AGV cluster scheduling system and method based on task and resource matching

Abstract

The application discloses a kind of based on task and resource matching AGV cluster scheduling system and method, system includes task analysis and decomposition module, AGV dynamic capability image module, multi-mode task-resource matching module, space-time synchronous path planning module, predictive resource scheduling module and dynamic adjustment module.The multi-mode task-resource matching module executes single machine and cooperation matching in parallel and single machine is prior;The predictive resource scheduling module inserts charging task in adjacent task idle window;The space-time synchronous path planning module generates following trajectory by space translation and time offset with main AGV trajectory as benchmark;The dynamic adjustment module executes flexible self-healing, promotes following AGV when main AGV fails and keeps progress recovery.This application realizes the fusion of heterogeneous AGV cluster flexible scheduling, charging and task execution, and adaptive expansion and abnormal recovery, significantly improves resource utilization and task continuity.

Description

Technical Field

[0001] This invention relates to the field of AGV scheduling technology, specifically to an AGV cluster scheduling system and method based on task and resource matching. Background Technology

[0002] With the rapid development of intelligent manufacturing and smart warehousing, modern factories and logistics centers typically operate a variety of heterogeneous equipment, such as forklift AGVs, roller conveyor AGVs, and stacker robots. These various types of equipment differ significantly in terms of load capacity, operating speed, and applicable aisle conditions, forming a heterogeneous AGV cluster. How to efficiently schedule this heterogeneous cluster to collaboratively complete material handling tasks is a core technical challenge currently facing the field of smart logistics.

[0003] The existing AGV cluster scheduling scheme has the following problems in practical applications.

[0004] I. The task-resource matching mode is too simplistic and lacks flexibility. Existing solutions generally adopt a hard matching rule of "task type → specified AGV type". When all AGVs of the matched type are busy, even if the combined capacity of multiple idle AGVs of other types can meet the task requirements, the system cannot automatically activate the multi-machine collaboration mode, resulting in wasted idle resources and continuous task backlog.

[0005] Second, charging scheduling and task scheduling are separated. When the AGV's battery level falls below a threshold, the current task is interrupted and the AGV proceeds to charging. For multi-machine collaborative tasks, if a single AGV exits midway, the entire formation will come to a standstill, severely affecting task continuity.

[0006] In summary, existing AGV cluster scheduling solutions urgently need a unified scheduling system and method that can achieve flexible multi-mode task-resource matching, charging scheduling and task scheduling through deep integration. Summary of the Invention

[0007] The purpose of this invention is to provide an AGV cluster scheduling system and method based on task and resource matching, thereby solving the problems existing in the prior art.

[0008] The technical solution of the present invention is as follows:

[0009] An AGV cluster scheduling system based on task and resource matching, characterized in that it includes:

[0010] The task parsing and decomposition module is used to receive logistics tasks issued by the upper-level system, extract task attributes, dynamically decompose the task into a standardized atomic task sequence containing pickup, transfer, and unloading, and determine the resource matching requirements of each atomic task.

[0011] The AGV dynamic capability profiling module is used to collect and maintain the static capability parameters and dynamic status parameters of each AGV in the cluster in real time, and integrate them into a capability vector in a unified format.

[0012] The multi-mode task-resource matching module is used to match and select the best between single-machine execution and multi-machine collaboration modes based on the resource matching requirements and the capability vector, and output a scheduling scheme that includes the execution mode, the list of participating AGV identifiers, and the master-slave role allocation.

[0013] The spatiotemporal synchronization path planning module is used to generate spatiotemporal trajectories with time windows for each participating AGV according to the scheduling scheme.

[0014] The predictive resource scheduling module is used to predict the future task density distribution for pre-deployment of idle AGVs, and to schedule charging between tasks based on the task plan and power consumption trend of each AGV.

[0015] The dynamic adjustment module is used to detect the positional deviation of the following AGV in multi-machine collaborative tasks and perform closed-loop dynamic compensation, as well as to perform rescheduling when the AGV fails.

[0016] Furthermore, the multi-mode task-resource matching module includes a single-machine matching submodule, a collaborative combination search submodule, and a scheme selection submodule. The single-machine matching submodule and the collaborative combination search submodule are executed in parallel. The scheme selection submodule follows the single-machine scheme priority principle: a collaborative scheme is selected only when the single-machine scheme is empty or the estimated completion time of the single-machine scheme is significantly longer than that of the collaborative scheme.

[0017] Furthermore, the collaborative combination search submodule searches for multi-machine combination schemes among AGVs with collaborative function identifiers and in an idle state. The search conditions include: the sum of the rated loads of each AGV in the group is greater than or equal to the weight constraint; the wireless communication delay between each AGV in the group is lower than a preset collaborative delay threshold; and the maximum difference between the distances of each AGV in the group to the picking station does not exceed a preset grouping distance threshold.

[0018] Furthermore, when the execution mode is multi-machine collaboration, the multi-mode task-resource matching module selects the AGV with the largest actual acceleration at full load and the smallest instruction response delay to serve as the master AGV, and the remaining AGVs serve as follower AGVs.

[0019] Furthermore, the spatiotemporal synchronization path planning module includes a main trajectory generation submodule and a follower trajectory generation submodule. The follower trajectory generation submodule uses the main AGV reference spatiotemporal trajectory as a reference and performs spatial translation and temporal offset on the coordinates of each time point on the main AGV reference spatiotemporal trajectory to generate an individual spatiotemporal trajectory for each follower AGV. The spatial translation refers to adding the formation position offset vector corresponding to the follower AGV to the coordinates of each time point on the main AGV reference trajectory. The temporal offset refers to shifting the main AGV reference trajectory on the time axis by a time advance. The time advance is composed of the sum of the response delay difference and acceleration difference between the follower AGV and the main AGV.

[0020] Furthermore, the predictive resource scheduling module includes an intelligent charging scheduling submodule. For AGVs marked as having low power risk, the intelligent charging scheduling submodule arranges their currently executing tasks and assigned subsequent tasks in chronological order to construct a task timeline. It then iterates through the gaps between adjacent tasks on the task timeline. If the gap duration is greater than or equal to the minimum charging duration of the idle time window, a charging task is inserted within that window. If no idle time window meets the conditions, the AGV is marked as having a charging risk state and its ability to receive new tasks is restricted.

[0021] Furthermore, the dynamic adjustment module includes a task deviation detection submodule, which uses a closed-loop dynamic compensation method to calculate the compensation speed. The calculation formula for the compensation speed is:

[0022] ;

[0023] In the formula: For the first The platform follows the AGV at all times Compensation speed command ( ); For the first The platform follows the AGV at all times The nominal planned speed is taken as a preset proportion of the current speed of the main AGV as the safe tracking speed. ); For the first The platform follows the AGV at all times The positional deviation, i.e., the Euclidean distance between the actual position and the preset formation position ( ); For the first The proportional gain coefficient of the AGV following the platform; For the first The differential gain coefficient of the AGV following the platform; For monitoring period ( ).

[0024] Furthermore, the dynamic adjustment module includes a dynamic rescheduling submodule. When the dynamic rescheduling submodule detects a failure of the main AGV or a heartbeat timeout, it executes the following elastic self-healing process: if there is an AGV among the original following AGVs that has the capacity to continue executing the task, it promotes it to the new main AGV; if an AGV needs to be added due to load requirements, it automatically searches for the nearest available idle AGV, completes rapid calibration, and adds it to the formation; it retains the completed progress and only replans the remaining transfer path from the current stopping position.

[0025] A method for scheduling AGV clusters based on task and resource matching, characterized by the following steps:

[0026] Step S1: Receive logistics tasks, extract task attributes, dynamically decompose the tasks into a standardized sequence of atomic tasks including pickup, transfer, and unloading, and determine the resource matching requirements of each atomic task.

[0027] Step S2: Collect and maintain the static capability parameters and dynamic status parameters of each AGV in the cluster in real time, and integrate them into a capability vector in a unified format;

[0028] Step S3: Perform single-machine matching and cooperative combination search in parallel, and output the final scheduling scheme according to the principle of prioritizing single-machine schemes. The final scheduling scheme includes the execution mode, the list of participating AGV identifiers, and the master-slave role allocation.

[0029] Step S4: Generate a reference spatiotemporal trajectory with differentiated node priority weights based on the task timeliness level of the AGV or the main AGV. For multi-machine collaboration mode, take the main AGV reference trajectory as the benchmark, and perform spatial translation and time offset on the reference trajectory according to the differences between each following AGV and the main AGV in command response delay and acceleration to generate the individual spatiotemporal trajectory of each following AGV.

[0030] Step S5: Predict the task density distribution of each area in the future, schedule idle and fully charged AGVs to the expected high task density area in advance, and deduce the power change curve of each AGV. Insert charging tasks during task gaps before the power is predicted to fall below the safety threshold.

[0031] Step S6: Real-time acquisition of the position and task progress of each AGV, detection of position deviation of the following AGV in multi-machine collaborative tasks and closed-loop dynamic compensation, and incremental rescheduling of affected and incomplete atomic tasks when AGV failure or new emergency task is detected.

[0032] Furthermore, step S2 also includes a plug-and-play process for the new AGV: when the new AGV first accesses the network, it automatically broadcasts the device serial number and type code, queries the device model library to automatically match the default static parameters, and issues a standard calibration task to complete the capability calibration. The calibration result is automatically updated to the capability vector, and the initial value of the collaborative synchronization parameter adopts the average value of the same model of AGV.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. This invention uses a multi-mode task-resource matching module to perform single-machine matching and collaborative combination search in parallel, and follows the principle of single-machine priority. When all single machines with sufficient capacity are busy, it automatically evaluates and enables a multi-machine collaboration scheme, making full use of idle AGV resources to complete tasks that would otherwise require waiting, thus significantly improving the overall resource utilization of heterogeneous AGV clusters.

[0035] 2. This invention uses a predictive resource scheduling module to predict the power change trend of each AGV. Before the power drops below the safety threshold, it actively inserts charging tasks during the idle time window between adjacent tasks, fundamentally avoiding AGVs being forced to exit midway through the execution of tasks. In particular, it ensures that all AGVs in the formation have sufficient power during the execution of multi-machine collaborative tasks, and eliminates the problem of the entire formation stopping due to the departure of a single AGV midway.

[0036] 3. This invention utilizes the elastic self-healing mechanism in the dynamic rescheduling submodule to automatically promote a capable following AGV to the new main AGV when the main AGV fails, automatically search for and calibrate the replacement AGV, retain the completed progress, and resume the task from the current position, thereby maximizing the continuity of tasks and improving the robustness and scalability of the system. Attached Figure Description

[0037] Figure 1 This is a diagram showing the overall architecture of the system of the present invention. Detailed Implementation

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1

[0040] like Figure 1 As shown, an AGV cluster scheduling system based on task and resource matching is deployed on a central scheduling server. The system communicates with the on-board controllers of each AGV through an industrial wireless network and consists of six core modules and their respective sub-modules.

[0041] (1) Task Analysis and Decomposition Module

[0042] The task parsing and decomposition module receives logistics tasks from the upper-level task management system, extracts task attributes, and dynamically decomposes the tasks into standardized atomic tasks based on the real-time status of each AGV at the scheduling time, determining the required execution mode. This module includes a task attribute extraction submodule and an atomic task decomposition submodule.

[0043] The task attribute extraction submodule extracts the cargo weight, cargo dimensions, vehicle type, pickup station coordinates, unloading station coordinates, and task timeliness level from the received logistics task.

[0044] The atomic task decomposition submodule dynamically decomposes logistics tasks into a standardized sequence of atomic tasks, including pickup, transfer, and unloading, based on cargo attributes and handling processes. It generates a globally unique task identifier for each atomic task and distributes the task identifier, along with constraints such as cargo weight and external dimensions, as resource matching requirements to the multi-mode task-resource matching module and the dynamic adjustment module.

[0045] (2) AGV dynamic capability profiling module

[0046] The AGV dynamic capability profiling module is used to collect and maintain the static capability parameters and dynamic status parameters of each AGV in the cluster in real time, and integrate the parameters into a unified capability vector for other modules to use. This module includes a static parameter management submodule and a dynamic status acquisition submodule.

[0047] The static parameter management submodule stores the static capability attributes of each AGV, including AGV type (forklift, transport, stacker), rated load, equipment dimensions (length, width), minimum turning radius, and collaborative function identifier (whether multi-machine team collaboration is supported).

[0048] The dynamic status acquisition submodule is based on The frequency collects real-time status data packets reported by each AGV through the industrial wireless network. The data packets include AGV identifier, current coordinates, running speed, currently executing task identifier, and hardware status code. The above data is aggregated in time series to generate and update a status snapshot covering the execution status of all AGVs and tasks in the field in real time, while monitoring hardware fault codes and heartbeat timeouts.

[0049] In addition, this submodule also calculates two dynamic capability parameters. Actual acceleration under full load: This is obtained by multiplying the AGV's no-load calibrated acceleration by the ratio of the current load to the rated load, and then multiplying by a reduction factor. The reduction factor reflects the nonlinear attenuation effect of increased load on acceleration and is stored in the AGV's local controller through factory calibration.

[0050] Command response delay: Calculated by the moving average of the time difference between the time when the system issues a motion command and the time when the AGV actually starts moving, based on feedback from the last 50 system motion commands.

[0051] This module also includes a new AGV rapid calibration subroutine: when a new AGV connects to the network for the first time, it automatically broadcasts the device serial number and type code. The static parameter management submodule queries the device model library to automatically match and load the default static parameters for that model. The system issues a set of standard calibration tasks to the AGV: no-load acceleration test (accelerating from a standstill to 1.0 m / s, recording the actual acceleration); full-load acceleration test (loading a standard counterweight, recording the full-load acceleration); command response test (issuing 20 consecutive start / stop commands, recording the response delay); and cooperative communication test (exchanging heartbeat packets with neighboring AGVs, recording the communication delay). The calibration results are automatically updated to the capability vector. The initial values ​​of the new AGV's cooperative synchronization parameter subset are the average values ​​of AGVs of the same model, requiring no manual intervention throughout the process.

[0052] (3) Multi-mode task-resource matching module

[0053] The multi-mode task-resource matching module is responsible for receiving standardized atomic task sequences and resource matching requirements, and performing centralized resource matching and optimization between single-machine execution and multi-machine collaboration modes. This module includes a single-machine matching submodule, a collaborative combination search submodule, and a solution optimization submodule, with the collaborative combination search submodule executing in parallel with the single-machine matching submodule.

[0054] The single-machine matching submodule receives standardized atomic tasks and corresponding resource matching requirements from the atomic task decomposition submodule. It selects AGVs with rated loads that meet weight constraints and are currently idle from all AGVs as candidate single machines. Combining the real-time position, picking station coordinates, motion parameters, and power consumption rate of the candidate AGVs, it evaluates the estimated execution time and estimated total energy consumption of each candidate single machine to complete the task and generates a list of candidate single-machine solutions. If no AGV meets the conditions, it returns an empty set.

[0055] The collaborative combination search submodule receives standardized atomic tasks and corresponding resource matching requirements from the atomic task decomposition submodule. It then searches for multi-machine combination schemes among AGVs with collaborative function identifiers that are currently idle. Search conditions include: the sum of the rated loads of all AGVs in the group is greater than or equal to the weight constraint; the wireless communication delay between all AGVs in the group is lower than a preset collaborative delay threshold; and the distance from each AGV in the group to the picking station is calculated based on the real-time position of each AGV, with the maximum difference between these distances not exceeding a preset grouping distance threshold. If all the above conditions are met, a list of candidate collaborative schemes is output; otherwise, an empty set is returned.

[0056] The scheme selection submodule receives the outputs from the single-machine matching submodule and the collaborative combination search submodule, and selects the best scheme according to the principle of prioritizing single-machine schemes: the collaborative scheme is selected only when the single-machine scheme is empty, or when the estimated completion time of the single-machine scheme is significantly longer than that of the collaborative scheme due to the candidate AGV being too far away; if both are empty, a notification of no available resources is returned to the upper-level system.

[0057] The final scheduling scheme outputs the execution mode, the list of participating AGV identifiers, and the master-slave role assignment. When the execution mode is cooperative, the AGV with the largest actual acceleration at full load and the smallest response latency is selected as the master AGV, and the remaining AGVs serve as follower AGVs.

[0058] (4) Spatiotemporal synchronization path planning module

[0059] The spatiotemporal synchronization path planning module is responsible for transforming the final scheduling scheme output by the scheme selection submodule into a sequence of executable driving paths and speed commands for each AGV. This module includes a main trajectory generation submodule, a following trajectory generation submodule, and a spatiotemporal conflict resolution submodule.

[0060] The main trajectory generation submodule generates a reference spatiotemporal trajectory for the AGV or the main AGV based on the task's timeliness level. Timeliness level is quantified as path node priority weights; urgent tasks are assigned a multiple of the standard weight (e.g., 1.5 times), while ordinary tasks are assigned a baseline weight. During path search, the AGV's maximum safe speed and acceleration under full load are used as motion constraints. A path search algorithm considering node priority weights is employed to generate a spatiotemporal trajectory with a time window from the current position through the picking station to the unloading station.

[0061] The trajectory generation submodule operates in multi-machine collaborative mode. Using the main AGV's reference spatiotemporal trajectory as a benchmark, it generates individual spatiotemporal trajectories for each following AGV in the formation. During generation, it first determines the formation position offset vector of each following AGV relative to the main AGV based on the cargo's external dimensions and the total number of AGVs involved in the collaborative transport. Subsequently, it performs spatial translation and temporal offset on the coordinates of each time point on the main AGV's spatiotemporal trajectory.

[0062] Spatial translation refers to adding the formation position offset vector of the following AGV to the coordinates of each time point on the main AGV reference trajectory in a two-dimensional plane coordinate system, thus obtaining the target position of the following AGV at the corresponding time point. Time offset refers to shifting the main AGV reference trajectory on the time axis by a time advance, which is the sum of the difference in response delay and acceleration time between the following AGV and the main AGV. The calculation formula is as follows:

[0063] ;Mode

[0064] In the formula: For the first The time lead of the platform following the AGV ( ); For the first The delay in the response to instructions from the AGV (Automated Guided Vehicle) ); Command response delay of the main AGV ( ); To achieve a unified target speed for coordinated formation ( ); For the first The actual acceleration of the AGV under full load (followed by the platform) ); The actual acceleration of the main AGV under full load ( ).

[0065] After spatial translation and time offset, each following AGV maintains the preset formation after its own dynamic response, and the residual deviation is corrected by the closed-loop dynamic compensation of the dynamic adjustment module.

[0066] The formation position offset vector refers to a two-dimensional vector formed by the longitudinal and lateral distances of the preset target position of the following AGV relative to the main AGV in a local coordinate system with the current position of the main AGV as the origin and the driving direction of the main AGV as the vertical axis.

[0067] After all AGV trajectories are planned, the spatiotemporal conflict resolution submodule performs cross-checking on the time axis and spatial paths of all trajectories. When two or more trajectories are detected to need to occupy the same path node (such as intersections or shared passages) in the same time period, priority is determined according to the task timeliness level—urgent tasks have higher priority than ordinary tasks—AGVs with lower priority will insert a yielding waiting time at the nearest safe stopping point on their path, while AGVs with higher priority will continue on their original trajectories.

[0068] (5) Predictive resource scheduling module

[0069] The predictive resource scheduling module is used to predict future task density distribution based on historical task arrival patterns in each region, combined with AGV hardware status, physical location, and real-time track node occupancy, in order to pre-deploy idle AGVs. It also predicts low-battery risks based on each AGV's task plan and power consumption trends, and schedules charging during task intervals. This module includes a task demand prediction submodule, an AGV status prediction submodule, and an intelligent charging scheduling submodule.

[0070] The task demand prediction submodule collects and maintains historical task arrival data for each functional area according to preset time periods. Functional areas are pre-defined regions with clear two-dimensional coordinate boundaries, based on the physical operation attributes of warehousing, including receiving areas, storage areas, sorting areas, and shipping areas. Within each scheduling cycle, this submodule matches the task arrival frequency (number of tasks generated per unit time) of each area in the historical database for the current time period, predicting the area with the highest historical task arrival frequency as the high-demand area for the next cycle.

[0071] The AGV status prediction submodule extrapolates the power change curve of each AGV within a preset future time window based on the real-time remaining power of each AGV, the estimated time of the assigned task sequence, and the power consumption rate per unit time. When the extrapolation results indicate that the power of an AGV will fall below the preset safe charging threshold before completing the current task or subsequent tasks, the AGV is marked as having a low power risk.

[0072] The intelligent charging scheduling submodule is used to perform predictive charging and idle AGV pre-deployment scheduling. Predictive charging scheduling: For AGVs marked as having low power risk, their currently executing tasks and assigned subsequent tasks are arranged in chronological order to construct a task timeline; the gaps between adjacent tasks on the task timeline are traversed, and the gap duration is calculated (i.e., the time difference between the estimated completion time of the previous task plus the estimated time to move to the starting point of the next task and the planned start time of the next task); if there is an idle time window with a duration greater than or equal to the minimum charging duration, a charging task is inserted within this window, and a charging path to the nearest available charging station is generated; if there is no idle time window that meets the conditions, the AGV is marked as having a charging risk state and its ability to receive new tasks is restricted.

[0073] Idle AGV Pre-deployment: Based on the high-density task area prediction value output by the task requirement prediction submodule, AGVs with remaining power above the safe charging threshold and currently idle in the cluster are selected. Pre-deployment routes are generated, and the AGVs are navigated to preset standby points in the high-density task area in advance. When the remaining power of an idle AGV is lower than the preset safe charging threshold, the charging scheduling priority is higher than the pre-deployment scheduling priority, and charging is performed before deployment.

[0074] (6) Dynamic adjustment module

[0075] The dynamic adjustment module is responsible for closed-loop monitoring and anomaly response. This module includes a task deviation detection submodule and a dynamic rescheduling submodule.

[0076] The task deviation detection submodule detects the actual relative position deviation (Euclidean distance) between each following AGV and the main AGV in a multi-machine collaborative task at a preset frequency (e.g., 5Hz).

[0077] The deviation detection method is as follows: the master AGV reports its own position in real time, and the preset formation position of each following AGV is calculated by adding the formation offset vector corresponding to the master AGV to its position. The actual position of the following AGV is obtained from its reported real-time position, and the deviation is the Euclidean distance between the preset position and the actual position. When the deviation of any following AGV exceeds the preset safety deviation threshold (0.05m), closed-loop dynamic compensation control is triggered.

[0078] The compensation adopts a closed-loop dynamic compensation method, which calculates the velocity compensation amount based on the real-time detected position deviation and its rate of change. The formula for calculating the compensation velocity is as follows:

[0079] ;Mode

[0080] In the formula: For the first The platform follows the AGV at all times Compensation speed command ( ); For the first The platform follows the AGV at all times The nominal planned speed is taken as a preset proportion of the current speed of the main AGV as the safe tracking speed. ); For the first The platform follows the AGV at all times The positional deviation, i.e., the Euclidean distance between the actual position and the preset formation position ( ); For the first The proportional gain coefficient of the AGV following the platform; For the first The differential gain coefficient of the AGV following the platform; For monitoring period ( ).

[0081] proportional gain coefficient and differential gain coefficient Adaptive adjustments are made based on the historical collaborative performance of the following AGV: a subset of collaborative synchronization parameters is maintained for each AGV in the AGV dynamic capability profiling module, recording the most recent... If the deviation recovery time of the following AGV shows a decreasing trend in the most recent M collaborative tasks, then the deviation exceeding the standard and the average recovery time should be considered. and To make it respond faster; if the deviation recovery time shows an upward trend or overshoot oscillation occurs, then appropriately reduce the speed. and This enhances stability.

[0082] Before outputting the compensation speed command, a limiting process is performed. The limiting methods include: when the compensation speed exceeds the maximum safe speed of the following AGV, the maximum safe speed is used as the output value; when the corresponding acceleration exceeds the maximum acceleration of the following AGV, the speed command is limited to the speed increment corresponding to the current speed plus the maximum allowable acceleration.

[0083] The dynamic rescheduling submodule is used to perform incremental rescheduling of affected tasks in the event of a failure, communication timeout, severe task timeout, or preemption by a new urgent task. When a failure or heartbeat timeout is detected in the main AGV, the dynamic rescheduling submodule executes the following elastic self-healing process: If the rated load capacity of one of the original following AGVs meets the conditions for continuing to execute the task, it is promoted to the new main AGV; when the original main AGV stops, the parking brake is automatically activated, and all following AGVs decelerate synchronously to a stop; the starting coordinates of the unfinished atomic tasks are updated to the current stopping position of the formation; if an AGV needs to be added due to load requirements, the system automatically searches for the nearest available idle AGV, completes the rapid calibration through the new AGV rapid calibration subroutine of the AGV dynamic capability profile module, and then adds it to the formation; the completed progress is retained, and only the remaining transfer path is replanned from the current stopping position.

[0084] Example 2

[0085] An AGV cluster scheduling method based on task and resource matching is presented below. The complete execution process of this method is illustrated using a large logistics warehouse as an application scenario. The warehouse area is approximately 12,000 m². 2 The system is divided into four functional areas: receiving area, storage area, sorting area, and shipping area. It operates 30 heterogeneous AGVs, with wireless communication latency between AGVs ranging from 15ms to 40ms. The scheduling server operates a scheduling cycle at a frequency of 10Hz. The method of this invention includes the following steps:

[0086] Step S1: Task Reception and Dynamic Decomposition. The task parsing and decomposition module receives the logistics task, extracts the cargo weight, picking station coordinates, unloading station coordinates, and task timeliness level, and dynamically decomposes the task into a standardized atomic task sequence of picking, transferring, and unloading. Subsequently, the generated atomic task sequence and corresponding resource matching requirements are sent to the multi-mode task-resource matching module, which determines whether to use a single-machine execution mode or a multi-machine collaborative execution mode.

[0087] Step S2: AGV Dynamic Capability Profile Update. The AGV dynamic capability profile module collects static capability parameters and dynamic status parameters of each AGV at a fixed frequency, integrating them into a unified capability vector for subsequent steps. When a new AGV is connected, a fast calibration subroutine is automatically triggered to complete automatic identification, capability calibration, and parameter inheritance.

[0088] Step S3: Multi-mode matching decision. The multi-mode task-resource matching module performs single-machine matching and collaborative combination search in parallel, following the principle of prioritizing single-machine solutions: a collaborative solution is selected only when a single-machine solution is empty, or when the estimated completion time of a single-machine solution is significantly longer than that of a collaborative solution due to the distance of candidate AGVs being too far. The final scheduling plan is output, including the execution mode, the list of participating AGV identifiers, and the master-slave role allocation.

[0089] Step S4: Spatiotemporal Synchronization Path Planning. The main trajectory generation submodule generates a reference spatiotemporal trajectory with differentiated node priority weights based on the task timeliness level of the AGV or the main AGV. For multi-machine collaborative mode, the follower trajectory generation submodule uses the main AGV reference trajectory as a benchmark and performs spatial translation and temporal offset on the reference trajectory based on the differences in command response latency and acceleration between each follower AGV and the main AGV to generate the individual spatiotemporal trajectory of each follower AGV. The spatiotemporal conflict resolution submodule performs conflict detection on all AGV trajectories and coordinates the passage order according to task priority.

[0090] Step S5: Predictive Resource Scheduling. The predictive resource scheduling module runs at a preset cycle, predicting the future task density distribution in each area and pre-scheduling idle AGVs with sufficient power to areas expected to have high task density. It also calculates the power change curve of each AGV and inserts charging tasks in the idle window before the predicted power level falls below the safety threshold. When the power level of an idle AGV is lower than the safe charging threshold, charging scheduling takes priority over pre-deployment scheduling.

[0091] Step S6: Perform monitoring and dynamic rescheduling. The execution status acquisition submodule acquires the position and task progress of each AGV in real time. The task deviation detection submodule continuously detects the actual relative position deviation between each following AGV and the main AGV in multi-machine collaborative tasks. When the deviation exceeds the limit, the compensation speed command is calculated according to formula (2), and the gain coefficient is adaptively adjusted according to the historical collaborative performance. When an AGV failure, communication timeout, or new emergency task is detected, incremental rescheduling is performed on the affected and incomplete atomic tasks.

[0092] Example 3

[0093] A task and resource matching-based AGV cluster scheduling method differs from Embodiment 2 in that this embodiment specifically verifies the "plug-and-play" and adaptive optimization process of collaboration synchronization parameters when a new AGV is added to the cluster.

[0094] When a new AGV connects to the network for the first time, it automatically broadcasts its serial number and type code. The AGV dynamic capability profiling module queries the device model database to automatically match the default static parameters and issues a standard calibration task to complete the capability calibration. The calibration results are automatically updated to the capability vector, and the initial values ​​of the collaborative synchronization parameters are the average values ​​of AGVs of the same model.

[0095] After the AGV participates in a collaborative task for the first time, its collaborative synchronization parameter subset records the deviation recovery time. After several collaborative tasks, the proportional gain coefficient and differential gain coefficient are automatically adjusted to personalized optimal values, and the deviation recovery time is gradually shortened, demonstrating the effect of adaptive optimization.

[0096] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AGV cluster scheduling system based on task and resource matching, characterized in that, include: The task parsing and decomposition module is used to receive logistics tasks issued by the upper-level system, extract task attributes, dynamically decompose the task into a standardized atomic task sequence containing pickup, transfer, and unloading, and determine the resource matching requirements of each atomic task. The AGV dynamic capability profiling module is used to collect and maintain the static capability parameters and dynamic status parameters of each AGV in the cluster in real time, and integrate them into a capability vector in a unified format. The multi-mode task-resource matching module is used to match and select the best between single-machine execution and multi-machine collaboration modes based on the resource matching requirements and the capability vector, and output a scheduling scheme that includes the execution mode, the list of participating AGV identifiers, and the master-slave role allocation. The spatiotemporal synchronization path planning module is used to generate spatiotemporal trajectories with time windows for each participating AGV according to the scheduling scheme. The predictive resource scheduling module is used to predict the future task density distribution for pre-deployment of idle AGVs, and to schedule charging between tasks based on the task plan and power consumption trend of each AGV. The dynamic adjustment module is used to detect the positional deviation of the following AGV in multi-machine collaborative tasks and perform closed-loop dynamic compensation, as well as to perform rescheduling when the AGV fails.

2. The AGV cluster scheduling system based on task and resource matching according to claim 1, characterized in that, The multi-mode task-resource matching module includes a single-machine matching submodule, a collaborative combination search submodule, and a solution selection submodule. The single-machine matching submodule and the collaborative combination search submodule are executed in parallel. The solution selection submodule follows the single-machine solution priority principle: a collaborative solution is selected only when a single-machine solution is empty or the estimated completion time of a single-machine solution is significantly longer than that of a collaborative solution.

3. The AGV cluster scheduling system based on task and resource matching according to claim 2, characterized in that, The collaborative combination search submodule searches for multi-machine combination schemes among AGVs with collaborative function identifiers and in an idle state. The search conditions include: the sum of the rated loads of each AGV in the group is greater than or equal to the weight constraint; the wireless communication delay between each AGV in the group is lower than a preset collaborative delay threshold; and the maximum difference between the distances of each AGV in the group to the picking station does not exceed a preset grouping distance threshold.

4. The AGV cluster scheduling system based on task and resource matching according to claim 1, characterized in that, When the execution mode is multi-machine collaboration, the multi-mode task-resource matching module selects the AGV with the largest actual acceleration at full load and the smallest instruction response delay to serve as the master AGV, and the remaining AGVs serve as follower AGVs.

5. The AGV cluster scheduling system based on task and resource matching according to claim 4, characterized in that, The spatiotemporal synchronization path planning module includes a main trajectory generation submodule and a follower trajectory generation submodule. The follower trajectory generation submodule uses the main AGV reference spatiotemporal trajectory as a reference and performs spatial translation and temporal offset on the coordinates of each time point on the main AGV reference spatiotemporal trajectory to generate an individual spatiotemporal trajectory for each follower AGV. The spatial translation refers to adding the formation position offset vector corresponding to the follower AGV to the coordinates of each time point on the main AGV reference trajectory. The temporal offset refers to shifting the main AGV reference trajectory on the time axis by a time advance. The time advance is composed of the sum of the response delay difference and acceleration difference between the follower AGV and the main AGV.

6. The AGV cluster scheduling system based on task and resource matching according to claim 1, characterized in that, The predictive resource scheduling module includes an intelligent charging scheduling submodule. For AGVs marked as having low power risk, the intelligent charging scheduling submodule arranges their currently executing tasks and assigned subsequent tasks in chronological order to construct a task timeline. It then iterates through the gaps between adjacent tasks on the task timeline. If the gap duration is greater than or equal to the minimum charging duration of the idle time window, a charging task is inserted within that window. If no idle time window meets the conditions, the AGV is marked as having a charging risk state and its ability to receive new tasks is restricted.

7. The AGV cluster scheduling system based on task and resource matching according to claim 1, characterized in that, The dynamic adjustment module includes a task deviation detection submodule, which uses a closed-loop dynamic compensation method to calculate the compensation speed. The calculation formula for the compensation speed is as follows: ； In the formula: For the first The platform follows the AGV at all times Compensation speed command ( ); For the first The platform follows the AGV at all times The nominal planned speed is taken as a preset proportion of the current speed of the main AGV as the safe tracking speed. ); For the first The platform follows the AGV at all times The positional deviation, i.e., the Euclidean distance between the actual position and the preset formation position ( ); For the first The proportional gain coefficient of the AGV following the platform; For the first The differential gain coefficient of the AGV following the platform; For monitoring period ( ).

8. The AGV cluster scheduling system based on task and resource matching according to claim 1, characterized in that, The dynamic adjustment module includes a dynamic rescheduling submodule. When the dynamic rescheduling submodule detects a failure of the main AGV or a heartbeat timeout, it executes the following elastic self-healing process: if there is an AGV among the original following AGVs that has the capacity to continue executing the task, it promotes it to the new main AGV; if an AGV needs to be added due to load requirements, it automatically searches for the nearest available idle AGV, completes rapid calibration, and adds it to the formation; it retains the completed progress and only replans the remaining transfer path from the current stopping position.

9. A method for scheduling AGV clusters based on task and resource matching, characterized in that, Includes the following steps: Step S1: Receive logistics tasks, extract task attributes, dynamically decompose the tasks into a standardized atomic task sequence that includes pickup, transfer, and unloading, and determine the resource matching requirements of each atomic task. Step S2: Collect and maintain the static capability parameters and dynamic status parameters of each AGV in the cluster in real time, and integrate them into a capability vector in a unified format; Step S3: Perform single-machine matching and cooperative combination search in parallel, and output the final scheduling scheme according to the principle of prioritizing single-machine schemes. The final scheduling scheme includes the execution mode, the list of participating AGV identifiers, and the master-slave role allocation. Step S4: Generate a reference spatiotemporal trajectory with differentiated node priority weights based on the task timeliness level of the AGV or the main AGV. For multi-machine collaboration mode, take the main AGV reference trajectory as the benchmark, and perform spatial translation and time offset on the reference trajectory according to the differences between each following AGV and the main AGV in command response delay and acceleration to generate the individual spatiotemporal trajectory of each following AGV. Step S5: Predict the task density distribution of each area in the future, schedule idle and fully charged AGVs to the expected high task density area in advance, and deduce the power change curve of each AGV. Insert charging tasks during task gaps before the power is predicted to fall below the safety threshold. Step S6: Real-time acquisition of the position and task progress of each AGV, detection of position deviation of the following AGV in multi-machine collaborative tasks and closed-loop dynamic compensation, and incremental rescheduling of affected and incomplete atomic tasks when AGV failure or new emergency task is detected.

10. The AGV cluster scheduling method based on task and resource matching according to claim 9, characterized in that, Step S2 also includes a plug-and-play process for the new AGV: when the new AGV first connects to the network, it automatically broadcasts the device serial number and type code, queries the device model library to automatically match the default static parameters, and issues a standard calibration task to complete the capability calibration. The calibration result is automatically updated to the capability vector, and the initial value of the collaborative synchronization parameter adopts the average value of the same model of AGV.

Images