Multi-AGV cooperative path planning and scheduling method for chip intelligent storage

Abstract

The invention discloses a multi-AGV cooperative path planning and scheduling method for chip intelligent storage, and belongs to the field of high-precision electronic component intelligent storage. The method comprises the following steps: running according to a highway guide strategy, and pre-allocating tasks of each robot by using an MTSP problem according to different task types before path planning; an improved A * algorithm is provided, a global thermodynamic diagram congestion prediction and turning waiting heuristic method is introduced to establish a space-time joint search model, and a transportation path is optimized by greatly reducing the number of nodes needing to be searched and the turning and waiting times of the AGV, so that the efficiency is improved; in addition, a series of priority rules are also provided, and appearing conflicts are eliminated. Experiments verify the effectiveness of the method, high-reliability and low-vibration dust-free workshop AGV collaborative scheduling can be realized, and an efficient and safe warehousing automation solution is provided for semiconductor manufacturing.

Description

Technical Field

[0001] The present invention relates to a multi-AGV collaborative path planning and scheduling method for chip intelligent warehousing, belonging to the field of intelligent warehousing of high-precision electronic components. Background Art

[0002] With the rapid development of the semiconductor industry, the requirements for automation, precision, and reliability in chip manufacturing and storage are becoming increasingly stringent. As high-value and highly sensitive electronic components, chips require their warehousing systems to meet strict environmental control requirements such as dust prevention, anti-static, constant temperature, and constant humidity. At the same time, efficient and accurate storage, sorting, and traceability functions need to be achieved. The traditional manual warehousing method is difficult to meet the cleanliness requirements of nanoscale processes in chip manufacturing, and is inefficient and prone to introducing contamination or electrostatic damage due to human operation.

[0003] Currently, leading international semiconductor factories have gradually introduced automated warehousing systems, such as using AGVs to achieve automated handling of wafer boxes or chip carriers. However, the particularity of the chip warehousing scenario poses higher challenges to multi-AGV collaborative scheduling:

[0004] High-precision environmental constraints: AGVs need to operate in a dust-free workshop, and path planning should avoid vibrations caused by frequent starts, stops, or sharp turns to prevent affecting chip stability.

[0005] High-density storage requirements: Chip warehouses usually use multi-layer high-density anti-static shelves, with narrow aisles and complex layouts, making it easy for AGVs to cause congestion and path conflicts when passing through.

[0006] Task priorities: In the chip manufacturing process, the urgency of different batches of chips varies (such as the timeliness requirements of wafers after lithography), and the task allocation strategy needs to be dynamically adjusted.

[0007] Existing path planning methods have significant deficiencies in the chip warehousing scenario:

[0008] Poor environmental adaptability: Traditional algorithms ignore anti-static path constraints and the requirements for the running stability of AGVs, resulting in uneven paths and increasing the risk of chip damage.

[0009] Low conflict resolution efficiency: In a high-density environment, AGVs frequently avoid each other, increasing the probability of system deadlocks and affecting the overall throughput.

[0010] High computational complexity: Centralized algorithms are difficult to respond to the dynamic changes of large-scale AGV systems in real time, while distributed algorithms lack global optimization capabilities. Summary of the Invention

[0011] In order to improve the stability of multi-AGV collaborative path planning, enhance the conflict resolution efficiency, and at the same time reduce the computational complexity problem, the present invention provides a multi-AGV collaborative path planning and scheduling method for chip intelligent warehousing. The technical solution is as follows:

[0012] Step 1: Model and initialize the chip intelligent warehousing environment;

[0013] Step 2: Based on the highway path guidance strategy, set up highways on the chip intelligent warehousing environment model established in Step 1;

[0014] Step 3: Obtain the tasks to be assigned, as well as the starting coordinates and task point coordinates of all AGVs;

[0015] Step 4: Model the task assignment as a multi-traveling salesman problem, and use an improved genetic algorithm to remove the crossover operator. Generate a task assignment scheme through tournament selection and mutation operations. The goal of task assignment is to minimize the total path cost for all AGVs to complete the tasks:

[0016]

[0017] where m represents the number of AGVs, n represents the number of tasks to be assigned, S i represents the current position of the i-th AGV, G j represents the target point position of the j-th task, x ij is a binary variable indicating whether the j-th task is assigned to the i-th AGV, D(S i , G j ) represents the Euclidean distance path cost from the current position S i of the i-th AGV to the target point position G j of the j-th task;

[0018] Step 5: Use a multi-AGV path planning algorithm based on conflict search to plan the AGV paths. The lower layer of the algorithm performs path planning to generate initial paths; the upper layer of the algorithm traverses the paths to check whether there are conflicts between the paths. If there are no conflicts, output the conflict-free paths and the AGVs execute the tasks; if there are path conflicts, perform conflict resolution and re-call the lower layer algorithm for path planning until there are no conflicts between the paths.

[0019] Optionally, in the lower layer search of Step 5, an improved A* algorithm is used for the single path planning of each AGV. The number of turns and waiting times in the generated paths are added to the heuristic function, and a heat map congestion prediction is introduced. The improved A* algorithm is expressed as:

[0020] f'(n) = g(n) + h'(n)

[0021]

[0022] Among them, f'(n) represents the improved A* algorithm cost function, g(n) represents the actual cost from the starting point to the current node n, h'(n) represents the improved heuristic function, h(n) represents the heuristic function of the traditional A* algorithm, β is the turning waiting times penalty term, C(n,t') represents the congestion prediction value, turns&waits represents the number of turns and waiting times in the path, and t max represents the maximum time step;

[0023] The congestion prediction value includes a dynamic update mechanism, which is expressed as:

[0024]

[0025] Among them, r represents the heat value attenuation coefficient; ΔC is the fixed heat growth value, and π i (t) represents the path of the AGV i

[0026] Optionally, step 5 adopts a priority-oriented conflict resolution strategy, and dynamically assigns priorities by combining AGV types, path costs, and avoidance strategies. The priorities are set as follows:

[0027] Priority 1: Loaded AGV with a unique shortest path;

[0028] Priority 2: Unloaded AGV with a unique shortest path;

[0029] Priority 3: Loaded AGV with multiple shortest paths;

[0030] Priority 4: Unloaded AGV with multiple shortest paths.

[0031] Optionally, when the starting point coordinates of the loaded AGV meet the following conditions, there are multiple shortest paths:

[0032] |x start -x goal | > x c

[0033] |y start -y goal | > y c

[0034] When the starting point coordinates of the unloaded AGV meet the following conditions, there are multiple shortest paths:

[0035] |x start -x goal | > 0

[0036] |y start -y goal ​| > 0

[0037] Among them, (x start , y start ) is the starting point coordinate of the AGV, and (x goal , y goal ) is the end point coordinate of the AGV; x c is the horizontal length of a row of storage shelves, and y c is the vertical length of a row of storage shelves.

[0038] The second object of the present invention is to provide a multi-AGV collaborative path planning and scheduling system for chip intelligent warehousing, including:

[0039] A model construction module configured to model and initialize the chip intelligent warehousing environment;

[0040] A highway setting module configured to set highways on the chip intelligent warehousing environment model established in the above step 1 based on the highway path guidance strategy;

[0041] A task acquisition module configured to acquire tasks to be assigned, as well as the starting point coordinates and task point coordinates of all AGVs;

[0042] A task assignment module configured to model task assignment as a multi-traveling salesman problem, adopt an improved genetic algorithm, remove the crossover operator, and generate a task assignment scheme through tournament selection and mutation operations. The goal of task assignment is to minimize the total path cost for all AGVs to complete tasks:

[0043]

[0044] Among them, m represents the number of AGVs, n represents the number of tasks to be assigned, S i represents the current position of the i-th AGV, G j represents the target point position of the j-th task, x ij is a binary variable indicating whether the j-th task is assigned to the i-th AGV, and D(S i , G j ) represents the Euclidean distance path cost from the current position S i of the i-th AGV to the target point position G j of the j-th task;

[0045] The path planning module is configured to plan the AGV path using a multi-AGV path planning algorithm based on conflict search. The lower layer of the algorithm generates an initial path through path planning. The upper layer of the algorithm traverses the paths to check if there are conflicts between the paths. If there are no conflicts, it outputs a conflict-free path and the AGV executes the task. If there are path conflicts, conflict resolution is performed and the lower layer algorithm is called again for path planning until there are no conflicts between the paths.

[0046] Optionally, the path planning module uses an improved A* algorithm in the lower layer search for single-path planning of each AGV, adds the number of turns and waiting times in the generated path to the heuristic function, and introduces heatmap congestion prediction. The improved A* algorithm is expressed as:

[0047] f'(n) = g(n) + h'(n)

[0048]

[0049] where f'(n) represents the cost function of the improved A* algorithm, g(n) represents the actual cost from the starting point to the current node n, h'(n) represents the improved heuristic function, h(n) represents the heuristic function of the traditional A* algorithm, β is the turning and waiting times penalty term, C(n,t') represents the congestion prediction value, turns&waits represents the number of turns and waiting times in the path, and t max represents the maximum time step;

[0050] The congestion prediction value includes a dynamic update mechanism, which is expressed as:

[0051]

[0052] where r represents the heat value decay coefficient; ΔC is the fixed value of heat growth, and π i (t) represents the path of the AGV i .

[0053] Optionally, the path planning module adopts a priority-oriented conflict resolution strategy, dynamically assigns priorities by combining AGV types, path costs, and avoidance strategies. The priority settings are as follows:

[0054] Priority 1: Loaded AGV with a unique shortest path;

[0055] Priority 2: Unloaded AGV with a unique shortest path;

[0056] Priority 3: Loaded AGV with multiple shortest paths;

[0057] Priority 4: Unloaded AGV with multiple shortest paths.

[0058] Optionally, in the path planning module, when the starting point coordinates of the load AGV satisfy the following conditions, there are multiple shortest paths:

[0059] |x start -x goal | > x c

[0060] |y start -y goal | > y c

[0061] When the starting point coordinates of the unloaded AGV satisfy the following conditions, there are multiple shortest paths:

[0062] |x start -x goal | > 0

[0063] |y start -y goal | > 0

[0064] Among them, (x start , y start ) are the starting point coordinates of the AGV, and (x goal , y goal ) are the end point coordinates of the AGV; x c is the horizontal length of a row of storage shelves, and y c is the vertical length of a row of storage shelves.

[0065] The third object of the present invention is to provide a multi-AGV collaborative path planning and scheduling device for chip intelligent warehousing, including a memory and a processor;

[0066] The memory is used to store a computer program;

[0067] The processor is used to implement the multi-AGV collaborative path planning and scheduling method for chip intelligent warehousing as described in any one of the above when executing the computer program.

[0068] The fourth object of the present invention is to provide a computer-readable storage medium, characterized in that a computer program is stored on the storage medium, and when the computer program is executed by a processor, the multi-AGV collaborative path planning and scheduling method for chip intelligent warehousing as described in any one of the above is implemented.

[0069] The beneficial effects of the present invention are:

[0070] The multi-AGV collaborative path planning and scheduling method proposed by the present invention optimizes the rationality of task allocation in terms of task allocation by modeling tasks as multi-traveling salesman problems and combining AGV status and shelf area constraints. Secondly, the path planning algorithm introduces a highway path guidance strategy on the basis of CBS, strictly restricts the moving direction of AGVs, and reduces the possibility of conflicts. The task allocation strategy based on the highway guidance strategy and the MTSP problem proposed by the present invention effectively reduces path conflicts and computational complexity.

[0071] Furthermore, the present invention proposes an improved A* algorithm, which introduces global heat map congestion prediction and turning waiting heuristic methods to establish a spatio-temporal joint search model. By greatly reducing the number of nodes to be searched, as well as the turning and waiting times of AGVs, the transportation path is optimized, thereby improving efficiency. A series of priority rules are also proposed to resolve the conflicts that occur, optimize the quality of path planning, and reduce the risk of goods damage.

[0072] The experimental results show the effectiveness of the multi-AGV collaborative optimization scheduling method proposed by the present invention. Compared with the baseline algorithm, there is a significant performance improvement, which can achieve high-reliability and low-vibration AGV collaborative scheduling in a dust-free workshop, providing an efficient and safe warehousing automation solution for semiconductor manufacturing. Brief Description of the Drawings

[0073] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0074] Figure 1 It is a visualization effect diagram of the path guidance strategy in the second embodiment of the present invention.

[0075] Figure 2 It is a visualization effect diagram of the MTSP task allocation in the second embodiment of the present invention.

[0076] Figure 3 It is a visualization effect diagram of the global heat map in the second embodiment of the present invention.

[0077] Figure 4 It is a visualization effect diagram of the multi-AGV collaborative scheduling system for chip intelligent warehousing in the second embodiment of the present invention. Detailed Embodiments

[0078] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the drawings.

[0079] Embodiment 1:

[0080] This embodiment provides a multi-AGV collaborative path planning and scheduling method for chip intelligent warehousing, including:

[0081] Step 1: Model and initialize the chip intelligent warehousing environment, and set basic parameters such as the number of AGVs, the size of the scenario, and the number of shelves.

[0082] Step 2: Based on the highway path guidance strategy, set up highways on the chip intelligent warehousing environment model established in Step 1.

[0083] For the chip intelligent warehousing environment in this embodiment, an undirected graph G=(V, E) of the warehousing environment is defined, and a directed subgraph G h =(V h , E h ) is created in the finite vertex set and the finite edge set composed of it, where V h contains the vertex positions involved in the highway, and E h represents the directed edge along the traffic direction of the highway. The highway setting can be defined by specifying the moving direction of each highway. The position sequence P={v0,..., v n} is called a highway of length n. For each position sequence P∈E h , its two endpoints (v0, v n )∈V h , and it satisfies that if and only if for i = 1→n-1, V i is connected to exactly two positions V i-1 and V i+1 according to the edge set E, that is, there is only one entrance and one exit for each vertex position. The starting point V0 and the ending point V n of the highway can be the same vertex, forming a circular highway. For each highway, there is a direction d∈D, which is an integer value representing the running direction of the corridor, and this direction definition is the same as the direction definition of the AGV running state. In addition, the intersection, which is the intersection of two or more highways, belongs to the intersection of the directions of the highways occupied by the vertex, that is, d c =d i ∩d j . Finally, given a set of directions D={d0,..., d m} (m is the number of highways), the highway can be represented as a subgraph For the remaining complementary graph In addition to the direction definition of the highway itself, the direction d of all other edges is uniformly defined as 0, indicating that the moving directions of these edges are not specified or unconstrained. This approach does not restrict the connectivity of adjacent vertices in the idle positions, ensuring the flexibility of such areas.

[0084] By defining the global highway, the purpose is to ensure that the paths of AGVs can share local cooperation rules, block path planning in the reverse highway direction, relieve congestion by resolving conflicts, and ensure the consistency of the planning results.

[0085] Step 3: Obtain the tasks to be assigned, as well as the starting coordinates and task point coordinates of all AGVs.

[0086] Step 4: Perform task point allocation based on the MTSP problem.

[0087] In this embodiment, the task allocation is modeled as a multi-traveling salesman problem (MTSP). Combining the AGV status (unloaded / loaded) and the occupancy of the shelf area, the optimization goal is to minimize the total path cost. An improved genetic algorithm GA is used to remove the crossover operator. For the optimization model established in this embodiment, the crossover operator is highly destructive and rarely improves the solution. A task allocation scheme is generated through tournament selection and mutation operations (flip, swap, slide) to ensure the optimal matching of tasks and AGVs.

[0088] Assume that there are m AGVs and n tasks in the system, and the current position of each AGV is S i , and the target point of each task is G j . The goal of task allocation is to minimize the total path cost for all AGVs to complete the tasks:

[0089]

[0090] where x ij is a binary variable indicating whether task j is assigned to AGV i ; D(S i , G j ) represents the Euclidean distance path cost from the current position S i of AGV i to the task target point G j .

[0091] Step 5: Based on the obtained task information, a multi-AGV path planning algorithm based on Conflict-Based Search (CBS) is used to plan the path. The lower layer of the algorithm performs path planning to generate an initial path; the upper layer of the algorithm traverses the paths to check whether there are conflicts between the paths. If there are no conflicts, the conflict-free paths are output and the AGVs execute the tasks; if there are path conflicts, conflict resolution is performed according to the proposed priority conflict resolution strategy and the underlying algorithm is called again for path planning until there are no conflicts between the paths.

[0092] The CBS multi-AGV path planning algorithm is a solution algorithm for Multi-agent Pathfinding (MAPF). The CBS algorithm adopts an upper and lower layer search structure. In the upper layer search, a binary tree structure is used to manage the conflict constraints among AGVs. If a conflict is found, corresponding constraints will be added for the lower layer search. In the lower layer search, the algorithm decomposes the MAPF problem into multiple single-AGV path planning problems, and uses a single-AGV path planning algorithm to plan paths for each AGV separately.

[0093] CBS solves the path conflict problem brought by lower layer path planning by adding constraints in the upper layer. The conflict types are divided into two types: point conflict and edge conflict.

[0094] For example, for two AGVs, AGV1 and AGV2, if the two AGVs need to pass through node v at the same time at time t, this constitutes a point conflict, which can be represented by <A1, A2, v, t>. The upper layer of CBS adds <A1, v, t> and <A2, v, t> constraints to the two AGVs respectively.

[0095] The edge conflict is that two AGVs exchange positions from node v1 to node v2 in the time period from time t to t + 1. AGV1 goes from node v1 to node v2, and AGV2 goes from node v2 to node v1; this conflict can be represented by <A1, A2, v1, v2, t>. The upper layer of CBS adds <A1, v1, v2, t> and <A2, v2, v1, t> constraints to the two AGVs respectively.

[0096] The CBS algorithm uses the data structure of Constraint Tree (CT) in the upper layer for conflict search and addition of path constraints. CT is a binary tree, and each node N includes three elements:

[0097] (1) N.Constraint: A set of constraint sets, composed of the imposed constraints;

[0098] (2) N.Solution: A solution that satisfies the constraint conditions, including multiple AGV paths. The paths are obtained in the lower layer single-AGV path planning search, and the number of paths is equal to the number of AGVs. When all paths satisfy all the added constraints, it means the solution is feasible;

[0099] (3) N.Cost: The total cost of the currently obtained solution, that is, the sum of the cost of all generated paths.

[0100] During the upper-level search, the root node of the CT is an empty constraint set. Subsequent nodes will inherit the constraints from the parent node and add new constraints based on new conflicts. If the N.Solution found by a certain node next is feasible, it means that the node is a conflict-free target node, and there is no AGV path conflict in the solution of this node. If the N.Solution of this node is not feasible, it means that there is a path conflict problem in the solution of this node. Then, two child nodes will be branched at this node, and new constraints will be added to them according to the type of path conflict. The two child nodes will generate a conflict-resolution solution and the total cost based on the constraints they add. The upper-level CT starts the best-first search and sorts the nodes in the CT according to the total cost of the solution. If there is a node with a feasible solution, the solution node with the minimum total cost will be selected according to the sorting. If the solutions generated by the two child nodes are both not feasible, child nodes will continue to be expanded based on the two child nodes until a feasible solution is found.

[0101] In the lower-level search, the improved A* algorithm is used for the single-path planning of each AGV. The main role of the lower level is to search for the shortest path for a single AGV to reach the target point. The traditional A* algorithm only considers the spatial dimension, while the improved A* algorithm proposed in this embodiment introduces the time dimension and heatmap congestion prediction to optimize the quality and efficiency of path planning. The search of the A* algorithm in the CBS algorithm needs to consider the constraints added in the upper-level search and consider the state of the AGV waiting in place.

[0102] In the actual intelligent warehousing environment, the quality of the generated path needs to be considered. When the solution contains a large number of right-angle turns and waiting nodes, this will cause the AGV to start and stop frequently, increasing energy consumption and mechanical wear, and even leading to failures. Therefore, the number of turns and waiting times in the generated path are added to the heuristic function to guide the algorithm to search for a smoother path. Further, a spatio-temporal joint search model is constructed by introducing a global heatmap. The formula is as follows:

[0103] f'(n) = g(n) + h'(n) (3)

[0104]

[0105] Among them, h(n) is the heuristic function of the traditional A* algorithm, β is the turning and waiting times penalty term, C(n, t') is the congestion prediction value, indicating the congestion situation in the grid at the current moment and including a dynamic update mechanism. f'(n) represents the improved A* cost function, g(n) represents the actual cost from the starting point to the current node n, turns represents the number of turns in the path, waits represents the number of waiting times in the path, t max represents the maximum time step, and n represents the current node.

[0106] The heat map model formula is as follows:

[0107]

[0108] Among them, C(n,t + 1) represents the node heat value update formula, and r represents the heat value decay coefficient.

[0109] When the AGV generates a path, the heat value of the grid position at each time step in its path π i is increased by a fixed value ΔC.

[0110] In the multi-AGV path planning problem, conflict resolution is the key to ensuring the efficient operation of the system. In the traditional conflict-based search (CBS) algorithm, when a conflict is encountered, the upper layer of the algorithm will add constraints to both sides of the conflict and generate two child nodes respectively. The generated solution may still have other conflicts after resolving the current conflict, and then the node with the lower total cost is selected to continue the expansion. However, this undirected search strategy is less efficient when the search space is large and cannot fully utilize the behavior characteristics and path selection flexibility of AGVs.

[0111] Combined with the intelligent warehousing scenario, this embodiment proposes a priority-oriented conflict resolution strategy, which dynamically allocates priorities by combining AGV types, path costs, and avoidance strategies to optimize the conflict resolution efficiency. The specific steps are as follows: The goal of the CBS algorithm is to find a conflict-free path with a smaller total cost. The total cost of the algorithm is the total number of time steps of all AGV paths in the solution. Therefore, the direction of the priority-oriented search is also the direction with a smaller increase in the cost value. AGVs are divided into loaded AGVs and unloaded AGVs, the conflict types are point conflicts and edge conflicts, and the avoidance strategies are waiting and path replanning. For the type of AGV, since the path selection of the loaded AGV is restricted and the task urgency is relatively high, a higher priority is given. If there is a unique shortest path in the solution, it has a higher priority compared to having multiple shortest paths. The waiting strategy requires the AGV to wait in place, which will inevitably increase the cost. Whether the cost of path replanning increases is related to the state of the AGV and the starting point. The path replanning cost of the loaded AGV needs to consider the structured scenario of the warehousing environment. Only when the difference in the horizontal coordinates between the starting point and the ending point of the AGV is greater than the horizontal grid length of the obstacle shelf, and the difference in the vertical coordinates is greater than the vertical grid length of the obstacle shelf, the AGV will have multiple shortest paths, and the path replanning cost will not increase. Let the horizontal length of a row of warehousing shelves be x c , and the vertical length be y c , the starting point coordinates of the AGV be (x start , y start ), and the ending point coordinates be (x goal , y goal ). When the following formula is satisfied:

[0112] |x start -x goal |>x c (6)

[0113] |y start -y goal |>y c (7)

[0114] It indicates that there are multiple shortest paths for this AGV.

[0115] For an empty-load AGV, the shelves in the scenario are not obstacles. Therefore, as long as the coordinate differences between the starting and ending points of the AGV in the horizontal and vertical directions are greater than 0, there are multiple shortest paths, and the following formula needs to be satisfied:

[0116] |x start -x goal |>0(8)

[0117] |y start -y goal |>0(9)

[0118] When an AGV with multiple shortest paths encounters a conflict, it can choose to switch paths without increasing the cost of the solution. Therefore, the AGV with a single shortest path is set to a high priority, and the AGV with multiple shortest paths is required to give way. The empty-load AGV with multiple paths has more path selection space than the loaded AGV, so the loaded AGV is set to have a higher priority. The priority settings are shown in Table 1, where the smaller the number, the higher the priority.

[0119] Table 1 AGV Priority Settings

[0120] AGV type Priority Unique shortest path load 1 Unique shortest path no-load 2 Multiple shortest paths load 3 Multiple shortest paths no-load 4

[0121] The multi-AGV collaborative path planning and scheduling method proposed in this embodiment has been improved in multiple aspects. First, in terms of task allocation, by modeling the task as a multi-traveling salesman problem (MTSP) and combining the AGV status and shelf area constraints, the rationality of task allocation is optimized. Second, in the path planning algorithm, a highway path guidance strategy is introduced based on CBS to strictly limit the moving direction of the AGV and reduce the possibility of conflicts. In addition, the improved A* algorithm introduces the time dimension and heat map congestion prediction, and optimizes the heuristic function to reduce the number of turns and waiting times. In terms of conflict resolution, through the dynamic priority strategy, the priority is adjusted according to the AGV status and path characteristics to give priority to ensuring the passage of the loaded AGV, improving the overall efficiency and throughput of the system.

[0122] Embodiment 2:

[0123] This embodiment verifies the technical effect of the multi-AGV collaborative path planning and scheduling method proposed by the present invention, and the specific process is as follows.

[0124] As Figure 1 shown, in this embodiment, the modeling of the chip intelligent warehousing environment and the setting of the highway path guidance strategy are first carried out. The scene map is set as a grid map with a size of 61*29, including 800 shelves placed in two columns and five rows. Some areas at the bottom are not used for placing anti-static shelves and are available for AGVs to drive freely. Eight picking areas are set, and the aisle width between the shelf areas can only pass one AGV. The shelf and AGV simulate a chip warehouse. The AGV lifts the shelf to realize the goods-to-person warehousing mode, and the unloaded AGV can pass under the shelf.

[0125] First, the system is initialized, mainly including AGV status information, shelf status information, task information, etc. The AGV status is divided into unloaded and loaded, which are represented by a white circular with a direction indicator and a red circular with a direction indicator respectively. The shelf is represented by a rectangle. The white rectangle represents the shelf area without goods, and the red rectangle represents that the current area is occupied by the shelf. The picking area is represented by a gray solid rectangle. The tasks are divided into the task of the unloaded AGV going to the shelf area with goods to perform the picking task and the task of the loaded AGV going to the shelf area without goods to perform the shelf storage task. The highway is set as a directed graph with the same size as the map, including the information of the running direction of the passage path. The forced guidance strategy is adopted, and all AGVs on the highway must run according to the path direction.

[0126] The implementation steps of this embodiment include: first, initialize, set basic parameters such as the number of AGVs, the size of the experimental scene, and the number of shelves; secondly, set the highway according to the experimental scene; randomly generate task instances, including the starting coordinates of the AGV and the coordinates of the task points; perform task point allocation based on the MTSP problem; hand over the obtained task information to each corresponding AGV, and the lower layer of the algorithm performs path planning to generate the initial path; the upper layer of the algorithm traverses the paths to find whether there are conflicts between the paths. If there are no conflicts, the conflict-free paths are output and the AGVs execute the tasks. If there are path conflicts, the conflict resolution is carried out according to the proposed priority conflict resolution strategy and the lower layer algorithm is called again for path planning until there are no conflicts between the paths.

[0127] In addition to the total path cost, the algorithm calculation time, and the number of expanded nodes, the performance evaluation also verifies the effectiveness of the method for the improved design evaluation indicators in a specific direction, such as the number of turning waiting times and the number of collisions generated in the path planning.

[0128] Adopting the technology provided by the present invention has the following characteristics:

[0129] (1) The task allocation strategy based on the highway guidance strategy and the MTSP problem proposed by the present invention reduces path conflicts and computational complexity.

[0130] (2) The improved A* algorithm proposed by the present invention combines heat map congestion prediction and turning waiting heuristic methods, reduces the expansion acceleration algorithm of nodes, optimizes the quality of path planning, and reduces the risk of cargo damage.

[0131] (3) The experimental results in the intelligent warehousing of chip semiconductors show the effectiveness of the multi-AGV collaborative optimization scheduling method proposed by the present invention, and there is a significant performance improvement compared with the baseline algorithm.

[0132] The specific performance is shown in Tables 1-4 as follows:

[0133] Table 1 Comparison of algorithm performance by path guidance strategy

[0134]

[0135]

[0136] Table 2 Comparison of algorithm performance by priority strategy

[0137]

[0138] Table 3 Comparison of algorithm performance by global heat map heuristic

[0139]

[0140] Table 4 Comparison of algorithm performance in the intelligent warehousing environment of chips

[0141] Evaluation index CBS The present invention Performance improvement Path cost 739.7 690.9 6.60％ Expanded node 97.6 23.4 76.02％ Algorithm running time 220.711 5.165 97.66％ Turn and wait count 136 35.2 74.12％ Collision count 56.578 12.2 78.44％

[0142] This embodiment verifies the effectiveness of different improvement strategies in improving the efficiency of multi-AGV path planning and scheduling in intelligent warehousing. As shown in Table 1, five instances are verified in the cases of 15 AGVs and 20 AGVs respectively. It can be seen that although the path cost increases slightly, there is a significant reduction in the number of node expansions and the algorithm running time. In the case of 15 AGVs, the algorithm running time is shortened by 82.3% compared with the benchmark algorithm, and in the case of 20 AGVs, it is shortened by 85.8%. It can be seen that the path guidance strategy can very effectively accelerate the algorithm.

[0143] Table 2 shows the use of the priority conflict resolution strategy. Since the load situation of the robot is considered, the unloaded AGV is allowed to run at the bottom of the shelf at this time, which results in a reduction in path cost. Through a series of priority definitions and conflict resolution strategies, when there are 15 AGVs, the number of node expansions and the algorithm running time are reduced by 70.5% and 68.5% respectively compared with the benchmark algorithm. When there are 20 AGVs, the number of node expansions and the algorithm running time are reduced by 75.5% and 75.6% respectively compared with the benchmark algorithm. The priority strategy can further improve the algorithm efficiency.

[0144] Table 3 shows the use of the global heat map heuristic strategy. By generating a heat map of the paths of global AGVs, the underlying algorithm can use the heat value as a heuristic value for path search to avoid searching in congested areas. The heat map method proposed in the present invention aims to reduce the conflicts generated in path search. It can be seen that after ten experiments, the number of collisions generated in path search is reduced by 73.8%, and there are also varying degrees of improvement in path cost, the number of expanded nodes, and algorithm running time.

[0145] Table 4 shows the experimental results of the effectiveness of the improved strategy proposed in the present invention for the multi-AGV chip warehousing system. In this embodiment, in the case of 20 AGVs, the experiments on ten task instances are compared. It can be seen that there is a performance improvement in each evaluation index. Compared with the baseline algorithm, the indicators of algorithm expanded nodes and running time are improved by 76.02% and 97.66% respectively. The improved strategy effectively reduces the computational complexity. For the evaluation indexes of path quality, the number of turns and waiting times is reduced by 74.12%. Thanks to the heuristic method proposed in the present invention, the path smoothness is improved, the risk of goods damage is reduced, and it meets the actual operation requirements of AGVs. The collisions generated in path search are also reduced by 78.44%. This enables the algorithm to consider the running conditions of other AGVs when searching for paths, avoid conflicts in advance, reduce the search nodes of algorithm expansion, thereby accelerating the algorithm. Combining the proposed path guidance strategy and the priority-based conflict resolution strategy further accelerates the algorithm.

[0146] In summary, through the multi-dimensional optimization strategy, the present invention provides an efficient collaborative path planning solution for the multi-AGV scenario of semiconductor chip intelligent warehousing.

[0147] Some steps in the embodiments of the present invention can be implemented by software, and the corresponding software program can be stored in a readable storage medium, such as a CD or a hard disk, etc.

[0148] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-AGV collaborative path planning and scheduling method for chip intelligent warehousing, characterized in that The method includes: Step 1: Model and initialize the chip intelligent warehousing environment; Step 2: Based on the highway path guidance strategy, set up highways on the chip intelligent warehousing environment model established in Step 1; Step 3: Obtain the tasks to be allocated, as well as the starting coordinates and task point coordinates of all AGVs; Step 4: Model the task allocation as a multi-traveling salesman problem, and use an improved genetic algorithm to remove the crossover operator. Generate a task allocation scheme through tournament selection and mutation operations. The goal of task allocation is to minimize the total path cost for all AGVs to complete the tasks: Among them, m represents the number of AGVs, n represents the number of tasks to be assigned, S i represents the current position of the i-th AGV, G j represents the target point position of the j-th task, x ij is a binary variable indicating whether the j-th task is assigned to the i-th AGV, D(S i , G j ) represents the Euclidean distance path cost from the current position S i of the i-th AGV to the target point position G j of the j-th task; Step 5: Use a multi-AGV path planning algorithm based on conflict search to plan the AGV paths. The lower layer of the algorithm performs path planning to generate an initial path; the upper layer of the algorithm traverses the paths to check whether there are conflicts between the paths. If there are no conflicts, output the conflict-free paths and the AGVs execute the tasks; if there are path conflicts, perform conflict resolution and re-call the lower-layer algorithm for path planning until there are no conflicts between the paths.

2. The method according to claim 1, wherein In Step 5, an improved A* algorithm is used for single-path planning of each AGV in the lower-layer search. The number of turns and waiting times in the generated paths are added to the heuristic function, and heat map congestion prediction is introduced. The improved A* algorithm is expressed as: f'(n) = g(n) + h'(n) Among them, f'(n) represents the cost function of the improved A* algorithm, g(n) represents the actual cost from the starting point to the current node n, h'(n) represents the improved heuristic function, h(n) represents the heuristic function of the traditional A* algorithm, β is the penalty term for the number of turning waits, C(n,t') represents the congestion prediction value, turns&waits represents the number of turns and waits in the path, and t max represents the maximum time step; The congestion prediction value includes a dynamic update mechanism, which is expressed as: Among them, r represents the thermal value decay coefficient; ΔC is the fixed value of thermal growth, and π i (t) represents the path of the AGV i .

3. The method according to claim 1, wherein In Step 5, a priority-oriented conflict resolution strategy is adopted, and priorities are dynamically allocated by combining AGV types, path costs, and avoidance strategies. The priority settings are as follows: Priority 1: Loaded AGVs with a unique shortest path; Priority 2: Unloaded AGVs with a unique shortest path; Priority 3: Loaded AGVs with multiple shortest paths; Priority 4: Unloaded AGVs with multiple shortest paths.

4. The method according to claim 3, characterized in that, When the starting point coordinates of the loaded AGV satisfy the following conditions, there are multiple shortest paths: |x start -x goal |>x c |y start -y goal |>y c When the starting point coordinates of the unloaded AGV satisfy the following conditions, there are multiple shortest paths: |x start -x goal | > 0 |y start -y goal | > 0 Among them, (x start , y start ) is the starting point coordinate of the AGV, and (x goal , y goal ) is the end point coordinate of the AGV; x c is the horizontal length of a row of storage shelves, and y c is the vertical length of a row of storage shelves.

5. A multi-AGV collaborative path planning and scheduling system for chip intelligent warehousing, characterized in that The system includes: A model construction module configured to model and initialize the chip intelligent warehousing environment; A highway setting module configured to set up highways on the chip intelligent warehousing environment model established in Step 1 based on the highway path guidance strategy; A task acquisition module configured to obtain the tasks to be allocated, as well as the starting coordinates and task point coordinates of all AGVs; A task allocation module configured to model the task allocation as a multi-traveling salesman problem, and use an improved genetic algorithm to remove the crossover operator. Generate a task allocation scheme through tournament selection and mutation operations. The goal of task allocation is to minimize the total path cost for all AGVs to complete the tasks: Among them, m represents the number of AGVs, n represents the number of tasks to be allocated, S i represents the current position of the i-th AGV, G j represents the target point position of the j-th task, x ij is a binary variable indicating whether the j-th task is allocated to the i-th AGV, D(S i , G j ) represents the Euclidean distance path cost from the current position S i of the i-th AGV to the target point position G j of the j-th task; The path planning module is configured to plan the AGV path by using a multi-AGV path planning algorithm based on conflict search. The lower layer of the algorithm performs path planning to generate an initial path. The upper layer of the algorithm traverses the paths to check if there are conflicts between the paths. If there are no conflicts, it outputs a conflict-free path and the AGV executes the task. If there are path conflicts, conflict resolution is performed and the underlying algorithm is called again for path planning until there are no conflicts between the paths.

6. The system according to claim 5, characterized in that, In the lower layer search of the path planning module, the improved A* algorithm is used for single path planning of each AGV. The number of turns and waiting times in the generated path are added to the heuristic function, and heat map congestion prediction is introduced. The improved A* algorithm is expressed as: f'(n) = g(n) + h'(n) Among them, f'(n) represents the cost function of the improved A* algorithm, g(n) represents the actual cost from the starting point to the current node n, h'(n) represents the improved heuristic function, h(n) represents the heuristic function of the traditional A* algorithm, β is the penalty term for the number of turning waits, C(n,t') represents the congestion prediction value, turns&waits represents the number of turns and waits in the path, and t max represents the maximum time step; The congestion prediction value includes a dynamic update mechanism and is expressed as: Among them, r represents the thermal value attenuation coefficient; ΔC is the fixed thermal growth value, and π i (t) represents the path of the AGV i .

7. The system according to claim 5, wherein The path planning module adopts a priority-oriented conflict resolution strategy, dynamically allocating priorities by combining AGV types, path costs, and avoidance strategies. The priority settings are as follows: Priority 1: Loaded AGV with a unique shortest path; Priority 2: Unloaded AGV with a unique shortest path; Priority 3: Loaded AGV with multiple shortest paths; Priority 4: Unloaded AGV with multiple shortest paths.

8. The system according to claim 5, wherein In the path planning module, when the starting point coordinates of the loaded AGV meet the following conditions, there are multiple shortest paths: |x start -x goal |>x c |y start -y goal |>y c When the starting point coordinates of the unloaded AGV meet the following conditions, there are multiple shortest paths: |x start -x goal | > 0 |y start -y goal | > 0 Among them, (x start , y start ) is the starting point coordinate of the AGV, and (x goal , y goal ) is the end point coordinate of the AGV; x c is the horizontal length of a row of storage shelves, and y c is the vertical length of a row of storage shelves.

9. A multi-AGV collaborative path planning and scheduling device for chip intelligent warehousing, characterized in that, It includes a memory and a processor; The memory is used to store computer programs; The processor is used to implement the multi-AGV collaborative path planning and scheduling method for chip intelligent warehousing as described in any one of claims 1 to 4 when executing the computer program.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the storage medium, and when the computer program is executed by the processor, it implements the multi-AGV collaborative path planning and scheduling method for chip intelligent warehousing as described in any one of claims 1 to 4.

Images