An iterative optimization method for multi-vehicle assembly configuration in off-road terrain

Abstract

A multi-vehicle combination configuration iterative optimization method for outdoor terrain includes the following steps: (1) obtaining elevation map environment information and establishing a 3D terrain model; calculating the height difference between two adjacent points in the map; determining attributes and categories through feature analysis, generating corresponding sparse semantic maps, extracting detailed obstacle parameters, and obtaining accurate map matrix information; (2) obtaining the initial path; (3) encoding the configuration and determining the association matrix; the unique association matrix corresponds to the unique configuration; (4) performing a genetic algorithm; (5) displaying the configuration in a MATLAB visual GUI interface. This method combines a pre-set objective function and iterates multiple times to find a set of values ​​with the highest fitness to obtain the best unique configuration. This method improves the previous problems of incomplete map data processing, non-optimal optimal path, non-optimal configuration, and weak obstacle crossing ability. It aims to pass through complex terrain smoothly, energy-saving, safely, and quickly, laying the foundation for subsequent control movement.

Description

Technical Field

[0001] The invention relates to configuration optimization based on genetic algorithms, and in particular to a method for a multi-vehicle assembly to automatically find an optimal configuration to overcome obstacles in a complex mountain environment. Background Art

[0002] The four-wheeled vehicle platform is currently an important technical carrier in the field of robots. It undertakes many specific tasks, especially in complex mountain obstacle crossing. In the face of complex outdoor mountains, when a single four-wheeled vehicle platform cannot successfully pass through the terrain to complete the task, the advantages of the multi-vehicle combination are immediately apparent: when multiple four-wheeled vehicles are connected by hinged connecting rods, they pull and control each other during movement, while enhancing the stability and stability of the structure, so that multiple vehicles can complete the obstacle crossing task that a single vehicle cannot complete in complex terrain. The performance and advantages of the multi-vehicle combination under the premise of unmanned autonomy are greatly improved, and a more flexible unmanned autonomy purpose is achieved.

[0003] Furthermore, how to obtain the optimal configuration for different complex terrain conditions is a problem in the field of robotics research. The current configuration optimization technology is not sufficient to obtain the best configuration, or there are various problems in obtaining the configuration, which greatly reduces the feasibility and stability of the multi-vehicle combination configuration and makes it difficult to complete subsequent forward work. Summary of the invention

[0004] In view of the problem that it is difficult for a multi-vehicle combination to generate an optimal configuration to complete obstacle crossing in complex terrain, the present invention provides an iterative optimization method for the configuration of a multi-vehicle combination for field terrain based on a genetic algorithm. A sparse semantic map is obtained by processing and transforming the elevation map, and parameter constraints are extracted and traversed to find the optimal path. The genetic algorithm is combined to optimize the configuration parameters, and the configuration encoding is converted into a one-to-one association matrix. Finally, a simplified configuration simulation diagram is visualized.

[0005] The multi-vehicle assembly configuration iteration optimization method for outdoor terrain of the present invention, the multi-vehicle assembly refers to M known four-wheeled modular robot platforms, each platform is simplified into a rectangle, and only the left side and the upper side of the four sides are connecting rods, which can actively dock with the connected module grooves; the right side and the lower side are connected grooves; the multi-vehicle assembly is combined according to the characteristics of the terrain, so as to realize the selection and change of the configuration, which specifically includes the following steps:

[0006] (1) Obtain elevation map environmental information and build a 3D terrain model;

[0007] Combined with the actual mountain elevation terrain database, the overall contour data and profile map and other file information are imported, and the map parameters are obtained through data analysis, that is, the height difference between two adjacent points in the map is calculated; the attributes and categories are determined through feature analysis, the corresponding sparse semantic map is generated, and the detailed obstacle parameters are extracted to obtain accurate map matrix information.

[0008] The method to calculate the height difference between two adjacent points in the map is:

[0009] Suppose the actual elevation map is converted into a sparse semantic map, and the corresponding matrix element of the semantic map is the average height value of the point. The obstacle part has I horizontal elements, that is, there are I starting points. Each route has J longitudinal elements, that is, there are J waypoints. PATH[i] represents the i-th starting element, and traversing each element represents a starting point of the path. PATH[j] represents the j-th path element, and each element represents a waypoint in the path. Each route calculates the height difference between the next point PATH[j+1] and the current point PATH[j], and compares it with the next height difference. Then the maximum and minimum height differences under the terrain are obtained, and compared with the critical longitudinal height difference under the terrain friction factor to obtain the obstacle attributes.

[0010] The terrain analysis of obstacle properties includes:

[0011] ① High wall: Calculate the vertical height difference between two adjacent points. When the minimum height of the obstacle is greater than 60 units (dimensionless, corresponding to the map matrix value), it is considered to be a passable high wall obstacle. When (high wall height / single vehicle length)*single vehicle weight ≤ obstacle surface dynamic friction coefficient*single vehicle weight*configuration length, and high wall height <configuration length, the obtained high wall height is regarded as the map parameter condition.

[0012] ② Gentle slope: Calculate the longitudinal height difference between two adjacent points. When the maximum height of the obstacle is less than 60 units (dimensionless, corresponding to the map matrix value), it is considered to be a passable gentle slope obstacle. When the dynamic friction coefficient ≤ the tangent value of the slope angle, and the high wall height < configuration length * sine value of the slope angle, the obtained gentle slope height is regarded as the map parameter condition.

[0013] ③Gully: Calculate the vertical height difference between two adjacent points. When the heights of the two points are the same and the vertical distance is less than the maximum vertical length of the combination, it is considered that the gully obstacle can be passed. The vertical height difference obtained is regarded as the map parameter condition.

[0014] ④ Canyon: When the maximum horizontal height difference is less than 45 units (dimensionless, corresponding to the matrix value of the map), it is considered to be a canyon obstacle. The minimum horizontal width is regarded as the parameter condition.

[0015] (2) Obtaining the initial path;

[0016] By calculating the maximum vertical height difference (height) and the minimum horizontal height difference (width) between two adjacent points, the route with the lowest cost among the available routes is found, the starting point is determined, the final cost (COST) of crossing the obstacle is calculated, and the 3D modeling of the optimal path is obtained;

[0017] According to the matrix information in (1), the path is explored to determine the maximum obstacle slope of the terrain and the minimum lateral width on both sides of the path, as well as other obstacle parameters. These serve as the constraint set for the subsequent genetic algorithm, and the best starting point is determined. The straight lines are connected to determine the passing route.

[0018] The method for calculating the maximum vertical height difference between two adjacent points in the map is:

[0019] Assume that the starting point and the end point of the complex combined map are known. When entering the obstacle from the flat ground, the obstacle part has I horizontal elements, that is, there are I starting points. Each route has J longitudinal elements, that is, there are J waypoints. Given the critical slope "θ" under the terrain friction factor. PATH[i] represents the i-th starting element, and each element represents a starting point of the path. PATH[j] represents the j-th way element, and each element represents a waypoint in the path. Each route calculates the height difference between the next point PATH[j+1] and the current point PATH[j]. If each height difference of the route is less than "θ", it is considered to be accessible through i and stored in P; until the height difference is greater than "θ", it is considered that this road is blocked, and i=i+1 continues to traverse the next route horizontally. Finally, all "accessible" routes P[i] are stored and compared. The route i with the smallest maximum height difference in the route is considered to be the road with the lowest cost. The cost function is: Q=a*y(max). Where a is the longitudinal cost weight under the corresponding terrain, and y(max) is the maximum longitudinal height difference among all accessible routes. It can be seen that the route cost is positively correlated with the height difference. That is, determine the final starting point, let location = i, the coordinates are start = GGCC (1, location), and the end point coordinates are goal = GGCC (J, location). The maximum vertical height difference of the route is obtained and stored as "height". At the same time, the total length L of the route is obtained, which is the configuration length constraint of the subsequent genetic algorithm.

[0020] The method for calculating the minimum horizontal height difference between two adjacent points in the map is:

[0021] Taking location as the starting point, the route has J elements. Search to both sides until the edge ends. The number of steps on the left + the number of steps on the right = (I-1), that is, for each path point, solve the slope on both sides. Given the critical slope "ω" under the terrain friction factor. First search to the left, when the lateral slope is less than "ω", width = width + 1; when the left side is greater than "ω", the search on the left side stops and the current search width width is stored; then search to the right side, when the lateral slope is less than "ω", repeat width = width + 1; when the right side is greater than "ω", stop searching and store the current search steps "width"; the search on both sides ends to obtain the final width and store the data. Continue the iterative loop, then j = j + 1, repeat the above operation until all the widths on both sides of all the steps of the path are obtained, and the minimum lateral height difference is compared to obtain the minimum width, which is recorded as the minimum lateral accommodation width x (min), and the lateral cost weight is recorded as b (the longitudinal cost a accounts for a larger proportion, and the lateral cost b is relatively smaller). Width is recorded as the configuration width constraint of the subsequent genetic algorithm.

[0022] So far, the basic passing road parameters are obtained, and the total cost of passing the obstacle (the final cost of crossing the obstacle) is

[0023] (3) Encode the configuration and determine the correlation matrix; a unique correlation matrix for a unique configuration;

[0024] The encoding rules of the association matrix are as follows: Initialize an M*M zero square matrix, with the first to Mth rows corresponding to the car numbers, and the first to Mth columns corresponding to the connected car numbers. When the left side of car i is connected to the right side of car j, the corresponding position (i, j) of the association matrix is ​​set to 2, and (j, i) is still 0; when the upper side of car i is connected to the lower side of car j, the corresponding position (i, j) of the association matrix is ​​set to 1, and (j, i) is still 0. By analogy, we finally get the association matrix corresponding to the car number and connection method, which corresponds to the configuration one by one according to the rule. According to this rule, it can be input into the GUI visualization interface of MATLAB to realize the mutual conversion from matrix to configuration.

[0025] (4) Genetic algorithm, the process includes:

[0026] ① Design objective function y = h*(x(1)+L)-k*x(2); x(1) is the configuration length, x(2) is the configuration width, and y is the positive correlation of the time required to pass the obstacle. The three configurations are the output of the genetic algorithm. Optimize the target obstacle crossing time, while making y as small as possible, and also ensure that the width is moderate, taking into account stability and passability. The function weight coefficients h and k are changed according to the actual situation of different terrains, and L is the total length of the obstacle path.

[0027] ② Constraint condition set: Use the parameters height, width, and COST obtained under different terrains in step (2) and take the union as the overall constraint to input into the objective function constraint set.

[0028] ③ Initialize individuals, randomly generate individuals, select the best chromosomes, and then calculate the fitness. The formula is min(y).

[0029] ④Evolution process: select, cross, and mutate to obtain more chromosome possibilities, and calculate the fitness function respectively. Find the chromosomes with the minimum and maximum fitness and their population positions, and replace the best chromosomes in the last evolution, which is considered as continuous evolution.

[0030] ⑤ The results show that when the iteration reaches 800 generations, the curve converges completely. The figure image depicts: function value curve; termination algebra; evolution algebra; function value; variable. Output x(1), x(2), y, and get the optimal solution.

[0031] (5) MATLAB visual GUI interface configuration display;

[0032] The optimized configuration length x(1) and width x(2) parameters are transformed and input into an M*M zero matrix. The required connection positions are set to 1 or 2 at the corresponding positions to obtain the corresponding association matrix form. A GUI visualization page is designed, which is divided into an association matrix input part, a visualization confirmation button part, and a configuration display part.

[0033] The addition of multiple precise constraints in the above process can improve the optimization effect of the genetic algorithm. The genetic algorithm analyzes the required constraints from the map parameters and scales, combines the pre-set objective function, and iterates multiple times to find a set of values ​​with the largest fitness, which is the optimal solution. Then, the optimal solution parameters are converted into a visual configuration and simulated in conjunction with Vrep. So far, a multi-vehicle configuration optimization method for mountain obstacle crossing based on genetic algorithms has been obtained.

[0034] The present invention converts any actual elevation map into a 3D map model on the premise of obtaining it, and simultaneously obtains an accurate grid map as elevation map information preparation; at the same time, the present invention can combine with a genetic algorithm to convert terrain conditions into constraints of the genetic algorithm, and combine with a pre-set objective function to iterate multiple times to find a set of values ​​with the largest fitness, which is the optimal solution, thereby obtaining the optimal configuration of the combination.

[0035] The optimal path selection method of the present invention proposes to convert the dense elevation map into a sparse semantic map, and obtain map parameter constraints by calculating parameters such as height difference and slope value, and then traverse the route to obtain the path with the lowest cost, which is recorded as the optimal path; then, for known map data, iterative configuration optimization based on genetic algorithm is proposed. The connection form of the vehicle body and the configuration is encoded, displayed in the form of an association matrix, and combined with GUI visualization. Therefore, the best and only configuration is obtained, which improves the previous problems of incomplete map data processing, non-optimal optimal path, non-optimal configuration, and weak obstacle crossing ability. It aims to pass through complex terrain smoothly, energy-saving, safely, and quickly, laying the foundation for subsequent control movement. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The present invention is an overall flow chart of the iterative optimization method for multi-vehicle combination configuration for outdoor terrain.

[0037] Figure 2 A schematic diagram for obtaining the optimal path.

[0038] Figure 3 Schematic diagram of the configuration corresponding to the 6*6 association matrix.

[0039] Figure 4 Generate a visualization diagram for the GUI interface association matrix. DETAILED DESCRIPTION

[0040] Given M four-wheeled modular robot platforms, each platform is simplified into a rectangle, and only the left side and the upper side of the four sides are connecting rods, which can actively dock with the connected module grooves; the right side and the lower side are connected grooves. On this basis, combinations are made according to the characteristics of the terrain to achieve the selection and change of configuration.

[0041] Figure 1 The overall process of the multi-vehicle configuration optimization mountain obstacle crossing method based on genetic algorithm of the present invention is given, which specifically includes the following steps.

[0042] 1. Obtain elevation map environment information and build a 3D terrain model

[0043] Combined with the actual mountain elevation terrain database, the overall contour data and profile map and other file information are imported, and the map parameters are obtained through data analysis, that is, the horizontal and vertical height differences between adjacent points are calculated; the attributes and categories are determined through feature analysis, the corresponding sparse semantic map is generated, and the detailed obstacle parameters are extracted to obtain accurate map matrix information.

[0044] The method to calculate the height difference between two adjacent points in the map is:

[0045] Suppose the actual elevation map is converted into a sparse semantic map, and the corresponding matrix element of the semantic map is the average height value of the point. The obstacle part has I horizontal elements, that is, there are I starting points. Each route has J longitudinal elements, that is, there are J waypoints. PATH[i] represents i starting elements, and traversing each element represents a starting point of the path. PATH[j] represents the jth path element, and each element represents a waypoint in the path. Each route calculates the height difference between the next point PATH[j+1] and the current point PATH[j], and compares the next height difference. Then the maximum and minimum height differences under the terrain are obtained, and compared with the critical longitudinal height difference under the following four terrain friction factors, the obstacle attributes can be obtained.

[0046] The attribute analysis and parameter extraction transformation of four common terrain obstacles are as follows:

[0047] 1. High wall: Calculate the vertical height difference between two adjacent points. When the minimum height of the obstacle is greater than 60 units (dimensionless, corresponding to the map matrix value), it is considered to be a passable high wall obstacle. When (high wall height / single vehicle length)*single vehicle weight ≤ obstacle surface dynamic friction coefficient*single vehicle weight*configuration length, and high wall height <configuration length, the obtained high wall height is regarded as the map parameter condition.

[0048] 2. Gentle slope: Calculate the longitudinal height difference between two adjacent points. When the maximum height of the obstacle is less than 60 units, it is considered to be a passable gentle slope obstacle. When the dynamic friction coefficient ≤ the tangent value of the slope angle, and the high wall height < configuration length * sine value of the slope angle, the obtained gentle slope height is regarded as the map parameter condition.

[0049] 3. Gully: Calculate the vertical height difference between two adjacent points. When the heights of the two points are the same and the vertical distance is less than the maximum vertical length of the combination, it is considered that the gully obstacle can be passed. The vertical height difference obtained is regarded as the map parameter condition.

[0050] 4. Canyon: When the maximum lateral slope is less than 45 units, it is considered to be passable. The minimum lateral width is regarded as the parameter condition.

[0051] 2. Get the initial path

[0052] The whole process is like Figure 2 shown.

[0053] By calculating the maximum longitudinal height difference and the minimum lateral height difference between two adjacent points, the route with the lowest cost among the traversable routes is found, the starting point is determined, the final cost of crossing the obstacle is calculated, and the optimal path 3D modeling is obtained.

[0054] According to the matrix information in step 1, the path is explored to determine the maximum obstacle slope of the terrain, the minimum horizontal width on both sides of the path, and other obstacle parameters, which serve as the constraint set of the subsequent genetic algorithm. The best starting point is determined and the straight line is connected to determine the passing route.

[0055] The method for calculating the maximum vertical height difference between two adjacent points in the map is:

[0056] Assume that the starting point and the end point of the complex combined map are known. When entering the obstacle from the flat ground, the obstacle part has I horizontal elements, that is, there are I starting points. Each route has J longitudinal elements, that is, there are J waypoints. Given the critical slope "θ" under the terrain friction factor. PATH[i] represents the i-th starting element, and each element represents a starting point of the path. PATH[j] represents the j-th way element, and each element represents a waypoint in the path. Each route calculates the height difference between the next point PATH[j+1] and the current point PATH[j]. If each height difference of the route is less than "θ", it is considered to be accessible through i and stored in P; until the height difference is greater than "θ", it is considered that this road is blocked, and i=i+1 continues to traverse the next route horizontally. Finally, all "accessible" routes P[i] are stored and compared. The route i with the smallest maximum height difference in the route is considered to be the road with the lowest cost. The cost function is: Q=a*y(max). Where a is the cost weight under the corresponding terrain, and y(max) is the maximum longitudinal height difference among all accessible routes. It can be seen that the route cost is positively correlated with the height difference. That is, determine the final starting point, let location = i, the coordinates are start = GGCC (1, location), and the end point coordinates are goal = GGCC (J, location). The maximum vertical height difference of the route is obtained and stored as "height". At the same time, the total length L of the route is obtained, which is the configuration length constraint of the subsequent genetic algorithm.

[0057] The method for calculating the minimum horizontal height difference in the map is:

[0058] Taking location as the starting point, the route has J elements. Search to both sides until the edge ends. The number of steps on the left + the number of steps on the right = (I-1), that is, for each path point, solve the slope on both sides. Given the critical slope "ω" under the terrain friction factor. First search to the left, when the lateral slope is less than "ω", width = width + 1; when the left side is greater than "ω", stop searching on the left and store the current search width width; then search to the right, when the lateral slope is less than "ω", repeat width = width + 1; when the right side is greater than "ω", stop searching and store the current search steps "width"; the search on both sides ends to obtain the final width and store the data. Continue the iterative loop, then j = j + 1, repeat the above operation until all the widths on both sides of all the steps of the path are obtained, and the minimum lateral height difference is compared to obtain the minimum width, which is recorded as the minimum lateral accommodation width x (min), and the cost weight is recorded as b. Width is recorded as the configuration width constraint of the subsequent genetic algorithm.

[0059] So far, the basic passing road parameters are obtained, and the total cost of passing obstacles is The vertical cost a accounts for a larger proportion, while the horizontal cost b is relatively smaller.

[0060] 3. Encode the configuration and determine the correlation matrix

[0061] Taking 6 modules as an example, the association matrix is ​​a 6*6 square matrix, such as Figure 3 shown.

[0062] To perform genetic algorithm calculations, the configuration must first be encoded. The unique association matrix can uniquely correspond to the length, width, midpoint coordinates and other data of the configuration for the unique configuration.

[0063] The encoding rules of the association matrix are as follows: initialize a 6*6 zero square matrix, the first to sixth rows correspond to the car numbers, and the first to sixth columns are the connected car numbers. When the left side of car i is connected to the right side of car j, the corresponding position (i, j) of the association matrix is ​​set to 2, and (j, i) is still 0; when the upper side of car i is connected to the lower side of car j, the corresponding position (i, j) of the association matrix is ​​set to 1, and (j, i) is still 0. And so on, finally get the association matrix corresponding to the car number and connection mode, according to the rule, it corresponds to the configuration one by one. According to this rule, it can be input into the GUI visualization interface of MATLAB to realize the mutual conversion from matrix to configuration.

[0064] 4. Genetic Algorithm Process

[0065] Genetic algorithm is a search algorithm based on natural selection and population genetic mechanism. It simulates the reproduction, hybridization and mutation phenomena in natural selection and natural genetic process. When using genetic algorithm to solve this problem, each possible solution of the configuration is encoded into a "chromosome", that is, an individual, and several individuals constitute a group. The multi-car configuration is described by the association matrix so that different association matrices uniquely correspond to one configuration. At the beginning of the genetic algorithm, some individuals (initial solutions) are randomly generated, and each individual is evaluated according to the predetermined objective function to give a fitness value. Based on this fitness value, some individuals are selected to produce the next generation. The selection operation embodies the principle of "survival of the fittest". "Good" configuration individuals are used to produce the next generation, and "bad" configuration individuals are eliminated, so that they gradually evolve towards the optimal solution of the configuration that adapts to the map. Therefore, the genetic algorithm can be regarded as a process of initial evolution of a group composed of feasible solutions.

[0066] 1. Design the objective function y = h*(x(1)+L)-k*x(2); where x(1) is the configuration length, x(2) is the configuration width, and y is the positive correlation of the time required to pass the obstacle. The three configurations are the output of the genetic algorithm. Optimize the target obstacle crossing time, while making y as small as possible, and also ensure that the width is moderate, taking into account stability and passability. The function weight coefficients h and k are changed according to the actual situation of different terrains, and L is the total length of the obstacle path.

[0067] 2. Constraint condition set: Use the parameters height, width, and COST obtained in step (2) under different terrain conditions, and take the union as the overall constraint to input into the objective function constraint set.

[0068] 3. Initialize the individuals, randomly generate individuals, select the best chromosomes and then calculate the fitness. The formula is min(y).

[0069] 4. Evolution process: select, cross, and mutate to obtain more chromosome possibilities, and calculate the fitness function respectively. Find the chromosomes with the minimum and maximum fitness and their population positions, and replace the best chromosomes in the last evolution, which is considered as continuous evolution.

[0070] 5. The results show that when the iteration reaches 800 generations, the curve converges completely. The figure image depicts: function value curve; termination algebra; evolution algebra; function value; variable. Output x(1), x(2), y, and get the optimal solution.

[0071] 5. MATLAB Visual GUI Interface Configuration Display

[0072] The optimized configuration length x(1) and width x(2) parameters are transformed and input into the M*M zero matrix. The positions of the required connections are set to 1 or 2 at the corresponding positions to obtain the corresponding association matrix form. The GUI visualization page is designed, which is divided into the association matrix input part, the visualization confirmation button part, and the obtained configuration display part, as shown in Figure 4 shown.

Claims

1. A multi-vehicle assembly configuration iterative optimization method for outdoor terrain, wherein the multi-vehicle assembly refers to M known four-wheeled modular robot platforms, each platform is simplified into a rectangle, and only the left side and the upper side of the four sides are connecting rods, which actively dock with the grooves of the connected modules, and the right side and the lower side are connected grooves. The multi-vehicle assembly is combined according to the characteristics of the terrain, thereby realizing the selection and change of the configuration; its characteristics are: The following steps are involved: (1) Obtain elevation map environmental information and build a 3D terrain model; Combined with the actual mountain elevation terrain database, the overall contour data and profile map file information are imported, and map parameters are obtained through data analysis, that is, the height difference between two adjacent points in the map is calculated; Determine attributes and categories through feature analysis, generate corresponding sparse semantic maps, extract detailed obstacle parameters, and obtain accurate map matrix information; (2) Obtaining the initial path; By calculating the maximum longitudinal height difference and the minimum lateral height difference between two adjacent points, the route with the lowest cost among the available routes is found, the starting point is determined, the final cost of crossing obstacles is calculated, and the 3D modeling of the optimal path is obtained; According to the matrix information in step (1), the path is explored to determine the maximum obstacle slope of the terrain and the minimum lateral width on both sides of the path and other obstacle parameters. These are used as the constraint set of the subsequent genetic algorithm, and the best starting point is determined. The straight lines are connected to determine the passing route; (3) Encode the configuration and determine the correlation matrix; a unique correlation matrix for a unique configuration; (4) Perform genetic algorithm; (5) MATLAB visual GUI interface configuration presentation.

2. The method for iterative optimization of multi-vehicle assembly configuration for outdoor terrain according to claim 1 is characterized in that: The method for calculating the height difference between two adjacent points in the map in step (1) is: Suppose the actual elevation map is converted into a sparse semantic map. The corresponding matrix element of the semantic map is the average height value of the point. The obstacle part has I horizontal elements, that is, there are I starting points. Each route has J longitudinal elements, that is, there are J waypoints. PATH[i] represents the i-th starting element. Traversing each element represents a starting point of the path. PATH[j] represents the j-th way element. Each element represents a waypoint in the path. For each route, the height difference between the next point PATH[j+1] and the current point PATH[j] is calculated and compared with the next height difference. Then the maximum and minimum height differences under the terrain are obtained, and compared with the critical longitudinal height difference under the terrain friction factor to obtain the obstacle attributes.

3. The method for iterative optimization of multi-vehicle assembly configuration for outdoor terrain according to claim 2 is characterized in that: The terrain analysis of obstacle properties includes: ① High wall: Calculate the longitudinal height difference between two adjacent points. When the minimum height of the obstacle is greater than 60 units, it is considered that the high wall obstacle can be passed. When (high wall height / single vehicle length)*single vehicle weight ≤ obstacle surface dynamic friction coefficient*single vehicle weight*configuration length, and the high wall height <configuration length, the obtained high wall height is regarded as the map parameter condition. ② Gentle slope: Calculate the longitudinal height difference between two adjacent points. When the maximum height of the obstacle is less than 60 units, it is considered to be a passable gentle slope obstacle. When the dynamic friction coefficient ≤ the tangent value of the slope angle, and the high wall height < configuration length * sine value of the slope angle, the obtained gentle slope height is regarded as the map parameter condition; ③Gully: Calculate the vertical height difference between two adjacent points. When the heights of the two points are the same and the vertical distance is less than the maximum vertical length of the combination, it is considered that the gully obstacle can be passed, and the obtained vertical height difference is regarded as the map parameter condition; ④ Canyon: When the maximum lateral height difference is less than 45 units, it is considered that the canyon obstacle can be passed, and the minimum lateral accommodation width obtained is regarded as the parameter condition.

4. The method for iterative optimization of multi-vehicle assembly configuration for outdoor terrain according to claim 1 is characterized in that: The method for calculating the maximum vertical height difference between two adjacent points in the map in step (2) is: Assume that the starting point and the end point of the complex combination map are known. When entering the obstacle from the flat ground, assume that the obstacle part has I lateral elements, that is, there is I starting point; Each route has J longitudinal elements, that is, J waypoints; given the critical slope "θ" under the terrain friction factor; PATH[i] represents the i-th starting element, each element represents a starting point of the path; PATH[j] represents the j-th path element, each element represents a waypoint in the path, and each route calculates the height difference between the next point PATH[j+1] and the current point PATH[j]. If each height difference of the route is less than "θ", it is considered to be passable and stored in P. Until the height difference is greater than "θ", it is considered that this road is blocked, i=i+1 continues to traverse the next route horizontally; finally, all "passable" routes P[i] are stored and compared The route i with the smallest maximum height difference in the route is regarded as the road with the smallest cost, and the cost function is: Q = a*y(max); where a is the cost weight under the corresponding terrain, and y(max) is the maximum longitudinal height difference among all passable routes; the route cost is positively correlated with the height difference, that is, the final starting point is determined, location = i, the coordinates are marked as start = GGCC(1,location), and the end point coordinates are goal = GGCC(J,location). The maximum longitudinal height difference of the route is obtained and stored as "height". At the same time, the total length L of the route is obtained, which is the configuration length constraint of the subsequent genetic algorithm.

5. The method for iterative optimization of multi-vehicle assembly configuration for outdoor terrain according to claim 1 is characterized in that: The method for calculating the minimum horizontal height difference between two adjacent points in the map in step (2) is: Taking location as the starting point, the route has J elements, and the search ends on both sides, the number of steps on the left + the number of steps on the right = (I-1), that is, for each waypoint, the slope is solved on both sides respectively, given the critical slope "ω" under the terrain friction factor, first search to the left, when the lateral slope is less than "ω", then width = width + 1; when the left side is greater than "ω", the search on the left side is stopped, and the current search width width is stored; then search to the right side, when the lateral slope is less than "ω", repeat width = width + 1; when the right side is greater than "ω", the search is stopped, and the current search step "width" is stored; the search on both sides ends to obtain the final width, and the data is stored; continue the iterative loop, then j = j + 1, repeat the above operation until all the widths on both sides of all the steps of the path are obtained, and the minimum lateral height difference is compared to obtain the minimum width, which is recorded as the minimum lateral accommodation width x (min), and the cost weight is recorded as b; width is recorded as the configuration width constraint of the subsequent genetic algorithm.

6. The multi-vehicle assembly configuration iterative optimization method for outdoor terrain according to claim 1 is characterized by: The final cost of crossing the obstacle in step (2) is Where a is the longitudinal cost weight under the corresponding terrain, b is the lateral cost weight, y(max) is the maximum longitudinal height difference among all passable routes, and x(min) is the minimum lateral width for passage.

7. The method for iterative optimization of multi-vehicle assembly configuration for outdoor terrain according to claim 1 is characterized in that: The correlation matrix encoding rules in step (3) are as follows: Initialize an M*M zero matrix, where the first to Mth rows correspond to the car numbers, and the first to Mth columns are the connected car numbers; when the left side of car i is connected to the right side of car j, the corresponding position (i, j) of the association matrix is ​​set to 2, and (j, i) is still 0; when the upper side of car i is connected to the lower side of car j, the corresponding position (i, j) of the association matrix is ​​set to 1, and (j, i) is still 0; and so on, finally we get the association matrix corresponding to the car number and connection method, which corresponds to the configuration one by one according to the rule.

8. The method for iterative optimization of multi-vehicle assembly configuration for outdoor terrain according to claim 1 is characterized in that: The process of the genetic algorithm in step (4) includes: ① Design objective function y = h*(x(1)+L)-k*x(2); x(1) is the configuration length, x(2) is the configuration width, and y is the positive correlation of the time required to pass the obstacle. The three configuration genetic algorithms are the output results; the function weight coefficients h and k are changed according to the actual conditions of different terrains, and L is the total length of the obstacle path; ② Constraint condition set: Use the parameters height, width, and COST obtained in step (2) under different terrain conditions, and take the union as the overall constraint to input into the constraint set of the objective function; height is the maximum vertical height difference, width is the minimum horizontal height difference, and COST is the final cost of crossing the obstacle; ③ Initialize individuals, randomly generate individuals, select the best chromosomes and then calculate the fitness. The formula is min(y); ④Evolution process: select, cross, and mutate to obtain more chromosome possibilities, calculate the fitness function respectively, find the chromosomes with the minimum and maximum fitness and their population positions, and replace the best chromosomes in the last evolution, which is considered as continuous evolution; ⑤The results show that when the iteration reaches 800 generations, the curve converges completely, and the figure image depicts: function value curve; termination algebra; evolution algebra; function value; variable; output x(1), x(2), y, and the optimal solution is obtained.

9. The method for iterative optimization of multi-vehicle assembly configuration for outdoor terrain according to claim 1 is characterized by: The process of displaying the MATLAB visual GUI interface configuration in step (5): The optimized configuration length x(1) and width x(2) parameters are transformed and input into an M*M zero matrix. The required connection positions are set to 1 or 2 at the corresponding positions to obtain the corresponding association matrix form. A GUI visualization page is designed, which is divided into an association matrix input part, a visualization confirmation button part, and a configuration display part.

Images