RRT path planning method and system based on limited random point generation

Abstract

The invention discloses an RRT path planning method and system based on limited random point generation, and the method comprises the steps: setting two random trees, and enabling the two random trees to alternately carry out sampling extension and attempt to be connected in a path searching process; in the sampling expansion process, a dynamic multilayer circular ring sampling area is introduced to select sampling points, a guide target point is set based on a starting point and an ending point, and the dynamic multilayer circular ring sampling area is constructed according to the guide target point; when a preset condition is met, one sampling area is selected from the multiple layers of circular ring sampling areas through the sampling probability to serve as a key sampling area, and sampling points are obtained in the key sampling area. Meanwhile, the problem that a connecting line of a sampling point and an adjacent point generated in path planning collides with an obstacle is solved through a fan-shaped swing expansion strategy, and the efficiency, the robustness and the path quality of path planning of the intelligent agent are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of path planning, and in particular relates to an RRT path planning method and system based on restricted random point generation. Background Art

[0002] With the rapid development of artificial intelligence (AI), autonomous navigation has been widely applied in various fields. Path planning is a core technology in this field. Existing path planning methods include the RRT (Rapid Random Tree) algorithm, the A* algorithm, the artificial potential field method, and the ant colony algorithm. The RRT algorithm is a type of path planning method based on random sampling and is particularly well-suited for handling high-dimensional state space problems. Its basic idea is to construct a search tree by continuously sampling in the feasible space and expanding it to new points, thereby achieving a collision-free path from the starting point to the target point.

[0003] However, the traditional RRT method still has the following defects: 1. Sampling is highly blind: The sampling direction is completely random and lacks guidance, resulting in low search efficiency and a large number of iterations; 2. Easy to get stuck in dense obstacle areas: In scenarios with complex obstacle distribution, node expansion frequently fails, and sampling efficiency drops sharply; 3. The path is long and tortuous: the generated path is not smooth enough, the nodes are discrete, and it is difficult to use it directly for trajectory tracking; 4. Poor adaptability to environmental changes: The algorithm lacks a dynamic feedback mechanism and cannot flexibly adjust its strategy based on local spatial complexity; 5. Dual-tree connection difficulty: In a bidirectional RRT, if the two trees cannot be efficiently merged, the path search will be stuck in a long wait.

[0004] Although some improvement methods such as target bias sampling, dynamic step size adjustment, and regional heuristic search have been proposed, most methods still rely on statically set parameters, lack dynamic adjustment capabilities, and are not adaptable enough to complex obstacle scenarios. Summary of the Invention

[0005] The object of the present invention is to provide a RRT path planning method and system based on restricted random point generation.

[0006] In a first aspect, the present invention provides an RRT path planning method based on restricted random point generation, which comprises the following steps: Obtain the environment information around the agent and build a coordinate system; set up two random trees, each with the starting point and end point of the path planning as the root node; sample and expand the two random trees respectively; The guide target point is set based on the starting point and the end point, and a dynamic multi-layer circular sampling area is constructed according to the guide target point; the method for constructing the dynamic multi-layer circular sampling area is as follows: with the guide target point as the center, multiple concentric circles with different radii are constructed; the circular ring formed by adjacent concentric circles is used as the sampling area; the area within the smallest concentric circle is used as the sampling area , the area outside the largest concentric circle is used as the sampling area , the rest of the sampling areas are sorted in order; among them, is the number of sampling areas; set the initial sampling probability of each sampling area respectively; A probability threshold is set, and different sampling methods are selected based on the probability threshold to obtain sampling points; the sampling methods include global random sampling and sampling based on dynamic multi-layer circular sampling areas; the dynamic multi-layer circular sampling area sampling selects a sampling area as a key sampling area in the multi-layer circular sampling area through the sampling probability, and obtains sampling points in the key sampling area; based on the sampling points, a new node is obtained to be added to the random tree, and the sampling probability of each sampling area is updated according to the sampling area where the new node is located, to obtain the updated sampling probability of each sampling area; all nodes in the two random trees are detected respectively, and if there is a node in the random tree in the sampling area , then the random tree obtains sampling points through combined sampling in the next iteration; Repeat the process of acquiring new nodes in the two random trees until the two random trees are connected and the optimal path of the agent from the starting point to the end point is obtained.

[0007] As a preferred method, the method of setting the guidance target point is as follows: Connect the starting point and the end point. If the midpoint of the line connecting the starting point and the end point is outside the obstacle range, then take the midpoint as the guidance target point; otherwise, draw a perpendicular line along the midpoint and move the vertical line at the midpoint in the direction of the perpendicular line with a step length of Δ. x Deviation is made to both sides until an offset point is found outside the obstacle range, and the offset point is used as the guiding target point.

[0008] As a preferred method, the method of selecting key sampling areas is: according to the sampling probability of each sampling area Get the corresponding cumulative probability vector ; Randomly generate index probability ; Select the cumulative probability vector Greater than index probability The sampling area corresponding to the minimum cumulative probability vector is taken as the key sampling area.

[0009] Preferably, the cumulative probability vector is the sum of the sampling probabilities of the sampling area and its inner sampling areas.

[0010] Preferably, the method for obtaining the updated sampling probability is as follows: Get the number corresponding to the sampling area where the new node is located in the current iteration process and the previous iteration process and ;like , the original sampling probability is used as the updated sampling probability; otherwise, the attenuated sampling area The sampling probability of the outer sampling area and the attenuated sampling probability are added to the sampling area and sampling area The updated sampling probability of each sampling area is obtained from the sampling probability of the inner sampling area.

[0011] Preferably, the method of adding the attenuated sampling probability in the sampling area is: setting the weight of each sampling area based on the number of the sampling area, the larger the number, the smaller the weight; splitting the attenuated sampling probability according to the weight of each sampling area, and adding it to the corresponding sampling probability to obtain the updated sampling probability.

[0012] As a preference, obtain sampling points in key sampling areas The method is as follows: in, and Respectively represent the horizontal and vertical coordinates of the sampling points; and are the horizontal and vertical coordinates of the guidance target point respectively; The line connecting the sampling point and the center of the circle is the coordinate system X The angle formed by the axes, ; is a uniformly distributed random number; is the distance between the sampling point and the center of the circle, and its expression is: in, is the inner ring radius of the key sampling area; is the outer ring radius of the key sampling area; is a uniformly distributed random number; If the key sampling area is the sampling area , then the inner ring radius Zero, outer ring radius For sampling area The radius of the sampling area; if the key sampling area is the sampling area , then the inner ring radius is the radius of the largest concentric circle, the outer ring radius Angle The distance from the corresponding boundary to the center of the circle.

[0013] As a preferred method, the method of obtaining new nodes based on sampling points is as follows: The node with the smallest distance from the sampling point in the random tree is selected as the neighboring point. If the line connecting the sampling point and the neighboring point does not touch the obstacle, the sampling point is selected as the new node. Otherwise, the line connecting the neighboring point and the sampling point is used as the sampling line segment. The sampling line segment is rotated around the neighboring point, alternating between the two sides, with the swing angle gradually increasing relative to the initial state until the sampling line segment does not intersect the obstacle or the maximum number of swings is reached. If the sampling line segment does not intersect the obstacle, the end of the sampling line segment relative to the neighboring point is selected as the new node. If the maximum number of swings is reached, the sampling point is resampled until a new node is obtained.

[0014] As a preferred method of combining sampling to obtain sampling points: when there are one or more nodes in a random tree in the sampling area When , the sampling points of the random tree are obtained based on the new nodes of another random tree and the global random points.

[0015] In a second aspect, the present invention provides an RRT path planning system based on restricted random point generation, which is used to execute the above-mentioned RRT path planning method; the RRT path planning system includes a signal acquisition module and a signal processing module; the signal acquisition module is used to collect environmental information of the intelligent agent and input it into the signal processing module to obtain the optimal path for the intelligent agent to move from the starting point to the end point.

[0016] The present invention has the following beneficial effects: 1. The present invention sets up dynamic multi-layer circular sampling areas to select sampling points. By adding new sampling positions in the random tree, the sampling probabilities of different sampling areas are dynamically adjusted, so that the acquired sampling points can quickly approach the set guidance target point according to the environment in which the intelligent agent is located, thereby improving the speed of connecting two random trees.

[0017] 2. The present invention uses a fan-shaped swing expansion strategy to solve the problem of collision between the lines connecting the sampling points and adjacent points generated in path planning and obstacles, which greatly avoids the invalid global random sampling in traditional path planning, thereby reducing the generation of redundant nodes, significantly shortening the algorithm convergence time, and improving the efficiency, robustness and path quality of path planning for intelligent agents. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the overall flow chart of the present invention.

[0019] Figure 2 Schematic diagram of the dynamic sampling area constructed in the present invention.

[0020] Figure 3 Generate an extension point diagram in the present invention.

[0021] Figure 4Schematic diagram of the optimal paths generated by the present invention and the traditional bidirectional fast random tree algorithm in a simple environment; among them, (a) is the optimal path generated by the present invention; (b) is the optimal path generated by the traditional bidirectional fast random tree algorithm.

[0022] Figure 5 Schematic diagram of the optimal paths generated by the present invention and the traditional bidirectional fast random tree algorithm in a complex environment; among them, (a) is the optimal path generated by the present invention; (b) is the optimal path generated by the traditional bidirectional fast random tree algorithm. DETAILED DESCRIPTION

[0023] The present invention will be further described below with reference to the accompanying drawings.

[0024] like Figure 1 As shown, an RRT path planning method based on restricted random point generation is used. The RRT path planning system adopted by the RRT path planning method includes a signal acquisition module and a signal processing module; the signal acquisition module is used to collect environmental information of the intelligent agent and input it into the signal processing module to obtain the optimal path for the intelligent agent to move from the starting point to the end point.

[0025] The RRT path planning method includes the following steps: Step 1: Initialize the environment and parameter settings Obtain information about the agent's surroundings and construct a two-dimensional static map of 800×800 pixels. Obstacles in the map are represented as grayscale images. Initialize the random trees T_start and T_goal, which form a bidirectional RRT structure. The random tree T_start has the starting point Xstart as its root node, while the random tree T_goal has the ending point Xgoal as its root node. During the path search, the two random trees alternately sample and expand, attempting to connect them.

[0026] Step 2: Set the guidance target point Connect the starting point and the end point. If the midpoint of the line connecting the starting point and the end point is outside the obstacle range, then take the midpoint as the guidance target point Vt; otherwise, draw a perpendicular line along the midpoint of the line connecting the starting point and the end point, and at the midpoint, follow the perpendicular line in the direction of the step length Δ x Deviation is made to both sides until an offset point is found outside the obstacle range, and the offset point is used as the guiding target point.

[0027] Step 3: Construct a dynamic multi-layer ring sampling area like Figure 2 As shown in the figure, with the guiding target point Vt as the center, five concentric circles with different radii are constructed; the area within the smallest concentric circle is used as the sampling area. The rings formed by the concentric circles from the inside to the outside are used as sampling areas in turn. , sampling area , sampling area , sampling area ; The area outside the largest concentric circle is used as the sampling area . Set the initial sampling probability of each area separately .

[0028] In this embodiment, the radii of the five concentric circles are 100, 200, 300, 400, and 500 respectively; the initial sampling probability of each area They are 0.30, 0.22, 0.18, 0.15, 0.10 and 0.05 respectively.

[0029] Step 4: Set the probability threshold , and randomly generate probability distribution If the probability distribution Greater than or equal to the probability threshold , then global random sampling is used to obtain the sampling points Otherwise, the sampling points are obtained based on the dynamic multi-layer circular sampling area. The specific process is as follows: a. According to the sampling probability of each area Get the corresponding cumulative probability vector , whose expression is: in, For sampling area The sampling probability of .

[0030] b. Randomly generate index probability ; Select the cumulative probability vector Greater than index probability The sampling area corresponding to the minimum cumulative probability vector is sampled as the key sampling area to obtain the sampling point , sampling point The horizontal and vertical coordinates and Respectively expressed as: in, and are the horizontal and vertical coordinates of the center of the circle respectively; The line connecting the sampling point and the center of the circle is X The angle formed by the axes, ; is a uniformly distributed random number; is the distance between the sampling point and the center of the circle, and its expression is: in, is the inner ring radius of the key sampling area; is the outer ring radius of the key sampling area; is a uniformly distributed random number.

[0031] If the key sampling area is the sampling area , then the inner ring radius Zero, outer ring radius For sampling area The radius of the sampling area; if the key sampling area is the sampling area , then the inner ring radius is the radius of the largest concentric circle, the outer ring radius Angle The distance from the corresponding boundary to the center of the circle.

[0032] In this embodiment, the probability threshold is 0.8.

[0033] Step 5: Figure 3 As shown, for the sampling points Perform collision detection and obtain new nodes based on the collision detection results , the specific process is as follows: Get the random tree and sampling points The node with the smallest distance is considered as the neighboring point If the sampling point With neighboring points If the line connecting the two points does not touch the obstacle, the sampling point As a new node Join the random tree; otherwise, use the fan swing expansion strategy to obtain new nodes, the process is as follows: At nearby points With sampling point The sampling line segment is intercepted on the line connecting the two points. The length of the sampling line segment is the preset expansion step. One endpoint of the sampling line segment is the adjacent point. , and the other end point is the extension point. As the rotation center, it swings alternately to both sides, and the swing angle relative to the initial state gradually increases. The sampling points are obtained during each swing. Corresponding extension points , which is expressed as: in, are the coordinates of the neighboring points; To expand the step length; For extension points and adjacent points The angle formed with the X-axis of the coordinate system is expressed as: in, For sampling points and neighboring points The angle formed with the X-axis of the coordinate system; is the swing angle interval; ; is the maximum number of swings.

[0034] If the maximum number of swings is reached Before, there was an extension point With neighboring points If the line connecting the two points does not touch the obstacle, the closest sampling point is used. Extension points As a new node Add the random tree; otherwise, abandon the expansion and re-acquire the sampling points until the new node is added to the random tree .

[0035] In this embodiment, the swing angle interval The value is 5° or 10°.

[0036] Step 6: Update dynamic sampling probability Get the new nodes in the current iteration and the previous iteration The number corresponding to the sampling area and ; If the new node obtained in the current iteration process , compared to the new nodes obtained in the previous iteration Perform a more outer sampling area or two new nodes In the same sampling area, , then the original sampling probability As the updated sampling probability ; Otherwise, based on the sampling area The sampling probability of each sampling area is updated. The specific process is as follows: Sampling area Outer sampling area ( ) is updated, and its updated sampling probability The expression is: in, For sampling area The original sampling probability of is the attenuation factor.

[0037] Based on sampling area Update sampling probability Constructing transition probabilities , whose expression is: in, is the number of sampling areas.

[0038] Set up sampling areas separately and sampling area The weight of each sampling area in the inner sampling area ( ), whose expression is: According to weight The transfer probability Assign to sampling area and sampling area Inner sampling area, improve the inner sampling probability, so as to obtain the sampling area Update sampling probability , whose expression is: in, is the total weight, .

[0039] Step 7: Detect all nodes in the two random trees respectively. If there is a node in the random tree that is in the sampling area In (i.e. ), in order to improve the efficiency of bidirectional RRT in the terminal connection stage, the random tree that meets this condition obtains sampling points through combined sampling in the next iteration process, guiding the sampling points of the random tree to move closer to the other random tree, so as to achieve rapid convergence of the two trees; the random tree that does not meet this condition obtains sampling points through the original sampling method; among them, is the node coordinate in the random tree; is the coordinate of the guiding target point; For sampling area radius.

[0040] The specific process of combined sampling is as follows: The sampling points of the random tree Based on another random tree, new nodes and global random points are obtained; in order to enhance the guidance of the sampling points while retaining randomness, the sampling points The expression is: in, is the new node coordinate of another random tree; Points are uniformly randomly generated across the entire image; It is the combined sampling weight coefficient to control the bias degree of the sampling points.

[0041] This formula effectively guides one random tree to converge toward another random tree while maintaining global randomness.

[0042] Step 8. Repeat steps 4 to 7 until a new node of the random tree is reached. The node closest to the new node in another random tree The distance between them is less than the preset connection threshold, and the new node With node If the line connecting the two random trees does not touch any obstacles, the two random trees are considered to be connected successfully. The paths of the two trees are traced back and spliced ​​together to obtain the optimal path from the starting point to the end point.

[0043] Step 9: Use the traditional bidirectional RRT algorithm and the present invention to perform path planning in simple and complex environments respectively. The optimal path after planning is compared with Figure 4 and 5 shown; from Figure 4 and 5 It can be seen that the random tree generated by the present invention has fewer sampling points and higher path quality. It can be seen that the present invention can not only avoid obstacles and reach the target point, but also effectively reduce the number of random tree expansions.

Claims

1. A RRT path planning method based on restricted random point generation, characterized in that: The method includes: Obtain the environment information around the agent and construct a coordinate system; set up two random trees, with the starting point and the end point as the root nodes respectively; alternately sample and expand the two random trees; The guide target point is set based on the starting point and the end point, and a dynamic multi-layer circular sampling area is constructed according to the guide target point; the method for constructing the dynamic multi-layer circular sampling area is as follows: with the guide target point as the center, multiple concentric circles with different radii are constructed; the circular ring formed by adjacent concentric circles is used as the sampling area; the area within the smallest concentric circle is used as the sampling area , the area outside the largest concentric circle is used as the sampling area , the rest of the sampling areas are sorted in order; among them, is the number of sampling areas; set the initial sampling probability of each sampling area respectively; A probability threshold is set, and different sampling methods are selected based on the probability threshold to obtain sampling points; the sampling methods include global random sampling and sampling based on dynamic multi-layer circular sampling areas; the dynamic multi-layer circular sampling area sampling selects a sampling area as a key sampling area in the multi-layer circular sampling area through the sampling probability, and obtains sampling points in the key sampling area; based on the sampling points, a new node is obtained to be added to the random tree, and the sampling probability of each sampling area is updated according to the sampling area where the new node is located, to obtain the updated sampling probability of each sampling area; all nodes in the two random trees are detected respectively, and if there is a node in the random tree in the sampling area , then the random tree obtains sampling points through combined sampling in the next iteration; Repeat the process of acquiring new nodes in the two random trees until the two random trees are connected, and the optimal path of the agent from the starting point to the end point is obtained.

2. The RRT path planning method based on restricted random point generation according to claim 1, characterized in that: The method for setting the guidance target point is as follows: Connect the starting point and the end point. If the midpoint of the line connecting the starting point and the end point is outside the obstacle range, then take the midpoint as the guidance target point; otherwise, draw a perpendicular line along the midpoint and move the vertical line at the midpoint in the direction of the perpendicular line with a step length of Δ. x Deviation is made to both sides until an offset point is found outside the obstacle range, and the offset point is used as the guiding target point.

3. The RRT path planning method based on restricted random point generation according to claim 1, characterized in that: The method of selecting key sampling areas is as follows: according to the sampling probability of each sampling area Get the corresponding cumulative probability vector ; Randomly generate index probability ; Select the cumulative probability vector Greater than index probability The sampling area corresponding to the minimum cumulative probability vector is taken as the key sampling area.

4. The RRT path planning method based on restricted random point generation according to claim 3, characterized in that: The cumulative probability vector is the sum of the sampling probabilities of the sampling area and its inner sampling areas.

5. The RRT path planning method based on restricted random point generation according to claim 1, characterized in that: The method for obtaining the updated sampling probability is as follows: Get the number corresponding to the sampling area where the new node is located in the current iteration process and the previous iteration process and ;like , the original sampling probability is used as the updated sampling probability; otherwise, the attenuated sampling area The sampling probability of the outer sampling area and the attenuated sampling probability are added to the sampling area and sampling area The updated sampling probability of each sampling area is obtained from the sampling probability of the inner sampling area.

6. The RRT path planning method based on restricted random point generation according to claim 5, characterized in that: The method of adding the attenuated sampling probability in the sampling area is as follows: the weight of each sampling area is set based on the number of the sampling area, and the larger the number, the smaller the weight; the attenuated sampling probability is split according to the weight of each sampling area, and added to the corresponding sampling probability to obtain the updated sampling probability.

7. The RRT path planning method based on restricted random point generation according to claim 1, characterized in that: Acquire sampling points in key sampling areas The method is as follows: in, and Respectively represent the horizontal and vertical coordinates of the sampling points; and are the horizontal and vertical coordinates of the guidance target point respectively; The line connecting the sampling point and the center of the circle is the coordinate system X The angle formed by the axes, ; is a uniformly distributed random number; is the distance between the sampling point and the center of the circle, and its expression is: in, is the inner ring radius of the key sampling area; is the outer ring radius of the key sampling area; is a uniformly distributed random number; If the key sampling area is the sampling area , then the inner ring radius Zero, outer ring radius For sampling area The radius of the sampling area; if the key sampling area is the sampling area , then the inner ring radius is the radius of the largest concentric circle, the outer ring radius Angle The distance from the corresponding boundary to the center of the circle.

8. The RRT path planning method based on restricted random point generation according to claim 1, characterized in that: The method for obtaining new nodes based on sampling points is as follows: The node with the smallest distance from the sampling point in the random tree is obtained as the neighboring point; if the line connecting the sampling point and the neighboring point does not touch the obstacle, the sampling point is used as the new node; otherwise, a sampling line segment is intercepted on the line connecting the neighboring point and the sampling point; the length of the sampling line segment is the expansion step, one endpoint of the sampling line segment is the neighboring point, and the other endpoint is used as the extension point; the sampling line segment is swung alternately to both sides with the neighboring point as the rotation center, and the swing angle relative to the initial state is gradually increased until the sampling line segment does not intersect with the obstacle or the maximum number of swings is reached; if the sampling line segment does not intersect with the obstacle, the extension point is used as the new node; if the maximum number of swings is reached, the sampling point is resampled until a new node is obtained.

9. The RRT path planning method based on restricted random point generation according to claim 1, characterized in that: The method of combined sampling to obtain sampling points is: when there are one or more nodes in a random tree in the sampling area When , the sampling points of the random tree are obtained based on the new nodes of another random tree and the global random points.

10. A RRT path planning system based on restricted random point generation, characterized by: Used to execute the RRT path planning method based on restricted random point generation as described in claim 1; the RRT path planning system includes a signal acquisition module and a signal processing module; the signal acquisition module is used to collect environmental information of the intelligent agent and input it into the signal processing module to obtain the optimal path for the intelligent agent to move from the starting point to the end point.

Images