A mobile robot path planning method based on an improved bidirectional RRT algorithm

Abstract

The application relates to a mobile robot path planning method based on an improved bidirectional RRT algorithm, and belongs to the technical field of mobile robot path planning. a b The bidirectional random tree T a b is constructed by taking a starting point and an ending point as root nodes, adaptive sampling is realized by adopting a directional deviation heuristic function containing a bias angle sampling strategy, node distribution is optimized by combining a redundant point removal strategy based on adaptive adjustment of an environmental complexity threshold, and the random tree is directionally expanded by introducing an improved artificial potential field method. The application can quickly generate a safer optimal path which is shorter and smoother, significantly improves the efficiency and robustness of path planning, and is suitable for complex scene path planning tasks of various mobile robots such as service robots and unmanned toy cars.

Description

Technical Field

[0001] This invention relates to a mobile robot path planning method based on an improved bidirectional RRT algorithm, belonging to the field of mobile robot path planning technology. Background Technology

[0002] In recent years, with the rapid development of sensor systems and computing technologies, autonomous wheeled mobile robot technology has made significant progress and its applications in service, industry, and autonomous driving fields have become increasingly widespread. Path planning, as the core technology of autonomous navigation for mobile robots, aims to plan a collision-free, optimal, or suboptimal passable path from the starting point to the target point for the robot.

[0003] Existing path planning algorithms can be mainly divided into three categories: The first category is grid-based algorithms, typically represented by Dijkstra's algorithm and A* algorithm. Dijkstra's algorithm can solve the shortest path, has a simple structure and good convergence, but it requires traversing a large number of nodes, resulting in high time complexity and low planning efficiency. A* algorithm improves the blind search problem of Dijkstra's algorithm by introducing a heuristic function, thus improving planning efficiency, but it still suffers from excessively high computational cost in complex environments. The second category is intelligent optimization algorithms, such as genetic algorithms. These algorithms simulate the biological evolution process, possess global search capabilities, and can handle complex nonlinear optimization problems, but they are prone to getting trapped in local optima, and their performance is highly sensitive to parameters such as inertia weights and learning factors. The third category is planning algorithms based on random sampling, represented by the RRT algorithm. Its core idea is to randomly generate nodes with the starting point as the root node and connect them into a tree until the target point is found. This algorithm has probabilistic completeness and can quickly find feasible solutions, but the generated paths are jagged, have poor path quality, and cannot guarantee convergence to the optimal path, making it unsuitable for complex obstacle scenarios.

[0004] The bidirectional randomized recursive root-targeting (RRT) algorithm, as an improved version of the RRT algorithm, enhances search efficiency by simultaneously constructing two random trees at the start and end points. However, the traditional bidirectional RRT algorithm still suffers from problems such as blind sampling, high node redundancy, and insufficient optimization of connection strategies, making it difficult to meet the path planning requirements of mobile robots in complex environments. Therefore, there is an urgent need to propose an improved bidirectional RRT path planning method that combines high planning efficiency, excellent path quality, and strong environmental adaptability. Summary of the Invention

[0005] To overcome the shortcomings of existing path planning algorithms, this invention provides a mobile robot path planning method based on an improved bidirectional RRT algorithm. This method addresses the deficiencies of traditional RRT and bidirectional RRT algorithms, such as zigzag paths, low planning efficiency, weak adaptability to complex environments, and susceptibility to local optima. The improved bidirectional RRT algorithm can quickly generate safe and optimal paths, improve planning efficiency and adaptability, and enhance the path planning performance of mobile robots in complex obstacle scenarios.

[0006] A path planning method for a mobile robot based on an improved bidirectional RRT algorithm includes the following steps:

[0007] Step 1: Building a map model: Install the ROS system on the embedded device Jetson Nano, perform SLAM mapping, obtain the experimental scene map and obstacle location information, determine the starting point and target point of the path, and collect environmental data of the mobile robot's surroundings.

[0008] Step 2: Initialize the bidirectional random tree: Set up two random trees T a T b Using the starting point and the target point as root nodes respectively, the bias angle sampling expansion operation is alternately performed on two random trees. The bias angle sampling expansion introduces a directional deviation heuristic function, which, based on the distribution of obstacles in the map, samples in the direction of the obstacle-free area during the sampling process to improve the efficiency of initial path generation.

[0009] Step 3: Redundant point removal: Taking into account the complexity of the environment, the node selection threshold is adaptively and dynamically adjusted to remove redundant nodes in the random tree, ensuring that the random search tree expands towards the target point as a whole.

[0010] Step 4: Selecting Extended Nodes: Based on the random sampling points, select extended nodes for the random tree using an improved artificial potential field method strategy, and set the potential field attraction coefficient K. r Repulsion coefficient K a And the range of the repulsive field;

[0011] Step 5: Set the search step size: Based on the environment information, configure a search step size that matches the environment complexity, which will serve as the basis for each step of the random tree expansion;

[0012] Step Six: Parent Node Filtering and Update: Based on the extended node, find its farthest reachable parent node; for the reachable parent node, generate the node closest to the obstacle between it and the extended node using a bisection method; perform bisection again on the node closest to the obstacle and the extended node, and after verifying that the connection is a safe path, create a new parent node for the extended node; set the hypersphere range with the new parent node as the center, and filter the better connection path within the sphere according to the triangle inequality, thus completing the update of the path list and node list;

[0013] Step 7: Pruning and Bidirectional Optimization of Connections: Find nodes q that can be connected to using the FindReachest process. new And the parent node q without collision reachest The node q closest to the obstacle is generated using a binary search method through the CreateNode process. create Set a termination threshold as the connection range between two random trees, and use a bidirectional optimization connection strategy to complete the connection between the two trees; set the pruning radius and the maximum number of iterations for path update Iter, and perform pruning and reconnection optimization on the generated initial path to obtain the optimal path.

[0014] The deflection angle sampling expansion in step two specifically includes:

[0015] A directional bias heuristic function is introduced. During the random tree expansion process, if the newly generated node q... new With target point q goal If there are obstacles on the path of connection, the sampling angle is changed by using a directional deviation heuristic function to avoid q. new With q goal Obstacles between;

[0016] Two parameters are defined to quantify the effect of obstacles, namely q new to q goal The percentage of obstacles along the route R V-n Obstacle area ratio R V-s ,in, , N V-obs For q new With q goal The number of obstacles between them, N obs S represents the total number of obstacles on the map. V-obs For q new With q goal The area of ​​the obstacle between them, S obs This represents the total area of ​​obstacles within the map.

[0017] Based on R V-n With R V-s Calculate the deflection angle Set the initial angle of the sampling direction deflection. The sampling angle range is determined to be { , If the sampling point angle is within this range, sampling is performed to achieve obstacle avoidance; if it exceeds this range, target bias sampling is performed, causing the random tree to expand in a directional direction towards the target point. Let be the initial angle of the sampling direction deflection, representing the angle from q. start to q goalThe initial angle formed by the line connecting the two points and the horizontal direction of the constructed map. Indicates the deviation of the sampling angle from the initial angle The deviation. Used to calculate the maximum and minimum angles. (The last part, "greater than the maximum angle," appears to be a separate, unrelated statement and is left untranslated.) and less than the minimum angle Within a certain range, the sampling function has a high sampling probability.

[0018] The redundancy point removal in step three specifically includes:

[0019] A threshold constraint parameter α is introduced to ensure that the node selection threshold is within a reasonable range;

[0020] Define the current iteration number I and the total iteration number I. ter Total environmental area S, number of obstacles N obs The ratio of obstacle area to total environmental area ;

[0021] According to the formula Calculate the node selection threshold d; α is a limiting parameter for the threshold, ensuring that the node selection threshold remains within a reasonable range, if random sampling points q rand If the distance to an adjacent node in the random tree is less than a threshold d, the sampling point is determined to be a redundant point and removed; otherwise, q is removed. rand Add to the random tree.

[0022] By comprehensively considering environmental complexity and algorithm optimization, the node selection threshold is adaptively and dynamically adjusted, effectively improving path quality and global optimality while ensuring search efficiency. First, the distribution of obstacles in the environmental map is analyzed to determine the initial value of the node selection threshold. This initial threshold can be adaptively set according to different environmental conditions, thus avoiding performance degradation caused by uniform parameter configuration in different scenarios. Furthermore, this allows the algorithm to quickly cover a large search space in the initial stage, improving global exploration efficiency; while in the later stage, the threshold is reduced to increase the accuracy of node selection, thereby more fully exploring potential high-quality paths. Therefore, the algorithm's path optimization capability and global exploration performance are significantly improved.

[0023] The specific process of the improved artificial potential field method strategy in step four is as follows:

[0024] The angle μ between the attractive force of the target point and the repulsive force of the obstacle is used as the key adjustment parameter;

[0025] When cosμ∈[−1 / 2,0], the target point and the nearest node q of the random sampling point nearest Applying attraction, the obstacle to q nearest When a repulsive force is applied, the resultant force of the three forces points towards the new node q.new The direction of generation;

[0026] When the included angle μ exceeds the set angle threshold λ, a projection factor P is introduced. d Dynamically adjust gravity F att The magnitude and amplitude of the repulsive force F are constrained, and the repulsive force component F is discarded. rep2 Only the repulsive force component F, which is orthogonal to the direction of gravity, is retained. rep1 This ensures that the direction of the resultant force maintains its tendency to move toward the target point while avoiding obstacles.

[0027] The algorithm treats the mobile robot's motion space as a potential field space, within which exist a gravitational potential field formed by the target point and a repulsive potential field formed by obstacles. The magnitudes of the gravitational and repulsive forces are the negative gradients of the gravitational and repulsive field functions, respectively. These three potential fields map to three forces: attraction, repulsion, and resultant force. The relationship between these three forces is shown in the following formula:

[0028]

[0029]

[0030]

[0031] Using the angle μ between the attractive force and the repulsive force of the obstacle at the target point as the key adjustment parameter, when cosμ is in the interval [-1 / 2, 0], it indicates that the directional relationship between the attractive and repulsive forces is relatively reasonable, and both the target point and the random sampling point are relatively stable. nearest Applying attraction, the obstacle to q nearest When a repulsive force is applied, the combined direction of the three forces is q. new The direction of generation. When the angle μ continues to increase and exceeds the set angle threshold λ, an angle is introduced. Dynamically adjust gravity F att The magnitude of the resultant force F is determined, and its amplitude is reasonably constrained; total Direction such as Figure 5 As shown in (b), it interacts with the repulsive force F. req The directions are orthogonal, which effectively avoids obstacles while preserving the tendency to move toward the target point to the greatest extent.

[0032]

[0033]

[0034]

[0035]

[0036] When μ∈[0,π], the optimized resultant force F att Attraction F ′ att The adjustment method is defined by the formula above. The threshold λ corresponds to the cosine value of the boundary angle, while P... d This represents the given projection factor. When μ exceeds this boundary, P is introduced. d Discarding repulsive force F rep2 Only its orthogonal component F is retained. rep1 By dynamically adjusting the force field, this method can improve obstacle avoidance efficiency, accelerate the robot's movement towards the target, and significantly improve overall motion efficiency. Simultaneously, the adjusted attractive force F... ′ att This ensures that the combined force remains within a safe range, thus preventing the robot from deviating from the predetermined path or falling into a local optimum due to unreasonable force application.

[0037] The improved pruning and reconnection strategy is as follows:

[0038] The method, based on the Rewire process of RRT*, splits the ChooseParent process into two processes: FindReachest and CreateNode.

[0039] To improve algorithm efficiency, the FindReachest process eliminates the concept of a hypersphere and stops operating on every "neighboring" node. Instead, it calculates the relationship between nodes globally and q. new Find the distance and locate the nearest node q. nearest like Figure 5 As shown in the node identifier in (b), based on this, we continuously search q. nearest The parent node of the node is checked for collisions until a connectable node q is found. new The parent node that does not collide is denoted as q. reachest The FindReachest process is now complete.

[0040] The CreateNode process uses a binary search method to generate a node q that is closest to the obstacle. create It can connect to q simultaneously new With q reachest The parent node has no collisions. To prevent infinite binary search, a D is added. dichotomy Parameters. For a detailed binary search process of CreateNode, see [link to CreateNode documentation]. Figure 6 As shown, this process will occur in q reachest With Parent(q) reachest The algorithm continuously generates intermediate nodes in a loop, thereby creating a node closer to the obstacle, shortening the path cost, and further improving the efficiency of the algorithm.

[0041] For the hypersphere in the rewire process, keeping the hypersphere's radius constant may yield good results in the initial stages of path planning. However, as planning progresses, the number of nodes within the hypersphere continuously increases, significantly increasing computational costs. Therefore, an adaptive strategy is proposed: as the number of nodes increases, R... near Decrease to reduce computational complexity. Adaptive R near The process is described by the following formula:

[0042]

[0043] The bidirectional optimized connection strategy in step seven specifically includes: the connection method is no longer limited to the starting point direction, but instead generates candidate nodes q along the target point direction. creat , q reachest The parent node and q creat Make connections to optimize and adjust the paths.

[0044] The bidirectional optimization strategy is as follows:

[0045] The bidirectional search tree strategy optimizes the traditional unidirectional model: it preserves the original, starting point q start Starting search tree T a Based on this, an additional tree is introduced with the target point q as the basis. goal Search tree T with root node b This forms a bidirectional parallel expansion logic of "starting point → intermediate region" and "target → intermediate region". In the bidirectional search, the connected path optimized by trigonometric functions cannot be further optimized using the CreatNode and FindReachest operations. The core reason is that, according to the original optimization logic, it is necessary to follow the path from the starting point q... start Generate new candidate nodes q in the direction creat But from Figure 7 As can be seen from scenario (a), this direction is restricted by obstacles and other factors, making it impossible to generate an effective q. creat Since nodes cannot be optimized using CreateNode (creating nodes) and FindReachest (finding the nearest node), a bidirectional optimization strategy is designed to address this issue: instead of being limited to the direction of the starting point, it instead follows the path along the target point q. creat Direction to generate candidate nodes q goal Then "q" reachest The parent node (parent-q) reachest )” and the newly generated q creat This allows for the optimization and adjustment of the path. From Figure 7(b) The effect of this strategy can also be seen intuitively: originally unable to generate q due to directional limitations. creat After being generated along the direction of the target point, (parent −q) reachest ) and q creat The connection successfully avoided obstacles while bringing the path closer to the target point, achieving the optimization goal.

[0046] Preferably, the preset distance is one step of the robot, but due to the existence of the bias strategy, the step is not a constant value.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] This invention significantly shortens path planning time by introducing a bidirectional extended tree construction mechanism, resulting in shorter and smoother paths. The optimization model, which combines a redundant point removal strategy with the artificial potential field method, significantly enhances obstacle avoidance capabilities. It solves the adaptability and efficiency problems of path planning algorithms in complex obstacle scenarios. The method of this invention can be widely applied to path planning tasks for service robots, driverless toy cars, and mobile robots in complex environments.

[0049] Improving planning efficiency: The bias angle sampling strategy achieves directional obstacle avoidance sampling through the directional bias heuristic function, reducing the number of invalid samplings; the redundant point removal strategy adaptively removes redundant nodes, reducing computational complexity. The combination of the two significantly shortens the path planning time.

[0050] Optimize path quality: Improve the artificial potential field method to avoid local optima and ensure that the path extends to the target point; parent node optimization and pruning and reconnection strategies eliminate path jaggedness, resulting in shorter and smoother paths.

[0051] Enhanced environmental adaptability: The bidirectional optimized connection strategy breaks through the limitations of traditional unidirectional connections and improves the success rate of dual-tree docking in complex obstacle scenarios; the parameters of each strategy can be adaptively adjusted according to the complexity of the environment, making it suitable for various scenarios such as simple, chaotic, and maze-like environments.

[0052] Highly practical: This invention is implemented based on Jetson Nano embedded devices and ROS systems, with low hardware costs and convenient deployment. It can be widely used in path planning tasks for service robots, driverless toy cars, and mobile robots in complex environments. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is an overall flowchart of a mobile robot path planning method based on an improved bidirectional RRT algorithm according to the present invention.

[0055] Figure 2 This invention relates to a simple, chaotic, maze-like map for a mobile robot path planning method based on an improved bidirectional RRT algorithm.

[0056] Figure 3 This is a schematic diagram of the bias angle sampling in a mobile robot path planning method based on an improved bidirectional RRT algorithm according to the present invention.

[0057] Figure 4 This is a redundancy point removal strategy diagram in a mobile robot path planning method based on an improved bidirectional RRT algorithm according to the present invention.

[0058] Figure 5 This invention relates to an improved artificial potential field strategy diagram in a mobile robot path planning method based on an improved bidirectional RRT algorithm.

[0059] Figure 6 This invention relates to a pruning and reconnection graph in a mobile robot path planning method based on an improved bidirectional RRT algorithm.

[0060] Figure 7 This is a bidirectional optimized connection strategy diagram in a mobile robot path planning method based on an improved bidirectional RRT algorithm according to the present invention.

[0061] Figure 8 This is a comparison chart of the improvement of various algorithms before and after improvement in a simple environment in the mobile robot path planning method based on the improved bidirectional RRT algorithm of the present invention.

[0062] Figure 9 This is a comparison chart of various algorithms before and after improvement in a chaotic environment in a mobile robot path planning method based on an improved bidirectional RRT algorithm according to the present invention.

[0063] Figure 10 This is a comparison diagram of the various algorithms before and after improvement in a maze environment in the mobile robot path planning method based on the improved bidirectional RRT algorithm of this invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Example:

[0066] The simulation environments used in this invention are all regions with a size of 100×100 pixels, such as... Figure 2 As shown, black pixels represent the space occupied by obstacles, white pixels represent free space, and blue or red pixels represent a random search tree with a tree-like expansion. Initial position q start = (2, 2) is located at the bottom left corner, and the target position is q. goal = (98, 98) is located in the upper right corner;

[0067] Step 1: Initialize two random trees T a T b ;T a With the starting point as the root node, T b The target point is the root node;

[0068] Step Two: To ensure the overall progress towards the target point, in each iteration, each tree adopts... Figure 3 The bias angle sampling strategy shown generates random points; a tree is expanded in the environment to generate random points q. rand The other tree is sampled; for each random tree, the distance from the random point q is found. rand The nearest tree node q nearest And try to extend along the direction from the random point to the node by a certain step size to generate a new node q. new If the node is close to the target point q goal If there are obstacles in the path, the function can avoid them by changing the sampling angle. new With q goal The obstacles between them. Two parameters are designed to quantify the impact of the obstacles, representing q respectively. new to q goal The percentage of obstacles along the route R V-n Obstacle area ratio R V-s， 。 Then based on R V-n With R V-s Generate a deflection angle. The initial angle of the sampling direction deflection. Indicates the deviation of the sampling angle from the initial angle The deviation. For the minimum angle, The maximum angle is determined. If the sampling points satisfy the angle relationship, sampling is performed, which is beneficial for obstacle avoidance. If the angle relationship is not satisfied, target-biased sampling is performed, prompting the random tree to expand in a directional direction towards the target point.

[0069] Step 3: To avoid multiple samplings in a certain area, a redundant point removal strategy is introduced, such as... Figure 4 As shown, once a q is detected rand Then calculate qrand The distance between it and its neighboring nodes. If the distance exceeds the node selection threshold d, then q will be selected. rand Add it to the random search tree; otherwise, prune the node to avoid unnecessary expansion.

[0070] Step 4: Generate new node q based on the improved artificial potential field method new .like Figure 5 As shown in (a), the angle μ between the attractive force and the repulsive force of the obstacle is used as the key adjustment parameter. When cosμ is in the interval [-1 / 2, 0], it indicates that the directional relationship between the attractive and repulsive forces is relatively reasonable, and both the target point and the random sampling point are relatively stable. nearest Applying attraction, the obstacle to q nearest When a repulsive force is applied, the combined direction of the three forces is q. new The direction of generation. When the angle μ continues to increase and exceeds the set angle threshold λ, an angle is introduced. Dynamically adjust gravity F att The magnitude of the resultant force F is determined, and its amplitude is reasonably constrained; total Direction such as Figure 5 As shown in (b), it interacts with the repulsive force F. req The directions are orthogonal, effectively avoiding obstacles while preserving the tendency to move towards the target point to the greatest extent. When μ∈[0,π], the optimized resultant force F att Attraction F ′ att The adjustment method is as follows Figure 5 As shown in (c). When μ exceeds this boundary, P is introduced. d Discarding repulsive force F rep2 Only its orthogonal component F is retained. rep1 By dynamically adjusting the force field, this method can improve obstacle avoidance efficiency, accelerate the robot's movement towards the target, and significantly improve overall motion efficiency. Simultaneously, the adjusted attractive force F... ′ att This ensures that the resultant force is always within a safe area, thereby preventing the robot from deviating from the predetermined path or falling into a local optimum due to unreasonable force action.

[0071] Step 5: Based on the identified extended node, trace its associated parent nodes in the path tree, filter out all reachable parent nodes, and finally determine the reachable parent node that is furthest from the extended node, such as... Figure 6 As shown in (b), after obtaining the intermediate node closest to the obstacle, a bisection method is used to further refine the segmentation between this intermediate node and the initial extended node. Finally, an optimal new parent node is redefined and created for the initial extended node. The result of generating this new parent node can be found in [reference needed]. Figure 6 As shown in (d). Wherein, Figure 6 In the example, (a) represents the ChooseParent procedure of other existing methods. Figure 6 (c) shows the rewire procedure for other existing methods;

[0072] Step Six: After the new parent node is created, a hypersphere region with a fixed radius is defined centered on the new parent node. Then, using the triangle inequality rule, it is determined whether the connection paths between other existing nodes within the hypersphere and the new parent node are better paths. If so, the connection relationships of the relevant nodes are updated, and the corresponding path data in the path table and the parent node association information in the node list are corrected simultaneously. If not, the node is ignored, and no connection operation is performed.

[0073] Step 7: Repeat the above steps. When the distance between a new node generated in one random tree and the nearest node in another random tree is less than a preset connection threshold, and there are no obstacles in the connection between the new node and the nearest node, the two random trees are considered successfully connected. Then, by backtracking the paths of the two random trees and completing the bidirectional path optimization connection, as shown... Figure 7 As shown in (b), the optimal path for the mobile robot from the starting point to the end point is finally obtained;

[0074] The following section will use Matlab to simulate the above control algorithm. Given the randomness of the sampling process, each algorithm needs to be run 50 times under the same environment and parameter conditions. That is, all algorithms use the exact same sampling sequence in each comparison experiment. The parameter settings for each algorithm are detailed below, and the specific explanations of each parameter name are as follows:

[0075] q start : Coordinates of the starting point for path planning;

[0076] q goal : Path planning endpoint coordinates;

[0077] P goal : Indicates the probability of sampling the target point, causing the overall path to shift towards the target point;

[0078] : Indicates that the sampling angle deviates from the initial angle The value; in connecting q start and q goal There are many, but small, obstacles near the route. The smaller value should be taken; conversely, A larger value should be chosen;

[0079] d: Node selection threshold; calculate qrand The distance to all nodes in the random tree; if this value is less than a set threshold d, it means q rand The region to which the point belongs has already been explored, and the sampling point is redundant; therefore, the point should be discarded. When the distance is greater than the threshold d, the space to which the point belongs has not been explored, and this point should be expanded into the random tree.

[0080] Numofnode: The number of nodes in the search tree; D[Numofnode] is used to calculate the distance from the Numofnode node to the sampling point;

[0081] S dis The expansion step size of the RRT tree;

[0082] The following comparison charts of operation under three different environments demonstrate the feasibility of the method of this invention;

[0083] Figure 8 This is a path comparison chart of the various algorithms run 50 times in a simple environment. Figure 8 (a) shows the results of the RRT* algorithm. Figure 8 (b) shows the results of the RRT*-Connect algorithm. Figure 8 (c) shows the results of running the PF-RRT* algorithm. Figure 8 (d) The running result of the improved algorithm of the present invention is shown in the figure. It can be seen from the figure that the method of the present invention has the shortest path length, the smoothest path, the fewest number of iterations, and the fewest number of generated nodes, indicating that it works well in a simple environment.

[0084] Figure 9 This is a comparison chart of the paths of each algorithm in 50 runs under a chaotic environment. Figure 9 (a) shows the results of the RRT* algorithm. Figure 9 (b) shows the results of the RRT*-Connect algorithm. Figure 9 (c) shows the results of running the PF-RRT* algorithm. Figure 9 (d) The running result of the improved algorithm of the present invention is shown in the figure. It can be seen from the figure that the method of the present invention has the shortest path length, the smoothest path, the fewest number of iterations, and the fewest number of generated nodes, indicating that it works well in a cluttered environment.

[0085] Figure 10 This is a path comparison chart of the various algorithms run 50 times in a maze environment. Figure 10 (a) shows the results of the RRT* algorithm. Figure 10 (b) shows the results of the RRT*-Connect algorithm. Figure 10 (c) shows the results of running the PF-RRT* algorithm. Figure 10(d) The result of the improved algorithm of the present invention is shown in the figure. It can be seen from the figure that the method of the present invention has the shortest path length, the smoothest path, the fewest iterations, and the fewest generated nodes, indicating that it works well in maze environments.

[0086] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A path planning method for a mobile robot based on an improved bidirectional RRT algorithm, characterized in that: Includes the following steps: Step 1: Building a map model: Install the ROS system on the embedded device Jetson Nano, perform SLAM mapping, obtain the experimental scene map and obstacle location information, determine the starting point and target point of the path, and collect environmental data of the mobile robot's surroundings. Step 2: Initialize the bidirectional random tree: Set up two random trees T a T b Using the starting point and the target point as root nodes respectively, the bias angle sampling expansion operation is alternately performed on two random trees. The bias angle sampling expansion introduces a directional deviation heuristic function, which samples the obstacle-free area in the direction of the sampling process according to the distribution of obstacles in the map, thereby improving the efficiency of initial path generation. Step 3: Redundant point removal: Taking into account the complexity of the environment, the node selection threshold is adaptively and dynamically adjusted to remove redundant nodes in the random tree, ensuring that the random search tree expands towards the target point as a whole. Step 4: Selecting Extended Nodes: Based on the random sampling points, select extended nodes for the random tree using an improved artificial potential field method strategy, and set the potential field attraction coefficient K. r Repulsion coefficient K a And the range of the repulsive field; Step 5: Set the search step size: Based on the environment information, configure a search step size that matches the environment complexity, which will serve as the basis for each step of the random tree expansion; Step Six: Parent Node Filtering and Update: Based on the extended node, find its farthest reachable parent node; for the reachable parent node, generate the node closest to the obstacle between it and the extended node using a binary search method; Based on the near-obstacle node and the extended node, perform the binary segmentation again. After verifying that the connection is a safe path, create a new parent node for the extended node. Set the range of the hypersphere with the new parent node as the center, and select the better connection path within the sphere according to the triangle inequality to complete the update of the path list and node list. Step 7: Pruning and Bidirectional Optimization of Connections: Find nodes q that can be connected to using the FindReachest process. new And the parent node q without collision reachest The node q closest to the obstacle is generated using a binary search method through the CreateNode process. create Set a termination threshold as the connection range between two random trees, and use a bidirectional optimization connection strategy to complete the connection between the two trees; set the pruning radius and the maximum number of iterations for path update Iter, and perform pruning and reconnection optimization on the generated initial path to obtain the optimal path.

2. The mobile robot path planning method based on the improved bidirectional RRT algorithm according to claim 1, characterized in that: The deflection angle sampling expansion in step two specifically includes: During the random tree expansion process, if the newly generated node q new With target point q goal There are obstacles on the connection path. The sampling angle is changed by using a directional deviation heuristic function to avoid the obstacles. Two parameters are defined to quantify the effect of obstacles, namely q new to q goal The percentage of obstacles along the route R V-n Obstacle area ratio R V-s ,in, , N V-obs For q new With q goal The number of obstacles between them, N obs S represents the total number of obstacles on the map. V-obs For q new With q goal The area of ​​the obstacle between them, S obs This represents the total area of ​​obstacles within the map. Based on R V-n With R V-s Calculate the deflection angle Set the initial angle of the sampling direction deflection. The sampling angle range is determined to be { , If the angle of the sampling point is within this range, sampling is performed to achieve obstacle avoidance; if it is outside this range, target bias sampling is performed to cause the random tree to expand in a directional manner toward the target point.

3. The mobile robot path planning method based on the improved bidirectional RRT algorithm according to claim 1, characterized in that: The redundancy point removal in step three specifically includes: A threshold constraint parameter α is introduced to ensure that the node selection threshold is within a reasonable range; Define the current iteration number I and the total iteration number I. ter Total environmental area S, number of obstacles N obs The ratio of obstacle area to total environmental area ; According to the formula Calculate the node selection threshold d; if the random sampling point q rand If the distance to an adjacent node in the random tree is less than a threshold d, the sampling point is determined to be a redundant point and removed; otherwise, q is removed. rand Add to the random tree.

4. The mobile robot path planning method based on the improved bidirectional RRT algorithm according to claim 1, characterized in that: The specific process of the improved artificial potential field method strategy in step four is as follows: The angle μ between the attractive force of the target point and the repulsive force of the obstacle is used as the key adjustment parameter; When cosμ∈[−1 / 2,0], the target point and the nearest node q of the random sampling point nearest Applying attraction, the obstacle to q nearest When a repulsive force is applied, the resultant force of the three forces points towards the new node q. new The direction of generation; When the included angle μ exceeds the set angle threshold λ, a projection factor P is introduced. d Dynamically adjust gravity F att The magnitude and amplitude of the repulsive force F are constrained, and the repulsive force component F is discarded. rep2 Retain the repulsive force component F that is orthogonal to the direction of gravity. rep1 This ensures that the direction of the resultant force maintains its tendency to move toward the target point while avoiding obstacles.

5. A mobile robot path planning method based on an improved bidirectional RRT algorithm according to claim 1, characterized in that: The bidirectional optimization connection strategy in step seven specifically includes: changing the connection method to generate candidate nodes q along the target point direction. creat , q reachest The parent node and q creat Make connections to optimize and adjust the paths.

6. A mobile robot path planning method based on an improved bidirectional RRT algorithm according to claim 1, characterized in that: The parent node selection in step six specifically includes: Based on the identified extended node, trace its associated parent nodes in the path tree, filter out all reachable parent nodes, and finally determine the reachable parent node that is farthest from the extended node. After obtaining the intermediate node that is close to the obstacle, continue to use the binary search method to further refine the segmentation between the intermediate node and the initial extended node, and finally redetermine and create an optimal new parent node for the initial extended node.

7. A mobile robot path planning method based on an improved bidirectional RRT algorithm according to claim 1, characterized in that: The parent node update in step six specifically includes: after the new parent node is created, a hypersphere region with a fixed radius is defined centered on the new parent node. Using the triangle inequality rule, it is determined whether the connection paths between other existing nodes within the hypersphere and the new parent node are better paths. If so, the connection relationships of the relevant nodes are updated, and the corresponding path data in the path table and the parent node association information in the node list are corrected simultaneously. If not, the node is ignored and no connection operation is performed.

Images