3: Calculate the K value of each area. According to the magnitude of the K value of each area, divide the areas into open areas, narrow passage areas, and obstacle-dense areas, and denote the corresponding circular area radii as R open 、R nar 、R den ; Perform adaptive adjustment on each region radius, and the corresponding calculation method is as follows: In the formula, μ is the region expansion coefficient, and v1, v2 are the region contraction coefficients; Step 5.4: Find the path segment not covered by the circular region, find the earliest - generated parent node and the last - generated child node in this segment, and use the Euclidean distance between these two characteristic nodes as the radius of the supplementary circular region; Step 5.5: Calculate the coincidence degree O and K - value difference ΔK of adjacent regions, merge adjacent regions with a large coincidence degree or small K - value difference, and use the Euclidean distance between the earliest - generated parent node and the last - generated child node of the region as the radius of the merged circular region.

7. The robot intelligent path planning method based on K - value adaptive multi - algorithm collaboration according to claim 1, characterized in that, In the bidirectional A* algorithm in step 6, the optimized heuristic function is: f(n) = g(n) + w1D(n) + w2C(n) In the formula, g(n) is the current path cost, D(n) is the Euclidean distance from the node to the target, C(n) is the direction cost, and w1, w2 are dynamic weight coefficients; Adopt a two-way search strategy with dynamic priorities, and its specific implementation process is as follows: Calculate the expansion index to dynamically determine which search tree to expand preferentially, where d(n) is the shortest Euclidean distance from the current node to the opposite search tree; if E start <E goal , preferentially expand the starting point tree; if E start >E goal , preferentially expand the ending point tree; If E start = E goal , then alternately expand the two trees.

8. A robot intelligent path planning method based on K - value adaptive multi - algorithm collaboration as described in claim 1, characterized in that, Specifically, the method for dynamically adjusting the step size according to the surrounding obstacle environment information in step 6 is as follows: where base step is the basic step size, d obs is the distance from the sampling point to the obstacle, d safe is the set safety distance, k is the direction influence factor, θ goal is the angle between the current expansion direction and the target direction.

Images