Improved path planning method based on RRT algorithm fused with artificial potential field method

Abstract

The invention belongs to the technical field of automatic guided vehicle path planning, and discloses an improved path planning method based on an RRT algorithm fused with an artificial potential field method, and the method comprises the steps: 1) constructing a state space, extracting an obstacle edge, and discretizing the obstacle edge into sampling points; 2) establishing a double-field cooperation model of a gravitational field (target point direction dynamic adjustment) and a repulsion field (obstacle sampling point superposition), and balancing exploration and obstacle avoidance through a weight ratio alpha / beta; 3) path nodes are generated based on a roulette selection method and an adaptive step length, and the search efficiency is improved; and 4) iteratively optimizing the path by adopting a greedy search strategy in combination with a triangular inequality, eliminating redundant nodes and verifying the security. According to the method, a large amount of redundant exploration of an RRT algorithm in random extension tree extension is effectively improved, the length of a planned path is shortened, and the problem that a target point cannot be reached due to conflict between virtual resultant force and an obstacle under a traditional artificial potential field algorithm is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning of Automated Guided Vehicles (AGVs) in a topological map, and specifically relates to a path planning method improved by fusing the artificial potential field method with the RRT algorithm. This method is applicable to scenarios such as intelligent robots and driverless vehicles that require efficient obstacle avoidance and path optimization. Background Art

[0002] Intelligent vehicles have great research value due to their huge advantages and broad application prospects in the civilian and military fields. An intelligent vehicle needs to analyze the environment around the vehicle body and make various decisions based on this to achieve functions such as real-time obstacle avoidance and road tracking. Compared with traditional vehicles, Automated Guided Vehicles (AGVs) have lower maintenance costs, can operate 24 hours a day with very little human intervention, and the improvement of their safety and the reduction of labor costs significantly enhance the economy. AGVs powered mainly by lithium batteries reduce energy consumption and greenhouse gas emissions, making them more environmentally friendly. The main task of path planning is to plan a reasonable, collision-free path from the starting point to the ending point based on the environmental data of the fusion understanding layer. Path planning is the key to reflecting the intelligence of the vehicle.

[0003] Regarding the path planning method for AGVs, many achievements have been made. Commonly used forward graph search algorithms and intelligent optimization algorithms are stable and usually can find the optimal solution. However, in practical applications, they are restricted by the degrees of freedom of the robot system and the dimensions of the state space. When considering differential constraints or other constraint conditions, they are prone to generating multiple dimensions, thus triggering the curse of dimensionality and resulting in an explosion of the computational amount. In fact, the planned path does not require the optimal solution. Usually, it only needs to find a sub-optimal path that meets various constraint conditions, which can avoid the use of dynamic programming and the problem of system space allocation. The advantages of random sampling planning methods are highlighted here. Rapidly-exploring Random Trees (RRT) is a single-query motion planning method based on optimal control theory, non-holonomic planning, and random path planning. Its basic idea is based on control theory. Within a short time interval, it incrementally grows from the initial state to a new state by inputting control. Each vertex in the random tree represents a state, and each directed edge represents the process of transitioning from the previous state to the new state. When a certain vertex reaches the target area, the random tree represents the open-loop trajectory starting from the initial state.

[0004] As an efficient data structure and algorithm, RRT has strong probabilistic completeness. However, the blind randomness during node expansion leads to a series of problems. For example, the state mutation between path nodes is relatively large, the acquisition of random sampling points does not consider the distance relationship with obstacles and the target point, and the efficiency during node expansion is low. Therefore, it is not easy to obtain a safe and reliable planning result and the algorithm is unstable. Each search takes a long time, and the planned path results may be very different. Summary of the Invention

[0005] In view of this, aiming at the problem of blind search in the existing RRT algorithm for path planning, the present invention provides an improved RRT algorithm that determines the direction based on the roulette wheel selection method combined with the probability distribution under the concept of virtual gravitational field in the artificial potential field. The present invention can effectively improve the large amount of redundant exploration in the random expansion tree expansion of the RRT algorithm, effectively shorten the length of the planned path, and avoid the problem that the target point is unreachable due to the conflict between the virtual resultant force and obstacles under the traditional artificial potential field algorithm.

[0006] The technical solution of the present invention is a path planning method based on the improvement of the RRT algorithm by integrating the artificial potential field method, and the specific steps are as follows:

[0007] Step 1: First, it is necessary to obtain environmental information to construct a geometric map. Define N as the state space, which consists of the obstacle region N obs and the free space N free and there is N free = N\N obs , where N free = N\N obs means that the free space is the part of the state space N excluding the obstacle region N obs . That is, N free is the search area for the feasible path. In Matlab, convert the obstacle image into a grayscale image, and take the edge information and the target point therein as the main bodies affecting the path planning. At the same time, store the edge information of each obstacle in a cell array in the form of an n-row and 2-column array. In order to express the schematic effect as clearly as possible, further extract one sample every 20 elements from the array. The set of cell arrays with less information obtained in this way is used as N obs . Starting from the initial point X init , explore the space by constructing a random expansion tree, and the representation of the tree is shown in formula (1):

[0008] G = (N, E) (1)

[0009] In the formula, N is the set of nodes, E is the connection relationship between nodes, and the set of edges.

[0010] Step 2: Take the initial node n init(or the current point) is included in the path Path, referring to the concepts of the repulsive field function and the attractive field function in the Artificial Potential Field (APF) algorithm, and designing an artificial potential field-guided probability distribution model for this point. All sampled elements in the obstacle information Edge and the target point n goal For n near (n near represents the node in the tree closest to the random point) have independent influences, which are respectively denoted as U rep and U att . First, extract the edge information of independent obstacles in the environmental information, sample every 20 points as a group and number them, and store them in a cell array Edge. Among them, n m,n ∈N obs (n m,n is the nth sampled point in the obstacle numbered m). According to the position coordinates of the current point and the target point coordinates n goal , the obstacle area N obs , design a probability distribution diagram based on the direction and distance relationship. Finally, determine the expansion direction of the child node according to this distribution diagram using the roulette wheel selection method.

[0011] Next, further describe U att [[ID=2G]]and U rep as well as the construction of the probability distribution diagram. In terms of U att , generate a random node n rand ∈N free in the state space N, and determine the node n rand closest to n near in the expansion tree through the shortest distance principle. The distance d rand from each node in the tree to the sampled point n i is represented by the Euclidean distance, as shown in formula (2):

[0012]

[0013] In the formula, dis represents the Euclidean distance (unit: meter), (x i , y i ) are the coordinates of the node nb i in the tree, i = 1, 2,..., k, k is the total number of current nodes, not greater than the maximum iteration T, (x rand , y rand ) are the coordinates of the random node n rand ;

[0014] Taking the direction dir and distance dis of the current point relative to the target point as influencing factors, the parameters μ and σ in the normal distribution are adjusted, and this part provides the probability of approaching the target point. Similarly, in U rep Regarding this, the probability of the target point colliding with an obstacle is reduced.

[0015] Step 3: In the case of expanding a fixed step length S outward in a certain direction to determine a new node n near Regarding the problem of, according to the additivity of the normal distribution. Taking the normal distribution as the basic element, construct n new The probability density functions of the expansion direction are respectively expressed as: near

[0016]

[0017] U = α * U att + β * U rep (5)

[0018] Among them, U att represents the probability density function of the normal distribution regarding the direction influence constructed by n goal for n near The probability density function of the normal distribution regarding the direction influence constructed by all obstacles accumulated for n rep represents the probability density function of the normal distribution regarding the direction influence constructed by all obstacles accumulated for n near The probability density function of the total normal distribution of the expansion direction of the child node is represented by U near α represents the weighting value for U att β represents the weighting value for U rep Among them, the ratio of α to β is dynamically adjusted according to the step length S: α / β = K·S, where K is a proportionality constant, set to 0.5 according to experiments. When S is small, increase α to strengthen the gravitational force. When S is large, increase β to enhance obstacle avoidance. Because the setting of the step length S means the fineness of the path planning. The smaller the step length, the easier it is for the artificial potential field to fail in obstacles in the shape of a U, and the search range needs to be expanded. Conversely, the larger the step length, the smaller the possibility of this problem occurring. M represents the number of obstacles in the map, N represents the number of sampling points for each obstacle every 20 samples, and σ m,n represents the σ value of the normal distribution constructed by the nth sampling point on the mth obstacle acting on n near μ m,n represents the μ value of the normal distribution constructed by the nth sampling point on the mth obstacle acting on n near

[0019] For the values of σ and μ in U att Regarding the method based on the virtual gravitational force in the artificial potential field method, it is expressed as:

[0020] ​​

[0021] μ = dir(n near , n goal ) (7)

[0022] Among them, ζ is the gravitational field adjustment coefficient, and its value range is 0.1 to 1.0. The value range of ζ (0.1 to 1.0) is determined through simulation experiments. When ζ = 0.5, the path convergence speed and obstacle avoidance effect are optimal. dis(n near , n goal ) represents the distance between n near and n goal , and dir(n near , n goal ) represents the direction of n goal with respect to n near .

[0023] According to the obtained U, increase the probability that the next new node n new approaches the target point n goal and does not touch the obstacle area N obs until a target point n new exists in the neighborhood of a certain new node n goal , and store n goal into the path array path.

[0024] Step 4: In the above steps, a feasible path path from the starting node n init to the target point n goal is determined. However, there are still some unnecessary nodes in the obtained path result, resulting in a large number of turns and a long distance.

[0025] Here, on the basis of the obtained path solution, a greedy search strategy is introduced. Starting from n goal as the starting direction, find n init through the greedy search strategy. Starting from n goal , with the goal of the shortest connected distance between nodes and the constraint of not touching obstacles during connection. After each child node is determined, directly connect this node to n init , and determine whether the connection line collides with the obstacle. Determine redundant nodes through the triangle inequality. If the following conditions are met, the intermediate node is removed, as shown in formula (8):

[0026] dis(path i , path i+2 ) ≤ dis(path i , path i+1 ) + dis(path i+1 , path i+2 ) (8)

[0027] In the formula, always pathi Represents the i-th node in the path array. i+2 is not greater than the length M of the path array.

[0028] If no collision occurs, it means that the init , n init The nodes and their connections are added to the random tree, and the expansion ends; otherwise, the next expansion cycle is entered until all unnecessary nodes in the path array are eliminated.

[0029] Based on the above steps, the RRT algorithm is implemented for path planning: (1) The entire space is initialized by the set map of the environment map constructed in step 1, and the starting point, end point, step length between points, target point threshold, virtual gravity weight α and virtual repulsion weight β are defined to obtain the coordinate information of the obstacle edge stored in the grayscale image in the cell array; (2) According to the position information of the current point of the AGV, the target point information, and the obstacle information, a probability distribution map is constructed. First, the direction dir and distance dis of the current point and the obstacle information Edge are used as the parameters affecting μ and σ in the normal distribution and the virtual repulsion weight β, and the probability distribution of the obstacle edge in the obstacle area N is obtained. obs The probability density function U of the normal distribution under the influence rep This section provides the deviation from the obstacle area N obs The probability density function of . Similarly, through the target point n goal The direction dir, distance dis and virtual gravity weight α of the current point position information are used to obtain the virtual gravity U from step 4. att , which provides the guidance target point n goal The probability density function of . This reduces the probability of collision between the target point and the obstacle; (3) Randomly generate random nodes n in space rand ; (4) Find the distance n from a random point in the node set of the known random expansion tree rand The nearest node n near , 4. From n near to n rand In the straight line direction from n near Intercept point n with step size S new ; (5) Determine n near and n new Is there an obstacle between them? If there is an obstacle, discard the point; (6) new Add to the tree set; (7) loop (2) (6), the loop ends when: there is a new node n new There is a target point n in the set neighborhood goal Then stop the growth of random expansion tree.

[0030] The technical effects and advantages of the present invention are as follows:

[0031] (1) The present invention guides the node expansion direction of RRT through an artificial potential field, significantly reducing the generation of invalid nodes and reducing redundant exploration.

[0032] (2) Compared with the traditional RRT algorithm, the path generated by the present invention is shorter and smoother, optimizing the path length.

[0033] (3) The present invention can dynamically adjust the gravitational and repulsive force parameters, solves the problem of unreachable targets in the artificial potential field method, and avoids local minima.

[0034] (4) The present invention can still quickly plan high-quality paths in multi-obstacle and complex environments, with high efficiency and adaptability. Description of the Drawings

[0035] Figure 1 It is a flowchart of the path planning algorithm of the improved RRT based on the artificial potential field of the present invention;

[0036] Figure 2 It is the influence of virtual gravity on n in the simulation example of the present invention; goal influence;

[0037] Figure 3 It is the influence of virtual repulsive force on n in the simulation example of the present invention; goal influence;

[0038] Figure 4 It is the image of the normal distribution probability density function U under the influence of α and β weights in the simulation example of the present invention;

[0039] Figure 5 It is a comparison chart between the original RRT and the improved RRT of the present invention;

[0040] Figure 6 It is a comparison chart of the algorithm paths of the simulation example of the present invention under different algorithms and different static maps;

[0041] Figure 7 It is a comparison chart of the scale of the random expansion tree under the static map of the simulation example of the present invention;

[0042] Figure 8 It is a comparison chart of the scale of the discarded nodes under the static map of the simulation example of the present invention;

[0043] Figure 9 It is a comparison chart of the algorithm for the planned path length under the static map of the simulation example of the present invention;

[0044] Figure 10 The present invention introduces a greedy search strategy based on the improved RRT algorithm;

[0045] Figure 11 It is a comparison chart of the algorithm for the planned path length under the narrow channel map of the simulation example of the present invention. Detailed Embodiments

[0046] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] As Figure 1 shown: A path planning method improved by fusing the artificial potential field method based on the RRT algorithm of the present invention, and the algorithm flow includes four stages: 1. Environment modeling (corresponding to step one): Extract the obstacle edges through grayscale image processing to generate the state space N obs and N free ; 2. Probability distribution generation (corresponding to step two): The probability density distributions of the gravitational field and the repulsive field; 3. Node expansion (corresponding to step three): Improve the node expansion process of RRT; 4. Path optimization (corresponding to step four): Compare the paths optimized by the greedy strategy.

[0048] Specifically, it is implemented through the following steps:

[0049] Step one: First, it is necessary to obtain environmental information to construct a geometric map. Define N as the state space, which consists of the obstacle region N obs and the free space N free , and there is N free =N\N obs , where N free =N\N obs means that the free space is the part of the state space N excluding the obstacle region N obs , that is, N free is the search area for the feasible path. Convert the obstacle image to a grayscale image in Matlab, and take the edge information and the target point therein as the main body affecting the path planning. At the same time, store the edge information of each obstacle in a cell array in the form of an n-row 2-column array. In order to express the schematic effect as clearly as possible, further extract one sample for every 20 elements of the array, and use the set of cell arrays with simpler information obtained in this way as N obs . Starting from the initial point X init , construct a form of a random tree to explore the space, and the representation of the tree is shown in formula (1):

[0050] G=(N,E) (1)

[0051] In the formula, N is the set of nodes, E is the connection relationship between nodes, and the set of edges.

[0052] Step two: Use the starting position coordinates as the initial node ninit , as the first node of the path array path, and randomly generate a random node n within the state space N rand ∈N free ;

[0053] Include the initial node n init (or the current point) into the path Path, and design a probability distribution model guided by the artificial potential field for this point. Refer to the repulsive field function U rep (RepulsiveField) and the attractive field function U att (AttractiveField) in the Artificial Potential Field (APF) algorithm.

[0054] The construction of the attractive field function U att and the repulsive field function U rep is based on the following mathematical principles:

[0055] 1. The construction of the attractive field function U att :

[0056] U att is driven by the direction and distance from the target point n goal to the current node n near , and its expression is:

[0057]

[0058] where μ goal is the target point direction angle, and σ goal is a function of the distance dir(n near , n goal ), and the specific relationship is:

[0059]

[0060] Here, ζ is the attractive field adjustment coefficient, and its value range is 0.1 - 1.0.

[0061] 2. The construction of the repulsive field function U rep :

[0062] U rep is composed of the superposition of obstacle sampling points, and the expression is:

[0063]

[0064] where μ m,n and σ m,n are the influence parameters of the nth sampling point of the mth obstacle on the current node respectively.

[0065] All sampling elements and the target point n in the obstacle information Edge goal For n near the influence is independent, denoted as U rep and U att respectively. As shown by Figure 2 and Figure 3 , the simulation effect can be seen. First, extract the edge information of independent obstacles in the environmental information, sample every 20 points as a group and number them, and store them in a cell array Edge. Among them, n m,n ∈N obs (n m,n is the nth sampling point in the obstacle numbered m). According to the position coordinates of the current point and the target point coordinates n goal , the direction and distance relationship with the obstacle area N obs , a probability distribution map is designed. Finally, according to this distribution map, the roulette wheel selection method is used to determine the next direction at the current point.

[0066] Next, a further description of the construction of U att and U rep and the probability distribution map is given:

[0067] In terms of U att , the node n rand nearest to n near in the tree is determined by the shortest distance principle. The distance d rand from each node in the tree to the sampling point n i is represented by the Euclidean distance, as shown in formula (2):

[0068]

[0069] In the formula, dis represents the Euclidean distance (unit: meter), (x i , y i ) are the coordinates of the node n i in the tree, i = 1, 2,..., k, k is the total number of current nodes, not greater than the maximum iteration T, (x rand , y rand ) are the coordinates of the random node n rand ;

[0070] Taking the direction dir and distance dis of the current point relative to the target point as influencing factors, adjust the parameters of μ and σ in the normal distribution. This part provides the probability of approaching the target point. Similarly, in terms of U rep , the probability of the target point colliding with the obstacle is reduced.

[0071] Step 3: Expand a fixed step length S in a certain direction from n near to determine a new node n newRegarding the problem, according to the superposition property of the normal distribution. Taking the normal distribution as the basic element, n is constructed near The probability density functions for the expansion directions are respectively expressed as:

[0072]

[0073]

[0074] U = α * U att + β * U rep (5)

[0075] Among them, U att represents the probability density function of the normal distribution for the direction influence constructed by n goal for n near constructed for the direction influence, U rep represents the probability density function of the normal distribution for the direction influence constructed by the accumulation of all obstacles on n near constructed for the direction influence, U represents n near the total probability density function of the normal distribution for the expansion direction of the child node, as shown by Figure 4 : α represents the weighting value for U att and β represents the weighting value for U rep .

[0076] The ratio of α to β is dynamically adjusted according to the step size S. The specific rule is α / β = K·S (K is a proportionality constant, set to 0.5 according to experiments). The experimental basis for this rule is as follows:

[0077] In the simulation experiment, when the step size S is small (e.g., S = 0.1m), α is increased to strengthen the gravitational field effect and avoid the path falling into local extrema due to too small a step size; when the step size S is large (e.g., S = 1.0m), β is increased to enhance the repulsive field effect and ensure obstacle avoidance safety. Through 50 comparison experiments with different maps, the path length and collision rate are comprehensively optimal when K = 0.5 (see Table 1 for details).

[0078] When S is small, α is increased to strengthen the gravity, and when S is large, β is increased to enhance obstacle avoidance. M represents the number of obstacles in the map, N represents the number of sampling points for each obstacle every 20 samples, and σ m,n represents the σ value of the normal distribution constructed by the nth sampling point on the mth obstacle for the action on n near . μ m,n represents the μ value of the normal distribution constructed by the nth sampling point on the mth obstacle for the action on n near .

[0079] For the values of σ and μ in U att based on the method of taking values for the virtual gravity in the artificial potential field method, it is expressed as:

[0080]

[0081] μ = dir(n near , n goal ) (7)

[0082] Where ζ is the gravitational field adjustment coefficient, and its value range is 0.1 to 1.0. The value range of ζ (0.1 to 1.0) is determined by simulation experiments. When ζ = 0.5, the path convergence speed and obstacle avoidance effect are optimal. dis(n near , n goal ) represents the distance between n near and n goal , and dir(n near , n goal ) represents the direction of n goal with respect to n near .

[0083] According to the obtained U, the probability that the next new node n new approaches the target point n goal and does not touch the obstacle area N obs is increased until a target point n new exists in the neighborhood of a certain new node n goal . Then n goal is stored in the path array path. The comparison of the paths obtained by the original RRT algorithm and the improved RRT algorithm is as Figure 5 shown: The improved RRT algorithm (solid line) has a shorter and smoother path compared to the original RRT (dashed line).

[0084] Considering that the RRT algorithm is a random algorithm, in order to verify that the experimental results have general adaptability, 50 trials are carried out under three maps respectively. The comparison of the algorithm paths under different static maps is as Figure 6 shown: Among them, taking Map3 as an example, the specific information of the scale of the randomly expanded tree nodes, the number of discarded nodes, and the length of the finally obtained path is observed as Figure 7 , Figure 8 , Figure 9 shown: The obtained data records are shown in Table 1 below:

[0085] Table 1 Comparison of algorithm experimental results under different environmental maps (comparison between improved RRT and original RRT)

[0086]

[0087] Note: The path length shortening rate is calculated based on the original RRT.

[0088] It can be seen that under two different metrics, the improved RRT is slightly superior and relatively stable to the traditional RRT algorithm in terms of the scale of the expansion of the random expansion tree. After integrating the greedy search algorithm, a significant improvement can be seen in the finally obtained path. Different from the traditional RRT algorithm that obtains the path planning result by spreading nodes throughout the map, the main strategy of the improved RRT algorithm is to make the randomly growing node tree extend towards the target and avoid obstacles purposefully by combining the concept of virtual gravity in the artificial potential field, thereby reducing a large number of unnecessary nodes. It can be clearly found from Table 1 that the number of nodes in the random expansion tree is significantly reduced, indicating that the combined action of the virtual gravity field and the virtual repulsive force field effectively avoids a large number of ineffective searches. Finally, the greedy strategy is combined to filter out redundant turning nodes between necessary turning nodes, further optimizing the effect. The path planning effect in narrow channels is as Figure 11 shown.

[0089] Step 4: In the above steps, a feasible path path from the starting node n init to the target point n goal is determined. However, there are still some unnecessary nodes in the obtained path result, resulting in a large number of turning angles and redundant paths in the journey.

[0090] Here, based on the obtained path solution, a greedy search strategy is introduced. Starting from n goal , n init is found through the greedy search strategy. Starting from n goal , with the goal of the shortest connected distance between nodes and the constraint of not touching obstacles during connection. After each child node is determined, directly connect this node to n init , and judge whether the connection line collides with obstacles. Redundant nodes are determined through the triangle inequality. If the following conditions are met, the intermediate nodes are removed, as shown in formula (8):

[0091] dis(path i ,path i+2 )≤dis(path i ,path i+1 )+dis(path i+1 ,path i+2 ) (8)

[0092] In the formula, path i always represents the i-th node in the path array path. i + 2 is not greater than the length M of the path array path.

[0093] If there is no collision, it means that n init has been reached, and n initand its connection are added to the random tree, and the expansion ends; otherwise, the next expansion cycle is entered. Until all unnecessary nodes in the path array path are eliminated, the path with redundant nodes is removed. Figure 10 shown.

[0094] The specific implementation algorithm of step four is summarized as follows:

[0095] 1. Take the target point n goal As the starting point, traverse the path array path in reverse;

[0096] 2. From the current node path i Try connecting directly to path i+2 , if the following conditions are met: dis(path i ,path i+2 )<dis(path i ,path i+1 )+dis(path i+1 ,path i+2 ) and the path path i →path i+2 If there is no collision with obstacles, the intermediate node path will be eliminated. i+1 ;

[0097] 3. Repeat step 2 until no further optimization is possible. Figure 10 shown.

[0098] Path planning is implemented based on the above steps: (1) The entire space is initialized by the collective map of the environment map constructed in step 1, and the starting point, end point, step length between points, target point threshold, virtual gravity weight α and virtual repulsion weight β are defined to obtain the coordinate information of the obstacle edge stored in the grayscale image in the cell array; (2) According to the position information of the current point of the AGV, the target point information, and the obstacle information, a probability distribution map is constructed. First, the direction dir and distance dis of the current point and the obstacle information Edge are used as the parameters affecting μ and σ in the normal distribution and the virtual repulsion weight β, and the probability distribution of the obstacle edge in the obstacle area N is obtained. obs The probability density function U of the normal distribution under the influence. This part provides the deviation from the obstacle area N obs The probability density function of . Similarly, through the target point n goal The direction dir, distance dis and virtual gravity weight α of the current point position information are used to obtain the virtual gravity U from step 4. att , which provides the guidance target point n goal The probability density function of . This reduces the probability of the target point colliding with the obstacle; (3) randomly generate a point n in space rand ; (4) Find the point n closest to this random point in the set of points in the known treenear , 4. In the linear direction from n near to n rand , intercept points n with a step size S near ; (5) Determine whether there is an obstacle between n new and n near . If there is an obstacle, discard this point; (6) Add n new to the set of the tree; (7) Loop (2)-(6). The loop termination condition: There is a new node n new and there is a target point n within the set neighborhood of the end point new , then stop the growth of the randomly expanding tree. goal

[0099] The present invention guides the node expansion direction of RRT through the artificial potential field, significantly reducing the generation of invalid nodes and reducing redundant exploration; compared with the traditional RRT algorithm, the generated path is shorter and smoother, and path length optimization is carried out; the present invention can dynamically adjust the gravitational and repulsive force parameters, solves the problem of target unreachability in the artificial potential field method, and avoids local extrema; it can still quickly plan high-quality paths in a multi-obstacle and complex environment, with high efficiency and adaptability.

[0100] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A path planning method improved by fusing the artificial potential field method based on the RRT algorithm, characterized in that, The method includes the following steps: Step 1: Construct the state space. Define N as the set containing the obstacle region N obs and the free space N free . Among them, N free = N \ N obs , extract the edge information of the obstacle and discretize it into sampling points; Step 2: Establish a dual-field collaborative model of the gravitational field and the repulsive field. The gravitational field is generated by the target point n goal and the repulsive field is generated by superimposing obstacle sampling points. The path exploration and obstacle avoidance are balanced by the dynamic weight ratio α / β, where the ratio of α and β is adaptively adjusted according to the step size S; Step 3: Generate a probability density function U based on the roulette wheel selection method, determine the expansion direction, and generate a new node n new , where the probability density function is composed of the superposition of the gravitational field and the repulsive field: U = α * U att + β * U rep Among them, U att and U rep are the normal distribution probability density functions of the target point and the obstacle for the current node, respectively; Step 4: Optimize the initial path using a greedy search strategy, determine and remove redundant nodes through the triangle inequality until the path satisfies the shortest distance and collision-free constraints.

2. The path planning method improved by fusing the artificial potential field method based on the RRT algorithm according to claim 1, wherein In the first step, the obstacle edge information is stored as a cell array in the form of an n×2 array, and the data volume is simplified by sampling every 20 elements; the mathematical representation of the random expansion tree G is: G=(N,E) where N is the set of nodes and E is the set of connecting edges between nodes.

3. A path planning method improved by integrating the artificial potential field method based on the RRT algorithm according to claim 1, characterized in that, In the second step: Gravitational field probability density function U att The construction formula is as follows: where μ goal is the direction angle of the target point, and σ goal is an adjustment parameter related to the distance between the current node and the target point; Repulsive force field probability density function U rep The construction formula is as follows: where μ m,n and σ m,n are the influence parameters of the n-th sampling point of the m-th obstacle; The dynamic weight ratio adjustment rule is: α / β = K·S where K is a proportionality constant, α is increased when the step size S is small, and β is increased when the step size S is large.

4. A path planning method improved by integrating the artificial potential field method based on the RRT algorithm according to claim 1, characterized in that The determination condition for redundant nodes in the fourth step is: dis(path i ,path i+2 ) ≤ dis(path i ,path i+1 ) + dis(path i+1 ,path i+2 ) where dis is the Euclidean distance and path i is the i-th node in the path array.

Images