Congested automatic warehouse multi-AGV path planning method based on conflict search algorithm

Abstract

The invention discloses a crowded automatic warehouse multi-AGV path planning method based on a conflict search algorithm. The method comprises the following steps: constructing a grid map model of an environment where AGVs are located; at the bottom layer of the conflict search algorithm, AGV environment information is obtained according to the grid map model, and a path solution of the AGV and conflict information carried by the path are obtained by using an A * algorithm; according to the top layer of the conflict search algorithm, an improved constraint tree is constructed according to the obtained AGV path solution and conflict information, a node with the minimum loss is selected in the constraint tree for conflict detection, planning succeeds if no conflict exists, corresponding constraints are applied to conflicting AGVs if conflicts exist, and the whole process is circulated till planning succeeds. According to the method, the searching process of corridor conflicts is simplified, the driving direction allowed by the corridor is adaptively determined through searching, the planning process is accelerated by introducing the conflict priority and bypassing the conflicts, the planning time and the quality of a path solution can be considered as much as possible, and the working efficiency of the AGV is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of path planning, and in particular to a multi-AGV path planning method based on an improved conflict search algorithm suitable for an automated warehouse in a crowded environment. Background Art

[0002] With the rapid development of the Internet of Things and artificial intelligence technologies, automated warehouses have gradually become a key link in improving logistics efficiency and reducing costs. As the core equipment in automated warehouses, automated guided vehicles (AGVs) can independently complete the tracking and driving of established routes by installing various sensors and non-contact navigation devices, thereby greatly reducing labor costs and improving work efficiency. They can not only be used in traditional manufacturing warehouses, e-commerce warehouses and other fields, but can also be expanded to special fields such as cold chain logistics and hazardous goods warehouses. Therefore, in this context, it is particularly important to develop a system suitable for AGVs to work efficiently.

[0003] One of the most critical technologies in the AGV system is its path planning. Most existing methods are based on the A* algorithm to improve the planning of a single AGV. However, in an automated warehouse, the transportation capacity of a single AGV is limited and it is difficult to meet the needs of multi-task efficient transportation. The collaborative work of multiple AGVs has become an inevitable trend.

[0004] For multi-AGV path planning, the conflict-based search (CBS) method is the most widely used, which has both optimality (the sum of the planned paths is the shortest) and completeness (if a path solution exists, it will be found, if no solution is found, it means that there is no path solution). However, in actual applications, in order to reduce costs as much as possible, most automated warehouses are often relatively crowded between shelves, forming a corridor-like layout, that is, only one AGV is allowed to pass through in one direction, such as Figure 1 , 2, 3. In this scenario, when two AGVs are traveling in opposite directions in a corridor and a corridor conflict occurs, CBS performs poorly. Although it can provide the optimal path solution, it requires a lot of search time, which greatly reduces the working efficiency of AGV.

[0005] The current solution is generally to introduce the idea of ​​highways into the planning, that is, to predetermine the driving direction of each corridor and only allow the AGV to move in a fixed direction in the corridor, so as to avoid corridor conflicts and directly avoid the path search process. This method can greatly improve the planning speed, but because it always fixes the driving direction of the corridor, there will be frequent detours, resulting in poor quality of the final path solution (the sum of the paths is large), and the actual driving time of the AGV becomes longer, which also affects the work efficiency of the AGV. Summary of the invention

[0006] In view of the above problems, the present invention provides a multi-AGV path planning method based on an improved conflict search algorithm, which combines the traditional method with the idea of ​​highways, simplifies the search process of corridor conflicts, and adaptively determines the allowed driving direction of the corridor through searching, so that the permanently fixed driving direction is changed to a phased fixed driving direction, and the corridor is transformed from a highway into a passage similar to a tidal lane, thereby reducing some unnecessary detours of AGVs. At the same time, the ideas of Prioritization of Conflicts (PC) and Bypass Conflict (BP) in Improved Conflict-Based Search (ICBS) are introduced to further accelerate the planning process. This method can take into account both the planning time and the quality of the path solution as much as possible, and effectively improve the working efficiency of AGVs.

[0007] In order to achieve the above object, the technical solution of the present invention is as follows:

[0008] The multi-AGV path planning method for a crowded automated warehouse based on a conflict search algorithm includes the following steps:

[0009] S1: Build a grid map model of the AGV's environment;

[0010] S2: Improve the bottom layer of CBS: Obtain AGV-related environmental information based on the grid map model of S1, use the improved A* algorithm to obtain the path solution and conflict information of a single AGV with constraints, and feed back the path solution and the conflict information carried by the path to S3; It should be noted that initially, the AGV does not carry constraints. As all steps are executed in a loop, the AGV will be subject to certain constraints in S3 of the previous loop. These constraints are dynamic constraints.

[0011] S3: Improve the top layer of CBS: Build an improved constraint tree based on the path solution and conflict information obtained by S2, select the node with the minimum loss in the constraint tree for conflict detection, if there is no conflict, the planning is successful, if there is a conflict, impose corresponding constraints on the conflicting AGV, call S2, and loop the entire process of S3-S2-S3 until the planning is successful.

[0012] Furthermore, the grid map model in step S1 mainly includes: the number of AGVs, the locations of the AGVs, the destinations of the AGVs, the size of the grid map, and the locations of obstacles.

[0013] Furthermore, in step S2, the improved A* algorithm is used to obtain the path of a single AGV, and the specific measures include:

[0014] S2.1 obtains the number of AGVs, the location of AGVs, the end point of AGVs, the size of the grid map, and the location of obstacles from S1; obtains the dynamic constraints of AGVs from S3 in the previous cycle. If it is the initial moment, there is no need to obtain the constraints;

[0015] S2.2 builds a conflict avoidance table (CAT), which stores the locations of all other AGVs at any time;

[0016] S2.3 calls the improved A* algorithm to search for the optimal path of a single AGV based on the relevant information obtained in S2.1, wherein the improved A* selects the optimal path with less conflicts based on the CAT constructed in S2.2 during the search process;

[0017] S2.4 obtains the specific conflict information carried by the optimal path according to the CAT constructed in S2.2, including the numbers of the conflicting AGVs, the types of conflicts and the conflict time; the conflict types include point conflicts, edge conflicts and the defined special conflict types: corridor conflicts, i.e. point conflicts and edge conflicts occurring in corridors. Whether the conflict is a corridor conflict is determined by judging whether the degree of the position of the AGV is two when the conflict occurs;

[0018] S2.5 feeds back the optimal path and the conflict information it carries to S3.

[0019] Furthermore, the specific measures for obtaining all conflict-free AGV paths using the improved constraint tree in step S3 include:

[0020] S3.1 initializes an empty set OPEN, initializes the root node according to the paths and conflict information of all AGVs obtained in S2, and puts the root node into OPEN;

[0021] S3.2 traverses OPEN to obtain the node N with the minimum cost, checks the conflict information of all AGVs in node N, and if there is no conflict in the current node, the planning is successful, and returns all AGV paths of the node N, and the loop ends;

[0022] S3.3 If there is a conflict, the corridor conflict is preferentially selected for processing based on the conflict information (PC); if there is no corridor conflict, the conflicts are processed in the order of their occurrence;

[0023] S3.4 expands subnodes N1 and N2 on node N, and applies corresponding constraints according to the type of conflict checked in S3.3, that is, point constraints are applied to point conflicts, edge constraints are applied to edge conflicts, and directional obstacle constraints are applied to corridor conflicts;

[0024] S3.5 calls S2 according to the constraints imposed by S3.4, and obtains the new AGV path and new conflict information fed back by S2 in the child node of S3.4. If there is a new path with the same cost as the original path and carries less conflict information, then according to the conflict bypass strategy (BP), the new path is used to directly replace the old path in node N and directly returns to S3.2. If it does not exist, the new path and new conflict information are used to update N1 and N2 and store them in the OPEN set, and then return to S3.2.

[0025] By adopting the above technical solution, the beneficial effects of the present invention are:

[0026] 1. The improved CBS obtains fewer conflicts in the paths at the bottom layer, which greatly relieves the pressure of the top layer to handle conflicts. At the same time, it omits the tedious conflict detection process at the top layer by providing additional conflict type and conflict time information.

[0027] 2. Instead of relying on traditional point constraints and edge constraints to resolve conflicts in corridors, a new directional obstacle constraint is used to simplify the process, allowing it to directly resolve conflicts by taking a detour, thus avoiding excessive search processes.

[0028] 3. Use the conflict bypassing strategy (BP) to reduce the expansion of the top-level constraint tree and improve operation efficiency.

[0029] 4. Use the strategy of prioritizing the expansion of cardinality conflicts and handling corridor conflicts first (PC) to further reduce the expansion of the top-level constraint tree and improve operational efficiency.

[0030] It is particularly noted that in previous studies and inventions

[00029] , a multi-value decision diagram (MDD) is often used to determine whether a conflict is a cardinality conflict, and then the corridor conflict in the cardinality conflict is selected for priority processing. However, in the crowded environment of the map considered by the present invention, corridor conflicts are cardinality conflicts in most cases. Therefore, the present invention omits the MDD process and directly prioritizes the corridor conflicts to speed up the operation efficiency. As a result, some redundant constraints may be imposed on some top-level nodes, resulting in an increase in the length of some paths. The present invention conducted experiments in Example 1, and the experiments proved that such processing will greatly speed up the efficiency of the algorithm, while the increase in path length is extremely small and almost negligible. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a typical grid map model of a small warehouse (with the start and end points of the AGV);

[0032] Figure 2It is a typical grid map model of a medium-sized warehouse (without the start and end points of the AGV);

[0033] Figure 3 It is a typical grid map model of a large warehouse (without the start and end points of the AGV);

[0034] Figure 4 It is a flowchart of the underlying algorithm of a multi-AGV path planning method suitable for crowded automated warehouses based on an improved conflict search algorithm of the present invention;

[0035] Figure 5 This is a comparison chart of the path costs of a multi-AGV path planning method for crowded automated warehouses based on a conflict search algorithm and a CBS algorithm and a CBS algorithm that introduces the idea of ​​a highway in a small warehouse;

[0036] Figure 6 This is a comparison chart of the path costs of a multi-AGV path planning method for crowded automated warehouses based on a conflict search algorithm and a CBS algorithm and a CBS algorithm that introduces the idea of ​​a highway in a medium-sized warehouse;

[0037] Figure 7 This is a comparison chart of the path costs of a multi-AGV path planning method for crowded automated warehouses based on a conflict search algorithm and a CBS algorithm and a CBS algorithm that introduces the idea of ​​a highway in a large warehouse; DETAILED DESCRIPTION

[0038] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0039] Step S1: Abstract the warehouse environment where the AGV is located into a grid map model; in this embodiment, Figure 1 As shown, specifically:

[0040] Figure 1 This is a grid map abstracted for a small warehouse. The circular pattern in the grid represents the starting point of the AGV, the star pattern represents the end point of the AGV, and the black area is the obstacle area, which can also be regarded as the shelf area. From this grid map, we can clearly know the number of AGVs, the starting point of the AGV, the end point of the AGV, the map size, the obstacle area and other information. This information will be used as the input of the subsequent planning algorithm and is a necessary condition for the subsequent planning steps.

[0041] Step S2: The bottom layer of the planning algorithm is responsible for planning the path of a single AGV under constraints; Figure 2 The following is the overall flow chart of the underlying algorithm, specifically:

[0042] S2.1 loads the environment information of S1 and the related constraints passed in by S3; initializes the OPEN and CLOSED sets;

[0043] Each node n in the OPEN and CLOSED sets has the following four variables:

[0044] The actual cost (ng), expressed as the length of the path from the starting point to node n;

[0045] Estimated cost (nh), expressed as the Manhattan distance from node n to the target point;

[0046] The estimated cost (nf), expressed as the sum of the actual cost and the estimated cost, that is, nf = n.h + ng, is the main basis for node selection;

[0047] Conflict information (n.conflict information), which indicates the conflict information carried by the path from the starting point to the node n, is used as a secondary basis for node selection. Specifically, it includes the number of the AGV that conflicts with the path, the type of conflict, and the conflict time. The conflict types include point conflict, i.e., two AGVs arrive at the same location at the same time; edge conflict, i.e., two AGVs exchange their positions at the same time; and corridor conflict, i.e., point conflict and edge conflict occurring in the corridor.

[0048] S2.2 Initialize the starting node start and put start into the OPEN set;

[0049] S2.3 Take the best node with the smallest f value from the OPEN set. If there are multiple best nodes, further compare their conflict counts, which are represented by the length of best.conflict information. Then select the node with the smallest conflict count. If the OPEN set is empty, it means that the planning fails and the bottom layer ends.

[0050] S2.4 deletes best from the OPEN set and adds best to the CLOSED set; determines whether best is the end point. If so, finds the entire path through reverse search and returns relevant information. Planning is successful, the bottom layer ends, and the loop is exited. If not, puts all neighbor nodes of the best node that meet the constraints and environmental information into neighbors;

[0051] S2.5 If neighbors is empty, return to S2.3, otherwise select a node nb at random from neighbors;

[0052] S2.6 If nb is already in the OPEN set, it means that it has been a neighbor node of other nodes. At this time, it is necessary to determine the relationship between the g value of the new nb and the g value of the old nb. If the g value of the new nb is larger, no operation is performed and the process returns to S2.5 directly. If the new g value is smaller, the old nb is replaced with the new nb, and its new h value, f value and conflict formation value are calculated, and its parent node is set to best.

[0053] S2.7 If nb is in the CLOSED set, it means it has been expanded, and it returns directly to S2.5;

[0054] S2.8 If nb is neither in OPEN nor CLOSED, initialize nb as in the initialization starting point start, calculate its g value, h value, f value and conflict information value, set its parent node to best, and store it in the OPEN set, and finally return to S2.5;

[0055] Step S3: Conflicts are processed at the top level until a conflict-free path solution is found and planning is successful. The pseudo code for the implementation process is as follows:

[0056]

[0057] S3.1 initializes the root node Root of the constraint tree and stores Root in the OPEN set. Root and each node in the constraint tree have the following four variables:

[0058] Constraint (Root.constraint), each AGV has its corresponding point constraint (a i ,v1,t) means that AGVi is prohibited from occupying grid v1 at time t. The edge constraint (a i ,v1,v2,t) means that AGVi is prohibited from moving from v1 to v2 at time [t,t+1]; in particular, a special directional obstacle constraint (a i ,v1,v2) to quickly handle corridor conflicts, that is, prohibiting AGVi from moving from v1 to v2 at any time period, thereby forcing it to directly bypass the corridor. The top layer changes the path of AGV by imposing these constraints to avoid conflicts;

[0059] Solution (Root.solution), the path solution of all AGVs, obtained by accumulating the underlying algorithm;

[0060] Conflict information (Root.conflict information): conflict information between all AGV paths, as a secondary basis for node selection, is accumulated by calling the underlying algorithm, including the conflicting AGV number, conflict type and conflict time, which is used to judge conflicts and select nodes;

[0061] Total cost (Root.cost), the total cost of the current solution, as the main basis for node selection;

[0062] S3.2 Select the node N with the minimum cost from the OPEN set. If there are multiple nodes in N, further compare their conflict numbers, which are expressed as the length of N.conflict information, and then select the node with the minimum conflict number.

[0063] S3.3 detects N.conflict information. If there is no conflict, N.solution is returned, the path solution is found, and the planning is successful. If there is a conflict, corridor conflict is preferentially selected for processing (PC), and the conflict is represented by c.

[0064] S3.4 performs the following steps on the two AGVs that collide:

[0065] A) Copy N nodes to A, and impose corresponding constraints based on the conflict c, that is, point constraints are imposed on point conflicts, edge constraints are imposed on edge conflicts, and directional obstacle constraints are imposed on corridor conflicts;

[0066] B) calling step S2 to obtain a new path and new conflict information under the new constraints;

[0067] C) If the cost of the new path is consistent with the old path and carries less conflicts, execute the conflict bypass strategy (BP), that is, delete the constraints imposed by node A in A), then store node A in the OPEN set and return to 3.2, otherwise store node A in the OPEN set and return to 3.4, until both AGVs have completed the corresponding steps and then return to 3.2;

[0068] Embodiment 1:

[0069] In such Figure 2 50 random instances are run in the medium-sized grid map shown in Table 1. Table 1 shows the total number of nodes expanded by the top layer of CBS in the two cases of prioritizing corridor conflicts and processing conflicts in chronological order under different numbers of AGVs. Table 2 shows the total cost of the above two schemes under different numbers of AGVs.

[0070] Table 1 Total number of nodes expanded at the top level of CBS

[0071] Table 2 Total cost

[0072] It can be seen from the above table that giving priority to handling corridor conflicts will greatly reduce the number of nodes expanded at the top level, thereby reducing the algorithm running time and improving the planning success rate, while the cost increase is extremely small and almost negligible. Therefore, it is quite necessary for the algorithm proposed in this invention to give priority to corridors.

[0073] Embodiment 2:

[0074] Respectively in Figure 1 , 50 random instances are run in grid maps of different sizes shown in Tables 2 and 3. Tables 3, 4 and 5 show the planning success rates of the CBS algorithm, the CBS+highways algorithm with the highway idea, and the CBS+improvedhighways algorithm proposed in the present invention under different maps with different numbers of AGVs (planning is successful if the path solution is found within the time limit, and the time limits for the three maps are 30s, 60s and 90s respectively). All algorithms introduce the BP strategy to improve the running speed for easy comparison.

[0075] Table 3 Planning success rate of three algorithms on small warehouse map

[0076] Table 4 Planning success rate of three algorithms on medium-sized warehouse map

[0077] Table 5 Planning success rate of three algorithms on large warehouse maps

[0078] It can be seen that with the increase of the map size and the number of AGVs, the success rates of different algorithms show a downward trend, but the success rates of CBS+highways and the algorithm CBS+improved highways proposed in the present invention are significantly better than the traditional CBS method. The success rates of the two algorithms only differ to a certain extent when the map is large and the number of AGVs is large.

[0079] Figure 5 , 6, and 7 are the comparison diagrams of the total path cost of the three algorithms in different maps with different numbers of AGVs.

[0080] It can be seen from the figure that the traditional CBS algorithm has the minimum cost because of its optimality. Although the planning time of CBS+highways is short and the planning success rate is high, its total path cost is extremely high due to frequent detours. Although the CBS+improved highways proposed in the present invention cannot provide the optimal path solution, the total cost of the solution is very close to the total cost of the optimal solution. The total path costs of both are much better than the CBS+highways algorithm.

[0081] From the above analysis, it can be seen that the improved algorithm proposed in the present invention can provide the same planning success rate as the CBS+highways algorithm and the same total path cost as the traditional CBS algorithm in most cases. It effectively utilizes the advantages of both and achieves a good balance between planning time and path solution quality, which can effectively improve the working efficiency of AGV in automated warehouses.

Claims

1. A multi-AGV path planning method for crowded automated warehouses based on a conflict search algorithm, characterized in that: The following steps are involved: S1: Build a grid map model of the AGV's environment; S2: The bottom layer of the conflict search algorithm: According to the grid map model of S1, the environmental information related to the AGV is obtained, and the path solution and conflict information of a single AGV with constraints are obtained using the improved A* algorithm, and the path solution and the conflict information carried by the path are fed back to S3; the constraints of the AGV are dynamic constraints. Initially, the AGV does not carry constraints. As all steps are executed in a loop, the AGV will be subject to certain constraints in S3 of the previous loop; S3: The top layer of the conflict search algorithm: Based on the path solution and conflict information obtained by S2, an improved constraint tree is constructed, and the node with the minimum loss is selected in the constraint tree for conflict detection. If there is no conflict, the planning is successful. If there is a conflict, the corresponding constraints are imposed on the conflicting AGVs, and S2 is called. The entire process of S3-S2-S3 is looped until the planning is successful.

2. The method for multi-AGV path planning in a crowded automated warehouse based on a conflict search algorithm according to claim 1 is characterized in that: The grid map model in step S1 includes: the number of AGVs, the locations of the AGVs, the destination of the AGVs, the size of the grid map, and the locations of obstacles.

3. The method for multi-AGV path planning in a crowded automated warehouse based on a conflict search algorithm according to claim 1, characterized in that: In step S2, the improved A* algorithm is used to obtain the path of a single AGV, including: S2.1 obtains the number of AGVs, the location of AGVs, the end point of AGVs, the size of the grid map, and the location of obstacles from S1; obtains the dynamic constraints of AGVs from S3 in the previous cycle. If it is the initial moment, there is no need to obtain the constraints; S2.2 builds a conflict avoidance table CAT, which stores the positions of all other AGVs at any time; S2.3 calls the improved A* algorithm to search for the optimal path of a single AGV based on the relevant information obtained in S2.1, wherein A* selects the optimal path with less conflicts based on the CAT constructed in S2.2 during the search process; S2.4 obtains the specific conflict information carried by the optimal path based on the CAT constructed in S2.2, including the numbers of the conflicting AGVs, the types of conflicts and the conflict time; the conflict types include point conflicts, edge conflicts and a newly defined special conflict type: corridor conflicts, i.e. point conflicts and edge conflicts occurring in corridors. Whether the conflict is a corridor conflict is determined by judging whether the degree of the position of the AGV is two when the conflict occurs; S2.5 feeds back the optimal path and the conflict information it carries to S3.

4. The method for multi-AGV path planning in a crowded automated warehouse based on a conflict search algorithm according to claim 1, characterized in that: The specific measures for obtaining all conflict-free AGV paths using the improved constraint tree in step S3 include: S3.1 initializes an empty set OPEN, initializes the root node according to the paths and conflict information of all AGVs obtained in S2, and puts the root node into OPEN; S3.2 traverses OPEN to obtain the node N with the minimum cost, checks the conflict information of all AGVs in node N, and if there is no conflict in the current node, the planning is successful, and returns all AGV paths of the node N, and the loop ends; S3.3 If there is a conflict, the corridor conflict is processed first based on the conflict information; if there is no corridor conflict, the conflicts are processed in the order of their occurrence; S3.4 expands subnodes N1 and N2 on node N, and applies corresponding constraints according to the type of conflict checked in S3.3, that is, point constraints are applied to point conflicts, edge constraints are applied to edge conflicts, and directional obstacle constraints are applied to corridor conflicts; S3.5 calls S2 according to the constraints imposed by S3.4, and obtains the new AGV path and new conflict information fed back by S2 in the child node of S3.

4. If there is a new path with the same cost as the original path and carries less conflict information, then the new path is used to directly replace the old path in node N according to the conflict bypass strategy and directly returns to S3.

2. If it does not exist, the new path and new conflict information are used to update N1 and N2 and store them in the OPEN set, and then return to S3.

2.

5. The method for multi-AGV path planning in a crowded automated warehouse based on a conflict search algorithm according to claim 1, characterized in that: In S2, S2.1 loads the environment information of S1 and the related constraints passed in by S3; initializes the empty sets OPEN and CLOSED; Each node n in the OPEN and CLOSED sets has the following four variables: The actual cost ng, expressed as the length of the path from the starting point to node n; The estimated cost nh is expressed as the Manhattan distance from node n to the target point; The estimated cost nf, expressed as the sum of the actual cost and the estimated cost, that is, nf = n.h + ng, is used as the basis for node selection; Conflict information n.conflict information, which indicates the conflict information carried by the path from the starting point to the node n, is used as a secondary basis for node selection. Specifically, it includes the number of the AGV that conflicts with the path, the type of conflict, and the conflict time. The conflict types include point conflict, i.e., two AGVs arrive at the same location at the same time; edge conflict, i.e., two AGVs exchange their positions at the same time; and corridor conflict, i.e., point conflict and edge conflict occurring in the corridor. S2.2 Initialize the starting node start and put start into the OPEN set; S2.3 Take the best node with the smallest f value from the OPEN set. If there are multiple best nodes, further compare their conflict counts, which are represented by the length of best.conflict information. Then select the node with the smallest conflict count. If the OPEN set is empty, it means that the planning fails and the bottom layer ends. S2.4 deletes best from the OPEN set and adds best to the CLOSED set; determines whether best is the end point. If so, finds the entire path through reverse search and returns relevant information. Planning is successful, the bottom layer ends, and the loop is exited. If not, puts all neighbor nodes of the best node that meet the constraints and environmental information into neighbors; S2.5 If neighbors is empty, return to S2.3, otherwise select a node nb at random from neighbors; S2.6 If nb is already in the OPEN set, it means that it has been a neighbor node of other nodes. At this time, it is necessary to determine the relationship between the g value of the new nb and the g value of the old nb. If the g value of the new nb is larger, no operation is performed and the process returns to S2.5 directly. If the new g value is smaller, the old nb is replaced with the new nb, and its new h value, f value and conflict information value are calculated, and its parent node is set as best. S2.7 If nb is in the CLOSED set, it means it has been expanded, and it returns directly to S2.5; S2.8 If nb is neither in OPEN nor CLOSED, initialize nb as in the initialization starting point start, calculate its g value, h value, f value and conflict information value, set its parent node to best, and store it in the OPEN set, and finally return to S2.

5.

6. The method for multi-AGV path planning in a crowded automated warehouse based on a conflict search algorithm according to claim 1, characterized in that: S3 is specifically: S3.1 initializes an empty set OPEN, initializes the root node Root of the constraint tree, and stores Root in the OPEN set. Root and each node in the constraint tree have the following four variables: Constraint Root.constraint, each AGV has its corresponding point constraint (a i ,v1,t) means that AGVi is prohibited from occupying grid v1 at time t. The edge constraint (a i ,v1,v2,t) means that AGVi is prohibited from moving from v1 to v2 at time [t,t+1]; a special directional obstacle constraint (a i ,v1,v2) to quickly handle corridor conflicts, that is, prohibiting AGVi from moving from v1 to v2 at any time period, thereby forcing it to directly bypass the corridor. The top layer changes the path of AGV by imposing these constraints to avoid conflicts; Solution Root.solution, the path solution of all AGVs, is accumulated by calling the underlying algorithm; Root.conflict information: conflict information between all AGV paths. It is the secondary basis for node selection and is accumulated by calling the underlying algorithm. It includes the conflicting AGV number, conflict type, and conflict time, which is used to judge conflicts and select nodes. Total cost Root.cost, the total cost of the current solution, serves as the basis for node selection; S3.2 Select the node N with the minimum cost from the OPEN set. If there are multiple nodes in N, further compare their conflict numbers, which are expressed as the length of N.conflict information, and then select the node with the minimum conflict number. S3.3 detects N.conflict information. If there is no conflict, N.solution is returned, the path solution is found, and the planning is successful. If there is a conflict, the corridor conflict is preferentially selected for processing, and the conflict is represented by c. S3.4 performs the following steps on the two AGVs that collide: A) Copy N nodes to A, and impose corresponding constraints based on the conflict c, that is, point constraints are imposed on point conflicts, edge constraints are imposed on edge conflicts, and directional obstacle constraints are imposed on corridor conflicts; B) calling step S2 to obtain a new path and new conflict information under the new constraints; C) If the cost of the new path is consistent with the old path and carries fewer conflicts, the conflict bypass strategy is executed, that is, the constraint imposed by node A in step A) is deleted, and then node A is stored in the OPEN set and returns to S3.

2. Otherwise, node A is stored in the OPEN set and returns to 3.4 until both AGVs have executed the corresponding steps and then return to S3.2.

Images