Sub-multiple unmanned platform task allocation and cooperative planning method based on region segmentation

Abstract

This invention discloses a task allocation and collaborative planning method for multiple unmanned platforms based on region segmentation. By embedding the target point region with Voronoi tessellation, the target point is divided into several optimal regions. After appropriately merging dense parts, these regions are allocated as the working areas of the mother-daughter vehicle system, improving the working efficiency of the unmanned vehicle system under high-load environments. The method employs a mother-daughter vehicle configuration, where the mother vehicle carries the goods and transports the child vehicles to parking points in each region. The child vehicles are responsible for traveling to the target point and completing the predetermined tasks there, maximizing robot scheduling efficiency, fully utilizing robot endurance, and significantly improving the practicality of the multi-unmanned vehicle cluster system. Furthermore, a conflict-based search (CBS) algorithm is adopted and improved, giving the algorithm good scalability and high computational speed for large inputs, while avoiding collisions between unmanned vehicles, ensuring safety and reliability during task execution.

Description

Technical Field

[0001] This invention belongs to the field of robotics technology, specifically relating to a method for task allocation and collaborative planning of multiple unmanned platforms based on region segmentation. Background Technology

[0002] To address the existing centralized multi-unmanned vehicle task allocation and path planning framework, this invention incorporates a task region segmentation method based on region nesting and adopts a heterogeneous robot structure of parent and child vehicles, which greatly improves the stability and robustness of the multi-unmanned vehicle system under large map and multi-obstacle conditions.

[0003] Collaborative multi-robot systems have wide applications in real-world scenarios such as industrial manufacturing, mining, inspection, and warehousing logistics. In these scenarios, deploying a single robot to perform complex tasks consumes significant time and energy. However, multi-robot systems, by coordinating multiple robots to collaboratively execute and complete a large number of tasks, offer a flexible solution characterized by automatic reconfiguration, low failure rate, high fault tolerance, and high robustness. In practical applications, increasing the number of tasks, the size of the target map, and the distribution of obstacles on the map can all impact the accuracy of task allocation and path planning in multi-robot systems, potentially even leading to search failures. Therefore, we need a region segmentation-based RECBS-TA algorithm, which expands the applicability of the CBS-TA algorithm and reduces the number of execution units required.

[0004] Research on heterogeneous multi-robot systems using mother-daughter vehicle architecture is still in its infancy, with sporadic research findings in the field of unmanned vehicles both domestically and internationally. However, research on task allocation for mother-daughter vehicle multi-robot systems is almost nonexistent. Therefore, we summarize existing work on task allocation and path planning for multi-robot systems and discuss the possibility of applying region segmentation to solve the task allocation and path planning problems of multi-robot systems in complex situations.

[0005] Task allocation issue:

[0006] In the task allocation problem, robots are divided into multiple groups, and each group is assigned a series of tasks. Current task allocation methods for multi-robot systems are mainly divided into centralized and distributed approaches. In distributed task allocation, each robot is assigned its own task. This scheme requires neither a globally connected network nor a central server. Each robot obtains the status of neighboring robots within its observation range, compares its own execution capabilities with those of its neighbors, and assigns itself the optimal task. This method is suitable for large robot swarms in environments with weak communication. Centralized task allocation schemes use a central server to monitor the relative positions, task status, and remaining energy of each robot, and identify the robot with the highest task execution efficiency. This method is more susceptible to the effects of fragile communication environments, but compared to distributed task allocation, it can achieve consensus on task allocation. Currently, the mainstream multi-robot task allocation methods include: market-based task allocation, behavior-based task allocation, optimization-based task allocation, and cluster task allocation. Cluster task allocation strategies group similar or adjacent tasks into clusters and then assign the clusters to specific robots. This method can reduce the average maneuver distance of the robot system and is suitable for scenarios such as field search, emergency rescue, and warehouse logistics. Based on the idea of ​​cluster task allocation, we propose a task region segmentation method.

[0007] Path planning problem:

[0008] In the Multi-Robot System Path Planning (MAPF) problem, most research is based on labeled cases, where each robot's goal is pre-assigned. MAPF research typically optimizes the sum of path costs for all robots or the total time taken for all robots in the cluster to reach the goal. This problem has been proven to be NP-hard. Optimal solutions to this type of MAPF problem can be found using methods such as Conflict-Based Search (CBS), M*, combinatorial auctions, or ILP solvers. Unlabeled MAPF problems can be solved using distributed auction-based or token-based methods, but solutions cannot be computed within a user-specified suboptimal range. Our method, based on the TA-CBS algorithm, solves the unlabeled path planning problem. This approach is an extension of the CBS method, using arbitrary assignment matrices, does not require the number of tasks to be the same as the number of robots, and has good scalability for large-scale robot systems. Summary of the Invention

[0009] In view of this, the purpose of this invention is to provide a method for task allocation and collaborative planning of multiple unmanned platforms based on region segmentation, which can reduce the number of actuators required in a multi-task environment and improve the practicality of multi-unmanned vehicle systems.

[0010] A method for task allocation and collaborative planning among multiple unmanned platforms based on region segmentation includes:

[0011] Step 1: Calculate the target point set P = {p1, p2, ... p} n The Voronoi mosaic of}, where p n Let n represent the position of the target point n; divide the target point set P into m polygonal regions and solve the set covering problem to obtain the simplest polygonal tessellation set A′={a0,a1,...a1}. m};

[0012] Step 2: Use the set of geometric centers of the tessellated polygon set as the original parking position set of the mother car, and optimize it using the attraction optimization algorithm to obtain the mother car parking position set; solve the traveling salesman problem for the mother car parking positions to obtain the optimal driving route of the mother car; perform task assignment for each mother car parking point to obtain the set of target points that the child cars need to be responsible for.

[0013] Step 3: Plan the driving route of the mother car between each parking point to avoid obstacles, and then assign the tasks that the daughter car needs to perform after the mother car stops and plan the obstacle avoidance route.

[0014] Preferably, the method in step one includes:

[0015] Step 1.1: Perform Voronoi partitioning on the points within the target point set to obtain a set of Voronoi polygons. Let the vertices of the polygons be V. edges This includes two target points, V. vpoints A convex polygon and a V containing 3 or more target points vpoints A convex polygon;

[0016] Step 1.2: Determine the number of target points within the convex polygon.

[0017] If the number of target points is 2, it is insufficient to calculate the Voronoi mosaic. In this case, for V vpoints Choose any un-epiterated target point x from the two target points in the array, and find the set of nearest neighbors V to target point x. nears Let s be the line segment formed by connecting the nearest vertex to the target point x. e Find a circle R centered at the nearest vertex and with the working range of the subcarriage as its radius, and determine the line segment s. e The two intersection points between the circle and the target point are connected by a line to generate the Voronoi edge required to cover the target point. This Voronoi edge and the V... vpoints Another target point in the graph is subjected to Delauney triangulation and transformed into a Voronoi mosaic.

[0018] If the number of target points is greater than 2, then calculate the Delauney triangulation based on the target points and convert it into a Voronoi mosaic.

[0019] Step 1.3: For each target point, calculate the distance between each target point t and each vertex v in the Voronoi mosaic. Then, select the vertex v closest to the target point t as the optimal vertex and compare its result with the nearest neighbor set V of the target point t. nears The nearest vertex in the graph is compared, and the nearest vertex is taken as the nearest vertex in the final output. Then the target point t is divided into the Voronoi mosaic graph to which the nearest vertex belongs.

[0020] Step 1.4: Apply the set covering solver. For the tessellation obtained in Step 1.2, and based on the tessellation partitioning of each target point in Step 1.3, trim unnecessary polygonal regions to obtain a minimal set of polygonal regions A′={a0,...,a...} that covers the entire target point set. m}

[0021] Preferably, the method in step two includes:

[0022] Step 2.1: Optimize the parking location of the mother car: For any region in region A′, use the gravity optimization algorithm to set a gravity point G within the set of target points to be explored. p The centroid, midpoint, or starting point of the point set can be used as the gravitational point G. p It attracts target points within its area and finds the point with the greatest attraction to that area based on the attraction algorithm. The points with the greatest attraction found in all areas of A′ form the parking point set V″.

[0023] Step 2.2: Solve the Traveling Salesman Problem for the parking positions of the mother car and optimize its route: For the set of parking points of the mother car V″={v0,...,V m} and the distance between any two vertices Use a genetic algorithm to find the shortest path that traverses each vertex and returns to the set starting point;

[0024] Step 2.3: Assign subtasks to each mother car parking point to optimize the final parking position of the mother car: Substitute the polygon domain set obtained from solving the set covering problem and the mother car path sequence obtained from the genetic algorithm into the objective function f = g + h established by the A* algorithm to calculate the sub-path of the child car at each mother car parking point; where the cost function g is the cumulative distance of the child car to the target point, and the heuristic function h is the distance of the remaining target points that the child car needs to visit; when calculating each sub-path, the mother car parking point v is regarded as the starting node p. startBased on this, nodes containing the following information are constructed: cost function value g, heuristic function value h, and evaluation function value t. The specific method includes the following steps:

[0025] Step 2.3.1: Check if the target point has been traversed. For each untraversed node p... cur All target points within the working range of the sub-vehicles around the parking point can be used as p. cur (Initially p) start The corresponding candidate adjacent points, each of which may become the next target point of the child vehicle;

[0026] Step 2.3.2, starting from the starting point p start Begin by finding the current node p. cur The candidate adjacent nodes are obtained, resulting in child nodes {p1, p2, ... p... i Then, find the candidate adjacent points of each child node, and so on, to obtain several sub-vehicle path sequences composed of target points;

[0027] Step 2.3.3, Determine p cur With p start If they are equal, it means that the sub-vehicle stopped at the mother vehicle for charging and resupply, and the value of g is initialized; otherwise, the sub-vehicle accumulates mileage normally.

[0028] Step 2.3.4: For each target point in each sub-vehicle path, determine whether the distance after reaching the target point is sufficient for the sub-vehicle to return. If not, it means that the sub-vehicle's working range is insufficient to reach the target point and return safely, and the target is invalid. If it is sufficient, include the node in the path result. Repeat the above process until all target points of the sub-vehicle path have been traversed.

[0029] Step 2.3.5: Determine whether the objective function f corresponding to the sub-vehicle path starting from the current stop point is smaller than the objective function f of another path starting from the same stop point that was previously calculated. If it is smaller, it means that the current result is better than the previous result. Update the trajectory cost function value g and the evaluation function value f. In this way, traverse all paths starting from all stops and update the nodes in the list of path results that have not been traversed.

[0030] Step 2.3.6: Repeat steps 2.3.1 to 2.3.5 until every stop point in the mother car's path has been traversed. The algorithm finally outputs a set of target point sequences that minimizes the sum of the child car's objective function f.

[0031] Preferably, the method in step three includes:

[0032] The A* algorithm is used to plan the obstacle avoidance route of the mother car between each parking point, and the CBS-TA algorithm is used to assign tasks to the sub-vehicles and perform path planning. Based on the task objectives to be performed by each group of sub-vehicles, the high-level search of CBS is expanded. Two key-value pairs are added to each node in the high-level search: one describing whether the current node is a root node, and the other describing whether the current task is used for the low-level search. The CBS-TA algorithm builds a search forest and expands the root nodes as needed. First, a root node with an optimal task assignment is retrieved, without considering potential conflicts between robots. Then, the search expands to other root nodes, simultaneously creating the next root node with an optimal task assignment. The optimal task assignment for sub-robot a is calculated without considering other robots. i To target g j The cost matrix C is obtained by calculating the cost of the shortest route used. i×j This allows for the division of solutions, the removal of certain allocation schemes, or the forced addition of certain solutions, so that the allocation schemes cover the entire solution space.

[0033] The present invention has the following beneficial effects:

[0034] To address the low success rate of centralized task allocation and path planning algorithms in large-scale, multi-obstacle scenarios, this invention proposes a task allocation and collaborative planning method for multiple unmanned platforms based on region segmentation, in order to improve the stability and robustness of multi-unmanned vehicle systems in complex environments.

[0035] To address the issue that traditional CBS algorithms require a large number of robots to perform complex tasks, a mother-daughter robot-based task allocation and path planning method is proposed.

[0036] This invention divides the target point into several optimal regions by Voronoi mosaicking of the target point region, and after appropriately merging the dense parts, allocates them as the working area of ​​the mother-daughter vehicle system, thereby improving the working efficiency of the unmanned vehicle system under high load environment.

[0037] The system adopts a mother-daughter vehicle configuration, where the mother vehicle carries the goods and transports the child vehicles to parking points in each area; the child vehicles are responsible for going to the target point and completing the scheduled tasks there. This maximizes the efficiency of robot scheduling and execution, makes full use of the robot's endurance, and greatly improves the practicality of the multi-unmanned vehicle cluster system.

[0038] The collision-based search (CBS) algorithm was adopted and improved, which enabled the algorithm to have better scalability and higher computation speed for large inputs, while avoiding collisions between autonomous vehicles and ensuring safety and reliability during task execution. Attached Figure Description

[0039] Figure 1 This is a flowchart of the method of the present invention;

[0040] Figure 2 This is a schematic diagram of the target point clustering to generate the mother vehicle parking point in this invention;

[0041] Figure 3 This is a schematic diagram illustrating the optimized driving trajectory of the mother vehicle in this invention;

[0042] Figure 4 This is a schematic diagram illustrating the generation of the mother vehicle's driving trajectory according to the present invention;

[0043] Figure 5 This is a schematic diagram illustrating the generation of the vehicle's driving trajectory in this invention;

[0044] Figure 6 This is a diagram illustrating the conflict avoidance effect of the CBS system of this invention. Detailed Implementation

[0045] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0046] like Figure 1 As shown, the present invention provides a method for task allocation and collaborative planning of multiple unmanned platforms based on region segmentation. A mother vehicle carries multiple child vehicles to a parking point, and the child vehicles depart from the parking point to execute preset tasks. The task allocation and path planning specifically include the following steps:

[0047] Step 1: Calculate the target point set P = {p1, p2, ... p} n The Voronoi mosaic of}, where p n Indicate the position of target point n; divide the target point set P into m groups, each containing m... i The set of polygonal regions for each target point is used to solve the set covering problem, resulting in the simplest polygon tessellation set A′={a0,a1,...a1}. m The specific steps include the following:

[0048] Step 1.1: Perform Voronoi partitioning on the points within the target point set to obtain a set of Voronoi polygons. Let the vertices of the polygons be V. edges This includes two target points, V. vpoints A convex polygon and a V containing 3 or more target points vpoints A convex polygon.

[0049] Step 1.2: Determine the number of target points inside the convex polygon.

[0050] If the number of target points is 2, it is insufficient to calculate the Voronoi mosaic. In this case, for V... vpoints Choose any un-epiterated target point x from the two target points in the array, and find the set of nearest neighbors V to target point x. nears Let s be the line segment formed by connecting the nearest vertex to the target point x. eFind a circle R centered at the nearest vertex and with the working range of the subcarriage as its radius, and determine the line segment s. e The two intersection points between the circle and the target point are connected by a line to generate the Voronoi edge required to cover the target point. This Voronoi edge and the V... vpoints Another target point in the graph is subjected to Delauney triangulation and transformed into a Voronoi mosaic.

[0051] If the number of target points is greater than 2, then the Delauney triangulation is calculated based on the target points and transformed into a Voronoi mosaic.

[0052] Step 1.3: For each target point, calculate the distance between each target point t and each vertex v in the Voronoi mosaic. Then, select the vertex v closest to the target point t as the optimal vertex and compare its result with the nearest neighbor set V of the target point t. nears The nearest vertex in the graph is compared, and the nearest vertex is taken as the nearest vertex in the final output. Then, the target point t is divided into the Voronoi mosaic graph to which the nearest vertex belongs.

[0053] Step 1.4: Apply the set covering solver. For the tessellation obtained in Step 1.2, and based on the tessellation partitioning of each target point in Step 1.3, trim unnecessary polygonal regions to obtain a minimal set of polygonal regions A′={a0,...,a...} that covers the entire target point set. m}

[0054] Step 2: Using the set of geometric centers of the tessellated polygon set as the original parking position set of the mother car, optimize it using the attraction optimization algorithm to obtain the parking position set {v0, v1, ..., v...} of the mother car. m Solve the traveling salesman problem for the parking position of the mother car to obtain the optimal driving route {v0,...,v}. m For each mother car parking point v, ..., v0}. j By performing task allocation, the set of target points {t1,...t} that the sub-vehicle needs to be responsible for is obtained. k Specifically, the steps include the following:

[0055] Step 2.1: Optimize the parking location of the mother car. For any region A′, use the gravity optimization algorithm to set a gravity point G within the set of target points to be explored. p The centroid, midpoint, or starting point of the point set can be used as the gravitational point G. p It exerts an attraction on target points within its area, and finds the point with the greatest attraction for that area based on the attraction algorithm. The points with the greatest attraction found in all areas of A′ form the parking point set V″.

[0056] Step 2.2: Solve the Traveling Salesman Problem for the parking positions of the mother car to optimize its route. For the set of parking points of the mother car, V″={v0,...,V m} and the distance between any two vertices Use a genetic algorithm to find the shortest path that traverses each vertex and returns to the set starting point.

[0057] Step 2.3: Assign subtasks to each mother car parking point to optimize the final parking position of the mother car. Substitute the polygon domain set obtained from solving the set covering problem and the mother car path sequence obtained from the genetic algorithm into the objective function f = g + h established by the A* algorithm to calculate the sub-path of the child car at each mother car parking point. Here, the cost function g is the cumulative distance the child car has reached the target point, and the heuristic function h is the distance to the remaining target points the child car needs to visit. When calculating each sub-path, the mother car parking point v is considered as the starting node p. start Based on this, nodes containing the following information are constructed: cost function value g, heuristic function value h, and evaluation function value t. The specific method includes the following steps:

[0058] Step 2.3.1: Check if the target point has been traversed. For each untraversed node p... cur All target points within the working range of the sub-vehicles around the parking point can be used as p. cur (Initially p) start The corresponding candidate adjacent points, each of which may become the next target point of the child vehicle.

[0059] Step 2.3.2, starting from the starting point p start Begin by finding the current node p. cur The candidate adjacent nodes are obtained, resulting in child nodes {p1, p2, ... p... i Then, find the candidate adjacent points of each child node, and so on, to obtain several sub-vehicle path sequences composed of target points.

[0060] Step 2.3.3, Determine p cur With p start If they are equal, it means that the sub-vehicle stopped at the mother vehicle for charging and resupply, and the value of g is initialized; otherwise, the sub-vehicle accumulates mileage normally.

[0061] Step 2.3.4: For each target point in each sub-vehicle path, determine whether the distance after reaching the target point is sufficient for the sub-vehicle to return. If not, it means that the sub-vehicle's working range is insufficient to reach the target point and return safely, and the target is invalid. If it is sufficient, include the node in the path result. Repeat the above process until all target points of the sub-vehicle path have been traversed.

[0062] Step 2.3.5: Determine whether the objective function f corresponding to the sub-vehicle path starting from the current stop point is smaller than the objective function f of another path starting from the same stop point that was previously calculated. If it is smaller, it means that the current result is better than the previous result. Update the trajectory cost function value g and the evaluation function value f. In this way, traverse all paths starting from all stops and update the nodes in the list of path results that have not been traversed.

[0063] Step 2.3.6: Repeat steps 2.3.1 to 2.3.5 until every stop point in the mother car's path has been traversed. The algorithm will eventually output a set of target point sequences that minimizes the sum of the child car's objective function f.

[0064] Step 3: For the mother car at each parking point {v0, v1, ..., v m The driving routes between} are planned to avoid obstacles, and then the tasks to be performed by the child vehicles after each stop of the mother vehicle are assigned {t1,...t}. k And plan obstacle avoidance routes.

[0065] The A* algorithm is used to plan the obstacle avoidance route of the mother robot between each parking point, and the CBS-TA algorithm is used to assign tasks to the child robots and perform path planning. Based on the task objectives to be performed by each group of child robots, the high-level search of CBS is expanded. Two key-value pairs are added to each node in the high-level search: one describing whether the current node is a root node, and the other describing whether the current task is used for the low-level search. The CBS-TA algorithm builds a search forest, expanding the root nodes as needed. First, a root node with the optimal task assignment is retrieved, without considering potential conflicts between robots. Then, expansion is extended to other root nodes, while simultaneously creating the next optimal task assignment root node. The calculation of the child robot a's task assignment is performed without considering other robots. i To target g j The cost matrix C is obtained by calculating the cost of the shortest route used. i×j This allows for the partitioning of solutions, the removal of certain allocation schemes, or the forced addition of certain solutions, ensuring that the allocation schemes cover the entire solution space.

[0066] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for task allocation and collaborative planning of multiple unmanned platforms based on region segmentation, characterized in that, include: Step 1: Calculate the target point set P = {p1, p2, ... p} n The Voronoi mosaic of}, where p n Let n represent the position of the target point n; divide the target point set P into m polygonal regions and solve the set covering problem to obtain the simplest polygonal tessellation set A′={a0,a1,...a1}. m }; Step 2: Use the set of geometric centers of the tessellated polygon set as the original parking position set of the mother car, and optimize it using the attraction optimization algorithm to obtain the mother car parking position set; solve the traveling salesman problem for the mother car parking positions to obtain the optimal driving route of the mother car; perform task assignment for each mother car parking point to obtain the set of target points that the child cars need to be responsible for. Step 3: Plan the driving route of the mother car between each parking point to avoid obstacles, and then assign the tasks that the slave car needs to perform after each parking stop and plan the obstacle avoidance route, including: The A* algorithm is used to plan the obstacle avoidance route of the mother car between each parking point, and the CBS-TA algorithm is used to assign tasks to the sub-vehicles and perform path planning. Based on the task objectives to be performed by each group of sub-vehicles, the high-level search of CBS is expanded. Two key-value pairs are added to each node in the high-level search: one describing whether the current node is a root node, and the other describing whether the current task is used for the low-level search. The CBS-TA algorithm builds a search forest and expands the root nodes as needed. First, a root node with an optimal task assignment is retrieved, without considering potential conflicts between robots. Then, the search expands to other root nodes, simultaneously creating the next root node with an optimal task assignment. The optimal task assignment for sub-robot a is calculated without considering other robots. i To target g j The cost matrix C is obtained by calculating the cost of the shortest route used. i×j This allows for the division of solutions, the removal of certain allocation schemes, or the forced addition of certain solutions, so that the allocation schemes cover the entire solution space.

2. The method for task allocation and collaborative planning of multiple unmanned platforms based on region segmentation according to claim 1, characterized in that, The method in step one includes: Step 1.1: Perform Voronoi partitioning on the points within the target point set to obtain a set of Voronoi polygons. Let the vertices of the polygons be V. edges This includes two target points, V. vpoints A convex polygon and a V containing 3 or more target points vpoints A convex polygon; Step 1.2: Determine the number of target points within the convex polygon. If the number of target points is 2, it is insufficient to calculate the Voronoi mosaic. In this case, for V vpoints Choose any un-epiterated target point x from the two target points in the array, and find the set of nearest neighbors V to target point x. nears Let s be the line segment formed by connecting the nearest vertex to the target point x. e Find a circle R centered at the nearest vertex and with the working range of the subcarriage as its radius, and determine the line segment s. e The two intersection points between the circle and the target point are connected by a line to generate the Voronoi edge required to cover the target point. This Voronoi edge and the V... vpoints Another target point in the graph is subjected to Delauney triangulation and transformed into a Voronoi mosaic. If the number of target points is greater than 2, then calculate the Delauney triangulation based on the target points and convert it into a Voronoi mosaic. Step 1.3: For each target point, calculate the distance between each target point t and each vertex v in the Voronoi mosaic. Then, select the vertex v closest to the target point t as the optimal vertex and compare its result with the nearest neighbor set V of the target point t. nears The nearest vertex in the graph is compared, and the nearest vertex is taken as the nearest vertex in the final output. Then the target point t is divided into the Voronoi mosaic graph to which the nearest vertex belongs. Step 1.4: Apply the set covering solver. For the tessellation obtained in Step 1.2, and based on the tessellation partitioning of each target point in Step 1.3, trim unnecessary polygonal regions to obtain a minimal set of polygonal regions A′={a0,...,a...} that covers the entire target point set. m } 3. The method for task allocation and collaborative planning of multiple unmanned platforms based on region segmentation according to claim 1, characterized in that, The method in step two includes: Step 2.1: Optimize the parking location of the mother car: For any region in region A′, use the gravity optimization algorithm to set a gravity point G within the set of target points to be explored. p The centroid, midpoint, or starting point of the point set can be used as the gravitational point G. p It attracts target points within its area and finds the point with the greatest attraction to that area based on the attraction algorithm. The points with the greatest attraction found in all areas of A′ form the parking point set V″. Step 2.2: Solve the Traveling Salesman Problem for the parking positions of the mother car and optimize its route: For the set of parking points of the mother car V″={v0,...,V m } and the distance between any two vertices (v i ,v j Using a genetic algorithm, the shortest path is found that traverses each vertex and returns to the set starting point. Step 2.3: Assign subtasks to each mother car parking point to optimize the final parking position of the mother car: Substitute the polygon domain set obtained from solving the set covering problem and the mother car path sequence obtained from the genetic algorithm into the objective function f = g + h established by the A* algorithm to calculate the sub-path of the child car at each mother car parking point; where the cost function g is the cumulative distance of the child car to the target point, and the heuristic function h is the distance of the remaining target points that the child car needs to visit; when calculating each sub-path, the mother car parking point v is regarded as the starting node p. start Based on this, nodes containing the following information are constructed: cost function value g, heuristic function value h, and evaluation function value t. The specific method includes the following steps: Step 2.3.1: Check if the target point has been traversed. For each untraversed node p... cur All target points within the working range of the sub-vehicles around the parking point can be used as p. cur (Initially p) start The corresponding candidate adjacent points, each of which may become the next target point of the child vehicle; Step 2.3.2, starting from the starting point p start Begin by finding the current node p. cur The candidate adjacent nodes are obtained, resulting in child nodes {p1, p2, ... p... i Then, find the candidate adjacent points of each child node, and so on, to obtain several sub-vehicle path sequences composed of target points; Step 2.3.3, Determine p cur With p start If they are equal, it means that the sub-vehicle stopped at the mother vehicle for charging and resupply, and the value of g is initialized; otherwise, the sub-vehicle accumulates mileage normally. Step 2.3.4: For each target point in each sub-vehicle path, determine whether the distance after reaching the target point is sufficient for the sub-vehicle to return. If not, it means that the sub-vehicle's working range is insufficient to reach the target point and return safely, and the target is invalid. If it is sufficient, include the node in the path result. Repeat the above process until all target points of the sub-vehicle path have been traversed. Step 2.3.5: Determine whether the objective function f corresponding to the sub-vehicle path starting from the current stop point is smaller than the objective function f of another path starting from the same stop point that was previously calculated. If it is smaller, it means that the current result is better than the previous result. Update the trajectory cost function value g and the evaluation function value f. In this way, traverse all paths starting from all stops and update the nodes in the list of path results that have not been traversed. Step 2.3.6: Repeat steps 2.3.1 to 2.3.5 until every stop point in the mother car's path has been traversed. The algorithm finally outputs a set of target point sequences that minimizes the sum of the child car's objective function f.

Images