Robot multi-target path planning method based on self-learning evolutionary algorithm

Abstract

The invention relates to the technical field of mobile robot path planning, in particular to a robot multi-target path planning method based on a self-learning evolutionary algorithm. Comprising the steps of initializing algorithm parameters; initializing a population; in an employed bee stage, performing path crossing, path variation, path shortening and path safety operation on each solution in the population; calculating a target value of each solution, and dividing the population into a non-dominated solution set and a dominated solution set; in the following bee stage, a learning mechanism based on cooperation acts on a non-dominated solution set, and a learning mechanism based on dominated guidance acts on a dominated solution set; combining the non-dominated solution set and the dominated solution set; in the reconnaissance bee stage, an individual body weight starting strategy is used for acting on a solution with the number of continuous evolution failures exceeding a set maximum threshold value; and updating the non-dominated solution set, and judging whether a termination condition is met or not. The application of the invention can effectively improve the path planning efficiency and enhance the path quality.

Description

Technical Field

[0001] The invention relates to the technical field of mobile robot path planning, and in particular to a robot multi-objective path planning method based on a self-learning evolutionary algorithm. Background Art

[0002] With the rapid development of science and technology, mobile robots have been widely used in many fields such as industrial manufacturing, medical surgery, agricultural production and autonomous driving. Path planning is a key technology to realize autonomous navigation of mobile robots. Its core task is to search for an optimal or suboptimal feasible path from the starting position to the target position in the configuration space. The path planning methods in the prior art usually only optimize a single path length target. However, when performing path planning in complex actual scenarios, multiple factors need to be considered comprehensively, especially potential dangerous obstacles in the environment. Therefore, the existing single-target path planning methods are difficult to generate high-quality paths that meet actual needs.

[0003] In addition, the existing methods for solving path planning lack specific optimization operations for problems and goals, resulting in poor solution accuracy. At the same time, most methods focus on the individual evolution of the population, but ignore the co-evolution between individuals, resulting in insufficient utilization of excellent information, thus affecting the quality of the final solution. Therefore, for multi-objective path planning problems, a multi-objective path planning method based on a self-learning evolutionary algorithm is urgently needed to further improve the quality of the solution. Summary of the invention

[0004] The purpose of the present invention is to provide a robot multi-objective path planning method based on a self-learning evolutionary algorithm, which solves the problems of minimizing path length and maximizing safety in the path planning process, so as to effectively improve the path planning efficiency and enhance the path quality.

[0005] The present invention provides a robot multi-objective path planning method based on a self-learning evolutionary algorithm, which is characterized by comprising the following steps: Step 1: Initialize algorithm parameters and set population size Ps , the maximum number of consecutive mutation failures in a path mutation operation C mf , the maximum number of consecutive evolution failures C ef , the end of operation time; Step 2: Initialize the population and use a mixed initialization strategy to generate Ps An initial solution; Step 3, the hired bee stage, performs path crossing operations and path mutation operations in sequence, explores each solution in the population, randomly selects to perform path shortening operations or path safety operations to evolve the solution, and accepts the evolved solution as the new solution; Step 4, calculate the target value of the solution of the population, perform non-dominated sorting on the solutions in the population, and divide the population into a non-dominated solution set and a dominated solution set; Step 5, the bee-following stage, uses two self-learning optimization mechanisms, including a collaborative learning mechanism and a dominant guidance learning mechanism, which act on the non-dominated solution set and the dominated solution set respectively to achieve a deep search of the solution space; Step 6, merge the non-dominated solution set and the dominated solution set; Step 7, the scout bee stage, uses the individual restart strategy to act on the solution whose number of consecutive evolution failures exceeds the set maximum threshold to enhance the vitality of the population; Step 8, update the non-dominated solution set and determine whether the termination condition is met. If so, end the evolution and output the non-dominated solution set. Otherwise, return to step 3.

[0006] Furthermore, in step 2, the process of the hybrid initialization strategy is as follows: the first solution is generated using a heuristic algorithm, and then the remaining solutions are generated using a fast exploration random tree algorithm and a space partitioning method according to the population size. Ps -1 solution; a solution represents a path, and the path is represented by a discrete sequence, Path = [ P 0 , ..., P i , ..., P n+1 ], Path Represents a path, P 0 Indicates the first node in the sequence, i.e. the starting node. P n+1 Indicates the last node in the sequence, i.e. the target node. represents an intermediate node, i Indicates the index number of the intermediate node, the value range is 1≤ i ≤ n , n Represents the number of intermediate nodes in the path; each solution corresponds to two target values, namely path length and path safety; the formula for calculating path length is: ,in Indicates the path Path Length, represents the Euclidean distance between two consecutive points in the path; the calculation formula for path security is ,in, represents the safety of the path, where is an obstacle in the configuration space, j Indicates the index number of the obstacle. m is the number of obstacles, is the minimum distance between a path and an obstacle in the configuration space.

[0007] Further, in step 3, a solution in the population is used as the first input solution of the path crossover operation, and a solution is randomly selected in the population as the second input solution. The path crossover operation includes the following steps: finding the intermediate node with the shortest distance between the two input solutions in the discrete sequences corresponding to the two input solutions, and setting the found node as the intermediate node in the first input solution. The second input solution ,in, q Check the selected node index in the sequence corresponding to the second input solution. and Whether the straight line segment between intersects with the obstacle, if not, two new solutions are generated: the first new solution is from the starting node of the sequence corresponding to the first input solution to the node , concatenate the sequence corresponding to the second input solution from node to the target node; the second new solution consists of the starting node to the node corresponding to the sequence of the second input solution , concatenate the sequence corresponding to the first input solution from node To the target node, perform non-dominated sorting on the two input solutions and the two new solutions, select the non-dominated solutions, and randomly select one solution from them to output; if they intersect, directly output the first input solution.

[0008] Furthermore, in step 3, the solution output by the path crossover operation is used as the input solution of the path mutation operation. The steps of the path mutation operation are to set a counter to record the number of consecutive mutation failures and set its initial value to 0, and randomly select an intermediate node in the sequence corresponding to the input solution. , and determine the node Minimum distance to surrounding obstacles d min , in the node is the center of the circle, d min Generate a new node randomly within a circle of radius P new ; Check node P new With Node and nodes Whether the straight line segment between nodes intersects with obstacles, if not, the new node P new Alternative Node A new solution is constructed, and the target value of the new solution is calculated. If the target value of the new solution is less than the target value of the input solution, the new solution is output and the path mutation operation ends. If the target value of the new solution is higher than the target value of the input solution, the counter is incremented by one and a new node is generated. If they intersect, the counter is incremented by one and a new node is generated. When the counter reaches the maximum number of consecutive mutation failures, C mf And no new solution with a smaller objective value is generated, the input solution is output.

[0009] Furthermore, in step 3, the solution output by the path mutation operation is used as the input solution of the path shortening operation. The steps of the path shortening operation are: setting the input solution to a new solution; randomly selecting two non-adjacent intermediate nodes in the sequence corresponding to the new solution P x and P y , check the node P x and P y Whether the straight line segment between the nodes intersects with the obstacle; if not, delete the node P x and P y The path shortening operation ends when a collision occurs between two nodes, and a new solution is output. If there is a collision, non-adjacent intermediate nodes are randomly selected until all two non-adjacent nodes are traversed and a new solution is output.

[0010] Furthermore, in step 3, the solution output by the path mutation operation is used as the input solution of the path safety operation. The steps of the path safety operation are: in the sequence corresponding to the input solution, two adjacent nodes form a line segment, and find the line segment closest to the obstacle. , and determine the line segment The position closest to the obstacle is set as the node P nearest ,Sure P nearest The node on the surface closest to the obstacle P obstacle ; Compute nodes P nearest With Node P obstacle The distance between d min , in the node P nearest As a starting point, is the direction, distance from the node P nearest d min Create a new node at the location Pnew , check the node P new With Node and nodes Whether the straight line segment between them intersects with the obstacle; if not, the node P new Insert into node and nodes Between them, a new solution is formed and output; if they intersect, the input solution is output.

[0011] Furthermore, in step 5, the collaborative learning mechanism is applied to the non-dominated solution set, and the dominant guidance-based learning mechanism is applied to the dominated solution set, wherein: The collaborative learning mechanism is to calculate the target value of each solution in the non-dominated solution set, and normalize these target values ​​using the following formula according to the target type: It refers to k The solution l target value, in addition, and Represents all solutions l The maximum and minimum values ​​of the targets, therefore, Indicates k The solution l The result of normalizing the target values; perform the following steps on each solution in the non-dominated solution set: if the current solution obtains the minimum value in path length or path safety, the path shortening operation is preferentially performed on it to generate a new solution; then the path safety operation is applied to the generated solution to further improve the safety of the solution; for the solution that does not obtain the minimum value in path length or path safety, the optimization operation is performed according to the size of the normalized value: if the normalized value of the path length is higher than the normalized value of the path safety, the path shortening operation is performed on the current solution; otherwise, the path safety operation is performed to obtain the optimized non-dominated solution; if the optimized non-dominated solution dominates the non-dominated solution before optimization, the non-dominated solution before optimization is replaced by the optimized non-dominated solution, and the optimized non-dominated solution set is generated and output; The learning mechanism based on domination guidance is that for each dominated solution, a non-dominated solution is randomly selected from the non-dominated solution set for comparison; if the path length and path safety of the non-dominated solution are both smaller than the path length and path safety of the current dominated solution, the path shortening operation and the path safety operation are performed on the current dominated solution in sequence; if the path length of the non-dominated solution is smaller than the path length of the current dominated solution, the path shortening operation is performed on the current dominated solution; otherwise, the path safety operation is performed on the current dominated solution to generate an optimized dominated solution. If the path length and path safety of the optimized dominated solution are smaller than the dominated solution before optimization, the optimized dominated solution is used to replace the dominated solution before optimization, and the optimized dominated solution set is generated and output.

[0012] Furthermore, in step 7, the individual restart strategy is to perform non-dominated sorting on the population generated in the follower bee stage, and divide the population into a non-dominated solution set and a dominated solution set; for each solution in the dominated solution set, detect whether the number of consecutive evolution failures exceeds the set maximum threshold. If satisfied, replace the current dominated solution with a new solution, and randomly select any of the following methods to generate a new solution: one is to select two solutions from the non-dominated solution set, one of which is a non-dominated solution that achieves the minimum value in path length or path safety, and the other is a random non-dominated solution, and perform path intersection operations on these two solutions to generate a new solution; the other is to use a fast random tree algorithm to generate five new solutions, and select the solution with the shortest path length as the new solution.

[0013] The robot multi-objective path planning method based on the self-learning evolutionary algorithm provided by the present invention optimizes the two objectives of path length and path safety at the same time for dangerous obstacles existing in the path planning process. Secondly, the present invention adopts a mixed initialization strategy to generate a high-quality initial population. Compared with some existing algorithms, the present invention proposes optimization operations for two objectives, improves the algorithm's ability to explore the solution space, and proposes two self-learning optimization mechanisms to guide the deep evolution of the solution. In addition, the individual restart strategy used by the present invention enhances the vitality of the population. In summary, the present invention has the advantages of effectively improving the quality of the solution and greatly improving the efficiency of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart for realizing the present invention; Figure 2 The home indoor scene used in the test of the present invention Figure 1 Schematic diagram of the characteristics; Figure 3 The wide area street scene used in the test of the present invention Figure 2 Schematic diagram of the characteristics; Figure 4 The present invention compares the existing algorithms in the Figure 1Distribution of solutions obtained in ; Figure 5 The present invention compares the existing algorithms in the Figure 2 Distribution of solutions obtained in ; Figure 6 The present invention compares the existing algorithms in the Figure 1 Convergence curve diagram on ; Figure 7 The present invention compares the existing algorithms in the Figure 2 Convergence curve diagram on ; Figure 8 For the present invention Figure 1 The obtained path length and path safety are displayed on the best solution; Fig. 9 For the present invention Figure 2 The obtained path length and the solution with the best path safety are shown in the graph. DETAILED DESCRIPTION

[0015] like Figure 1 As shown, the robot multi-objective path planning method based on self-learning evolutionary algorithm provided by the present invention is mainly implemented through the following steps.

[0016] Step 1: Initialize algorithm parameters. Set the population size Ps , the maximum number of consecutive mutation failures in a path mutation operation C mf , the maximum number of consecutive evolution failures C ef ,in, Ps =50, C ef =20, C mf =10, the cut-off running time is 50s, which is the termination condition of the algorithm.

[0017] Step 2: Initialize the population. Use a hybrid initialization strategy to generate Ps initial solutions. Specifically, a solution represents a path, and the path is represented by a discrete sequence. Path = [ P 0 , ..., P i , ..., P n+1 ], Path Represents a path, P 0 Indicates the first node in the sequence, i.e. the starting node. P n+1 Indicates the last node in the sequence, i.e. the target node. represents an intermediate node, iIndicates the index number of the intermediate node, the value range is 1≤ i ≤ n , n Represents the number of intermediate nodes in the path; each solution corresponds to two target values, namely path length and path safety; the formula for calculating path length is: ,in Indicates the path Path Length, represents the Euclidean distance between two consecutive points in the path; the calculation formula for path security is ,in, represents the safety of the path, where is an obstacle in the configuration space, j Indicates the index number of the obstacle. m is the number of obstacles, is the minimum distance between the path and the obstacle in the configuration space. The hybrid initialization strategy steps are as follows: using an efficient heuristic algorithm, the A* algorithm is used in the present invention to generate the first initial solution, and then the fast random tree algorithm and the space partitioning method are used to generate the remaining Ps -1 initial solution. Specifically, two candidate solutions are generated using the fast random tree algorithm and the space partitioning method respectively. The non-dominated solution is selected as the next initial solution, and finally a variety of initial solutions are generated, which provides a good foundation for subsequent optimization.

[0018] Step 3, the hired bee stage, performs path crossing operations and path mutation operations in sequence, explores each solution in the population, randomly selects to perform path shortening operations or path safety operations to evolve the solution, and accepts the evolved solution as the new solution. The random selection method in the technical solution of the present invention is: by using a random function rand () generates a random number and makes different choices based on the value of the random number. For example, in step 3, by using the random function rand () Generate a random number between 0 and 1. If the random number is greater than 0.5, perform a path shortening operation on the solution. Otherwise, perform a path security operation on the solution.

[0019] Specifically, the current solution in the population is used as the first input solution of the path crossover operation, and a solution is randomly selected in the population as the second input solution. The path crossover operation includes the following steps: finding the intermediate node with the shortest distance between the two input solutions in the discrete sequence corresponding to each other, and setting the found node as the intermediate node in the first input solution. The second input solution ,in, q Check the selected node index in the sequence corresponding to the second input solution. and Whether the straight line segment between intersects with the obstacle, if not, two new solutions are generated: the first new solution is from the starting node of the sequence corresponding to the first input solution to the node , concatenate the sequence corresponding to the second input solution from the node to the target node; the second new solution consists of the starting node to the node corresponding to the sequence of the second input solution , concatenate the sequence corresponding to the first input solution from node To the target node, perform non-dominated sorting on the two input solutions and the two new solutions, select the non-dominated solutions, and randomly select one solution from them to output; if they intersect, directly output the first input solution.

[0020] Specifically, the solution output by the path crossover operation is used as the input solution of the path mutation operation. The steps of the path mutation operation are to set a counter to record the number of consecutive mutation failures and set its initial value to 0, and randomly select an intermediate node in the sequence corresponding to the input solution. , and determine the node Minimum distance to surrounding obstacles d min , in the node is the center of the circle, d min Generate a new node randomly within a circle of radius P new ; Check node P new With Node and nodes Whether the straight line segment between nodes intersects with obstacles, if not, the new node P new Alternative Node A new solution is constructed, and the target value of the new solution is calculated. If the target value of the new solution is less than the target value of the input solution, the new solution is output and the path mutation operation ends. If the target value of the new solution is higher than the target value of the input solution, the counter is incremented by one and a new node is generated. If they intersect, the counter is incremented by one and a new node is generated. When the counter reaches the maximum number of consecutive mutation failures, C mf And if no new solution with a smaller objective value is generated, the input solution is output.

[0021] Specifically, the solution output by the path mutation operation is used as the input solution of the path shortening operation. The steps of the path shortening operation are: setting the input solution to a new solution; randomly selecting two non-adjacent intermediate nodes in the sequence corresponding to the new solution P x and P y , check the node P x andP y Whether the straight line segment between the nodes intersects with the obstacle; if not, delete the node P x and P y The path shortening operation ends when a collision occurs between two nodes, and a new solution is output. If there is a collision, non-adjacent intermediate nodes are randomly selected until all two non-adjacent nodes are traversed and a new solution is output.

[0022] Specifically, the solution output by the path mutation operation is used as the input solution of the path safety operation. The steps of the path safety operation are: in the sequence corresponding to the input solution, two adjacent nodes form a line segment, and find the line segment closest to the obstacle. , and determine the line segment The position closest to the obstacle is set as the node P nearest ,Sure P nearest The node on the surface closest to the obstacle P obstacle ; Compute nodes P nearest With Node P obstacle The distance between d min , in the node P nearest As a starting point, is the direction, distance from the node P nearest d min Create a new node at the location P new , check the node P new With Node and nodes Whether the straight line segment between them intersects with the obstacle; if not, the node P new Insert into node and nodes Between them, a new solution is formed and output; if they intersect, the input solution is output.

[0023] Step 4, calculate the target value of the solution of the population, perform non-dominated sorting on the solutions in the population, and divide the population into a non-dominated solution set and a dominated solution set.

[0024] Step 5, the bee-following stage, uses two self-learning optimization mechanisms, including a collaborative learning mechanism and a dominant guidance learning mechanism. The collaborative learning mechanism acts on the non-dominated solution set, and the dominant guidance learning mechanism acts on the dominated solution set. The collaborative learning mechanism is to calculate the target value of each solution in the non-dominated solution set, and normalize these target values ​​using the following formula according to the target type: It refers to k The solution l target value, in addition, and Respectively represent all solutions l The maximum and minimum values ​​of the targets, therefore, Expressing the k The solution l The result of normalizing the target values; for each solution in the non-dominated solution set, the following steps are performed: if the current solution obtains the minimum value in path length or path safety, the path shortening operation is preferentially performed on it to generate a new solution; then the path safety operation is applied to the generated solution to further improve the safety of the solution; for the solution that does not obtain the minimum value in path length or path safety, the optimization operation is performed according to the size of the normalized value: if the normalized value of the path length is higher than the normalized value of the path safety, the path shortening operation is performed on the current solution; otherwise, the path safety operation is performed to obtain the optimized non-dominated solution; if the optimized non-dominated solution dominates the non-dominated solution before optimization, the optimized non-dominated solution is used to replace the non-dominated solution before optimization, and the optimized non-dominated solution set is generated and output.

[0025] The learning mechanism based on domination guidance is that for each dominated solution, a non-dominated solution is randomly selected from the non-dominated solution set for comparison; if the path length and path safety of the non-dominated solution are both smaller than the path length and path safety of the current dominated solution, the path shortening operation and the path safety operation are performed on the current dominated solution in sequence; if the path length of the non-dominated solution is smaller than the path length of the current dominated solution, the path shortening operation is performed on the current dominated solution; otherwise, the path safety operation is performed on the current dominated solution to generate an optimized dominated solution. If the path length and path safety of the optimized dominated solution are smaller than the dominated solution before optimization, the optimized dominated solution is used to replace the dominated solution before optimization, and the optimized dominated solution set is generated and output.

[0026] Step 6: Merge the non-dominated solution set and the dominated solution set.

[0027] Step 7, in the scout bee stage, the individual restart strategy is used to act on the solutions whose number of consecutive evolution failures exceeds the set maximum threshold to enhance the vitality of the population. Specifically, the individual restart strategy is to perform non-dominated sorting on the population generated in the follower bee stage, and divide the population into a non-dominated solution set and a dominated solution set; for each solution in the dominated solution set, check whether the number of consecutive evolution failures exceeds the set maximum threshold. If it does, replace the current dominated solution with a new solution, and randomly select any of the following methods to generate a new solution: one is to select two solutions from the non-dominated solution set, one of which is a non-dominated solution that obtains the minimum value in path length or path safety, and the other is a random non-dominated solution, and perform path intersection operations on these two solutions to generate a new solution; the other is to use the fast random tree algorithm to generate five new solutions, and select the solution with the shortest path length as the new solution.

[0028] Step 8: Update the non-dominated solution set and determine whether the running time reaches 50 seconds. If so, end the evolution and output the non-dominated solution set, otherwise return to step 3.

[0029] The present invention proposes a robot multi-objective path planning method based on a self-learning evolutionary algorithm, which combines two self-learning mechanisms with an artificial bee colony algorithm framework and is named EABC. The method improves the quality of the initial population through a hybrid initialization strategy, improves the global exploration capability of the algorithm through path crossing, path mutation, path shortening and path safety operations in the employed bee stage, improves the local search capability of the algorithm through two self-learning optimization mechanisms in the follower bee stage to achieve deep evolution of the solution, and enhances the vitality of the population through a new individual restart strategy in the scout bee stage.

[0030] The present invention is further described below by listing two specific embodiments of the present invention.

[0031] like Figure 2 and Figure 3 As shown, Figure 2 It is a family interior scene. Figure 3 is a wide area street scene. The size of the two scene maps is standardized to 500×500 pixels, which is used to test and verify the effectiveness of the path planning method of the present invention. The present invention is experimentally compared with the improved artificial bee colony algorithm (IMO-ABC), the improved multi-objective optimization algorithm (LMOEA / D), the non-dominated sorting algorithm (NSGA-II), the improved multi-objective artificial bee colony algorithm (IMOABC) and the enhanced improved multi-objective particle swarm optimization algorithm (FIMOSO). The simulation experiments and results are as follows: In the two scenes, the present invention and the above five existing algorithms are used to run for 50s respectively, and the non-dominated solutions obtained are as follows: Figure 4 and Figure 5 As shown. Figure 4 and Figure 5In the figure, the horizontal axis represents the safety index value, and the vertical axis represents the length index value. It can be seen that the non-dominated solution set obtained by the method is closer to the lower left corner of the picture, indicating that the application of the present invention can obtain a non-dominated solution set with higher quality.

[0032] like Figure 6 and Figure 7 As shown in the figure, the comparison results of the present invention and the above five existing algorithms on the hypervolume metric are made. Hypervolume represents the volume of the hypercube formed by the non-dominated solution set and the reference point, where the reference point is the point corresponding to the worst value in path length and path safety during the algorithm operation. Hypervolume is used to evaluate the comprehensive performance of the algorithm, and the value range is [0,1]. The larger the hypervolume value, the better the overall performance of the algorithm. Figure 6 and Figure 7 In the figure, the horizontal axis represents the hypervolume value, and the vertical axis represents the running time. The algorithm of the present invention achieves the maximum hypervolume value in the two test scenarios, which reflects that the comprehensive performance of the method of the present invention is the best.

[0033] like Figure 8 and Fig. 9 As shown in the figure, the present invention generates optimal paths for different optimization objectives in two test scenarios. The figure intuitively shows the specific impact of the shortest path length and the highest path safety on the path planning results, thereby revealing the differences in path selection for different optimization objectives. Therefore, the path planning method proposed in the present invention can achieve a better balance between path length and path safety and generate high-quality solutions.

Claims

1. A robot multi-objective path planning method based on a self-learning evolutionary algorithm, characterized in that: The following steps are included: Step 1: Initialize algorithm parameters and set population size Ps , the maximum number of consecutive mutation failures in a path mutation operation C mf , the maximum number of consecutive evolution failures C ef , the end of operation time; Step 2: Initialize the population and use a mixed initialization strategy to generate Ps An initial solution; Step 3, the hired bee stage, performs path crossing operations and path mutation operations in sequence, explores each solution in the population, randomly selects to perform path shortening operations or path safety operations to evolve the solution, and accepts the evolved solution as the new solution; Step 4, calculate the target value of the solution of the population, perform non-dominated sorting on the solutions in the population, and divide the population solutions into a non-dominated solution set and a dominated solution set; Step 5, the bee-following stage, uses two self-learning optimization mechanisms, including a collaborative learning mechanism and a dominant guidance learning mechanism, which act on the non-dominated solution set and the dominated solution set respectively to achieve a deep search of the solution space; Step 6, merge the non-dominated solution set and the dominated solution set; Step 7, the scout bee stage, uses the individual restart strategy to act on the solution whose number of consecutive evolution failures exceeds the set maximum threshold to enhance the vitality of the population; Step 8, update the non-dominated solution set and determine whether the termination condition is met. If so, end the evolution and output the non-dominated solution set. Otherwise, return to step 3.

2. According to claim 1, a robot multi-objective path planning method based on a self-learning evolutionary algorithm is also characterized in that: In step 2, the hybrid initialization strategy is as follows: the first solution is generated using a heuristic algorithm, and then the remaining solutions are generated using a fast-exploring random tree algorithm and a spatial partitioning method based on the population size. Ps -1 solution; A solution represents a path, and the path is represented by a discrete sequence. Path = [ P 0, ..., P i , ..., P n+1 ], Path Represents a path, P 0 means the first node in the sequence, i.e. the starting node. P n+1 Indicates the last node in the sequence, i.e. the target node. represents an intermediate node, i Indicates the index number of the intermediate node, the value range is 1≤ i ≤ n , n Indicates the number of intermediate nodes in the path; each solution corresponds to two target values, namely path length and path security; The formula for calculating the path length is: ,in Indicates the path Path Length, Represents the Euclidean distance between two consecutive points in a path; The calculation formula of path safety is: ,in, represents the safety of the path, where is an obstacle in the configuration space, j Indicates the index number of the obstacle. m is the number of obstacles, is the minimum distance between a path and an obstacle in the configuration space.

3. The robot multi-objective path planning method based on self-learning evolutionary algorithm according to claim 2 is further characterized in that: A solution in the population is used as the first input solution of the path crossover operation, and a solution is randomly selected in the population as the second input solution. The path crossover operation includes the following steps: finding the intermediate node with the shortest distance between the two input solutions in the discrete sequence corresponding to each other, and setting the found node as the intermediate node in the first input solution. The second input solution ,in, q Check the selected node index in the sequence corresponding to the second input solution. and Whether the straight line segment between intersects with the obstacle, if not, two new solutions are generated: the first new solution is from the starting node of the sequence corresponding to the first input solution to the node , concatenate the sequence corresponding to the second input solution from node to the target node; the second new solution consists of the starting node to the node of the sequence corresponding to the second input solution , concatenate the sequence corresponding to the first input solution from node To the target node, the two input solutions and the two new solutions are non-dominated sorted, the non-dominated solutions are selected, and one solution is randomly selected for output; if they intersect, the first input solution is directly output.

4. The robot multi-objective path planning method based on self-learning evolutionary algorithm according to claim 3 is also characterized in that: The solution output by the path crossover operation is used as the input solution of the path mutation operation. The steps of the path mutation operation are to set a counter to record the number of consecutive mutation failures and set its initial value to 0. Randomly select an intermediate node in the sequence corresponding to the input solution. , and determine the node Minimum distance to surrounding obstacles d min , in the node is the center of the circle, d min Generate a new node randomly within a circle of radius P new ; Check node P new With Node and nodes Whether the straight line segment between nodes intersects with obstacles, if not, the new node P new Alternative Node A new solution is constructed, and the target value of the new solution is calculated. If the target value of the new solution is less than the target value of the input solution, the new solution is output and the path mutation operation ends. If the target value of the new solution is higher than the target value of the input solution, the counter is incremented by one and a new node is generated. If they intersect, the counter is incremented by one and a new node is generated. When the counter reaches the maximum number of consecutive mutation failures, C mf And no new solution with a smaller objective value is generated, the input solution is output.

5. The robot multi-objective path planning method based on self-learning evolutionary algorithm according to claim 4 is further characterized in that: The solution output by the path mutation operation is used as the input solution of the path shortening operation. The steps of the path shortening operation are as follows: set the input solution to the new solution; randomly select two non-adjacent intermediate nodes in the sequence corresponding to the new solution. P x and P y , check the node P x and P y Whether the straight line segment between them intersects with obstacles; If they do not intersect, delete the node P x and P y The path shortening operation ends when a collision occurs between two nodes, and a new solution is output. If there is a collision, non-adjacent intermediate nodes are randomly selected until all two non-adjacent nodes are traversed and a new solution is output.

6. The robot multi-objective path planning method based on self-learning evolutionary algorithm according to claim 5 is further characterized in that: The solution output by the path mutation operation is used as the input solution of the path safety operation. The steps of the path safety operation are: in the sequence corresponding to the input solution, two adjacent nodes form a line segment, and find the line segment closest to the obstacle. , and determine the line segment The position closest to the obstacle is set as the node P nearest ,Sure P nearest The node on the surface closest to the obstacle P obstacle ; Compute nodes P nearest With Node P obstacle The distance between d min , in the node P nearest As a starting point, is the direction, distance from the node P nearest d min Create a new node at the location P new , check the node P new With Node and nodes Whether the straight line segment between them intersects with obstacles; If they do not intersect, move the node P new Insert into node and nodes Between them, a new solution is formed and output; if they intersect, the input solution is output.

7. The robot multi-objective path planning method based on self-learning evolutionary algorithm according to claim 6 is further characterized in that: The learning mechanism based on collaboration is applied to the non-dominated solution set, and the learning mechanism based on dominance guidance is applied to the dominated solution set, where: The collaborative learning mechanism is to calculate the target value of each solution in the non-dominated solution set, and normalize these target values ​​using the following formula according to the target type: It refers to k The solution l target value, in addition, and Represents all solutions l The maximum and minimum values ​​of the targets, therefore, Indicates k The solution l The result of normalizing the target values; perform the following steps on each solution in the non-dominated solution set: if the current solution obtains the minimum value in path length or path safety, the path shortening operation is preferentially performed on it to generate a new solution; then the path safety operation is applied to the generated solution to further improve the safety of the solution; for the solution that does not obtain the minimum value in path length or path safety, the optimization operation is performed according to the size of the normalized value: if the normalized value of the path length is higher than the normalized value of the path safety, the path shortening operation is performed on the current solution; otherwise, the path safety operation is performed to obtain the optimized non-dominated solution; if the optimized non-dominated solution dominates the non-dominated solution before optimization, the non-dominated solution before optimization is replaced by the optimized non-dominated solution, and the optimized non-dominated solution set is generated and output; The learning mechanism based on domination guidance is that for each dominated solution, a non-dominated solution is randomly selected from the non-dominated solution set for comparison; if the path length and path safety of the non-dominated solution are both smaller than the path length and path safety of the current dominated solution, the path shortening operation and the path safety operation are performed on the current dominated solution in sequence; if the path length of the non-dominated solution is smaller than the path length of the current dominated solution, the path shortening operation is performed on the current dominated solution; otherwise, the path safety operation is performed on the current dominated solution to generate an optimized dominated solution. If the path length and path safety of the optimized dominated solution are smaller than the dominated solution before optimization, the optimized dominated solution is used to replace the dominated solution before optimization, and the optimized dominated solution set is generated and output.

8. The robot multi-objective path planning method based on self-learning evolutionary algorithm according to claim 7 is further characterized in that: The individual restart strategy is to perform non-dominated sorting on the population generated in the follower bee stage, and divide the population into a non-dominated solution set and a dominated solution set; for each solution in the dominated solution set, check whether the number of consecutive evolution failures exceeds the set maximum threshold. If it meets the threshold, replace the current dominated solution with a new solution, and randomly select any of the following methods to generate a new solution: one is to select two solutions from the non-dominated solution set, one of which is a non-dominated solution that achieves the minimum value in path length or path safety, and the other is a random non-dominated solution. Path intersection operations are performed on these two solutions to generate new solutions; the other is to use the fast random tree algorithm to generate five new solutions, and select the solution with the shortest path length as the new solution.

Images