Cluster scheduling system and scheduling method based on mapf

Abstract

The application belongs to the technical field of intelligent warehousing, and particularly discloses a cluster scheduling system based on MAPF and a scheduling method thereof, which comprises a task allocation module, a graph clustering grouping module, a grouped parallel scheduling module and a global deadlock resolution module. The task allocation module is used for allocating scheduling tasks to each mobile robot. The graph clustering grouping module dynamically groups the robots based on real-time position information. The grouped parallel scheduling module performs single-vehicle path planning for each group of robots in a multithreading manner, and resolves path conflicts within the group in a rolling time window through a MAPF solver. The global deadlock resolution module detects deadlock and congestion of the robots in the whole scene after merging the paths of each group, and performs corresponding processing. The application combines grouping and parallel scheduling with MAPF solving, effectively reduces the computational complexity, reduces the probability of deadlock and congestion, and improves the scheduling efficiency of large-scale robot clusters and the robustness of the system.

Description

Technical Field

[0001] This invention belongs to the field of intelligent warehousing technology, specifically relating to a cluster scheduling system and scheduling method based on MAPF. Background Technology

[0002] Cluster scheduling systems are core control systems used for task allocation and path planning for large-scale mobile robots, playing a crucial role in intelligent manufacturing, warehousing and logistics, and automated material handling. With the continuous improvement of industrial automation, the number of mobile robots in a single project scenario has increased from dozens to hundreds or even thousands, placing higher demands on the carrying capacity and scheduling robustness of cluster scheduling systems. Existing cluster scheduling systems typically consist of a task allocation module, a single-vehicle path planning module, and a multi-vehicle scheduling deadlock resolution module. After uniformly planning single-vehicle paths for all vehicles, all vehicle paths are traversed to check for conflicts. Based on the conflict verification results, deadlock resolution and congestion replanning are executed using rule-based algorithms. However, as the number of scheduled vehicles increases, the constraint computation required for conflict resolution grows exponentially, and the scheduling scale that the system can typically handle is usually no more than 200 robots. Furthermore, existing systems rely heavily on specific rule-based algorithms for deadlock and congestion handling in different scene maps, lacking adaptability and prone to frequent deadlock and congestion when switching scene maps. Summary of the Invention

[0003] To address the limitations of existing technologies in large-scale vehicle cluster scheduling, where the computational load for conflict resolution increases exponentially with the number of vehicles, leading to limited system capacity and insufficient robustness in handling deadlocks and congestion across different map scenarios, this invention proposes a cluster scheduling system and method based on MAPF. The invention provides the following technical solution: A cluster scheduling system based on MAPF includes a task allocation module, a graph clustering grouping module, a grouped parallel scheduling module, and a global deadlock resolution module. The task allocation module is used to allocate corresponding scheduling tasks to each mobile robot in the scheduling scenario. The graph clustering grouping module is used to obtain the real-time location information of all mobile robots in the scene, and dynamically group all mobile robots based on the graph clustering algorithm to obtain multiple robot groups. The number of mobile robots in each robot group does not exceed a preset upper limit. The group parallel scheduling module is communicatively connected to the graph clustering grouping module. It is used to perform single-vehicle path planning for the mobile robots in each robot group in a multi-threaded parallel manner, and input the single-vehicle path planning results of each robot group into the MAPF solver. Within a preset rolling time window, it completes the resolution of multi-robot path conflicts within the group and obtains conflict-free paths within each group. The global deadlock resolution module is communicatively connected to the grouped parallel scheduling module. It is used to merge the conflict-free paths within all robot groups, perform global deadlock detection and congestion detection on all mobile robots in the scene, and execute corresponding deadlock resolution and congestion relief processing.

[0004] Preferably, the graph clustering grouping module has built-in hierarchical clustering logic, and performs graph clustering in the hierarchical order of map identifier, floor identifier, area identifier, and coordinate position; the total number of robot groups is determined by the quotient of the total number of mobile robots and the preset upper limit of the number of a single group; the graph clustering algorithm includes k-means clustering algorithm and DBSCAN clustering algorithm, and the k-means clustering algorithm is used by default.

[0005] Preferably, the grouped parallel scheduling module has a built-in scenario adaptive solver configuration unit, which is used to adaptively switch the core solving algorithm of the MAPF solver according to the scheduling scenario type; when the scheduling scenario is a transportation scenario, it is configured to use the CBS algorithm to resolve path conflicts; when the scheduling scenario is a warehousing and fulfillment scenario, it is configured to use the PBS algorithm to resolve path conflicts.

[0006] Preferably, the scenario adaptive solver configuration unit is further configured to adjust the length of the rolling time window according to the scheduling scenario type. In the transportation scenario, the rolling time window is set to 30 to 35 basic time units, and in the warehousing fulfillment scenario, the rolling time window is set to 20 to 25 basic time units.

[0007] Preferably, when the grouped parallel scheduling module performs intra-group path conflict resolution, it only includes path conflicts generated by mobile robots within the same group in the solution constraints, and conflicts generated by mobile robots in other groups are not included in the constraint calculation; the path conflicts included in the solution constraints include three types: vertex conflicts, edge conflicts and envelope conflicts. The three types of path conflicts are respectively converted into solution constraints corresponding to the MAPF solver, and intra-group conflict-free paths are generated based on all solution constraints.

[0008] A cluster scheduling method based on MAPF includes the following steps: S1. Task Allocation: Assign corresponding scheduling tasks to all mobile robots within the scheduling scenario; S2. Dynamic clustering and grouping: Obtain the real-time location information of all mobile robots in the scene, and dynamically group all mobile robots based on graph clustering algorithm to obtain multiple robot groups. The number of mobile robots in each robot group does not exceed a preset upper limit. S3. Grouped parallel planning and intra-group conflict resolution: The single-robot path planning is performed on each robot group in a multi-threaded parallel manner. The single-robot path planning results of each group are input into the MAPF solver. The intra-group multi-robot path conflict resolution is completed within a preset rolling time window to obtain the intra-group conflict-free path of each group. S4. Global grouping to resolve deadlocks: Merge all conflict-free paths within each robot group, perform global deadlock and congestion detection on all mobile robots in the scene, and execute corresponding deadlock resolution and congestion relief processes.

[0009] Preferably, in step S1, a bipartite graph matching algorithm is used to bind the scheduling task to the mobile robot one by one, thus completing the task allocation.

[0010] Preferably, in step S3, a graph search algorithm is first used to perform single-vehicle path planning for each mobile robot in the group to generate an initial single-vehicle path; then, all the initial single-vehicle paths of each group are substituted into the MAPF solver to perform intra-group conflict resolution.

[0011] Preferably, in step S3, the single-vehicle path segments within the group within the rolling time window are extracted and substituted into the MAPF solver to perform conflict resolution; the default length of the rolling time window is 25 basic time units, and the length is adjusted according to the scheduling scenario type.

[0012] Preferably, in step S4, after merging the paths of all robot groups, all mobile robots in the scene are traversed, deadlock loops and congestion relationships are identified based on path conflicts, and deadlock resolution and congestion replanning are performed through a rule-based algorithm strategy.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. By using a bipartite graph matching algorithm in the task allocation module, a one-to-one binding between scheduling tasks and mobile robots is achieved, realizing efficient matching between tasks and robots, thereby improving overall scheduling efficiency; 2. The graph clustering grouping module dynamically groups all mobile robots based on the graph clustering algorithm, realizing effective grouping management of large-scale vehicle clusters, and making the grouping status dynamically adjusted according to the real-time location of the vehicles. 3. By using a grouped parallel scheduling module to perform single-vehicle path planning for each robot in a multi-threaded parallel manner, the computation time is significantly reduced, and the scheduling scale can be expanded to approximately 1,000 mobile robots. 4. By incorporating intra-group path conflicts into the MAPF solver to solve constraints and isolating the conflict calculations of other groups, the amount of conflict resolution computation is effectively reduced. The core solving algorithm of the MAPF solver is adaptively switched according to the scheduling scenario type by the scenario adaptive solver configuration unit, thereby improving the system's adaptability to different scenario maps and scheduling robustness. 5. By using a rolling time window mechanism to extract single-vehicle path segments within a fixed time interval for conflict resolution, the real-time performance and feasibility of path planning are guaranteed. The global deadlock resolution module performs global deadlock detection and congestion detection on all mobile robots in the scene and executes corresponding deadlock resolution and congestion relief processing to reduce the overall deadlock rate and congestion rate of the system. Attached Figure Description

[0014] Figure 1 This is a system framework diagram of the present invention.

[0015] Figure 2 This is a flowchart of the scheduling method of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. The directional terms mentioned in the following embodiments, such as "up", "down", "left", and "right", are only for reference to the directions in the accompanying drawings. Therefore, the directional terms used are for illustration and not for limiting the present invention.

[0017] like Figure 1 As shown, a cluster scheduling system based on MAPF includes a task allocation module, a graph clustering grouping module, a grouped parallel scheduling module, and a global deadlock resolution module.

[0018] The task allocation module is used to assign corresponding scheduling tasks to each mobile robot in the scheduling scenario. It uses a bipartite graph matching algorithm to achieve one-to-one binding between scheduling tasks and mobile robots, thereby improving the overall scheduling efficiency through efficient matching of tasks and robots.

[0019] The graph clustering grouping module is used to obtain the real-time location information of all mobile robots in the scene. Based on the graph clustering algorithm, all mobile robots are dynamically grouped to obtain multiple robot groups. The number of mobile robots in each robot group does not exceed the preset limit. Graph clustering is performed in the hierarchical order of map identifier, floor identifier, area identifier, and coordinate position. The total number of robot groups is determined by the quotient of the total number of mobile robots and the preset limit of the number of a single group. The graph clustering algorithm includes k-means clustering algorithm and DBSCAN clustering algorithm. The default is to use k-means clustering algorithm. Through real-time grouping, the group state is continuously dynamically adjusted with the real-time position of the vehicles, thereby adapting to the current full map state information and effectively reducing the computational complexity of resolving conflicts within the group.

[0020] The grouped parallel scheduling module communicates with the graph clustering grouping module. It performs single-robot path planning for each robot group in a multi-threaded parallel manner, and inputs the single-robot path planning results of each robot group into the MAPF solver. Within a preset rolling time window, it resolves multi-robot path conflicts within the group, obtaining conflict-free paths for each group. It includes a built-in scene-adaptive solver configuration unit, which adaptively switches the core solving algorithm of the MAPF solver according to the scheduling scene type. When the scheduling scene is a material handling scenario, it is configured to use the CBS algorithm for path conflict resolution; when the scheduling scene is a warehouse fulfillment scenario, it is configured to use the PBS algorithm. Simultaneously, it adjusts the length of the rolling time window according to the scheduling scene type: 30 to 35 basic time units for material handling scenarios and 20 to 25 basic time units for warehouse fulfillment scenarios. When resolving path conflicts, only path conflicts generated by mobile robots within the same group are included in the solution constraints. Conflicts generated by mobile robots in other groups are not included in the constraint calculation. The path conflicts included in the solution constraints include three types: vertex conflicts, edge conflicts, and envelope conflicts. The three types of path conflicts are converted into the solution constraints corresponding to the MAPF solver. Based on all solution constraints, conflict-free paths within the group are generated. The default upper limit of the preset number of robots in each group is 100, which can be adjusted down according to the hardware performance conditions of the deployed hardware, with a minimum of 50 robots. By parallel solving within the group and isolated calculation of conflict constraints, the overall system computation time is significantly reduced, and the scheduling scale can be expanded to about 1000 mobile robots. The hardware performance conditions are judged by the dimensions of CPU utilization, memory utilization, and average time of solving a single round. When the hardware load exceeds the preset threshold for multiple consecutive scheduling cycles, the logic of reducing the group size is automatically triggered.

[0021] The global deadlock resolution module communicates with the grouped parallel scheduling module and is used to merge the conflict-free paths within all robot groups. It performs global deadlock detection and congestion detection on all mobile robots in the scene and executes corresponding deadlock resolution and congestion relief processing. By traversing all mobile robots in the scene, it identifies deadlock loops and congestion relationships based on path conflicts, and executes deadlock resolution and congestion replanning through rule-based algorithm strategies to further reduce the overall deadlock rate and congestion rate of the system. like Figure 2 As shown, a cluster scheduling method based on MAPF includes the following steps: S1. Task Allocation: Assign corresponding scheduling tasks to all mobile robots in the scheduling scenario. Use a bipartite graph matching algorithm to bind scheduling tasks to mobile robots one by one. Improve overall scheduling efficiency through efficient matching of tasks and robots. S2. Dynamic Clustering Grouping: Obtain the real-time location information of all mobile robots in the scene, and dynamically group all mobile robots based on the graph clustering algorithm to obtain multiple robot groups. The number of mobile robots in each robot group does not exceed the preset limit. Graph clustering is performed in the hierarchical order of map identifier, floor identifier, area identifier, and coordinate position. The total number of robot groups is determined by the quotient of the total number of mobile robots and the preset limit of the number of a single group. The graph clustering algorithm includes k-means clustering algorithm and DBSCAN clustering algorithm. The default is to use k-means clustering algorithm. Through real-time grouping, the grouping state is continuously dynamically adjusted with the real-time position of the vehicle, thereby adapting to the current full map state information. S3. Grouped Parallel Planning and Intra-Group Conflict Resolution: Single-robot path planning is performed on each mobile robot within a robot group using a multi-threaded parallel approach. The single-robot path planning results of each group are input into the MAPF solver. Within a preset rolling time window, intra-group multi-robot path conflict resolution is completed, resulting in conflict-free paths for each group. First, a graph search algorithm is used to perform single-robot path planning on each mobile robot within the group (the graph search algorithm used for single-robot path planning includes, but is not limited to, the A-star algorithm; an appropriate path search algorithm can be selected based on the topology of the scene map), generating initial single-robot paths. Then, path segments within the rolling time window of all initial single-robot paths from each group are extracted and substituted into the MAPF solver to resolve intra-group conflicts. The system automatically resolves conflicts by adaptively switching the core solution algorithm of the MAPF solver based on the scheduling scenario type. In the material handling scenario, the CBS algorithm is used with a rolling time window of 30 to 35 basic time units, while in the warehouse fulfillment scenario, the PBS algorithm is used with a rolling time window of 20 to 25 basic time units. Only path conflicts generated by mobile robots within the same group are included in the solution constraints. Path conflicts include three types: vertex conflicts, edge conflicts, and envelope conflicts. The three types of path conflicts are converted into the solution constraints corresponding to the MAPF solver. Based on all solution constraints, conflict-free paths within the group are generated. By parallel solving within the group and isolated calculation of conflict constraints, the overall computation time of the system is greatly reduced, and the scheduling scale can be expanded to about 1,000 mobile robots. S4. Global grouping to resolve deadlocks: Merge all non-conflicting paths within each robot group, perform global deadlock and congestion detection on all mobile robots in the scene, and execute corresponding deadlock resolution and congestion relief processes. Traverse all mobile robots in the scene, identify deadlock loops and congestion relationships based on path conflicts, and execute deadlock resolution and congestion replanning through rule-based algorithm strategies to further reduce the overall deadlock rate and congestion rate of the system. The rolling time window mechanism uses a fixed-length time interval starting from the current time to extract single-vehicle path segments within a group for MAPF solver conflict resolution. The time window rolls forward with the scheduling cycle, ensuring the real-time performance and feasibility of path planning. The graph clustering grouping module reacquires the real-time coordinates of all mobile robots at the beginning of each scheduling cycle and re-executes the graph clustering algorithm based on the position information to update the grouping, ensuring that the grouping results always reflect the current vehicle distribution status. During the solution process, the MAPF solver establishes corresponding constraint equations for vertex conflicts, edge conflicts, and envelope conflicts. Vertex conflict constraints ensure that no two mobile robots occupy the same node at the same time, edge conflict constraints ensure that no two mobile robots travel in opposite or the same direction on the same edge at the same time, and envelope conflict constraints ensure the safe passage distance for mobile robots in turning or narrow passage scenarios. When switching between the CBS algorithm and the PBS algorithm, the scene adaptive solver configuration unit automatically inherits the current rolling time window settings and grouping parameters, completing the algorithm switch without manual intervention, improving the system's adaptability to different scene maps and scheduling robustness.

[0022] This invention merges conflict-free paths within all robot groups through a global deadlock resolution module, traverses all mobile robots in the scene, identifies deadlock loops and congestion relationships based on path conflicts, and executes deadlock resolution and congestion replanning through a rule-based algorithm strategy, continuously completing the full cluster scheduling cycle, which is highly efficient and adaptable.

[0023]

[0024] As shown in the table above, compared with the existing technology, this invention patent has made substantial improvements in terms of scheduling scale, computing efficiency and deadlock rate. It can effectively increase the scale of dispatched vehicles to thousands of vehicles. It can adjust its own scheduling algorithm for different scene maps by autonomously switching different core algorithms, thereby improving the system's robustness to different scene maps.

Claims

1. A cluster scheduling system based on MAPF, characterized in that, It includes a task allocation module, a graph clustering and grouping module, a grouped parallel scheduling module, and a global deadlock resolution module; The task allocation module is used to allocate corresponding scheduling tasks to each mobile robot in the scheduling scenario. The graph clustering grouping module is used to obtain the real-time location information of all mobile robots in the scene, and dynamically group all mobile robots based on the graph clustering algorithm to obtain multiple robot groups. The number of mobile robots in each robot group does not exceed a preset upper limit. The group parallel scheduling module is communicatively connected to the graph clustering grouping module. It is used to perform single-vehicle path planning for the mobile robots in each robot group in a multi-threaded parallel manner, and input the single-vehicle path planning results of each robot group into the MAPF solver. Within a preset rolling time window, it completes the resolution of multi-robot path conflicts within the group and obtains conflict-free paths within each group. The global deadlock resolution module is communicatively connected to the grouped parallel scheduling module. It is used to merge the conflict-free paths within all robot groups, perform global deadlock detection and congestion detection on all mobile robots in the scene, and execute corresponding deadlock resolution and congestion relief processing.

2. The cluster scheduling system based on MAPF according to claim 1, characterized in that, The graph clustering grouping module has built-in hierarchical clustering logic, which performs graph clustering in the hierarchical order of map identifier, floor identifier, area identifier, and coordinate position. The total number of robot groups is determined by the quotient of the total number of mobile robots and the preset upper limit of the number of a single group. The graph clustering algorithm includes k-means clustering algorithm and DBSCAN clustering algorithm, with k-means clustering algorithm used by default.

3. The cluster scheduling system based on MAPF according to claim 1, characterized in that, The grouped parallel scheduling module has a built-in scenario adaptive solver configuration unit, which is used to adaptively switch the core solving algorithm of the MAPF solver according to the scheduling scenario type. When the scheduling scenario is a transportation scenario, it is configured to use the CBS algorithm to resolve path conflicts; when the scheduling scenario is a warehousing and fulfillment scenario, it is configured to use the PBS algorithm to resolve path conflicts.

4. The cluster scheduling system based on MAPF according to claim 3, characterized in that, The scenario adaptive solver configuration unit is also used to adjust the length of the rolling time window according to the scheduling scenario type. The rolling time window is set to 30 to 35 basic time units in the transportation scenario and 20 to 25 basic time units in the warehousing fulfillment scenario.

5. The cluster scheduling system based on MAPF according to claim 1, characterized in that, When the grouped parallel scheduling module performs intra-group path conflict resolution, it only includes path conflicts generated by mobile robots within the same group in the solution constraints, while conflicts generated by mobile robots in other groups are not included in the constraint calculation. The path conflicts included in the solution constraints include three types: vertex conflicts, edge conflicts, and envelope conflicts. The three types of path conflicts are respectively converted into solution constraints corresponding to the MAPF solver, and a conflict-free path within the group is generated based on all solution constraints.

6. A cluster scheduling method based on MAPF, characterized in that, Includes the following steps: S1. Task Allocation: Assign corresponding scheduling tasks to all mobile robots within the scheduling scenario; S2. Dynamic clustering and grouping: Obtain the real-time location information of all mobile robots in the scene, and dynamically group all mobile robots based on graph clustering algorithm to obtain multiple robot groups. The number of mobile robots in each robot group does not exceed a preset upper limit. S3, Grouped Parallel Planning and Intra-Group Conflict Resolution; The single-robot path planning is performed on each robot group in a multi-threaded parallel manner. The single-robot path planning results of each group are input into the MAPF solver. The multi-robot path conflict resolution within the group is completed within a preset rolling time window to obtain the conflict-free path within each group. S4. Global grouping to resolve deadlocks: Merge all conflict-free paths within each robot group, perform global deadlock and congestion detection on all mobile robots in the scene, and execute corresponding deadlock resolution and congestion relief processes.

7. The cluster scheduling method based on MAPF according to claim 6, characterized in that, In step S1, a bipartite graph matching algorithm is used to bind the scheduling task to the mobile robot one by one, thus completing the task allocation.

8. The cluster scheduling method based on MAPF according to claim 6, characterized in that, In step S3, a graph search algorithm is first used to perform single-vehicle path planning for each mobile robot in the group to generate initial single-vehicle paths; then, all the initial single-vehicle paths of each group are substituted into the MAPF solver to perform intra-group conflict resolution.

9. The cluster scheduling method based on MAPF according to claim 6, characterized in that, In step S3, the single-vehicle path segments within the group within the rolling time window are extracted and substituted into the MAPF solver to perform conflict resolution; the default length of the rolling time window is 25 basic time units, and the length is adjusted according to the scheduling scenario type.

10. The cluster scheduling method based on MAPF according to claim 6, characterized in that, In step S4, after merging the paths of all robot groups, all mobile robots in the scene are traversed, deadlock loops and congestion relationships are identified based on path conflicts, and deadlock resolution and congestion replanning are performed through a rule-based algorithm strategy.

Images