Path planning method, electronic equipment and computer readable storage medium

Abstract

The invention discloses a path planning method, electronic equipment and a computer readable storage medium. The method comprises the steps that initial paths of all moving objects under different preset resource priority information are acquired; determining path collision information among the moving objects according to the initial paths of the moving objects under the same preset resource priority information; the path collision information corresponding to the preset resource priority information is input into the path optimization network model, multiple path collision-free probabilities are obtained, and the path collision-free probabilities represent the collision-free probabilities when the moving objects travel according to the corresponding initial paths under the same preset resource priority information; and in response to the condition that the target path collision-free probability meets the preset requirement, performing path expansion processing on the initial path of each moving object when the target path collision-free probability is obtained, so as to obtain the target path of each moving object. Therefore, the path planning efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of path planning processing technology, and in particular to a path planning method, electronic equipment and computer-readable storage medium. Background Art

[0002] In applications such as smart warehousing, multi-vehicle path planning methods have become the main method to solve the scheduling and planning problems of warehouse robots.

[0003] In multi-vehicle path planning methods, when the environment is dense and there are many mobile devices, the time requirement for path planning is high. Currently, the conflict-based search (CBS) method is widely used, but it usually cannot predict whether there is a solution to the path planning problem, resulting in the solution process continuing and taking too long, affecting the execution of subsequent programs. Summary of the invention

[0004] The main technical problem solved by the present application is to provide a path planning method, an electronic device and a computer-readable storage medium, which can improve the efficiency of path planning.

[0005] In order to solve the above technical problems, a technical solution adopted in the present application is: a path planning method is provided, the method comprising: obtaining the initial path of each moving object under different preset resource priority information; determining the path collision information between each moving object according to the initial path of each moving object under the same preset resource priority information; inputting the path collision information corresponding to each preset resource priority information into the path optimization network model to obtain multiple path collision-free probabilities, the path collision-free probability characterizing the probability of collision-free when each moving object expands its path based on the initial path in a preset path expansion manner under the same preset resource priority information; in response to the existence of a target path collision-free probability that meets the preset requirements, the initial path of each moving object when the target path collision-free probability is obtained is subjected to path expansion processing in the preset path expansion manner to obtain the target path of each moving object.

[0006] In order to solve the above technical problems, another technical solution adopted in the present application is: to provide an electronic device, including a memory and a processor, the memory storing program instructions, and the processor calling the program instructions from the memory to execute the above path planning method.

[0007] In order to solve the above technical problems, another technical solution adopted by the present application is: providing a computer-readable storage medium including program data stored therein, wherein the program data is used to implement the above path planning method when executed by a processor.

[0008] Compared with the current CBS method for obtaining the path planning of multiple moving objects, the above scheme is prone to the problem of too long solution time or even no solution. The present application proposes a path planning method, which includes: obtaining the initial path of each moving object under different preset resource priority information; determining the path collision information between each moving object according to the initial path of each moving object under the same preset resource priority information; inputting the path collision information corresponding to each preset resource priority information into the path optimization network model to obtain multiple path collision-free probabilities, and the path collision-free probability represents the probability of collision-free when each moving object follows the corresponding initial path under the same preset resource priority information; in response to the existence of the target path collision-free probability meeting the preset requirements, the initial path of each moving object when the target path collision-free probability is obtained is subjected to path extension processing to obtain the target path of each moving object. Therefore, before performing path extension, the path collision-free probability is generated for each preset resource priority information through the path optimization network model, and whether the initial path under the corresponding preset resource priority information needs to be subjected to path extension processing is determined according to the path collision-free probability, which can reduce unnecessary path extension and improve the efficiency of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0010] Figure 1 is a flowchart of an exemplary embodiment of a path planning method shown in the present application;

[0011] Figure 2 is a schematic diagram of an exemplary embodiment of preset resource priority information of each moving object;

[0012] Figure 3 yes Figure 1 A flow chart of an exemplary embodiment of step S140 in the path planning method is shown;

[0013] Figure 4 yes Figure 1 A flow chart of an exemplary embodiment of step S140 in the path planning method is shown;

[0014] Figure 5 It is a schematic diagram of the topological structure of the initial path of each moving object;

[0015] Figure 6 yes Figure 1A flow chart of an exemplary embodiment of step S130 in the path planning method is shown;

[0016] Figure 7 is a schematic diagram of a framework of an exemplary embodiment of a path optimization network model;

[0017] Figure 8 is a schematic diagram of a framework of an exemplary embodiment of the CBS method;

[0018] Fig. 9 is a schematic structural diagram of an exemplary embodiment of a path planning device shown in the present application;

[0019] Fig.10 It is a structural schematic diagram of an embodiment of an electronic device provided by the present application;

[0020] Fig.11 It is a structural schematic diagram of an embodiment of a computer-readable storage medium provided by the present application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It will be appreciated that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some but not all structures related to the present application are shown in the drawings. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the art without making creative work are within the scope of protection of the present application.

[0022] First of all, it should be noted that path planning for multiple moving objects refers to finding effective paths for multiple moving objects in a dynamic and complex environment, while ensuring that there are no conflicts or collisions between the paths of the moving objects. However, when there are multiple moving objects in a complex environment, even if each moving object has an obstacle avoidance strategy, it is easy to cause road congestion and blockage, and it will become more serious as the number of moving objects increases.

[0023] Therefore, it is crucial to find effective paths for each moving object quickly and efficiently. Currently, path planning is mainly based on the CBS method. CBS consists of two search processes. The lower-level search process is responsible for planning paths for moving objects based on the constraints determined by the upper layer. The upper layer constructs a constraint search tree. The information of each node in the constraint search tree includes: node cost value, node conflict and node path. Each time the node is expanded, the lower layer is called to update the path. However, when the environment is complex and there are many moving objects, the CBS method will have a long path planning time or even no solution, which will lead to road congestion and failure to execute normally.

[0024] Based on this, the present application proposes a path planning method, an electronic device, and a computer-readable storage medium, which optimize the CBS method in combination with a neural network model to obtain a path planning method based on a path optimization network model. The structure of this method can be expressed as NET-CBS (Neural Network - Conflict-Based Search method). For details, please refer to Figure 1 , Figure 1 is a schematic flowchart of an exemplary embodiment of the path planning method shown in the present application.

[0025] The execution subject of the path planning method can be a terminal device, a server, or other processing devices. Among them, the terminal device can be a user equipment (UE), a computer, a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. The execution subject of the path planning method can also be a path planning device. In some possible implementation manners, the path planning method can be implemented by a processor calling computer-readable instructions stored in a memory.

[0026] Specifically, the path planning method of this embodiment includes the following steps:

[0027] S110: Obtain the initial paths of each moving object under different preset resource priority information.

[0028] A moving object refers to an object that moves in a preset scenario. Exemplarily, the moving object can be an automated guided vehicle (AGV), a robot, etc. AGVs are usually used in warehouses and factories and are responsible for automatically transporting goods, raw materials, and products.

[0029] The preset resource priority information includes the priority arrangement order of each moving object. Path planning is also the search for the optimal path. By traversing the different priority arrangement orders of each moving object and determining the initial paths of each moving object according to the priority arrangement order. Take Figure 2 as an example. If the current conflict occurs during the movement of two moving objects, the different priority arrangement orders of the two moving objects can be traversed, that is, the priority of the left node is a1 < a2, and the priority of the right node is a2 < a1. The child nodes need to inherit all the priority relationships of the parent node. Each node corresponds to a preset resource priority information, and each child node is obtained by expanding the path of the parent node according to the corresponding preset resource priority information. The expanded path must satisfy the priority arrangement order of each moving object in the preset resource priority information, and the path of the moving object with a lower priority does not conflict with the paths of all moving objects with a higher priority than it.

[0030] The initial path refers to a path determined according to the preset resource priority information of each moving object. Under different preset resource priority information, the initial path of each moving object is also different. Exemplarily, the initial path can be a path obtained by adjusting the randomly generated path according to the preset resource priority information, or a path obtained by adjusting the path according to the preset resource priority information. Specifically, during the first path extension, a path is randomly generated according to the starting point and end point of each moving object, and then the randomly generated paths of each moving object are traversed to determine whether there is a collision between the paths of each moving object. If not, the path extension is not performed, and the randomly generated path is determined as the target path of each moving object; if so, the path of each moving object is adjusted according to each preset resource priority information to obtain the initial path of each moving object under each preset resource priority information; it is determined whether there is no collision in the initial path of each moving object under the same preset resource priority information, and if so, the initial path of each moving object under the corresponding preset resource priority information is determined as the target path; if not, the initial path of each moving object under each preset resource priority information is extended again according to the preset resource priority information, and the path after the path extension is determined as the initial path of this path extension. The process of path adjustment is to first plan the randomly generated path of the high-priority moving object to obtain the high-priority initial path; then plan the path of the low-priority randomly generated path according to the high-priority initial path to avoid collision with the high-priority initial path, and obtain the low-priority initial path.

[0031] The path planning device determines whether there is collision-free in the current path of each moving object. If there is collision-free, the current path of each corresponding moving object is determined as the target path; if there is no collision, the initial path of each moving object under different preset resource priority information is determined according to the preset resource priority information.

[0032] S120: Determine path collision information between each moving object according to the initial path of each moving object under the same preset resource priority information.

[0033] Path collision information refers to the collision information when each moving object moves along the corresponding initial path under the same preset resource priority information. Collisions include point collisions and edge collisions. Point collisions refer to two or more moving objects arriving at the same path point at the same time. For example, point collisions can be expressed as (a i ,a j ,v,t), where a i and a j Represent two moving objects respectively. The conflict means that at time t, a iMoving object and a j The moving objects occupy the same path point v. An edge conflict is when two or more moving objects swap positions between two path points in the same time period. For example, an edge conflict can be expressed as (a i ,a j ,v1,v2,t), where a i and a j Represent two moving objects respectively. The conflict is from time t-1 to time t. i and a j Passing the same edge in opposite directions, a i From v1 to v2, a j From v2 to v1. Since the initial paths of the moving objects under different preset resource priority information are different, the path collision information between the moving objects needs to be determined according to the initial paths under the same preset resource priority information. The number of path collision information is the same as the number of preset resource priority information.

[0034] The path planning device obtains the initial path of each moving object under different preset resource priority information, and determines the path collision information of each moving object under the corresponding preset resource priority information according to the initial path of each moving object under the same preset resource priority information.

[0035] S130: Inputting the path collision information corresponding to each preset resource priority information into the path optimization network model to obtain a plurality of path collision-free probabilities.

[0036] The path optimization network model is used to predict the probability of obtaining a collision-free path when the path is extended according to the initial path under each preset resource priority information. In other embodiments, the path optimization network model can also be used to predict the probability of obtaining a collision-free path when the path is extended according to the initial path under each preset resource priority information within a preset time period. The path optimization network model predicts the probability of finding a collision-free path from each preset resource priority information, and determines whether to perform path extension processing on the initial path under the corresponding preset resource priority information based on the collision-free probability of the path. If the probability is small, it means that the probability of finding a collision-free path from the corresponding preset resource priority information is very small, then the preset resource priority information can be discarded and no longer extended. Thereby reducing unnecessary path extensions and improving the efficiency of path planning.

[0037] The path collision-free probability refers to the probability of no collision when each moving object expands its path based on the initial path in a preset path expansion method under the same preset resource priority information determined by the path optimization network model. The preset path expansion method refers to the path expansion method determined according to each preset resource priority information, first planning the randomly generated path of the high-priority moving object to obtain a high-priority initial path; then planning the low-priority randomly generated path according to the high-priority initial path to avoid collision with the high-priority initial path, and obtaining a low-priority initial path.

[0038] S140: In response to the target path non-collision probability satisfying the preset requirement, the initial path of each moving object when the target path non-collision probability is obtained is subjected to path extension processing in a preset path extension manner to obtain the target path of each moving object.

[0039] The target path collision-free probability refers to the path collision-free probability that meets the preset requirements. The preset requirements can be the probability of obtaining the target path within a preset time that is preset based on experience. Exemplarily, the path collision-free probability of the initial path of each moving object under each preset resource priority information is obtained through the path optimization network model, and the path collision-free probability that meets the preset requirements is determined as the target path collision-free probability. The target path collision-free probability can be one or more, which is not specifically limited here.

[0040] The target path refers to a path that will not cause collision when each moving object moves. Each moving object moves according to the determined target path, thereby realizing path planning. Exemplarily, the target path of each moving object is obtained by performing path extension processing on the initial path of each moving object under the preset resource priority information that meets the preset requirements.

[0041] The path planning device obtains the target path collision-free probability that meets the preset requirements from the collision-free probability of each path output by the path optimization network model, and performs path expansion processing on the initial path of each moving object under the target path collision-free probability in a preset path expansion manner to obtain the target path of each moving object.

[0042] It can be seen that the path planning method of the embodiment of the present application obtains the initial path of each moving object under different preset resource priority information; determines the path collision information between each moving object according to the initial path of each moving object under the same preset resource priority information; inputs the path collision information corresponding to each preset resource priority information into the path optimization network model to obtain multiple path collision-free probabilities, and the path collision-free probability represents the probability of collision-free when each moving object follows the corresponding initial path under the same preset resource priority information; in response to the existence of a target path collision-free probability that meets the preset requirements, the initial path of each moving object when the target path collision-free probability is obtained is subjected to path extension processing to obtain the target path of each moving object. Therefore, before performing path extension, the path collision-free probability is generated for each preset resource priority information through the path optimization network model, and whether the initial path under the corresponding preset resource priority information needs to be subjected to path extension processing is determined according to the path collision-free probability, which can reduce unnecessary path extensions and improve the efficiency of path planning.

[0043] Based on the above embodiments, the embodiments of the present application adopt Figure 3 The flowchart explains in detail how to obtain the target path collision-free probability that meets the preset requirements from the path collision-free probability. Figure 3 , Figure 3 yes Figure 1 The flowchart of an exemplary embodiment of step S140 in the path planning method is shown. Specifically, in response to the existence of a target path collision-free probability that meets a preset requirement, step S140 performs path extension processing on the initial path of each moving object when the target path collision-free probability is obtained in a preset path extension manner, and the process of obtaining the target path of each moving object includes the following steps:

[0044] S310: Determine whether the path collision-free probability is greater than or equal to a preset path collision-free probability.

[0045] The collision-free probability of the preset path can be preset based on experience. For example, the collision-free probability of the preset path can be set to 0.2, 0.3, 0.4 or 0.5.

[0046] After obtaining the collision-free probability of each path, the path planning device compares the collision-free probability of each path with the preset collision-free probability of the path to determine whether the collision-free probability of the path is greater than or equal to the preset collision-free probability of the path.

[0047] S320: If yes, the corresponding path non-collision probability is determined as the target path non-collision probability, and the initial path of each moving object is obtained when the target path non-collision probability is obtained, and the path extension process is performed in a preset path extension manner to obtain the target path of each moving object.

[0048] After the path planning device determines that the path collision-free probability is greater than or equal to the preset path collision-free probability, the corresponding path collision-free probability is determined as the target path collision-free probability, and the initial path of each moving object under the target path collision-free probability is subjected to path expansion processing according to the corresponding preset resource priority information to obtain the target path of each moving object.

[0049] Therefore, the collision-free probability of the path output by the path optimization network model is conditionally selected, and the preset resource priority information for which a collision-free path cannot be obtained within the preset time is eliminated. Compared with the current method of blindly traversing all the preset resource priority information of each moving object to obtain the target path, the embodiment of the present application greatly reduces the computational burden and improves the computational efficiency.

[0050] In other embodiments, the step of obtaining the target path collision-free probability that meets the preset requirements from the path collision-free probability may also include: in response to the path collision-free probability being less than the preset path collision-free probability value, randomly generating a reserved probability value of the path collision-free probability; if the reserved probability value is less than or equal to the preset reserved probability value, determining the corresponding path collision-free probability as the target path collision-free probability, and obtaining the target path collision-free probability, performing path expansion processing on the initial path of each moving object in a preset path expansion manner to obtain the target path of each moving object. Thus, the path collision-free probability output by the path optimization network model that is less than the preset path collision-free probability is re-selected, which can reduce the probability of path extension errors caused by model errors and improve the accuracy of path planning.

[0051] Specifically, after a preliminary judgment is made on the path collision-free probability output by the path optimization network model, the path collision-free probability greater than the preset path collision-free probability is determined as the target path collision-free probability; for the remaining path collision-free probabilities less than the preset path collision-free probability, a random sampling value can be generated, that is, a randomly generated retention probability of the corresponding path collision-free probability is randomly generated, and when the retention probability is less than or equal to the preset retention probability, the corresponding path collision-free probability is determined as the target path collision-free probability; when the retention probability is greater than the preset retention probability, the path extension processing is not performed on the initial path of each moving object under the corresponding path collision-free probability. Exemplarily, the retention probability is a value randomly sampled from 0-1. The preset retention probability can be determined based on the corresponding path collision-free probability, and its calculation satisfies the following formula:

[0052]

[0053] Wherein, P represents the preset retention probability, c can be set in advance, for example, it can be set to 0.6, and v represents the collision-free probability of the path output by the path optimization network model.

[0054] Based on the above embodiments, the embodiments of the present application adopt Figure 4 a flowchart to elaborate in detail how to perform path extension processing on the initial paths of each moving object in a preset path extension manner. Please refer to Figure 4 , Figure 4 which Figure 1 is a schematic flowchart of an exemplary embodiment of step S140 in the path planning method shown. Specifically, the process of step S140 obtaining the target paths of each moving object by performing path extension processing on the initial paths of each moving object in a preset path extension manner includes the following steps:

[0055] To elaborate in detail how to perform path extension processing on the initial path in a preset path extension manner, the embodiments of the present application take a first moving object and a second moving object under a preset resource priority information that meets the preset requirements as an example, where the priority of the first moving object in the preset resource priority information corresponding to the collision-free probability of the target path is higher than that of the second moving object. It can be referred to Figure 2 , and the embodiments of the present application can perform path extension on the parent node to obtain a right node (a2 < a1) as an example, where the first moving object is a1 and the second moving object is a2.

[0056] S410: Determine the first extended path of the first moving object based on the initial path corresponding to the first moving object at the collision-free probability of the target path.

[0057] The first extended path is obtained by the first moving object performing path extension in a preset extension manner. The preset extension manner means that, according to the priorities of each moving object in the preset resource priority information from high to low, starting from the moving object corresponding to the highest category priority, the initial path of the second moving object is subjected to path avoidance processing in turn according to the first extended path of the first moving object in the adjacent priorities to obtain the second extended path of the second moving object. In other embodiments, path planning can also be performed according to the moving object with a lower priority among multiple high-priority moving objects, as long as the extended path of the lower priority does not conflict with the extended path of the high priority.

[0058] As an example, if the priority of the first moving object in the preset resource priority information is the highest, the initial path of the first moving object can be determined as the first extended path of the first moving object; the initial path of the second moving object is subjected to path avoidance processing according to the first extended path to obtain the second extended path. As another example, if there are other moving objects with higher priorities than the first moving object in the preset resource priority information, the first extended path of the first moving object can be obtained by performing path avoidance processing according to the extended paths of other moving objects; then the initial path of the second moving object is subjected to path avoidance processing according to the first extended path to obtain the second extended path.

[0059] The path planning device obtains an initial path of the first moving object under the probability of no collision with the target path, and performs path extension processing on the initial path of the first moving object according to corresponding preset resource priority information to obtain a first extended path of the first moving object.

[0060] S420: Performing path avoidance processing on an initial path corresponding to a second moving object with a target path non-collision probability according to the first extended path of the first moving object to obtain a second extended path of the second moving object.

[0061] The second extended path is obtained by performing path avoidance processing on the basis of the first extended path. Exemplarily, the path avoidance processing method may include the following steps: in response to the first extended path and the second moving object having the same target path point in the initial path corresponding to the target path without collision probability, respectively predicting the time points at which the first moving object and the second moving object arrive at the target path point; in response to the arrival time point of the first moving object and the arrival time point of the second moving object being the same, controlling the first moving object to take precedence over the second moving object at the target path point. Thus, path avoidance is achieved by controlling the low-priority moving object not to occupy the target path point of the high-priority moving object.

[0062] See also Figure 5 , Figure 5 The topological structure diagram shows that the first moving object moves from path point 1 to path point 5, and the second moving object moves from path point 5 to path point 6. The initial path of the first moving object can be motion_1_2, motion_2_3, motion_3_4, motion_4_5, and the initial path of the second moving object can be motion_5_4, motion_4_3, motion_3_9, motion_9_8, motion_8_7, motion_7_6. It can be seen that if the initial path of the first moving object and the initial path of the second moving object collide at path point 3, the initial path of the second moving object can be avoided. There can be many ways here. For example, the initial path of the second moving object can be adjusted to: motion_5_4, wait4, motion_4_3, motion_3_9, motion_9_8, motion_8_7, motion_7_6, that is, the second moving object stays at path point 4 until the first moving object releases path point 3. Of course, it can also stay at path point 5 or the middle position between path point 4 and path point 3, and it can also bypass path point 3.

[0063] In other embodiments, path collision also has edge conflict, and two adjacent path points constitute an edge. Edge conflict means that at the same time point, the first moving object moves from path point 2 to path point 3, and the second moving object moves from path point 3 to path point 2. At this time, the path avoidance processing method can also be: control the first moving object to take precedence over the second moving object on the corresponding edge. That is, the first moving object moves from path point 2 to path point 3 first, and the second moving object can enter path point 3 only after the first moving object releases the edge.

[0064] The path planning device determines the collision information between the first extended path of the first moving object and the initial path of the second moving object, performs path avoidance processing on the initial path of the second moving object according to the collision information to prevent collision with the first extended path of the first moving object, and obtains the second extended path of the second moving object.

[0065] S430: In response to the absence of path collision between the first extended path and the second extended path, determining the first extended path and the second extended path as the target path of the first moving object and the target path of the second moving object, respectively.

[0066] Taking two moving objects as an example here, when the path planning device obtains the target path with no collision probability and there is no path collision between the first extended path of the first moving object and the second extended path of the second moving object, it means that the feasible paths of the first moving object and the second moving object have been searched, and the first extended path and the second extended path can be determined as the target path of the first moving object and the target path of the second moving object, respectively.

[0067] When there is a path collision between the first extended path of the first moving object and the second extended path of the second moving object, it means that the target path has not been searched. At this time, the path collision information between the first extended path and the second extended path needs to be input into the path optimization network model to obtain the path collision-free probability between the first extended path and the second extended path; in response to the path collision-free probability meeting the preset requirements, the extended path of the first moving object and the extended path of the second moving object are subjected to path extension processing to obtain the target paths of the first moving object and the second moving object. In this way, the path collision-free probability prediction is performed at each extended node, which can reduce unnecessary path extensions and improve the efficiency of path planning.

[0068] In other embodiments, all paths of the first moving object and the second moving object are traversed, each path of the first moving object is compared with each path of the second moving object, and a pair of paths without conflict between the paths of the first moving object and the paths of the second moving object is selected, and then the path pair with the smallest cost is selected as the extended path of the first moving object and the extended path of the second moving object.

[0069] The path planning method of the embodiment of the present application performs path extension for moving objects of different priorities, ensuring that the extended path of the moving object with a low priority does not conflict with the extended path of the moving object with a high priority.

[0070] Based on the above embodiments, the embodiments of the present application adopt Figure 6 The flowchart details how to obtain the collision-free probability of a path through the path optimization network model. Figure 6 , Figure 6 yes Figure 1 The flowchart of an exemplary embodiment of step S130 in the path planning method is shown. Specifically, step S130 inputs the path collision information corresponding to each preset resource priority information into the path optimization network model, and the process of obtaining the collision-free probability of multiple paths includes the following steps:

[0071] S610: traverse the initial paths of the moving objects under the preset resource priority information, and construct a path collision matrix between the moving objects under the corresponding preset resource priority information.

[0072] The path collision information includes the path collision matrix constructed by each initial path, the path collision time, etc. The path collision information corresponds to the preset resource priority information, and each initial path in the same preset resource priority information is traversed to obtain the path collision information corresponding to the preset resource priority information.

[0073] The path collision matrix is ​​an N*M matrix, where M represents all initial path indexes, and N represents the initial path index that has a collision relationship with M, which is used to reduce the size of the matrix. The size of the path collision matrix column is determined by the number of initial paths, and the size of the row is determined by the maximum number of collisions contained in the initial path. As an example, the path collision matrix can be 1500*100, where 1500 is the number of initial paths, and 100 is the maximum number of collisions between initial paths. First, initialize each element of the path collision matrix to 0. In the initial path under each preset resource priority information, when the initial path corresponding to the column index and the initial path corresponding to the row index have a path collision, the element is determined to be 1, and the path collision matrix of each moving object under each preset resource priority information is obtained.

[0074] The path planning device first initializes the path collision matrix as an N*M matrix according to the collision relationship between the initial paths, and then determines the path collision matrix under each preset resource priority information according to the initial paths under the preset resource priority information.

[0075] S620: Inputting the path collision matrix corresponding to each preset resource priority information into the path optimization network model for probability prediction processing to obtain collision-free probabilities of multiple paths.

[0076] The path planning device inputs the path collision matrix corresponding to each preset resource priority information into the path optimization network model for probability prediction processing to obtain the path collision-free probability corresponding to each preset resource priority information.

[0077] Among them, in one embodiment, the path optimization network model includes a fully connected layer, a batch normalization layer and a nonlinear activation layer. Specifically, the path collision matrix is ​​fully connected through the path optimization network model to obtain the fully connected path collision matrix; the fully connected path collision matrix is ​​normalized to obtain the normalized path collision matrix; the normalized path collision matrix is ​​nonlinearly activated to obtain the path collision-free probability corresponding to each preset resource priority information. In this way, the path collision-free probability prediction can be achieved through a simple neural network structure, reducing the complexity of the path collision-free probability prediction.

[0078] In other embodiments, see Figure 7 , Figure 7 It is a schematic diagram of a framework of an exemplary embodiment of a path optimization network model. The path optimization network model includes a first network structure and a second network structure. The first network structure includes an input layer, a fully connected layer (1024-fully-connected), a batch normalization layer (batch-normalization), a dropout layer (Dropout) and a nonlinear activation layer (ReLu). The parameters of the input layer are 1500×100, and the parameters of the fully connected layer are 1024; the second network structure includes a fully connected layer (512-fully-connected), a batch normalization layer (batch-normalization), a dropout layer (Dropout), a nonlinear activation layer (ReLu) and an output layer (tanh), and the parameters of the fully connected layer are 512. Specifically, the path collision matrix is ​​input into the path optimization network model through the input layer. First, the path collision matrix is ​​processed by the first network structure in sequence by full connection, normalization, dropout and nonlinear activation. Then, the path collision matrix processed by the first network structure is processed by the second network structure by full connection, normalization, dropout and nonlinear activation, and the path collision probability output by the path optimization network model is obtained.

[0079] It should be noted that the above-mentioned path optimization network model is a trained network model, and the training process of the path optimization network model includes the following steps: obtaining a training sample set, the training samples in the training sample set include the historical extended paths when the target paths of each moving object are obtained within a preset time; inputting the historical extended paths into the path optimization network model to obtain the path collision-free probability output by the path optimization network model; calculating the loss value of the loss function between the path collision-free probability and the actual probability value; training the path optimization network model with the goal of reducing the loss value until the loss value meets the preset requirements.

[0080] The training samples can be obtained in the process of obtaining the target path by the CBS method. Specifically, the target path search process of each moving object is simulated by the CBS method. Figure 8 , Figure 8 It is a framework diagram of an exemplary embodiment of the CBS method, taking the conflict between two moving objects as an example. The root node includes a path randomly generated according to the starting point, end point and inherent obstacles of each moving object; if there is a collision in the randomly generated paths of each moving object in the root node, path extension processing is performed according to the randomly generated paths of each moving object. The path extension processing includes traversing the preset resource priority information of each moving object to generate child nodes of the root node, and determining the initial path of each moving object according to the preset resource priority information. Two moving objects can extend two child nodes, that is, the priority of the first moving object is higher than that of the second moving object and the priority of the first moving object is lower than that of the second moving object; then, it is determined whether there is a collision in the initial paths of each moving object in the two child nodes. If both exist, path extension processing is performed on the two child nodes to generate two child nodes of each child node, and thus it is continuously iterated until a target path without collision between the moving objects is found. In order to avoid the target path planning taking too long, a target path S without collision found within a preset time can be obtained. T The preset time can be 1 minute, 2 minutes, etc., so it can be seen that after obtaining the target path S T The historical expansion paths (nodes indicated by dashed arrows in the figure) are all better expansion paths, because the corresponding historical expansion paths can be used as training samples. At the same time, in order to facilitate the understanding of the present technical solution, it can be explained here that the above-mentioned path optimization network model is used to determine whether each node can find the target path within a preset time according to the initial path of each moving object under each node before each node is expanded. If not, there is no need to expand the path of the corresponding node, thereby reducing unnecessary path expansion and improving the efficiency of path planning.

[0081] The actual probability value is the target path S according to the target path search time in the CBS simulation. TThe reward value r, r∈[-1, 1], is obtained by scoring. The reward value is related to the average time and maximum time of the target path search. In each simulation, the training samples can be stored as<s,r> , where s represents the node state state, and the state is determined by the constraint (a i , a j , v k , v l , t) description, this constraint represents the moving object a i The initial path v k than the moving object a j The initial path v l The priority is high. Then the time dimension t is ignored and the elements in the path collision matrix are initialized to 0. The column size of the path collision matrix is ​​determined by the number of initial paths, and the row size is determined by the maximum number of collisions in the initial paths.

[0082] For the path optimization network model, the path optimization network model parameters θ are first initialized, and the model parameters θ are adjusted through training samples to minimize the error between the path collision-free probability output by the path optimization network model and the simulation result Z.

[0083] In other embodiments, in addition to the simulation results of CBS being used as training samples for the path optimization network model, the historical planning paths during operation contain the topological information of the map, so the historical planning paths can also be used as part of the training samples in the training sample set to continuously optimize the model parameter θ. This allows learning during operation to improve the efficiency and success rate of target path planning.

[0084] See also Fig. 9 , Fig. 9 It is a structural diagram of an exemplary embodiment of a path planning device shown in the present application. The video compression device 900 includes an acquisition module 910, a determination module 920, a probability prediction module 930 and a path extension module 940. The acquisition module 910 is used to obtain the initial path of each moving object under different preset resource priority information; the determination module 920 is used to determine the path collision information between each moving object according to the initial path of each moving object under the same preset resource priority information; the probability prediction module 930 is used to input the path collision information corresponding to each preset resource priority information into the path optimization network model to obtain multiple path collision-free probabilities, and the path collision-free probability represents the probability of collision-free when each moving object expands its path based on the initial path in a preset path extension manner under the same preset resource priority information; the path extension module 940 is used to respond to the existence of a target path collision-free probability that meets the preset requirements, and then the initial path of each moving object when the target path collision-free probability is obtained is processed in a preset path extension manner to obtain the target path of each moving object.

[0085] In the above scheme, the path planning device obtains the initial path of each moving object under different preset resource priority information; determines the path collision information between each moving object according to the initial path of each moving object under the same preset resource priority information; inputs the path collision information corresponding to each preset resource priority information into the path optimization network model to obtain multiple path collision-free probabilities, and the path collision-free probability represents the probability of collision-free when each moving object follows the corresponding initial path under the same preset resource priority information; in response to the existence of a target path collision-free probability that meets the preset requirements, the initial path of each moving object when the target path collision-free probability is obtained is subjected to path extension processing to obtain the target path of each moving object. Therefore, before performing path extension, the path collision-free probability is generated for each preset resource priority information through the path optimization network model, and whether the initial path under the corresponding preset resource priority information needs to be subjected to path extension processing is determined according to the path collision-free probability, which can reduce unnecessary path extensions and improve the efficiency of path planning.

[0086] Among them, the functions of each module can be found in the path planning method embodiment, and will not be repeated here.

[0087] In order to implement the path planning method of the above embodiment, the present application proposes another electronic device, for details, please refer to Fig.10 , Fig.10 It is a structural schematic diagram of an embodiment of an electronic device provided by the present application.

[0088] The electronic device 1000 includes a memory 1001 and a processor 1002 , wherein the memory 1001 and the processor 1002 are coupled.

[0089] The memory 1001 is used to store program data, and the processor 1002 is used to execute the program data to implement the path planning method of the above embodiment.

[0090] In this embodiment, the processor 1002 may also be referred to as a CPU (Central Processing Unit). The processor 1002 may be an integrated circuit chip having signal processing capabilities. The processor 1002 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. A general-purpose processor may be a microprocessor or the processor 1002 may also be any conventional processor, etc.

[0091] The present application also provides a computer-readable storage medium, such as Fig.11As shown, the computer-readable storage medium 1100 is used to store program data 1101. When the program data 1101 is executed by the processor, it is used to implement the path planning method in the method embodiment of the present application.

[0092] The method involved in the embodiment of the path planning method of the present application, when implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a device, such as a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0093] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A path planning method, characterized in that: The path planning method comprises: Obtaining the initial path of each moving object under different preset resource priority information; Determine path collision information between each moving object according to the initial path of each moving object under the same preset resource priority information; Inputting the path collision information corresponding to each preset resource priority information into the path optimization network model, and obtaining multiple path collision-free probabilities, wherein the path collision-free probabilities represent the probability of collision-free when each moving object expands its path in a preset path expansion manner based on the initial path under the same preset resource priority information; In response to the target path non-collision probability satisfying the preset requirement, the initial path of each moving object when the target path non-collision probability is obtained is subjected to path extension processing in the preset path extension manner to obtain the target path of each moving object.

2. The path planning method according to claim 1, characterized in that: In response to the existence of a target path collision-free probability that satisfies a preset requirement, the step of performing path extension processing on the initial path of each moving object when the target path collision-free probability is obtained in the preset path extension manner to obtain the target path of each moving object includes: Determining whether the path collision-free probability is greater than or equal to a preset path collision-free probability; If so, the corresponding path non-collision probability is determined as the target path non-collision probability, and the initial path of each moving object when the target path non-collision probability is obtained is subjected to path extension processing in the preset path extension manner to obtain the target path of each moving object.

3. The path planning method according to claim 1, characterized in that: In response to the existence of a target path collision-free probability that satisfies a preset requirement, the step of performing path extension processing on the initial path of each moving object when the target path collision-free probability is obtained in the preset path extension manner to obtain the target path of each moving object also includes: In response to the path collision-free probability being less than a preset path collision-free probability, randomly generating a retention probability of the path collision-free probability; If the retention probability is less than or equal to the preset retention probability, the corresponding path collision-free probability is determined as the target path collision-free probability, and the initial path of each moving object when the target path collision-free probability is obtained is subjected to path extension processing in the preset path extension manner to obtain the target path of each moving object.

4. The path planning method according to any one of claims 1 to 3, characterized in that: The moving objects include a first moving object and a second moving object, the first moving object has a higher priority in the preset resource priority information corresponding to the non-collision probability of the target path than the second moving object, and the step of performing path extension processing on the initial path of each moving object in the preset path extension mode when obtaining the non-collision probability of the target path, and obtaining the target path of each moving object also includes: Determining a first extended path of the first moving object based on an initial path corresponding to the first moving object under a non-collision probability of the target path; Performing path avoidance processing on an initial path corresponding to the second moving object under the non-collision probability of the target path according to the first extended path of the first moving object, to obtain a second extended path of the second moving object; In response to the absence of path collision between the first extended path and the second extended path, the first extended path and the second extended path are determined as a target path of the first moving object and a target path of the second moving object, respectively.

5. The path planning method according to claim 4, characterized in that: The step of performing path avoidance processing on an initial path corresponding to the second moving object under the non-collision probability of the target path according to the first extended path of the first moving object to obtain a second extended path of the second moving object comprises: In response to the first extended path and the initial path corresponding to the second moving object under the target path non-collision probability having the same target path point, respectively predicting the time point when the first moving object and the second moving object arrive at the target path point; In response to the arrival time point of the first moving object being the same as the arrival time point of the second moving object, the first moving object is controlled to take precedence over the second moving object at the target path point.

6. The path planning method according to claim 1, characterized in that: The path collision information includes a path collision matrix constructed by initial paths of each moving object. The step of inputting the path collision information corresponding to each preset resource priority information into the path optimization network model to obtain a plurality of path collision-free probabilities includes: Traversing the initial paths of each moving object under each preset resource priority information, and constructing a path collision matrix between each moving object under the corresponding preset resource priority information; The path collision matrix corresponding to each preset resource priority information is input into the path optimization network model for probability prediction processing to obtain the collision-free probability of multiple paths.

7. The path planning method according to claim 6, characterized in that: The step of inputting the path collision matrix corresponding to each preset resource priority information into the path optimization network model for probability prediction processing to obtain the collision-free probability of multiple paths includes: Performing full connection processing on the path collision matrix through the path optimization network model to obtain a path collision matrix after full connection processing; Normalizing the path collision matrix after the full connection processing to obtain a normalized path collision matrix; The normalized path collision matrix is ​​subjected to nonlinear activation processing to obtain the path collision-free probability corresponding to each preset resource priority information.

8. The path planning method according to claim 1, characterized in that: The method further comprises a step of training the path optimization network model, wherein the training step comprises: Acquire a training sample set, wherein the training samples in the training sample set include historical extended paths when a target path of each moving object is obtained within a preset time; Inputting the historical extended path into the path optimization network model to obtain the collision-free probability of the path output by the path optimization network model; Calculating the loss value of the loss function between the collision-free probability of the path and the actual probability value; The path optimization network model is trained with the goal of reducing the loss value until the loss value meets the preset requirement.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores program instructions, and the processor retrieves the program instructions from the memory to execute the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: include: Program data is stored, and when the program data is executed by a processor, it is used to implement the method according to any one of claims 1 to 8.

Images