A mobile robot multi-target path planning method based on RRT* and APF fusion

Abstract

The application discloses a mobile robot multi-target path planning method based on RRT* and APF fusion, comprising: environment modeling and parameter initialization; based on the artificial potential field method, sampling and node expansion are carried out; for a new reference node, parent node reselection and rewiring are carried out based on a comprehensive cost function; adaptive step adjustment is carried out; target accessibility is judged, and path optimization is carried out. The application draws the guiding sampling area by the resultant force direction of the APF attraction and repulsion through the fusion of the APF guiding sampling and parallel sampling strategy, and carries out parallel sampling in the guiding area and the global range at the same time, which reduces the invalid sampling probability, improves the search direction, improves the sampling efficiency through the parallel mode, and at the same time, a small amount of random sampling is reserved to guarantee the global search and avoid local blind area.

Description

Technical Field

[0001] This application belongs to the field of robot path planning technology, specifically relating to a multi-objective path planning method for mobile robots based on the fusion of RRT* and APF. Background Technology

[0002] Path planning is one of the key technologies in robot autonomous navigation systems. Its core objective is to plan a feasible path from the starting point to the target point for the robot in an environment with obstacles, while meeting specific performance requirements. With the continuous expansion of robot application scenarios, such as complex and dynamic environments like indoor warehousing, outdoor exploration, and disaster relief, single-objective path planning (such as simply pursuing the shortest path) can no longer meet practical needs. Multi-objective path planning (which needs to consider path length, safety, energy consumption, and motion stability simultaneously) has become a current research hotspot and challenge.

[0003] In existing technologies, various algorithms have been proposed for robot path planning, among which the Fast Search Random Tree Star (RRT*) algorithm and the Artificial Potential Field (APF) method are widely used. The RRT* algorithm, as an improved version of the RRT algorithm, introduces parent node reselection and rewiring, possessing asymptotic optimality and demonstrating advantages in high-dimensional path planning. The APF algorithm guides robot movement by simulating the "attraction" of target points and the "repulsion" of obstacles, featuring smooth paths and good real-time performance, excelling in local path planning. However, both still have significant shortcomings in multi-objective path planning scenarios in complex dynamic environments, making it difficult to meet practical application requirements on their own.

[0004] 1. Shortcomings of the Fast Search Random Tree Star Algorithm (RRT*): Although the RRT* algorithm has asymptotic optimality, it still has core shortcomings: the randomness of sampling leads to low search efficiency, many redundant nodes, and lag in response to complex dynamic environments; the traditional cost function is difficult to adapt to multi-objective requirements and has weak adaptability to complex scenarios; the generated path has poor smoothness and motion control is difficult; the step size is fixed, making it difficult to balance planning efficiency and safety.

[0005] 2. Shortcomings of Artificial Potential Field (APF) Method: Although the APF algorithm has smooth paths and strong real-time performance, it has shortcomings: First, it is prone to getting trapped in local optima, causing the robot to fail to reach the target point; Second, it is not good at guiding distant targets, and it is easy to wander aimlessly in the initial stage; Third, it is difficult to adapt to the needs of multi-objective optimization; Fourth, the fluctuation of the repulsive field update in dynamic environments can easily lead to path jitter.

[0006] Therefore, how to combine the global search capability and asymptotic optimality of the RRT* algorithm with the local guidance capability and smooth path generation advantage of the APF algorithm to make up for the shortcomings of each in complex dynamic environments, and at the same time solve problems such as path evaluation, search efficiency optimization, dynamic obstacle avoidance, and path quality improvement under multi-objective constraints, has become a technical problem that urgently needs to be solved in current robot path planning technology, and is also the core technical pain point that this invention aims to overcome. Summary of the Invention

[0007] This application provides a multi-objective path planning method for mobile robots based on the fusion of RRT* and APF, in order to solve the technical problems of random sampling, low search efficiency and poor path quality in the existing RRT* algorithm.

[0008] To address the aforementioned technical problems, this application adopts the following technical solution: a multi-objective path planning method for mobile robots based on the fusion of RRT* and APF, comprising:

[0009] S1. Based on the location information of the starting point, target point, and obstacles, perform environmental modeling and parameter initialization;

[0010] S2. Sampling and node expansion are performed based on the artificial potential field method;

[0011] S3. For the new baseline node, perform parent node reselection and rerouting based on the comprehensive cost function;

[0012] S4. Adaptive step size adjustment;

[0013] S5. Determine the reachability of the target and optimize the path.

[0014] Furthermore, the method in step S1 includes: constructing a grid map of the robot's motion environment using a grid method, defining the position information of the starting point, the target point, and obstacles; initializing a random tree with the starting point as the root node, configuring the basic attributes of the root node; and simultaneously setting the core control parameters for the growth of the random tree and the effect of the artificial potential field.

[0015] Furthermore, the method in step S2 includes:

[0016] S21. Using the most recently generated node in the random tree as the reference node, calculate the magnitude and direction of the gravitational force of the target point on the reference node, and the magnitude and direction of the repulsive force of the obstacle on the reference node;

[0017] S22. Based on the direction of the resultant force of attraction and repulsion, candidate sampling points are obtained using a parallel sampling method;

[0018] S23. Perform node expansion based on candidate sampling points.

[0019] Furthermore, the method in step S22 includes:

[0020] S221. Based on the direction of the resultant force, determine the guiding sampling area and the global random sampling area;

[0021] S222. Perform parallel uniform sampling within the guided sampling region to generate N1 guided candidate sampling points; simultaneously perform parallel random sampling within the global random sampling region to generate N2 random candidate sampling points; where N1+N2=N, and N is the preset total number of parallel samples;

[0022] S223. Calculate the cost from the current reference node to each candidate point generated by parallel sampling, taking into account path length, safety, and energy consumption, and normalize it by the effective distance traveled to the target point;

[0023] S224. Sort all candidate points by cost and select the candidate point with the lowest cost as the final candidate sampling point; if there are multiple candidate points with the same cost, the candidate point located in the guided sampling area shall be selected as the final candidate sampling point.

[0024] Furthermore, the method in step S23 includes:

[0025] S231. In the constructed random tree, find the nearest neighbor node to the candidate sampling point;

[0026] S232. Move from the neighboring nodes toward the candidate sampling point with the current step size to generate a new node;

[0027] S233. Based on the collision detection algorithm, detect whether there is an obstacle collision between the connection between the new node and the neighboring nodes; if there is a collision, discard the new node and return to step S22 for resampling; if there is no collision, the new node is the new reference node.

[0028] Furthermore, the method in step S3 includes:

[0029] S31. Taking the reference node as the center, define a neighborhood region with a preset radius, and select all nodes located in the neighborhood region from the constructed random tree that have no obstacles colliding with the reference node, forming a candidate parent node set.

[0030] S32. Based on each candidate parent node, calculate the comprehensive cost of the baseline node when the corresponding candidate parent node is the parent node;

[0031] S33. Sort the comprehensive costs of all candidate parent nodes and select the node with the smallest comprehensive cost as the optimal parent node; if there are multiple candidate parent nodes with the same comprehensive cost, prioritize the node that is closest to the benchmark node.

[0032] S34. Using the benchmark node as the core of optimization, optimize the cost of existing nodes in the nearest neighbor region of the benchmark node in the random tree.

[0033] Furthermore, the method in step S34 includes:

[0034] S341. Select existing nodes in the nearest neighbor region of the base node, excluding the parent node of the base node, from the constructed random tree to form a set of candidate nodes for rewiring.

[0035] S342. Perform collision detection and cost calculation sequentially for each rewiring candidate node to obtain the cumulative overall cost when the baseline node is the parent node;

[0036] S343. Compare the current cumulative total cost of the rerouting candidate node with the cumulative total cost when the reference node is the parent node; if the cumulative total cost when the reference node is the parent node is lower, then perform the rerouting operation and update the parent node of the rerouting candidate node to the reference node; if the cumulative total cost when the reference node is the parent node is not lower than the current cumulative total cost, then abandon the rerouting of the rerouting candidate node.

[0037] S344. After traversing all nodes in the set of candidate nodes for rerouting, the rerouting optimization is complete.

[0038] Furthermore, the method for adaptive step size adjustment in step S4 includes:

[0039] S41. Use the first distance from the current reference node to the obstacle and the second distance to the target point as the input parameters of the fuzzy controller, and the step size as the output parameter;

[0040] S42. Blur the first distance, the second distance, and the step size to obtain the input fuzzy quantity;

[0041] S43. Based on the input fuzzy quantity, a fuzzy rule base is formulated according to the environmental adaptability requirements. The Mamdani inference method is used to obtain the output fuzzy quantity, and the centroid method is used to convert the output fuzzy quantity into a precise step size value.

[0042] The beneficial effects of this application are as follows: This application integrates APF guided sampling and parallel sampling strategies, uses the resultant force direction of APF attraction and repulsion to define the guided sampling region, and simultaneously carries out parallel sampling in the guided region and the global scope. This not only reduces the probability of invalid sampling and improves the search directionality, but also improves the sampling efficiency through the parallel mode. At the same time, it retains a small amount of random sampling to ensure the globality of the search and avoid local blind spots. By constructing a comprehensive cost function covering path length, safety, and energy consumption, it achieves multi-objective collaborative optimization, which makes up for the limitations of single-objective optimization of traditional algorithms and better adapts to the needs of complex scenarios. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating an embodiment of the multi-objective path planning method for mobile robots based on the fusion of RRT* and APF in this application;

[0044] Figure 2 yes Figure 1 Step S1 in the embodiment is a grid map;

[0045] Figure 3 yes Figure 1 A flowchart illustrating step S2 of an embodiment;

[0046] Figure 4 yes Figure 3 A flowchart illustrating step S22 of an embodiment;

[0047] Figure 5 yes Figure 4 Step S221 is a schematic diagram of APF guided sampling in one embodiment;

[0048] Figure 6 yes Figure 3 A flowchart illustrating step S23 in one embodiment;

[0049] Figure 7 yes Figure 1 A flowchart illustrating step S3 in one embodiment;

[0050] Figure 8 yes Figure 1 A flowchart of an embodiment of step S4. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0052] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0053] See Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the multi-objective path planning method for mobile robots based on the fusion of RRT* and APF according to this application. The method includes:

[0054] S1. Based on the location information of the starting point, target point, and obstacles, perform environmental modeling and parameter initialization.

[0055] Specifically, a grid map of the robot's motion environment is constructed using the grid method, clearly defining the coordinates of the starting point and the target point, as well as the location information of obstacles; among them, passable areas in the grid map are marked with a preset value (such as 0), and obstacle areas are marked with another preset value (such as 1). Figure 2 The image shows a 20x20 grid map, where black areas represent obstacles, white areas represent passable areas, S is the starting point, and G is the target point. The root node of the initial random tree is set to the starting point S. The root node stores its own coordinates, the index of its parent node (initially -1, representing no parent node), and the accumulated comprehensive cost (initially 0). Core parameters are set, including the initial step size, maximum number of iterations, sampling area range, multi-objective weight coefficients, APF attraction coefficient, APF repulsion coefficient, repulsion threshold, and the radius of the target point's preset threshold area.

[0056] S2. Based on the Artificial Potential Field (APF) method, sampling and node expansion are performed. The aim is to guide the sampling by the direction of the resultant force of attraction and repulsion of the APF, reduce the randomness of sampling, improve the directionality of the search, and complete the node expansion at the same time.

[0057] For details, please refer to Figure 3 The method of step S2 includes:

[0058] S21. Using the most recently generated node in the random tree as the reference node, calculate the magnitude and direction of the gravitational force of the target point on the reference node, as well as the magnitude and direction of the repulsive force of the obstacle on the reference node.

[0059] Specifically, gravity calculation: the most recently generated node in the current random tree is used as the reference node. Calculate the target point G relative to the reference node. The magnitude and direction of gravity. The magnitude of gravity is positively correlated with the distance from the node to the target point. Using a distance-dependent gravity model, the calculation formula is:

[0060] (1);

[0061] in, The gravitational coefficient, for Euclidean distance to G, From The unit vector pointing to G indicates that the gravitational force always points towards the target point.

[0062] Repulsion calculation: Calculate the relationship between the obstacle and the reference node. The magnitude and direction of the repulsive force are related, and the magnitude of the repulsive force is negatively correlated with the distance from the node to the obstacle; the closer the distance, the greater the repulsive force. Using a distance-negative correlation repulsive force model, the repulsive force is calculated for each obstacle separately and then summed. The formula for calculating the repulsive force of a single obstacle is:

[0063] (2);

[0064] in, The repulsion coefficient is... for To the i-th obstacle The shortest distance, The threshold of repulsive force. From point to The unit vector of the repulsive force always points away from the obstacle. Total repulsive force , where n is the total number of obstacles.

[0065] S22. Based on the direction of the resultant force of attraction and repulsion, candidate sampling points are obtained by parallel sampling.

[0066] Specifically, the direction of the resultant force of the combined attraction and repulsion. Candidate sampling points are generated using a parallel sampling method.

[0067] See Figure 4 The method of step S22 includes:

[0068] S221. Based on the direction of the resultant force, determine the guiding sampling area and the global random sampling area.

[0069] For details, please refer to Figure 5 Centered on the direction of the resultant force, a ±30° guiding sampling area is defined, while a global random sampling area (covering the entire feasible area of ​​robot movement) is also defined.

[0070] S222. Perform parallel uniform sampling within the guided sampling region to generate... One guiding candidate sampling point; simultaneously, parallel random sampling is performed within the global random sampling region to generate... 10 random candidate sampling points; among them N is the preset total number of parallel samples.

[0071] S223. Calculate the cost from the current base node to each candidate point generated by parallel sampling, taking into account path length, security, and energy consumption. The specific formula is as follows:

[0072] (3);

[0073] in: As a reference node to candidate point The Euclidean distance reflects the length of the local path; for The closer the distance to the nearest obstacle, the higher the safety cost. Energy consumption coefficient per unit length The energy consumption coefficient per unit steering angle. for Parent node and Connect and and The angle between the lines; The effective distance traveled towards the target point is defined as... The difference between the straight-line distance to the target point and the straight-line distance from the candidate point to the target point, i.e. . The weighting coefficients for path length, safety, and energy consumption satisfy... .

[0074] S224. Cost of generating N candidate points from all parallel sampling. The points are sorted, and the candidate point with the lowest cost is selected as the final candidate sampling point. If multiple candidate points have the same cost, priority should be given to candidate points located within the guided sampling area to ensure a balance between the directionality and optimality of the sampling.

[0075] S23. Perform node expansion based on candidate sampling points.

[0076] For details, please refer to Figure 6 The method of step S23 includes:

[0077] S231. In the constructed random tree, find the candidate sampling point. Nearest neighbor node .

[0078] S232. From neighboring nodes To candidate sampling points Move in the current step direction and generate a new node. New node The coordinate calculation formula is: ; where step is the current step size, for point to , unit vector;

[0079] S233. Detecting new nodes based on collision detection algorithms. and Check if there are any obstacles or collisions in the connecting lines; if a collision occurs, discard the new node. Then return to step S22 for resampling; if no collision occurs, then the new node... This serves as the new baseline node.

[0080] S3. For the new baseline node, perform parent node reselection and rerouting based on the comprehensive cost function.

[0081] Specifically, the parent node reselection and rewiring based on the comprehensive cost function aims to select the optimal parent node through multi-objective comprehensive cost evaluation, and optimize the random tree structure through rewiring to reduce the overall path cost and ensure that the path length, security and energy consumption are optimally combined.

[0082] For details, please refer to Figure 7 The method of step S3 includes:

[0083] S31. Based on the reference node Centered on a target node, a nearest neighbor region with a preset radius is defined (the radius is set according to the complexity of the environment to balance search accuracy and efficiency). From the constructed random tree, all nodes within this nearest neighbor region that are connected to the base node without any obstacles or collisions are selected to form a candidate parent node set. The nearest neighbor nodes... Force inclusion into the candidate parent node set to ensure the validity of the basic expansion path.

[0084] S32. Based on each candidate parent node, calculate the comprehensive cost of the baseline node when the corresponding candidate parent node is the parent node.

[0085] The specific definition and calculation of the comprehensive cost function: Calculate for each candidate parent node X. The overall cost when it is the parent node is defined as follows:

[0086] (4);

[0087] in: express The cumulative total cost when X is the parent node. Let X be the cumulative total cost; For X and The Euclidean distance, i.e., the cost of the additional local path length; for The reciprocal of the distance to the nearest obstacle, i.e., the increment in safety cost; Energy consumption coefficient per unit length The energy consumption coefficient per unit steering angle. The line connecting the parent node of X to X and X is... and The angle between the lines.

[0088] S33. Sort the total cost of all candidate parent nodes and select the node with the lowest total cost as the parent node. The optimal parent node is determined by the baseline node; if multiple candidate parent nodes have the same overall cost, the node with the same cost is preferred. By selecting the nearest node, local path efficiency is further improved. The parent node index is updated to the index of the optimal parent node, its cumulative total cost is recorded, and then... Add to the random tree.

[0089] S34. Using the benchmark node as the core of optimization, optimize the cost of existing nodes in the nearest neighbor region of the benchmark node in the random tree.

[0090] For details, please refer to Figure 8 The method in step S34 includes:

[0091] S341. Rewire Candidate Node Selection: Select baseline nodes from the constructed random tree. Within the nearest neighbor area, excluding the reference node The existing nodes other than the parent node form a set of candidate nodes for rewiring.

[0092] S342. Rerouting Feasibility and Cost Assessment: Perform the following operations sequentially for each candidate node X for rerouting:

[0093] 1) Collision detection: Determine the baseline node Check whether the connection between the node and the candidate node X for rewiring is free from obstacle collisions to ensure the path is feasible.

[0094] 2) Cost Calculation: Calculate the candidate node X for rewiring based on the baseline node. The cumulative total cost when the node is the new parent is calculated as follows:

[0095] (5);

[0096] in, Indicates X as The cumulative total cost when it becomes the new parent node. for The cumulative total cost; for The Euclidean distance from X, i.e., the cost of the additional local path length; It is the reciprocal of the distance from X to the nearest obstacle, i.e., the increment of safety cost; Energy consumption coefficient per unit length The energy consumption coefficient per unit steering angle. for Parent node and Connect and The angle between the line connecting it to X and the line X.

[0097] S343. Rerouting Execution: Compare the current cumulative total cost of rerouting candidate node X with that of the baseline node. Cumulative total cost when acting as a parent node ;like If the value is lower, a rerouting operation is performed, updating the parent node of the rerouting candidate node to [the new value]. The cumulative comprehensive cost is updated synchronously. And update the cumulative total cost of all child nodes of X, where the cost of a child node = the updated cost of the parent node + the local cost between the parent and child nodes; if If the cost is not lower than the current cumulative total cost, then the rerouting of the candidate node for rerouting will be abandoned.

[0098] S344. Rerouting Termination: After traversing all nodes in the rerouting candidate node set, the rerouting optimization is complete. Rerouting can improve the random tree structure, reduce path redundancy costs, and provide a better tree foundation for subsequent iterative expansion.

[0099] S4. Adaptive step size adjustment aims to improve the algorithm's adaptability to different environments by dynamically adjusting the step size. The specific implementation process is as follows:

[0100] The step size is adjusted in real time using a fuzzy logic controller. The specific process is as follows:

[0101] S41. Use the first distance from the current reference node to the obstacle and the second distance to the target point as the input parameters of the fuzzy controller, and the step size as the output parameter.

[0102] Specifically, the input and output variables are determined as follows: the first distance from the current baseline node (the latest node in the random tree) to the obstacle. and the second distance to the target point The step size is used as the input parameter for the fuzzy controller and as the output parameter.

[0103] S42. Blur the first distance, the second distance, and the step size to obtain the input fuzzy quantity.

[0104] Specifically, the blurring process involves: reducing the first distance... Divide into three fuzzy subsets: "near," "medium," and "far," and use the second distance... The values ​​are divided into three fuzzy subsets: "near", "medium", and "far". The step function is further divided into three fuzzy subsets: "small", "medium", and "large". The triangular membership function is used to convert the precise values ​​into fuzzy values.

[0105] S43. Based on the input fuzzy quantity, a fuzzy rule base is formulated according to the environmental adaptability requirements. The Mamdani inference method is used to obtain the output fuzzy quantity, and the centroid method is used to convert the output fuzzy quantity into a precise step size value.

[0106] Specifically, fuzzy rule base formulation: rules are set based on environmental adaptability requirements, such as "if the first distance..." For the closest and second distance "If it is arbitrary, then step is small." "If the first distance..." For the farthest and second distance "If the distance is far, then step is large." "If the first distance..." For the middle and second distance If it is in the middle, then step is in the middle, etc.;

[0107] Fuzzy Inference and Defuzzification: The Mamdani inference method is used to obtain the output fuzzy quantity, which is then converted into a precise step size value using the centroid method. When the first distance... When the distance is relatively small (near obstacles or narrow passages), use small step sizes to ensure obstacle avoidance accuracy; when the first distance... Larger and second distance When the area is large (open area), use a larger step size to improve search efficiency.

[0108] S5. Determine the reachability of the target and optimize the path.

[0109] The goal reachability assessment verifies the positional deviation and connectivity between the planned path's endpoint and the preset target point to confirm that the path meets the goal reachability requirements. The specific steps are as follows:

[0110] Determine the new baseline node Whether to enter the preset threshold region of target point G; the threshold region is set as follows: a circular region is constructed with target point G as the center and the sum of the robot's own radius and the safety redundancy distance as the radius. If the reference node... If the coordinates fall within the circular area, it is determined that the region has entered the threshold region.

[0111] If the reference node If the threshold region is entered, an attempt is made from the baseline node. Directly connect to target point G to determine the reference node. Check if the line connecting to the target point G is free from obstacles and collisions; if not, plan a feasible path from S to the target point G (traverse the parent node index from the target point G back to S to obtain the original path node sequence); if the reference node... If the target point threshold region is not entered, or if a collision occurs between the line connecting the target point G and the target point G, return to step S2 and repeat the iteration until the maximum number of iterations is reached; if no feasible path is planned after reaching the maximum number of iterations, the planning is deemed to have failed and no feasible path information is output.

[0112] Post-processing optimization of the path involves pruning redundant nodes and smoothing the original planned path to simplify its structure and improve trajectory continuity. This ensures the final trajectory conforms to the robot's kinematic constraints, guaranteeing the stability of subsequent tracking motion. The specific steps are as follows:

[0113] Path pruning: First, the original path is recorded as a sequence of nodes including the starting point S and the target point G. An empty list is created and the starting point is added. Then, taking the last node in the list as the reference, the subsequent nodes are checked in turn to determine whether the straight line segment between the reference node and the subsequent node crosses an obstacle. The last node that can be connected without collision is recorded and added to the list, and the reference node is updated. This process is repeated until the reference node is the target point. The final list is the simplified path after removing redundant nodes. This can shorten the path length and reduce the computational complexity of subsequent smoothing processing.

[0114] Path smoothing: A cubic B-spline curve is used to fit the pruned path, eliminating inflection points and abrupt changes. The specific calculation method is as follows: using the nodes of the pruned path as control vertices, the basis functions of the B-spline curve are constructed. The formula for calculating the basis functions is:

[0115] (6);

[0116] (7);

[0117] Where u is a parameter variable, For k-th order B-spline basis functions, Given node vectors, a smoothed B-spline curve is obtained by weighting the vertices using basis functions, thus generating a smooth and continuous final path.

[0118] The above-mentioned technical solution of this application has significant advantages and beneficial effects: By integrating APF guided sampling and parallel sampling strategies, the guided sampling area is defined by the direction of the resultant force of APF attraction and repulsion, and parallel sampling is carried out simultaneously in the guided area and the global range. This not only reduces the probability of invalid sampling and improves the search directionality, but also improves the sampling efficiency through the parallel mode. At the same time, a small amount of random sampling is retained to ensure the globality of the search and avoid local blind spots. By constructing a comprehensive cost function covering path length, safety, and energy consumption, multi-objective collaborative optimization is achieved, which makes up for the limitations of single-objective optimization of traditional algorithms and better adapts to the needs of complex scenarios. Relying on the fuzzy logic controller to dynamically adjust the step size, small step size is used to achieve fine obstacle avoidance in complex and narrow environments, and large step size is used to advance quickly in open environments, effectively balancing planning efficiency and path safety, and enhancing the adaptability to environments of different complexities. By shortening the path length and eliminating path tortuosity through path pruning and smoothing, the path after B-spline fitting is smooth and continuous, reducing the number of starts, stops and turns during robot movement, reducing energy consumption while improving motion stability and extending the robot's service life.

[0119] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A multi-objective path planning method for mobile robots based on the fusion of RRT* and APF, characterized in that, include: S1. Based on the location information of the starting point, target point, and obstacles, perform environmental modeling and parameter initialization; S2. Sampling and node expansion are performed based on the artificial potential field method; S3. For the new baseline node, perform parent node reselection and rerouting based on the comprehensive cost function; S4. Adaptive step size adjustment; S5. Determine the reachability of the target and optimize the path.

2. The method according to claim 1, characterized in that, The method of step S1 includes: constructing a grid map of the robot's motion environment using a grid method, defining the position information of the starting point, the target point, and obstacles; initializing a random tree with the starting point as the root node, configuring the basic attributes of the root node; and setting the core control parameters for the growth of the random tree and the effect of the artificial potential field.

3. The method according to claim 1, characterized in that, The method of step S2 includes: S21. Using the most recently generated node in the random tree as the reference node, calculate the magnitude and direction of the gravitational force of the target point on the reference node, and the magnitude and direction of the repulsive force of the obstacle on the reference node; S22. Based on the direction of the resultant force of the attraction and the repulsion, candidate sampling points are obtained using a parallel sampling method; S23. Based on the candidate sampling points, perform node expansion.

4. The method according to claim 3, characterized in that, The method of step S22 includes: S221. Based on the direction of the resultant force, determine the guiding sampling region and the global random sampling region; S222. Perform parallel uniform sampling within the guided sampling area to generate... A number of guiding candidate sampling points are generated; simultaneously, parallel random sampling is performed within the global random sampling region to generate... 1 random candidate sampling point; among which, N is the preset total number of parallel samples; S223. Calculate the cost from the current reference node to each candidate point generated by parallel sampling, taking into account path length, safety, and energy consumption, and normalize it by the effective distance traveled to the target point; S224. Sort the costs of all the candidate points and select the candidate point with the lowest cost as the final candidate sampling point; wherein, if there are multiple candidate points with the same cost, the candidate point located in the guided sampling area shall be selected as the final candidate sampling point.

5. The method according to claim 3, characterized in that, The method of step S23 includes: S231. In the constructed random tree, find the nearest neighbor node to the candidate sampling point; S232. Move from the neighboring node toward the candidate sampling point with the current step size to generate a new node; S233. Based on the collision detection algorithm, detect whether there is an obstacle collision between the connection between the new node and the neighboring node; if there is a collision, discard the new node and return to step S22 for resampling; if there is no collision, the new node is the new reference node.

6. The method according to claim 1, characterized in that, The method of step S3 includes: S31. Taking the reference node as the center, define a neighborhood region with a preset radius, and select all nodes located in the neighborhood region and whose lines connecting to the reference node are free from obstacles from the constructed random tree to form a candidate parent node set. S32. Based on each candidate parent node, calculate the comprehensive cost of the base node when the corresponding candidate parent node is the parent node; S33. Sort the comprehensive costs of all candidate parent nodes and select the node with the smallest comprehensive cost as the optimal parent node; if there are multiple candidate parent nodes with the same comprehensive cost, prioritize the node that is closest to the benchmark node. S34. Using the benchmark node as the optimization core, perform cost optimization on the existing nodes in the nearest neighbor region of the benchmark node in the random tree.

7. The method according to claim 6, characterized in that, The method of step S34 includes: S341. Select existing nodes in the neighborhood of the reference node, excluding the parent node of the reference node, from the constructed random tree to form a set of candidate nodes for rewiring. S342. Perform collision detection and cost calculation sequentially for each rewiring candidate node to obtain the cumulative overall cost when the baseline node is the parent node; S343. Compare the current cumulative total cost of the candidate rerouting node with the cumulative total cost when the reference node is the parent node; if the cumulative total cost when the reference node is the parent node is lower, then perform a rerouting operation and update the parent node of the candidate rerouting node to the reference node; if the cumulative total cost when the reference node is the parent node is not lower than the current cumulative total cost, then abandon the rerouting of the candidate rerouting node. S344. After traversing all nodes in the set of candidate nodes for rerouting, the rerouting optimization is complete.

8. The method according to claim 1, characterized in that, The method for adaptive step size adjustment in step S4 includes: S41. Use the first distance from the current reference node to the obstacle and the second distance to the target point as the input parameters of the fuzzy controller, and the step size as the output parameter; S42. Blur the first distance, the second distance, and the step size step to obtain the input fuzzy quantity; S43. Based on the input fuzzy quantity, a fuzzy rule base is formulated according to the environmental adaptability requirements. The Mamdani inference method is used to obtain the output fuzzy quantity, and the centroid method is used to convert the output fuzzy quantity into a precise step size value.

Images