Improved RRT path planning method based on collision feedback-threshold information
An improved RRT path planning method, which introduces collision feedback and threshold information, solves the problems of search efficiency and path quality in complex environments, and generates more efficient and smoother paths.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI JIAOTONG UNIV
- Filing Date
- 2026-02-04
- Publication Date
- 2026-05-01
AI Technical Summary
The original RRT algorithm has low search efficiency and poor path quality in complex environments, especially in dynamic and multi-robot scenarios.
An improved RRT path planning method is introduced by incorporating collision feedback-threshold information. This method optimizes the path generation process through target point bias sampling, adaptive guidance weight coefficients, and node expansion strategies.
It improves the efficiency and quality of path planning, reduces invalid sampling, generates more direct and smooth paths, and maintains good obstacle avoidance performance.
Smart Images
Figure CN121957033A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of robot path planning technology, and particularly relates to an improved RRT path planning method based on collision feedback-threshold information. Background Technology
[0002] In intelligent systems such as autonomous mobile robots, drones, and autonomous driving, path planning is one of the core technologies for achieving environmental perception, motion control, and task execution. Among numerous path search algorithms, the Rapidly-exploring Random Tree (RRT) algorithm has been widely used in real-time path planning tasks for robots since its inception due to its applicability to high-dimensional, non-convex, and complex obstacle environments. RRT constructs a tree structure that approximately covers the state space in the configuration space through a random sampling strategy, gradually approaching the target point, and has good exploration efficiency and convergence. However, the original RRT algorithm still has some limitations in practical applications, such as low search efficiency, poor path quality (manifested as long and non-smooth paths), and insufficient exploration of certain scenarios, especially under dynamic, multi-robot, or complex constraints.
[0003] Compared to traditional graph-based path planning methods (such as A...), Dijkstra's RRT is better suited for solving continuous space search problems in unstructured environments. While the algorithm performs well on known maps, its reliance on graph heuristics and grid partitioning makes it difficult to extend to high-dimensional spaces. Although Dijkstra's algorithm can find the globally optimal path, its computational complexity struggles to cope with the exponential growth in high-dimensional environments. Summary of the Invention
[0004] To address the aforementioned technical problems, this invention proposes an improved RRT path planning method based on collision feedback-threshold information. This method can solve the problems of low search efficiency and poor path quality of the RRT algorithm in certain complex environments, thereby improving the efficiency of robot path planning.
[0005] To achieve the above objectives, this invention provides an improved RRT path planning method based on collision feedback-threshold information, comprising: Initialize a random tree based on the robot's starting point and target area, and set the loop stopping condition; Perform robot target point offset sampling to determine the sampling point; Based on the sampling point, the target point, and the first collision feedback-threshold information, the comprehensive distance of the existing nodes in the random tree is determined, and the parent node is determined according to the comprehensive distance; Based on the parent node, the sampling point, the target point, and the second collision feedback-threshold information, the expansion direction and expansion step size of the new node are determined, and the new node is generated. Collision detection is performed. If no collision occurs between the parent node and the new node, the new node is added to the random tree. If the nearest parent node of the current node in the random tree exceeds the specified distance limit or the cumulative number of iterations reaches the limit, the iteration process is terminated. When the iteration terminates and the cumulative number of iterations does not exceed the limit, a path from the starting point to the target point is generated by backtracking the random tree.
[0006] Optionally, the determination of the loop stopping condition includes: In each iteration, the distance to the nearest parent node of the current node in the random tree is calculated; The iteration process terminates when the distance to the nearest parent node exceeds the specified distance limit; the path generation fails when the cumulative number of iterations exceeds the iteration limit; and the iteration terminates when the new node enters the target area.
[0007] Optionally, perform robot target point offset sampling to determine the sampling points, including: Based on a preset probability threshold, the selection probability is randomly generated; When the selection probability is less than the preset probability threshold, the target point is determined as a sampling point; When the selection probability is greater than or equal to the preset probability threshold, sampling points are determined by random sampling in free space.
[0008] Optionally, determining the comprehensive distance of existing nodes in the random tree based on the sampling points, the target point, and the collision feedback-threshold information includes: Calculate the first distance between the sampling point and each node in the random tree; Calculate the second distance between the target point and each node in the random tree; Based on the collision feedback-threshold information, and combined with the sampling node guidance weight coefficient and the target point guidance weight coefficient, the first distance and the second distance are weighted and summed to determine the comprehensive distance.
[0009] Optionally, the first collision feedback-threshold information includes: the number of consecutive collisions, the number of consecutive no-collisions, and the iteration number threshold; When the number of consecutive collisions reaches a first threshold, the guidance weight coefficient of the sampling node and the guidance weight coefficient of the target point are adjusted based on a linear correction function. When the number of consecutive collisions reaches the second threshold, the guiding weight coefficient of the sampling node and the guiding weight coefficient of the target point are adjusted based on a nonlinear correction function. When the number of consecutive collision-free times reaches the third threshold, the target point guidance weight coefficient is increased. When the number of iterations reaches the iteration threshold, the initial values of the sampling node guidance weight coefficient and the target point guidance weight coefficient are nonlinearly adjusted based on the reduction factor.
[0010] Optionally, the expansion direction and expansion step size of the new node are determined, and the generation of the new node includes: The position of the new node is calculated based on the first direction vector from the parent node to the sampling point, the second direction vector from the parent node to the target point, the direction expansion coefficient of the sampling point, the direction expansion coefficient of the target point, and the expansion step size.
[0011] Optionally, the second collision feedback-threshold information includes: the number of consecutive collisions, the number of consecutive no-collisions, the iteration threshold, and the geometric angle; When the number of consecutive collisions does not reach the threshold of the complex region, the sampling point direction expansion coefficient, the target point direction expansion coefficient, and the expansion step size are adjusted based on the linear correction function. When the number of consecutive collisions reaches the threshold of the complex region, the sampling point direction expansion coefficient, the target point direction expansion coefficient, and the expansion step size are adjusted based on a nonlinear correction function. When the angle between the first direction vector and the second direction vector is an obtuse angle, increase the target point direction expansion coefficient and decrease the expansion step size; When the number of consecutive collision-free times reaches the fourth threshold, the directional expansion coefficient of the sampling point and the directional expansion coefficient of the target point are increased. When the number of iterations reaches the iteration threshold, the initial values of the sampling point directional expansion coefficient and the target point directional expansion coefficient are nonlinearly adjusted based on the reduction factor.
[0012] Optionally, when there are no obstacles between the parent node and the target point, the target point directional expansion coefficient is set to a first preset value, the sampling point directional expansion coefficient is set to a second preset value, and the expansion step size is adjusted according to the distance between the parent node and the target point.
[0013] Compared with the prior art, the present invention has the following advantages and technical effects: 1. Effectively improves exploration efficiency and reduces invalid sampling and expansion: By introducing a target point guidance mechanism and an adaptive guidance coefficient, this invention can expand more focusedly towards the target point, reduce useless branches generated in non-critical areas, thereby significantly improving the effective exploration ratio and accelerating the path search speed.
[0014] 2. Implement adaptive parameter adjustment to enhance algorithm stability and environmental adaptability: By utilizing an adaptive parameter adjustment mechanism that combines linear and nonlinear approaches, the node expansion strategy can be dynamically adjusted based on real-time collision feedback and threshold information, enabling the invention to maintain good exploration performance in various complex environments.
[0015] 3. Optimize path structure, reduce redundant movements, and lower path costs: In the process of expanding new nodes, a comprehensive adaptive expansion coefficient and step size control are introduced. Combined with geometric angle analysis, unnecessary detour behavior is effectively suppressed, and a more direct and smooth optimal path is generated.
[0016] 4. Improve path optimization while maintaining good obstacle avoidance performance: The improved node expansion strategy enhances path planning efficiency and quality while ensuring effective obstacle avoidance, and does not compromise path safety in pursuit of shorter paths. Attached Figure Description
[0017] 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 the improved RRT path planning method based on collision feedback-threshold information according to an embodiment of the present invention; Figure 2 This is a path diagram of an embodiment of the present invention; Figure 3 This is a box plot of the runtime of an embodiment of the present invention; Figure 4 This is a box plot of the path length according to an embodiment of the present invention; Figure 5 This is an effective exploration box plot of the present invention. Detailed Implementation
[0018] 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.
[0019] 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.
[0020] This embodiment proposes an improved RRT path planning method based on collision feedback-threshold information, such as... Figure 1 As shown, the specific steps include: Initialize a random tree based on the robot's starting point and target area, and set the loop stopping condition; Perform robot target point offset sampling to determine the sampling point; Based on the sampling point, the target point, and the first collision feedback-threshold information, the comprehensive distance of the existing nodes in the random tree is determined, and the parent node is determined according to the comprehensive distance; Based on the parent node, the sampling point, the target point, and the second collision feedback-threshold information, the expansion direction and expansion step size of the new node are determined, and the new node is generated. Collision detection is performed. If no collision occurs between the parent node and the new node, the new node is added to the random tree. If the nearest parent node of the current node in the random tree exceeds the specified distance limit or the cumulative number of iterations reaches the limit, the iteration process is terminated. When the iteration terminates and the cumulative number of iterations does not exceed the limit, a path from the starting point to the target point is generated by backtracking the random tree.
[0021] The detailed steps include: Step 1: Set the starting point as the root node of the tree, initialize an empty random tree, and set the loop stopping condition. Define the free space or obstacle environment and the target area.
[0022] Step 2: Target point bias sampling, select the target point as the sampling point with a certain probability.
[0023] Step 3: Parent node selection. Find the node in the existing tree that has the closest comprehensive distance to the random sampling point. In the selection process, not only the random point and existing nodes in the tree are considered. Distance also includes the distance between the target point and existing nodes in the tree. Distance. The two distances mentioned above are weighted and summed to obtain the node-parent node with the closest comprehensive distance. The weights are adaptive parameters based on collision feedback and threshold information.
[0024] Step 4: Expand the tree to create new nodes. Extend a certain distance from the nearest neighbor node in the direction of the combined effect of the random sampling point and the target point to generate a new node. The specific weights in the two directions are based on collision feedback-threshold information and geometric angle changes. It is worth noting that a direct connection strategy is used when there are no obstacles between the current node and the target point.
[0025] Step 5: Collision detection. Check whether the path between the newly generated node and the nearest node collides with obstacles to ensure the feasibility of the path. If a collision occurs, discard the node. If no collision occurs, record the node and add the newly generated node to the random tree.
[0026] Step 6: In each iteration, verify whether the nearest parent node of the current node in the tree structure has exceeded the specified distance limit. Once the condition is met, stop the iteration process; otherwise, continue to the next round of exploration.
[0027] Step 7: If the cumulative number of iterations has not exceeded the upper limit when the loop terminates, the path generation is successful. The final path can be constructed by backtracking to complete the entire process of the RRT algorithm.
[0028] Specifically, conducting research on improved versions of the RRT algorithm has significant theoretical and practical value. On the one hand, it helps to improve search efficiency, path quality, and support for complex tasks while maintaining the RRT algorithm's exploration capabilities and robustness. On the other hand, it also provides more efficient algorithmic support for dynamic environments, multi-constraint tasks, and scenarios with extremely high real-time requirements (such as autonomous driving, cooperative obstacle avoidance, and air traffic management).
[0029] Furthermore, the determination of the loop stopping condition includes: In each iteration, the distance to the nearest parent node of the current node in the random tree is calculated; The iteration process terminates when the distance to the nearest parent node exceeds the specified distance limit; the path generation fails when the cumulative number of iterations exceeds the iteration limit; and the iteration terminates when the new node enters the target area.
[0030] Further, perform robot target point bias sampling to determine the sampling points, including: Based on a preset probability threshold, the selection probability is randomly generated; When the selection probability is less than the preset probability threshold, the target point is determined as a sampling point; When the selection probability is greater than or equal to the preset probability threshold, sampling points are determined by random sampling in free space.
[0031] Specifically, in step 2, the target point bias sampling is performed according to equation (1).
[0032] (1); Furthermore, based on the sampling points, the target point, and the collision feedback-threshold information, determining the comprehensive distance of existing nodes in the random tree includes: Calculate the first distance between the sampling point and each node in the random tree; Calculate the second distance between the target point and each node in the random tree; Based on the collision feedback-threshold information, and combined with the sampling node guidance weight coefficient and the target point guidance weight coefficient, the first distance and the second distance are weighted and summed to determine the comprehensive distance.
[0033] Furthermore, the first collision feedback-threshold information includes: the number of consecutive collisions, the number of consecutive no-collisions, and the iteration number threshold; When the number of consecutive collisions reaches a first threshold, the guidance weight coefficient of the sampling node and the guidance weight coefficient of the target point are adjusted based on a linear correction function. When the number of consecutive collisions reaches the second threshold, the guiding weight coefficient of the sampling node and the guiding weight coefficient of the target point are adjusted based on a nonlinear correction function. When the number of consecutive collision-free times reaches the third threshold, the target point guidance weight coefficient is increased. When the number of iterations reaches the iteration threshold, the initial values of the sampling node guidance weight coefficient and the target point guidance weight coefficient are nonlinearly adjusted based on the reduction factor.
[0034] Specifically, in step 3, from the existing tree Find the distance to the random sampling point and distance from the target point The closest node in the overall distance As the parent node, it is executed according to (2).
[0035] (2); in, For sampling node guided weight coefficients, The target point is guided by a weighting coefficient.
[0036] Exploring areas with simple obstacles requires setting a strong exploration goal, but considering that the current node is relatively far from the goal point at the start, therefore, [the following settings are not specified]. A lower value is sufficient to meet the requirements. At this point, , The constraint relationship shown in equation (3) is satisfied.
[0037] (3); This indicates that the rapid exploration towards the target did not get stuck in a complex obstacle area, at which point the number of consecutive collisions was [data missing]. right A linear correction is performed, as shown in equation (4).
[0038] (4); in for Linear correction factor.
[0039] This indicates that the rapid exploration towards the target has entered a complex obstacle area, at which point the target point guidance weighting coefficient should be increased. The correction is performed using a nonlinear correction as shown in equation (5).
[0040] (5); in for The nonlinear correction coefficient.
[0041] When collisions cease, in order to more effectively guide the agent through complex obstacle areas, this embodiment proposes to base the number of consecutive collision-free events on... and their corresponding thresholds Dynamic adjustment , Initial value. The specific method is as follows: hour, (6); To prevent the algorithm from getting stuck in complex obstacle areas for too long, when the number of iterations reaches a threshold... At that time, the initial weighting coefficients A nonlinear adjustment will be made to obtain new weight coefficients. This mechanism can effectively increase the number of random exploration branches, thereby improving the obstacle avoidance ability of the algorithm, as shown in Equation (7).
[0042] , (7); in, This represents the reduction factor associated with the iteration threshold and the target point guiding weight coefficient.
[0043] When the system determines that it has left the complex obstacle area, then... To enhance the ability to quickly generate exploration branches towards the objective point, the expression can be represented as: (8); once After the update is complete, execute according to formulas (3), (4), and (5) again.
[0044] Furthermore, determining the expansion direction and expansion step size of the new node, and generating the new node includes: The position of the new node is calculated based on the first direction vector from the parent node to the sampling point, the second direction vector from the parent node to the target point, the direction expansion coefficient of the sampling point, the direction expansion coefficient of the target point, and the expansion step size.
[0045] Furthermore, the second collision feedback-threshold information includes: the number of consecutive collisions, the number of consecutive no-collisions, the iteration threshold, and the geometric angle; When the number of consecutive collisions does not reach the threshold of the complex region, the sampling point direction expansion coefficient, the target point direction expansion coefficient, and the expansion step size are adjusted based on the linear correction function. When the number of consecutive collisions reaches the threshold of the complex region, the sampling point direction expansion coefficient, the target point direction expansion coefficient, and the expansion step size are adjusted based on a nonlinear correction function. When the angle between the first direction vector and the second direction vector is an obtuse angle, increase the target point direction expansion coefficient and decrease the expansion step size; When the number of consecutive collision-free times reaches the fourth threshold, the directional expansion coefficient of the sampling point and the directional expansion coefficient of the target point are increased. When the number of iterations reaches the iteration threshold, the initial values of the sampling point directional expansion coefficient and the target point directional expansion coefficient are nonlinearly adjusted based on the reduction factor.
[0046] Specifically, in step 4, the new node expansion is as shown in equation (9): (9); in Represents the directional spread coefficient of the sampling points, Represents the target point directional expansion coefficient, To extend the step size, unlike RRT-APF, the extension coefficient is based on collision feedback-threshold information and adaptively changes the geometric angle.
[0047] Exploring areas with simple obstacles requires setting a strong target-oriented function during expansion; therefore, the setting is... At a high level, at this time, , The constraints embodied in equation (10) are satisfied.
[0048] (10); During the exploration, it should be noted that if there are no obstacles between the current nearest node and the target point, the weight coefficient of the nearest point to the target point direction is 1, the initial weight coefficient of the nearest point to the target point direction is 0, and the algorithm step size is adjusted according to formula (11).
[0049] (11); This indicates that the rapid exploration towards the target did not get stuck in a complex obstacle area, at which point the number of consecutive collisions was [data missing]. right Adaptive correction is performed, as shown in (12).
[0050] (12); It is the revised version. , When not entering a complex obstacle area linear correction coefficient, It is the revised version. , It is the linear correction coefficient for step size when not entering a complex obstacle region.
[0051] It is worth noting that if the sampling point deviates too much from the target direction before entering the complex area, it needs to be corrected. Here, the parameters can be corrected by geometric angle. Calculate the direction vector of the nearest point to the target point and the angle between the direction vector of the nearest point to the random point. If it is an obtuse angle, the random expansion direction deviates too much from the target. Then the target direction expansion coefficient should be increased appropriately and the initial step size should be reduced to make the expansion direction deviate from the target direction. The mathematical expression is shown in Equation (13).
[0052] (13); This indicates that the exploration towards the target is trapped in a complex obstacle area. In this case, a nonlinear correction is used to accelerate the reduction of the target direction coefficient. The specific correction is shown in Equation (14).
[0053] (14); in This is a nonlinear correction coefficient for the target point's directional expansion coefficient when entering complex regions. This is the step size nonlinearity correction coefficient when entering complex regions.
[0054] When collisions cease, in order to more effectively guide the agent through complex obstacle areas, this embodiment proposes to base the number of consecutive collision-free events on... and their corresponding thresholds Dynamic adjustment , Initial value. The specific method is as follows: hour, (15); To prevent the algorithm from getting stuck in complex obstacle areas for too long, when the number of iterations reaches a threshold... At that time, initial coefficients A nonlinear adjustment will be made to obtain new weight coefficients. This mechanism can effectively accelerate the search of the remaining areas of space, thereby improving the obstacle avoidance ability of the algorithm, as shown in Equation (16).
[0055] , (16); in, This represents the reduction factor associated with the iteration threshold and the target point weight expansion coefficient.
[0056] When the system determines that it has left the complex obstacle area, then... To perform a boost to rapidly expand toward the target point, the expression can be represented as: (17); once After the update is complete, execute according to formulas (10)-(17) again.
[0057] Furthermore, when there are no obstacles between the parent node and the target point, the target point directional expansion coefficient is set to a first preset value, the sampling point directional expansion coefficient is set to a second preset value, and the expansion step size is adjusted according to the distance between the parent node and the target point.
[0058] The following describes a specific application of this embodiment in a complex simulation environment. This embodiment implements the improved RRT algorithm using MATLAB simulation software. The simulation software version is MATLAB R2021a, the computer processor is an Intel(R) i5-13500H, the RAM is 16GB, and the operating system is Windows 11.
[0059] The specific implementation of this embodiment will be described in detail below with reference to specific examples.
[0060] To understand the simulation results, this embodiment first clarifies how to measure performance, specifically through four key metrics: running efficiency (the speed at which the algorithm runs), path length (the total distance of the generated effective paths), stability (measured by the change in path length before pruning), and effective exploration ratio, defined as the ratio of the final unpruned path length to all branch / trunk paths.
[0061] The complex environment is a dense complex area composed of multiple rectangular obstacles. Based on 100 simulation data, the algorithm has an average path cost of 701.3, an average running time of 0.09s, and an average effective exploration ratio of 35.33%.
[0062] Figure 2-5 Experimental results show that in complex environments with dense obstacles, this embodiment not only demonstrates high path exploration efficiency and excellent obstacle avoidance performance, but also exhibits good overall performance in terms of path length and feasibility.
[0063] 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. An improved RRT path planning method based on collision feedback-threshold information, characterized in that, include: Initialize a random tree based on the robot's starting point and target area, and set the loop stopping condition; Perform robot target point offset sampling to determine the sampling point; Based on the sampling point, the target point, and the first collision feedback-threshold information, the comprehensive distance of the existing nodes in the random tree is determined, and the parent node is determined according to the comprehensive distance; Based on the parent node, the sampling point, the target point, and the second collision feedback-threshold information, the expansion direction and expansion step size of the new node are determined, and the new node is generated. Perform collision detection. If no collision occurs between the parent node and the new node, add the new node to the random tree. If the nearest neighbor of the current node in the random tree exceeds the specified distance limit or the cumulative number of iterations reaches the limit, terminate the iteration process. When the iteration terminates and the cumulative number of iterations has not exceeded the upper limit, a path from the starting point to the target point is generated by backtracking the random tree.
2. The improved RRT path planning method based on collision feedback-threshold information according to claim 1, characterized in that, The determination of the loop termination condition includes: In each iteration, the distance to the nearest parent node of the current node in the random tree is calculated; The iteration process terminates when the distance to the nearest parent node exceeds the specified distance limit; the path generation fails when the cumulative number of iterations exceeds the iteration limit; and the iteration terminates when the new node enters the target area.
3. The improved RRT path planning method based on collision feedback-threshold information according to claim 1, characterized in that, Perform robot target point offset sampling to determine the sampling points, including: Based on a preset probability threshold, the selection probability is randomly generated; When the selection probability is less than the preset probability threshold, the target point is determined as a sampling point; When the selection probability is greater than or equal to the preset probability threshold, sampling points are determined by random sampling in free space.
4. The improved RRT path planning method based on collision feedback-threshold information according to claim 1, characterized in that, Based on the sampling points, the target point, and the collision feedback-threshold information, the comprehensive distance to the existing nodes in the random tree is determined as follows: Calculate the first distance between the sampling point and each node in the random tree; Calculate the second distance between the target point and each node in the random tree; Based on the collision feedback-threshold information, and combined with the sampling node guidance weight coefficient and the target point guidance weight coefficient, the first distance and the second distance are weighted and summed to determine the comprehensive distance.
5. The improved RRT path planning method based on collision feedback-threshold information according to claim 4, characterized in that, The first collision feedback-threshold information includes: the number of consecutive collisions, the number of consecutive no-collisions, and the iteration number threshold; When the number of consecutive collisions reaches a first threshold, the guidance weight coefficient of the sampling node and the guidance weight coefficient of the target point are adjusted based on a linear correction function. When the number of consecutive collisions reaches the second threshold, the guiding weight coefficient of the sampling node and the guiding weight coefficient of the target point are adjusted based on a nonlinear correction function. When the number of consecutive collision-free times reaches the third threshold, the target point guidance weight coefficient is increased. When the number of iterations reaches the iteration threshold, the initial values of the sampling node guidance weight coefficient and the target point guidance weight coefficient are nonlinearly adjusted based on the reduction factor.
6. The improved RRT path planning method based on collision feedback-threshold information according to claim 1, characterized in that, Determine the expansion direction and expansion step size of the new node, and generate the new node by including: The position of the new node is calculated based on the first direction vector from the parent node to the sampling point, the second direction vector from the parent node to the target point, the direction expansion coefficient of the sampling point, the direction expansion coefficient of the target point, and the expansion step size.
7. The improved RRT path planning method based on collision feedback-threshold information according to claim 6, characterized in that, The second collision feedback-threshold information includes: the number of consecutive collisions, the number of consecutive no-collisions, the iteration threshold, and the geometric angle; When the number of consecutive collisions does not reach the threshold of the complex region, the sampling point direction expansion coefficient, the target point direction expansion coefficient, and the expansion step size are adjusted based on the linear correction function. When the number of consecutive collisions reaches the threshold of the complex region, the sampling point direction expansion coefficient, the target point direction expansion coefficient, and the expansion step size are adjusted based on a nonlinear correction function. When the angle between the first direction vector and the second direction vector is an obtuse angle, increase the target point direction expansion coefficient and decrease the expansion step size; When the number of consecutive collision-free times reaches the fourth threshold, the directional expansion coefficient of the sampling point and the directional expansion coefficient of the target point are increased. When the number of iterations reaches the iteration threshold, the initial values of the sampling point directional expansion coefficient and the target point directional expansion coefficient are nonlinearly adjusted based on the reduction factor.
8. The improved RRT path planning method based on collision feedback-threshold information according to claim 7, characterized in that, When there are no obstacles between the parent node and the target point, the direction expansion coefficient of the target point is set to a first preset value, the direction expansion coefficient of the sampling point is set to a second preset value, and the expansion step size is adjusted according to the distance between the parent node and the target point.