A robot rapid path optimization method based on multi-strategy fusion

Abstract

The application relates to the field of mobile robot path planning, and particularly relates to a robot rapid path optimization method based on multi-strategy fusion, which comprises the following steps: step 1, initializing iteration greedy algorithm parameters, and generating an initial path by using a bidirectional rapid expansion random tree method; step 2, using a difference guided path optimization mechanism to destruct and reconstruct the path, and expanding a path exploration space; step 3, performing path pruning, node insertion and node mutation operations by selecting an adaptive adjustment strategy, and locally searching the current path; step 4, accepting the path according to a simulated annealing criterion; step 5, judging whether the algorithm termination time is reached; if the algorithm termination time is reached, the algorithm ends, and the current best path is output; otherwise, returning to step 2. The application effectively solves the problems of minimizing path length and path smoothness in robot path planning, and has the positive effects of improving path planning efficiency and optimizing path quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of mobile robot path planning, in particular to a robot fast path optimization method based on multi-strategy fusion. BACKGROUND

[0002] The research of robots involves the comprehensive cross-fusion of multiple modern disciplines, such as multi-sensor fusion and control, microelectronics, wireless communication, automation, etc. With the rapid development of 5G and the Internet of Things, robots are becoming an important foundation for emerging technologies. In recent years, significant progress has been made in robot technology at home and abroad, driving rapid development in the field. As research continues to deepen, the application field scenarios are also expanding rapidly, such as manufacturing, logistics transportation and distribution, military, agricultural production, environmental monitoring, and various aspects of social production and life.

[0003] Path planning technology is one of the core technologies of mobile robot research. Path planning refers to a safe, obstacle-free path planned by a mobile robot based on environmental perception information and using a certain specific algorithm. Excellent path planning technology not only significantly saves a lot of time, but also reduces the working time and cost of the mobile robot. As the complexity of the robot working scene increases, the performance requirements for mobile robots are also increasing, not only to meet some simple task completion, but also to pursue more efficient and safer problem solving for mobile robots. Improving the performance of path planning can significantly improve the efficiency of the robot, including reducing the path planning time, shortening the path length, and ensuring safe arrival at the destination. At the same time, optimizing path smoothness is also an important factor in improving the motion efficiency and stability of the robot.

[0004] Traditional path planning algorithms may face high computational complexity, slow convergence speed, and easy to fall into local optimum in complex environments, therefore, it is essential to study a new mobile robot path planning method to achieve efficient path planning and improve the navigation performance of the robot in complex environments. SUMMARY

[0005] The purpose of the present application is to provide a robot fast path optimization method based on multi-strategy fusion, which solves the problem of minimizing path length and path smoothness in the path planning process, to achieve the purpose of improving path planning efficiency and optimizing path quality.

[0006] The robot fast path optimization method based on multi-strategy fusion of the present application is characterized by the following steps,

[0007] Step 1, initialize the iteration greedy algorithm parameters, generate an initial path using the bidirectional rapid expansion random tree method, wherein the initial path is used as the initial current best path;

[0008] Step 2: Use a difference-guided path optimization mechanism to disrupt and reconstruct the path, expanding the path exploration space; where the path of the first disruption and reconstruction is the initial path, and the paths of the second and subsequent disruption and reconstructions are the paths accepted after the previous iteration.

[0009] Step 3: Select and execute path pruning, node insertion, and node mutation operations through adaptive adjustment strategies to perform a local search on the current path;

[0010] Step 4: Accept the path according to the simulated annealing criteria;

[0011] Step 5: Determine if the algorithm's termination time has been reached. If the termination time has been reached, the algorithm ends and the current best path is output; otherwise, return to step 2.

[0012] Furthermore, in step 1, the path π is composed of a series of discrete nodes, represented as π = [P1 ... P...]. i ...P n ], P1 represents the robot's starting node, P n P represents the target node. i Let be an intermediate node in the path, and let be the index of the intermediate node. The value of is within the range of ... , where n represents the number of all nodes in the path.

[0013] Furthermore, in step 2, the implementation process of the differentially guided path optimization mechanism is as follows:

[0014] The difference ε between the current path and the ideal path is calculated. The ideal path refers to the shortest path from the starting node to the target node under obstacle-free conditions. If ε is less than a preset lower threshold, it means the path is close to the ideal path, and the path remains unchanged. If ε is greater than or equal to a preset upper threshold, a new path is generated using a mirroring strategy. If ε is between the preset lower and upper thresholds, a new path is generated using a projection strategy. If the generated new path is infeasible, a path repair strategy is executed to make it feasible. The target value of the new path is calculated. If the target value of the new path is less than the target value of the original path, the new path is accepted directly. Otherwise, the difference probability between the new path and the original path is calculated. When the difference probability value is greater than a preset probability threshold, the new path is accepted, and the search space is expanded. Infeasibility refers to the existence of nodes within obstacles or intersections between the path and obstacles. The difference probability... The calculation formula is f(π') is the target value of the new path π', f(π) is the target value of the original path π, and d(π', π) represents the difference between the new path π' and the original path π.

[0015] Further, the mirror strategy is that the distance of each node in the calculation path from the ideal path is calculated, the node whose distance from the ideal path is greater than the preset minimum distance is removed from the path and stored in the node set, each node in the node set is mirror flipped, a new node is obtained by reversing the node coordinates, the new node is inserted into the corresponding position of the original node in the path before mirroring, and a new path is obtained;

[0016] The projection strategy is that z nodes are randomly selected from the path, the randomly selected z nodes are deleted from the path and stored in the node set, each node in the node set is mapped, a new node is generated by mapping the node coordinates, the new node is inserted into the corresponding position of the original node in the path before mapping, and a new path is obtained;

[0017] The path repair strategy is that the nodes inside the obstacle in the path are disturbed, the disturbance method is to generate a new node randomly within a fixed search radius range with the node inside the obstacle as the center, the new node is ensured to be located in the free area, the new node replaces the node inside the obstacle, and the repaired path does not intersect with the obstacle; for the path intersecting with the obstacle, the two end points of the path intersecting with the obstacle are selected as new starting nodes and target nodes, and a bidirectional rapid expansion random tree method is used for searching, and each time the new path is generated according to the preset search step.

[0018] Further, in step 3, the adaptive adjustment strategy is that the scoring mechanism is combined with the roulette wheel selection method to dynamically adjust the weights of the path pruning operation, the node insertion operation and the node mutation operation, the initial weights of the path pruning operation, the node insertion operation and the node mutation operation are set to 1, the initial score is set to 0, and the initial selection probabilities of the operations are the same, an operation is selected using the roulette wheel selection method according to the weights of the operations, a new path is generated by executing the selected operation on the path, the new path is scored according to the target value of the new path and the original path, the original path of the first operation is the path obtained after the destruction and reconstruction in step 2, the original path of the second and subsequent operations is the path after the dynamic adjustment of the last time, and the update formula of the score c is , , , The score parameter is denoted by c and satisfies > > , f is the target value, π' represents the new path, π represents the original path, π best is the current best path, after scoring, the weights of the operations are dynamically adjusted according to the scores of the operations and the calling times of the operations, and the update formula of the weight τ is λ∈(0,1) represents the weight update coefficient, τ is the current weight of each operation, c represents the score of each operation, and ψ is the number of times each operation is called. Based on the updated weights of each operation, the roulette wheel selection method is used to select operations. After the selected operation is executed, a new path is generated. If the target value of the new path is smaller than the target value of the original path, the new path replaces the original path. Otherwise, the original path remains unchanged. The above dynamic adjustment process is repeated until the maximum number of iterations preset by the adaptive adjustment strategy is reached.

[0019] The path pruning operation is to randomly select a node P from the current path. i Check from P i To other nodes P in the current path j Does the straight segment intersect the obstacle? If from P i To P j The straight segment does not intersect the obstacle, and P i With P j If the nodes between nodes are redundant, meaning they do not affect the connectivity of the path and do not change the shortest path, then delete P. i With P j All redundant intermediate nodes between P i With P j Connect directly to form an optimized new path; otherwise, leave the current path unchanged.

[0020] The node insertion operation is to randomly select a node P from the current path. i compute node P i With the previous node P i-1 Direction vectors between and node P i With the next node P i+1 Direction vectors between Node P i The newly generated node P is moved along prevdir and nextdir by a fixed step size, respectively. new1 and P new2 Determine from P new1 To P new2 Does the straight segment intersect the obstacle? If from P new1 To P new2 The straight segment does not intersect the obstacle, so P new1 Insert into P i Before that, P new2 Insert into P i After that, delete P. i Generate a new path; otherwise, keep the current path unchanged.

[0021] The node mutation operation is to randomly select a point in the current path as a mutation node P i , generate a random angle θ, calculate a direction vector dv=[cosθ, sinθ] according to θ, generate a random step m, and generate a new node P i around P new , respectively judge whether the straight line segment from P i-1 to P new and the straight line segment from P new to P i+1 intersect with the obstacle, if the straight line segment from P i-1 to P new and the straight line segment from P new to P i+1 do not intersect with the obstacle, replace P new with P i , generate a new path, otherwise keep the current path unchanged.

[0022] Further, in step 4, the simulated annealing criterion is that if the target value of the current path is smaller than the target value of the path before the destruction and reconstruction, the current path is accepted, if the target value of the current path is smaller than the target value of the current best path, the current path is replaced by the current best path, if the target value of the current path is greater than the target value of the path before the destruction and reconstruction, the current path is accepted with a probability r=exp{-(f(π c )-f(π0)) / T}, wherein exp is the natural exponential function, f(π c ) is the target value of the current path π c , f(π0) is the target value of the path π0 before the destruction and reconstruction, and T represents the current temperature.

[0023] Further, in step 5, the algorithm termination time is 5s.

[0024] The application provides a robot rapid path optimization method based on multi-strategy fusion, according to the problem characteristics, the application comprehensively considers the environmental factors and the path optimization demand, optimizes a path without collision with obstacles, and effectively reduces the length of the path and the number of turning points. By optimizing the path structure, unnecessary zigzag is reduced, so that the path smoothness of the mobile robot is improved. The shortening of the path also enables the robot to more efficiently perform tasks. The application generates an initial path by using an efficient bidirectional rapid expansion random tree method, explores more potential areas by combining a difference guided path optimization mechanism, and uses an adaptive adjustment strategy to improve the search ability. Compared with the traditional path planning method, the application can not only significantly improve the path planning efficiency, but also effectively avoid the algorithm from falling into local optimum, has stronger search ability, fast convergence speed, and can more effectively find the shortest path. In summary, the application effectively solves the problem of mobile robot path planning, and has a positive effect of improving the path planning efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is the implementation flowchart of the application;

[0026] Figure 2 is the indoor scene used in the test of the application Figure 1 ;

[0027] Figure 3 is the outdoor scene used in the test of the application Figure 2 ;

[0028] Figure 4 is the comparison chart of the path length index of the application and the comparison algorithm in the test Figure 1 ;

[0029] Figure 5 is the comparison chart of the path length index of the application and the comparison algorithm in the test Figure 2 ;

[0030] Figure 6 is the comparison chart of the path smoothness index of the application and the comparison algorithm in the test Figure 1 ;

[0031] Figure 7 is the comparison chart of the path smoothness index of the application and the comparison algorithm in the test Figure 2 ;

[0032] Figure 8 is the running path display chart of the application in the test Figure 1 ;

[0033] Figure 9 is the running path display chart of the comparison algorithm F-RRT* in the test Figure 1 ;

[0034] Figure 10 The running path display diagram of the comparative algorithm NMABC in the ground Figure 1 ;

[0035] Figure 11 The running path display diagram of the comparative algorithm VNIG in the ground Figure 1 ;

[0036] Figure 12 The running path display diagram of the comparative algorithm IGDLM in the ground Figure 1 ;

[0037] Figure 13 The running path display diagram of the present application in the ground Figure 2 ;

[0038] Figure 14 The running path display diagram of the comparative algorithm F-RRT* in the ground Figure 2 ;

[0039] Figure 15 The running path display diagram of the comparative algorithm NMABC in the ground Figure 2 ;

[0040] Figure 16 The running path display diagram of the comparative algorithm VNIG in the ground Figure 2 ;

[0041] Figure 17 The running path display diagram of the comparative algorithm IGDLM in the ground Figure 2 ;

[0042] Figure 18 The convergence curve diagram of the present application and the comparative algorithm in the path length index in the ground Figure 1 ;

[0043] Figure 19 The convergence curve diagram of the present application and the comparative algorithm in the path length index in the ground Figure 2 ;

[0044] Figure 20 The parameter factor level trend diagram of the preset minimum distance in the mirror strategy of the present application, the preset search step in the path repair strategy, the preset threshold lower limit and threshold upper limit in the difference guided path optimization mechanism. DETAILED DESCRIPTION

[0045] As Figure 1 shown, the robot rapid path optimization method based on multi-strategy fusion provided by the present application is mainly realized through the following steps.

[0046] Step 1: Initialize the parameters of the iterative greedy algorithm and generate the initial path using the bidirectional fast expanding random tree method. The initial path serves as the initial current best path.

[0047] Specifically, the path π consists of a series of discrete nodes, represented as π = [P1 ... P2]. i ... P n ], P1 represents the robot's starting node, P n P represents the target node. i Let be an intermediate node in the path, and let be the index of the intermediate node. The value of is within the range of ... Here, n represents the number of nodes in the path. The bidirectional fast expanding random tree method expands progressively from the starting node in a tree structure. It randomly samples points in space, finds the nearest node in the tree, and then expands one step in the direction from the nearest node to the sampled point, finding a new node. If the path has no collisions, the new node is added to the tree. The algorithm repeats sampling and expansion until a path connecting the target node is found or the maximum number of iterations preset by the bidirectional fast expanding random tree method is reached, thus constructing the desired path.

[0048] Step 2: Use a difference-guided path optimization mechanism to disrupt and reconstruct the path, expanding the path exploration space; where the path of the first disruption and reconstruction is the initial path, and the paths of the second and subsequent disruption and reconstruction are the paths accepted after the previous iteration.

[0049] Specifically, the implementation process of the difference-guided path optimization mechanism is as follows: Calculate the difference ε between the current path and the ideal path, where the ideal path refers to the shortest path from the starting node to the target node under obstacle-free conditions; if ε is less than a preset lower threshold, it indicates that the path is close to the ideal path, and the path remains unchanged; if ε is greater than or equal to a preset upper threshold, a new path is generated through a mirroring strategy; if ε is between the preset lower and upper thresholds, a new path is generated through a projection strategy; if the generated new path is infeasible, a path repair strategy is executed to repair the new path to feasibility, and the target value of the new path is calculated. If the target value of the new path is less than the target value of the original path, the new path is directly accepted; otherwise, the difference probability between the new path and the original path is calculated. When the difference probability value is greater than a preset probability threshold, the new path is accepted, and the search space is expanded. Here, infeasibility refers to the presence of nodes within obstacles or intersections between the path and obstacles; the difference probability... The calculation formula is f(π') is the target value of the new path π', f(π) is the target value of the original path π, and d(π', π) represents the difference between the new path π' and the original path π.

[0050] The difference probability comprehensively considers the target value difference and the shape difference of the path, which can not only encourage to accept high-quality paths, but also avoid excessive path changes. The specific analysis process is as follows.

[0051] The influence factor of the target value difference of the path on path acceptance is as follows: when f ( π' )< f ( π ) is satisfied, the target value of the new path is smaller, and the algorithm tends to accept the new path; when f ( π' )> f ( π ) is satisfied, the target value of the new path is larger, and the algorithm tends to reject the new path. In order to control the probability of accepting the new path when the target value is larger, an exponential decay function, exp( f ( π' )- f ( π )) is introduced. When f ( π' )< f ( π ) is satisfied, the exponential decay function is smaller, and the probability of accepting the new path is increased; when f ( π' )> f ( π ) is satisfied, the exponential decay function is increased, and the probability of accepting the new path is reduced.

[0052] The influence factor of the shape difference of the path on path acceptance is as follows: the difference d ( π' , π ) between the new path d and the original path π' measures the shape difference between the new path and the original path. π ( d , π' ) is realized by calculating the Euclidean distance between the nodes of the path. When π ( d , π' ) is smaller, it indicates that the shape difference between the new path and the original path is smaller, and the algorithm tends to accept the new path; when π ( ,

[0053] ) is larger, it indicates that the path changes dramatically, resulting in that the path is not smooth or there is a large deviation, and the algorithm tends to reject the new path.

[0054] By the difference probability calculation method of the present application, the target value difference of the path and the shape difference of the path are combined to comprehensively evaluate whether to accept the new path. When the target value difference of the path is small and the shape difference of the path is small, the resulting difference probability is larger, so the algorithm tends to accept the new path; when the target value difference is larger and the path shape difference is smaller, the resulting difference probability is smaller, so the algorithm tends to reject the new path; when the target value difference of the path is larger and the shape difference of the path is larger, the resulting difference probability is smaller, and the algorithm strongly rejects the new path; when the target value difference of the path is smaller and the path shape difference is larger, the path is not smooth or there is a large deviation, so the probability of the algorithm accepting the new path will also be reduced.

[0054] The mirror strategy is to calculate the distance of each node in the path from the ideal path, remove the nodes with a distance greater than the preset minimum distance from the path and store them in a node set, mirror each node in the node set, generate a new node by reversing the node coordinates, and insert the new node into the corresponding position of the original node in the path before mirroring to obtain a new path.

[0055] The projection strategy is to randomly select z nodes from the path, remove the randomly selected z nodes from the path and store them in a node set, map each node in the node set, generate a new node by mapping the node coordinates, and insert the new node into the corresponding position of the original node in the path before mapping to obtain a new path.

[0056] The path repair strategy is to disturb the nodes inside the obstacle in the path, the disturbance method is to generate a new node randomly within a fixed search radius range with the node inside the obstacle as the center, ensure that the new node is located in the free area, replace the node inside the obstacle with the new node, so that the repaired path does not intersect with the obstacle; for the path that intersects with the obstacle, select the two endpoints of the path that intersects with the obstacle as the new starting node and target node, and use the bidirectional rapid expansion random tree method for search, and generate a new path according to the preset search step each time.

[0057] Step 3, select the path pruning, node insertion, and node mutation operations to perform local search on the current path through the adaptive adjustment strategy.

[0058] Specifically, the adaptive adjustment strategy is to dynamically adjust the weights of the path pruning operation, the node insertion operation and the node mutation operation in a manner of combining a scoring mechanism with a roulette wheel selection method, set the initial weights of the path pruning operation, the node insertion operation and the node mutation operation to 1, set the initial scores to 0, and the initial probabilities of selection of each operation are the same, select an operation according to the weights of each operation using the roulette wheel selection method, generate a new path by executing the selected operation on the path, and score according to the target values of the new path and the original path. The original path of the first operation is the path after the destruction and reconstruction performed in step 2, the original path of the second and subsequent operations is the path after the last dynamic adjustment, the update formula of the score c is , , , , wherein c represents a score parameter and satisfies , , , f is a target value, π' represents a new path, π represents an original path, π best is a current best path, after scoring, dynamically adjust the weights of the operations according to the scores of each operation and the number of calls of each operation, and the update formula of the weight τ is , λ∈(0,1) represents a weight update coefficient, τ is the current weight of each operation, c represents the score of each operation, and ψ is the number of times each operation is called, continue to select operations using the roulette wheel selection method according to the updated weights of each operation, generate a new path after executing the selected operation, if the target value of the new path is smaller than that of the original path, replace the original path with the new path, otherwise, the original path remains unchanged, repeat the above dynamic adjustment process until the maximum number of iterations preset by the adaptive adjustment strategy is reached. The maximum number of iterations preset by the adaptive adjustment strategy is 50 times.

[0059] Specifically, the path pruning operation is to randomly select a node P i from the current path, check whether a straight line segment from P i to other nodes P j (j≠i) in the current path intersects with an obstacle, if the straight line segment from P i to P j does not intersect with the obstacle, and the nodes between P i and P j are redundant, the redundancy refers to that the nodes do not affect the connectivity of the path and do not change the shortest path, delete all redundant intermediate nodes between P i and P j , directly connect P i and P j , and form an optimized new path; otherwise, keep the current path unchanged.

[0060] The node insertion operation is to randomly select a node P from the current path. i compute node P i With the previous node P i-1 Direction vectors between and node P i With the next node P i+1 Direction vectors between Node P i The newly generated node P is moved along prevdir and nextdir by a fixed step size, respectively. new1 and P new2 Determine from P new1 To P new2 Does the straight segment intersect the obstacle? If from P new1 To P new2 The straight segment does not intersect the obstacle, so P new1 Insert into P i Before that, P new2 Insert into P i After that, delete P. i Generate a new path; otherwise, keep the current path unchanged.

[0061] The node mutation operation involves randomly selecting a point in the current path as the mutated node P. i Generate a random angle θ, calculate the direction vector dv=[cosθ, sinθ] based on θ, generate a random step size m, and then... i Around the direction vector dv, generate a new node P with a step size m. new Determine from P respectively i-1 To P new The straight line segment and from P new To P i+1 Does the straight segment intersect the obstacle? If from P i-1 To P new The straight line segment and from P new To P i+1 If none of the straight segments intersect the obstacle, then use P. new Replace P i Generate a new path; otherwise, keep the current path unchanged.

[0062] Step 4: Accept the path according to the simulated annealing criterion. Specifically, the simulated annealing criterion is as follows: if the target value of the current path is smaller than the target value of the path before reconstruction, then accept the current path; if the target value of the current path is smaller than the target value of the current best path, then replace the current path with the current best path; if the target value of the current path is larger than the target value of the path before reconstruction, then accept the path with probability r = exp{-(f(π)}. c)-f(π0)) / T} accepts the current path, where exp is the natural exponential function, f(π c ) is the current path π c The target value is f(π0), which is the target value of the path π0 before destruction and reconstruction, and T represents the current temperature. It is worth noting that if the path accepted according to the simulated annealing criterion does not terminate, the accepted path will re-enter step 2, and the difference-guided path optimization mechanism will be used again to destroy and reconstruct the accepted path, further expanding the path exploration space.

[0063] Step 5: Determine if the algorithm's termination time has been reached. If it has, the algorithm ends, and the current optimal path and target value are output; otherwise, return to Step 2. Specifically, the algorithm's termination time is 5 seconds.

[0064] This invention discloses a robot fast path optimization method based on multi-strategy fusion, namely MSIG. It generates an initial path through a bidirectional fast expanding random tree method; it improves the algorithm's global exploration capability by destructively reconstructing the path through a difference-guided path optimization mechanism; it performs local search on the path by adaptively adjusting path pruning, node mutation, and node insertion operations, reducing the number of path nodes through path pruning and optimizing the path structure through node mutation and insertion operations; and it avoids getting trapped in local optima through simulated annealing acceptance criteria, thereby obtaining a better path. In the specific application of path planning, this invention not only effectively shortens the path length but also optimizes the path smoothness. The invention is further described below through two specific embodiments.

[0065] like Figure 2 and Figure 3 As shown, Figure 2 It is an apartment interior environment. Figure 3 For the outdoor environment of the street. Figure 2 The distribution of obstacles in a confined space demonstrates the algorithm's path planning capability within a restricted area. Figure 3 The algorithm's path selection and optimization performance in open environments is demonstrated by showcasing obstacle distribution in relatively open spaces. Each map is uniformly 500 x 500 pixels, with pixel distance used as the unit of distance in path planning. It should also be noted that the values ​​of the preset minimum distance in the mirror strategy, the preset search step size in the path repair strategy, and the preset lower and upper thresholds in the difference-guided path optimization mechanism were obtained from the following experiments, the specific implementation process of which is as follows.

[0066] The minimum distance of the mirror strategy is used d min The distance unit represents the search step size in the path repair strategy.l step one distance unit. The threshold lower limit and the threshold upper limit in the difference-based path optimization mechanism are represented by v min and v max one distance unit, respectively. Taguchi experiment is used to select the optimal parameter settings. Each parameter is assigned four appropriate numerical values, and the levels of the parameters are represented by numbers (e.g. 1, 2, 3, 4, etc.). Each number under each parameter level represents a specific value of the parameter, so as to simplify the experimental design and analysis. The experimental parameter level table is shown in Table 1:

[0067] Table 1 Experimental parameter level table

[0068]

[0069] Orthogonal table is used to determine the parameter combinations. Each parameter combination is executed for 30 times, and a total of 480 results are obtained. The performance index is the relative percentage increase (RPI), wherein, f best represents the optimal target value obtained in the 480 experiments, and f i represents the target value obtained when running the i th experiment. The different parameter combinations and RPI response values are shown in Table 2:

[0070] Table 2 Table of different parameter combinations and RPI response values

[0071]

[0072] The average RPI response value of each parameter and the parameter level are shown in Table 3:

[0073] Table 3 Table of average RPI response value of each parameter and parameter level

[0074]

[0075] Among them, the rank in Table 3 represents the difference between the maximum and minimum RPI values of each parameter, reflecting the influence of each parameter on the performance of the algorithm, and the level is used for sorting to determine the importance of each parameter.

[0076] As Figure 20As shown, the trend of each parameter factor level is shown. The horizontal axis represents each level of each parameter, and the vertical axis represents the corresponding RPI, and the smaller the RPI, the better the algorithm performance. Therefore, by selecting the lowest point of each curve, the optimal value of each parameter at different levels is found. Specifically, the lowest point of each curve corresponds to the horizontal axis value (i.e., the parameter level), which is the optimal selection of each parameter in the experiment. For example, the lowest point of the curve of d min corresponds to the horizontal axis value of 2, and the row in Table 1 where d min is found, and the value of d min at level 2 is viewed, and the value of d min is 3, and so on, that is, the best parameter settings are obtained: the minimum distance in the mirror strategy d min = 3, the search step in the path repair strategy l step = 4, the lower threshold in the differential guidance path optimization mechanism v min = 25, the upper threshold in the differential guidance path optimization mechanism v max = 150.

[0077] Under each map environment, the optimal parameter settings are used by the present application to test whether the algorithm can normally run through 30 experiments by setting the maximum iteration number of the bidirectional rapid expansion random tree method preset. According to the experimental situation, the maximum iteration number of the bidirectional rapid expansion random tree method preset is 1x10 4 ~ 5x10 5 , and the specific value can be adaptively adjusted according to different map environments to ensure that the algorithm normally runs. The present application performs 30 experiments under different maps, and the experimental results show that when the probability threshold preset by the differential guidance path optimization mechanism is set to 0.5, the algorithm performs relatively stably.

[0078] The present application is compared with four existing comparative algorithms, namely, the fast RRT* algorithm (F-RRT*), the node moving artificial bee colony algorithm (NMABC), the variable neighborhood search iterative greedy algorithm (VNIG), and the double-layer mutation iterative greedy algorithm (IGDLM). In order to reduce the error of the experiment, each example of the present application and the comparative algorithm is repeatedly executed 30 times to generate statistical data. In two map environments, the comparative results of the present application and the four comparative algorithms in the path length index are shown in Figure 4 and Figure 5 . Figure 4 and Figure 5is a box plot, the horizontal axis represents the algorithm of the application and four kinds of comparative algorithms, the vertical axis represents the path length. The box in the box plot reflects the concentration of data, the median line represents the intermediate value of the data, the line segment extending from the upper quartile and the lower quartile of the box shows the normal range of the data, that is, the maximum and minimum data points excluding outliers, and the plus sign represents the outlier point, indicating that the algorithm is unstable in some cases. As can be seen from the figure, the box part of the application is smaller, indicating that the consistency of the application in path length is higher, and a shorter path can be stably generated. And the numerical value of the path length of the application is smaller, the difference between the maximum and minimum values of the path length is smaller, and the performance is more efficient.

[0079] Figure 6 and Figure 7 is also a box plot, the comparison results of the application and four kinds of comparative algorithms in the path smoothness index shown by Figure 6 and Figure 7 It can be seen from the comparison results of the application and four kinds of comparative algorithms in the path smoothness index shown by Figures 8-17 It can be seen from the comparison results of the application and four kinds of comparative algorithms in the path smoothness index shown by Figure 18 , Figure 19 In the convergence curve graph of the application and four kinds of comparative algorithms in two different map environments, the algorithm running time is 5 seconds, and the path length is recorded every 0.2 seconds. As can be seen from the figure, the application can optimize the shortest path more quickly than the existing four kinds of comparative algorithms under the same conditions, and has significant convergence ability. Therefore, the application can generate a short and smooth path in the practical application of path planning, thereby significantly improving the efficiency of path planning.

Claims

1. A robot rapid path optimization method based on multi-strategy fusion, characterized in that, The method comprises the following steps of: Step 1, initializing the parameters of the iterative greedy algorithm, and generating an initial path by using a bidirectional rapid expansion random tree method, wherein the initial path is used as an initial current best path; Step 2, performing damage reconstruction on the path by using a difference-guided path optimization mechanism, and expanding a path exploration space; wherein the path is damaged and reconstructed for the first time, the path is damaged and reconstructed for the second time and after the second time, and the path accepted after the last iteration is used as the path; wherein the difference-guided path optimization mechanism is implemented as follows, The difference ε between the calculated path and an ideal path, the ideal path being the shortest path from the start node to the target node under the condition of no obstacles; if ε is less than a preset lower threshold, indicating that the path is close to the ideal path, the path is kept unchanged; if ε is greater than or equal to a preset upper threshold, a new path is generated through a mirror strategy; if ε is between the preset lower threshold and the preset upper threshold, a new path is generated through a projection strategy; if the new path is infeasible after being generated, a path repair strategy is executed to repair the new path to be feasible, the target value of the new path is calculated, if the target value of the new path is less than the target value of the original path, the new path is directly accepted, otherwise, the difference probability between the new path and the original path is calculated, when the difference probability value is greater than a preset probability threshold, the new path is accepted, and the search space is expanded, wherein the infeasible refers to that a node in the path is located in an obstacle or the path intersects with an obstacle; the difference probability The calculation formula of the difference probability is f(π') is the target value of the new path π', f(π) is the target value of the original path π, and d(π', π) represents the difference between the new path π' and the original path π. Step 3, performing path pruning, node insertion and node mutation operations by using an adaptive adjustment strategy to perform local search on the current path; Step 4, accepting the path according to a simulated annealing criterion; Step 5, determining whether the algorithm termination time is reached, if the algorithm termination time is reached, the algorithm is ended, and the current best path is output, otherwise, returning to step 2.

2. The method of claim 1, wherein, In step 1, the path π is composed of a series of discrete nodes, represented as π = [P1...P2]. i ...P n ], P1 represents the robot's starting node, P n P represents the target node. i Let i be an intermediate node in the path, i is the position index of the intermediate node, and the value of i is in the range of 2≤i≤n−1, where n represents the number of all nodes in the path.

3. The method of claim 2, wherein the method further comprises: The mirror strategy is that distances between each node in the path and an ideal path are calculated, nodes with distances greater than a preset minimum distance are removed from the path and stored in a node set, each node in the node set is flipped by mirroring, a new node is obtained by reversing the node coordinates, the new node is inserted into a corresponding position of the original node in the path before the mirroring, and a new path is obtained; The projection strategy is that z nodes are randomly selected from the path, the randomly selected z nodes are removed from the path and stored in a node set, each node in the node set is mapped, a new node is generated by mapping the node coordinates, the new node is inserted into a corresponding position of the original node in the path before the mapping, and a new path is obtained; The path repair strategy is that nodes inside an obstacle in the path are disturbed, the disturbance is a new node generated within a fixed search radius range with the node inside the obstacle as a center, the new node is ensured to be located in a free area, the new node is used to replace the node inside the obstacle, and the repaired path does not intersect with the obstacle; For the path intersecting with the obstacle, two end points of the path intersecting with the obstacle are selected as new starting nodes and target nodes, a bidirectional rapid expansion random tree method is used for searching, and a new path is generated according to a preset search step length during each expansion.

4. The method of claim 3, wherein, In step 3, the adaptive adjustment strategy dynamically adjusts the weights of path pruning, node insertion, and node mutation operations using a combination of a scoring mechanism and a roulette wheel selection method. The initial weights of these operations are set to 1, and their initial scores are set to 0. Each operation has an equal initial probability of being selected. Based on the weights of each operation, a roulette wheel selection method is used to select one operation. This selected operation is then applied to the path to generate a new path. A score is calculated based on the target values ​​of the new path and the original path. The original path for the first operation is the path obtained in step 2 after the destructive reconstruction. The original paths for the second and subsequent operations are the paths after the previous dynamic adjustment. The update formula for the score c is... Let ℓ1, ℓ2, ℓ3 represent the scoring parameters satisfying ℓ1>ℓ2>ℓ3, f be the target value, π' represent the new path, and π represent the original path. best This is the current optimal path. After scoring, the weights of each operation are dynamically adjusted based on their scores and the number of times they are called. The update formula for the weight τ is: λ∈(0,1) represents the weight update coefficient, τ is the current weight of each operation, c represents the score of each operation, and ψ is the number of times each operation is called. Based on the updated weights of each operation, the roulette wheel selection method is used to select operations. After the selected operation is executed, a new path is generated. If the target value of the new path is smaller than the target value of the original path, the new path replaces the original path. Otherwise, the original path remains unchanged. The above dynamic adjustment process is repeated until the maximum number of iterations preset by the adaptive adjustment strategy is reached.

5. The method of claim 4, wherein, The path pruning operation is to randomly select a node P from the current path i , check whether the straight line segment from P i to other nodes P j in the current path intersects with the obstacle, if the straight line segment from P i to P j does not intersect with the obstacle, and the nodes between P i and P j are redundant, the redundancy refers to that the nodes do not affect the connectivity of the path and do not change the shortest path, then all the redundant intermediate nodes between P i and P j are deleted, P i and P j are directly connected, and the optimized new path is formed Otherwise, the current path is kept unchanged; The node insertion operation is to randomly select a node P from the current path. i compute node P i With the previous node P i-1 The direction vector between them is prevdir=P i -P i-1 and node P i With the next node P i+1 The direction vector between them is nextdir=P i+1 -P i Node P i The newly generated node P is moved along prevdir and nextdir by a fixed step size, respectively. new1 and P new2 Determine from P new1 To P new2 Does the straight segment intersect the obstacle? If from P new1 To P new2 The straight segment does not intersect the obstacle, so P new1 Insert into P i Before that, P new2 Insert into P i After that, delete P. i Generate a new path; Otherwise, the current path is kept unchanged; The node mutation operation is to randomly select a point in the current path as a mutation node P i , generate a random angle θ, calculate a direction vector dv=[cosθ, sinθ] according to θ, generate a random step m, generate a new node P i around P new , respectively judge whether the straight line segment from P i-1 to P new and the straight line segment from P new to P i+1 intersect with the obstacle, if the straight line segment from P i-1 to P new and the straight line segment from P new to P i+1 do not intersect with the obstacle, replace P new with P i , generate a new path, otherwise keep the current path unchanged.

6. The method of claim 5, wherein the method further comprises: In step 4, the simulated annealing criterion is that if the objective value of the current path is less than the objective value of the path before the destruction reconstruction, the current path is accepted, if the objective value of the current path is less than the objective value of the current best path, the current path is replaced by the current best path, and if the objective value of the current path is greater than the objective value of the path before the destruction reconstruction, the current path is accepted with a probability r = exp{-(f(π c )-f(π0)) / T}, where exp is the natural exponential function, f(π c ) is the objective value of the current path π c , f(π0) is the objective value of the path π0 before the destruction reconstruction, and T represents the current temperature.

7. The method of claim 1-6, wherein In step 5, the algorithm termination time is 5s.

Images