AMR cluster task planning cloud platform based on digital twinning

Abstract

The application relates to the technical field of industrial automation and intelligent logistics, and discloses an AMR cluster task planning cloud platform based on digital twinning. The platform comprises multiple modules such as task scheduling, twinning modeling, cluster state analysis, path planning and core operation. The task scheduling module decomposes job requirement data; the twinning modeling module generates two kinds of task topology models; the cluster state analysis module determines the scheduling level; the path planning module provides navigation parameters; and the core operation unit generates final scheduling instructions. In addition, modules such as equipment cooperation, instruction conversion and task archiving are also provided. The platform can accurately plan tasks, improve the work efficiency of an AMR cluster, realize optimal allocation of resources, solve the space-time contradiction between tasks, record the whole work process, and provide an effective solution for AMR cluster task planning in industrial production and logistics and warehousing scenarios.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and intelligent logistics technology, specifically to an AMR cluster task planning cloud platform based on digital twins. Background Technology

[0002] In modern industrial production and logistics warehousing, autonomous mobile robots (AMRs), with their high flexibility and autonomy, are gradually becoming key equipment for improving production efficiency and logistics delivery accuracy. As their application scale continues to expand, how to efficiently plan and schedule tasks for AMR swarms has become a crucial problem that urgently needs to be solved.

[0003] Traditional AMR task planning methods often rely on human experience or simple algorithms, which have many limitations when handling complex tasks and large-scale clusters. On the one hand, manual scheduling struggles to cope with rapidly changing production demands and complex task scenarios, easily leading to problems such as unreasonable task allocation, equipment idleness or overload, resulting in overall low efficiency. For example, in large e-commerce warehouses, order demands change constantly, and manual scheduling cannot promptly and rationally arrange AMRs for goods handling based on factors such as the urgency of orders and the storage location of goods, often causing goods backlog or delivery delays.

[0004] On the other hand, simple algorithms cannot achieve globally optimal task planning when dealing with complex situations such as multi-task concurrency, limited resources, and dynamic environmental changes. For example, early path planning algorithms based on fixed rules did not take into account dynamic factors such as obstacles and interference from other devices that the AMR may encounter during operation. This caused frequent path conflicts when the AMR was executing tasks, which not only reduced operating efficiency but also may have caused safety accidents such as equipment collisions.

[0005] Furthermore, traditional task planning systems lack adequate monitoring of AMR equipment status and resource management. They cannot accurately obtain real-time information such as equipment energy consumption, remaining power, and fault status, making it difficult to allocate tasks and schedule resources reasonably based on the actual situation of the equipment. This impacts equipment lifespan and overall system stability.

[0006] With the rise of digital twin technology, its application in the industrial field has gradually attracted attention. Digital twins, by constructing virtual models of physical entities, can reflect the state and behavior of physical entities in real time, providing strong support for optimization decisions. However, research on applying digital twin technology to AMR cluster task planning is still in the exploratory stage. Existing related systems and methods still have shortcomings in task decomposition, model building, resource matching, and scheduling optimization, failing to fully leverage the advantages of digital twin technology and making it difficult to meet the needs of modern industry and logistics for efficient AMR cluster task planning. Therefore, developing a cloud platform for AMR cluster task planning based on digital twins has significant practical implications. Summary of the Invention

[0007] The purpose of this invention is to provide a cloud platform for AMR cluster task planning based on digital twins, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a cloud platform for AMR cluster task planning based on digital twins, the platform comprising:

[0009] The task scheduling module is used to receive multi-source job requirement data and identify the task type, and divide the job requirement data into basic task data stream and extended task data stream based on preset task decomposition rules.

[0010] The twin modeling module is used to perform three-dimensional spatial meshing processing on the basic task data stream to generate a first task topology model, and to perform dynamic resource matching processing on the extended task data stream to generate a second task topology model.

[0011] The cluster status analysis module is used to map the first task topology model and the second task topology model to the corresponding scheduling level after data fusion according to the preset constraint condition matrix, and use the scheduling level as the task allocation benchmark for the current cluster.

[0012] The path planning module is used to call the optimal path scheme from the preset path database based on the scheduling level, and use the optimal path scheme as the navigation parameter of the current cluster.

[0013] The core computing unit is used to perform logical verification on the task allocation benchmark and the navigation parameters to generate the final scheduling instruction.

[0014] Preferably, the twin modeling module performs dynamic resource matching processing on the extended task data stream, including:

[0015] The task sequence in the extended task data stream is divided into fixed task groups and elastic task groups. Based on a preset dynamic load balancing algorithm, the resource utilization rate of the elastic task groups is calculated to generate a load feature set.

[0016] The device status data in the extended task data stream is processed hierarchically, and the energy consumption characteristics of each level of device are extracted and an energy consumption map is constructed.

[0017] The load feature set is dynamically associated and matched with the energy consumption map to generate the second task topology model.

[0018] Preferably, the preset task decomposition rules include a set of rigid constraints and a set of flexible constraints; the set of rigid constraints includes a time window identifier, a device capacity identifier, and a job priority identifier; the set of flexible constraints includes an energy consumption threshold identifier and a path conflict identifier, and each identifier corresponds to an independent task processing channel.

[0019] Preferably, the platform further includes a data interaction interface, which is used to enable the task scheduling module, the twin modeling module, the cluster status analysis module, and the path planning module to communicate with the physical device network respectively;

[0020] The task scheduling module divides the job requirement data based on the preset task decomposition rules, including:

[0021] The data interaction interface is used to obtain mixed data packets from the physical device network in real time, and the protocol tags of the mixed data packets are matched with the identifiers in the rigid constraint set to separate the basic task segments.

[0022] The extended tags of the hybrid data packets are traversed according to the identifiers in the set of flexible constraints to extract the elastic task segments;

[0023] The basic task segment and the elastic task segment are synchronized and aligned according to the task trigger time, and then written into the task storage area and the extended task buffer respectively.

[0024] Preferably, when the preset constraint matrix adopts a static analysis model, the scheduling level is an interval mapping result of the linear superposition values ​​of the first task topology model and the second task topology model;

[0025] When the preset constraint matrix adopts a dynamic adjustment model, the scheduling level is a continuous set of states that are corrected in real time by using an incremental learning algorithm to modify the associated data of the first task topology model and the second task topology model.

[0026] Preferably, it also includes a device collaboration module connected to the core computing unit, the device collaboration module being connected to the physical device resource pool through the data interaction interface;

[0027] The device coordination module is used to select available device queues from the physical device resource pool according to the device scheduling requirements in the final scheduling instruction, and generate a task execution sequence to optimize the cluster operation process.

[0028] Preferably, the device collaboration module generates a task execution sequence including:

[0029] Load a three-dimensional spatial topology model, and locate the spatial coordinates of each device in the available device queue within the topology model;

[0030] The optimal movement trajectory from the current position of each device to the target work area is calculated based on the dynamic programming algorithm, and the available device queue is sorted by performance according to the urgency of the task.

[0031] The optimal movement trajectory and the performance ranking are integrated into the topology model to generate a visual task execution sequence.

[0032] Preferably, when the core computing unit performs logical verification on the task allocation benchmark and navigation parameters, it adopts a dual verification mode of an anomaly detection mechanism and a constraint conflict verification mechanism. The anomaly detection mechanism is used to confirm the integrity of the equipment operation, and the constraint conflict verification mechanism is used to resolve spatiotemporal contradictions between tasks.

[0033] Preferably, it also includes an instruction conversion module connected to the core computing unit. The instruction conversion module is used to convert the final scheduling instruction into device control code and send the device control code to the designated device cluster through the data interaction interface to start the task program.

[0034] Preferably, it also includes a task archiving module connected to the core computing unit. The task archiving module is used to store the job requirement data, the first task topology model, the second task topology model, the scheduling level, and the final scheduling instruction, and to generate a complete job record chain according to the task lifecycle.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] In terms of task scheduling, the task scheduling module receives multi-source job requirement data and identifies the task type. Based on preset task decomposition rules, it divides the job requirement data into basic task data streams and extended task data streams. These preset task decomposition rules include a rigid constraint set and a flexible constraint set. The rigid constraint set includes time window identifiers, equipment capacity identifiers, and job priority identifiers, while the flexible constraint set includes energy consumption threshold identifiers and path conflict identifiers. This ensures the rationality and efficiency of task processing. For example, based on the time window identifier, urgent tasks can be prioritized to avoid delays; based on the equipment capacity identifier, tasks can be rationally allocated to prevent equipment overload. Furthermore, by acquiring mixed data packets in real time from the physical device network through a data interaction interface, it accurately separates basic task segments and extracts flexible task segments, aligns them synchronously according to task trigger times, and stores them, ensuring the timeliness and orderliness of task processing.

[0037] The design of the twin modeling module is crucial. It performs 3D spatial meshing on the basic task data stream to generate the first task topology model, providing a precise spatial foundation for subsequent task allocation and path planning. It then performs dynamic resource matching on the extended task data stream to generate the second task topology model. Specifically, this involves dividing the task sequence into fixed task groups and flexible task groups, calculating the resource occupancy rate of the flexible task groups based on a preset dynamic load balancing algorithm to generate a load feature set, and simultaneously performing hierarchical processing of equipment status data to construct an energy consumption map. These two sets are then dynamically correlated and matched. This allows task planning to fully consider task requirements and equipment resource status, achieving optimal resource allocation. For example, in a logistics warehouse scenario, tasks can be rationally allocated based on the urgency and weight of different cargo handling tasks, as well as the power and load capacity of each AMR (Automatic Mobile Transporter), preventing some AMRs from being overworked while others remain idle, thus improving overall operational efficiency.

[0038] The cluster state analysis module maps the data from two task topology models to corresponding scheduling levels based on a preset constraint matrix. When the preset constraint matrix uses a static analysis model, the scheduling level is an interval mapping result of the linear superposition of the two models; when using a dynamic adjustment model, a continuous set of states is obtained by real-time correction of the associated data through an incremental learning algorithm. This flexible scheduling level determination method adapts to different task scenarios and dynamically changing environments, providing a scientific and reasonable benchmark for task allocation.

[0039] The path planning module uses the optimal path from the preset path database based on the scheduling level as navigation parameters. By combining the actual task and equipment status to determine the path, path conflicts are effectively avoided, improving the AMR's operating efficiency. In complex production workshops or warehouse environments, the AMR can quickly and accurately reach the target location according to the optimal path, reducing operating time and energy consumption.

[0040] The core computing unit performs logical verification on the task allocation benchmark and navigation parameters, employing a dual verification mode of anomaly detection and constraint conflict verification mechanisms. The anomaly detection mechanism confirms the integrity of equipment operation and promptly identifies potential equipment failures; the constraint conflict verification mechanism resolves spatiotemporal contradictions between tasks, ensuring the safety and reliability of task execution. For example, when multiple AMRs execute tasks simultaneously, it can prevent them from arriving at the same location at the same time, thus preventing collisions.

[0041] The device collaboration module filters available device queues and generates task execution sequences based on the final scheduling instructions. By loading a 3D spatial topology model, it locates the spatial coordinates of devices, calculates the optimal movement trajectory based on a dynamic programming algorithm, and sorts the tasks by urgency to generate a visual task execution sequence. This makes collaborative operations between devices smoother, further optimizes the cluster operation process, and improves overall operation efficiency.

[0042] The instruction conversion module converts the final scheduling instructions into device control codes and sends them to the designated device cluster to start the task program, ensuring the accuracy of instruction transmission and the effectiveness of device execution. The task archiving module stores job requirement data, task topology model, scheduling level, and final scheduling instructions, and generates a complete job record chain according to the task lifecycle, facilitating subsequent querying, analysis, and optimization, and helping to continuously improve task planning strategies. Attached Figure Description

[0043] Figure 1 This is a schematic diagram illustrating the working principle of the AMR cluster task planning cloud platform based on digital twins as described in this invention.

[0044] Figure 2 A schematic diagram illustrating the working principle of the twin modeling module for processing extended task data streams.

[0045] Figure 3 A diagram illustrating the working principle of the task scheduling module dividing data based on preset task decomposition rules;

[0046] Figure 4 A schematic diagram illustrating the working principle of generating task execution sequences for the device collaboration module;

[0047] Figure 5 This is a schematic diagram illustrating the working principle of the core computing unit's logic verification. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0049] Please see Figures 1-5 This invention provides a cloud platform for AMR cluster task planning based on digital twins, and the specific implementation steps are as follows:

[0050] The task scheduling module receives multi-source job requirement data and identifies the task type. Based on preset task decomposition rules, it divides the job requirement data into a basic task data stream and an extended task data stream. Next, the twin modeling module performs 3D spatial meshing on the basic task data stream to generate a first task topology model, and performs dynamic resource matching on the extended task data stream to generate a second task topology model. Then, the cluster state analysis module, based on a preset constraint matrix, merges the first and second task topology models and maps them to the corresponding scheduling level, using this scheduling level as the task allocation benchmark for the current cluster. Next, the path planning module, based on the scheduling level, calls the optimal path scheme from a preset path database and uses this optimal path scheme as the navigation parameter for the current cluster. Finally, the core computing unit performs logical verification on the task allocation benchmark and the navigation parameters to generate the final scheduling instruction.

[0051] The present invention will be further described below with reference to Examples 1 to 5:

[0052] Example 1:

[0053] In this cloud platform, the twin modeling module employs a rigorous process when performing dynamic resource matching on extended task data streams. First, the task sequences within the extended task data stream are divided into fixed task groups and elastic task groups. In actual task processing scenarios, tasks vary in nature; some tasks have fixed execution modes and requirements, while others can be flexibly allocated based on actual resource availability. This is the basis for dividing fixed and elastic task groups. After division, resource utilization is calculated for the elastic task groups based on a preset dynamic load balancing algorithm. This algorithm comprehensively considers various resource factors, such as computing and storage resources, and generates a load feature set by analyzing and integrating the resource requirements of each task within the elastic task group. This load feature set comprehensively reflects the resource utilization and trends of the elastic task group during execution.

[0054] The device status data in the extended task data stream is processed hierarchically. During operation, devices at different levels have different functions and performance characteristics, resulting in variations in energy consumption. Hierarchical processing allows for a more detailed analysis of the energy consumption characteristics of each level. The process involves collecting, organizing, and analyzing device energy consumption data to extract data indicators representative of the energy consumption characteristics of each level, thereby constructing an energy consumption map. This energy consumption map visually displays the distribution and variation patterns of energy consumption across different levels of devices.

[0055] A second task topology model is generated by dynamically associating and matching the load feature set with the energy consumption map. During the matching process, the interrelationship between load features and energy consumption features is fully considered to find the optimal matching combination. This dynamic association and matching ensures that the generated second task topology model better reflects actual task requirements and resource conditions, improving the rationality and efficiency of task planning.

[0056] In a large e-commerce warehouse, a large number of orders are generated daily, requiring a cluster of AMRs (Autonomous Mobile Robots) to collaboratively complete the tasks of sorting, handling, and delivering goods. These task data are continuously input into the task scheduling module of the cloud platform, forming an extended task data stream.

[0057] The cloud platform's twin modeling module begins performing dynamic resource matching processing on the extended task data stream. The first step is task grouping. Some warehouse sorting tasks have fixed processes, such as sorting specific categories of goods from fixed shelves to fixed temporary storage areas; these tasks are grouped into fixed task groups. However, expedited sorting tasks to handle sudden orders, or replenishment tasks adjusted based on real-time inventory, are grouped into elastic task groups because their execution time and resource requirements are more flexible.

[0058] After dividing the task groups, a load feature set is generated by calculating the resource utilization of the elastic task groups based on a preset dynamic load balancing algorithm. For example, within a certain time period, there are three urgent order sorting tasks in the elastic task group. Each task has different usage time of the AMR, power consumption, and computational resource consumption of the warehouse management system. The preset dynamic load balancing algorithm will comprehensively consider these factors by analyzing and integrating the various resources expected to be used by each task during execution. For example, one task is expected to require the AMR to work continuously for 2 hours, consuming 30% of the power and occupying 10% of the system's computational resources; another task is expected to work for 1.5 hours, consuming 20% ​​of the power and occupying 8% of the computational resources, etc. After analyzing and processing this data, the algorithm finally generates a load feature set that comprehensively reflects the resource utilization of the elastic task group. This feature set includes information such as the resource utilization ratio and time distribution of each task, providing a basis for subsequent resource matching.

[0059] The device status data in the extended task data stream is hierarchically processed to extract energy consumption characteristics of each level of equipment and construct an energy consumption map. The equipment in the warehouse includes AMRs, rack lifts, and conveyor belts. Taking AMRs as an example, they are divided into levels based on factors such as battery capacity and motor power. AMRs at different levels consume different amounts of energy when performing the same task. For each level of AMR, energy consumption data is collected under different task scenarios, such as the power consumption when moving the same distance under full load and empty conditions, and the power consumption when sorting goods of different weights. After processing and analyzing this data, key energy consumption characteristics of each level of AMR are extracted, such as peak energy consumption and average energy consumption. Then, these energy consumption characteristics are presented in chart form to construct an energy consumption map, clearly showing the energy consumption distribution and variation patterns of AMRs at different levels.

[0060] The load feature set and energy consumption map are dynamically correlated and matched to generate a second task topology model. Combining the previously generated load feature set and energy consumption map, the cloud platform seeks the optimal matching combination between the two. Suppose that a certain expedited sorting task in the load feature set has specific requirements for AMR usage time and power consumption, and the energy consumption map shows that a certain level of AMR consumes the least energy and can efficiently complete the task while meeting the time requirements. The cloud platform will then match this task with the AMR at that level, while considering other relevant factors such as task priority and the current location of the equipment, ultimately generating a comprehensive and reasonable second task topology model. This model clarifies the matching relationship between each task and its corresponding equipment, as well as the execution order and dependencies between tasks, providing an accurate basis for subsequent task scheduling and execution.

[0061] Example 2:

[0062] Pre-defined task decomposition rules play a crucial role in the entire task planning process, and include a set of rigid constraints and a set of flexible constraints. The set of rigid constraints includes time window identifiers, equipment capacity identifiers, and job priority identifiers. The time window identifier clarifies the allowed time range for task execution, a vital constraint ensuring timely task completion. The equipment capacity identifier specifies the equipment capacity required to execute the task, preventing task failure due to insufficient capacity. The job priority identifier determines the priority order of different tasks, ensuring that important tasks are processed first.

[0063] The flexible constraint set includes energy consumption threshold identifiers and path conflict identifiers, with each identifier corresponding to an independent task processing channel. Energy consumption threshold identifiers limit the upper limit of energy consumption during task execution to achieve energy conservation. Path conflict identifiers are used to detect and avoid path conflict issues during task execution, ensuring the safety and efficiency of task execution.

[0064] In practical applications, when the task scheduling module divides job requirement data based on preset task decomposition rules, it obtains mixed data packets from the physical device network in real time through a data interaction interface. This data interaction interface is responsible for the communication between the task scheduling module and the physical device network, ensuring accurate data transmission. After obtaining the mixed data packets, the protocol tags of the mixed data packets are matched with the identifiers in the rigid constraint set to separate the basic task segments. The protocol tags contain key task information; by matching them with the identifiers in the rigid constraint set, the basic task segments can be accurately identified. Then, the extended tags of the mixed data packets are traversed according to the identifiers in the flexible constraint set to extract the elastic task segments. The extended tags record information related to the flexible constraints; by traversing these tags, the elastic task segments can be extracted. Finally, the basic task segments and elastic task segments are synchronized and aligned according to their task trigger times and written to the task storage area and extended task buffer, respectively, for subsequent processing.

[0065] Taking a large manufacturing plant as an example, the plant produces a variety of products and has a large number of daily needs for raw material handling, parts processing and finished product assembly. These needs are input into the AMR cluster task planning cloud platform based on digital twin in the form of work requirement data. In this scenario, this embodiment has the following specific implementation method.

[0066] Production operations in factories have strict time requirements, which highlights the role of time windows in rigid constraint sets. For example, the production of a batch of products requires the transportation of raw materials within a specific time frame to ensure that subsequent processing can proceed on time. If the transportation task is not completed within the specified time window, it will cause delays in the entire production process and increase production costs. A time window acts like a precise clock, setting strict time limits for task execution.

[0067] Equipment capacity markings are equally crucial. Different processing equipment has different capacity limitations during parts manufacturing. For example, a large CNC machining center has a limited number and size of parts it can process at a time. When scheduling processing tasks, the quantity of parts processed in each batch must be determined based on the equipment capacity markings to ensure that the equipment does not malfunction due to overload or affect processing accuracy. This is akin to setting a "capacity limit" for the equipment, ensuring it operates safely and efficiently.

[0068] Task prioritization plays a crucial role in coordinating multiple tasks. When a factory receives an urgent order, related production tasks are assigned higher priority. For example, if a major customer urgently needs a batch of parts with special specifications, all tasks related to the production of these parts, including raw material handling, processing, and finished product transportation, will be prioritized over other routine tasks. This ensures that urgent orders can be delivered quickly, meeting customer needs and maintaining the factory's reputation.

[0069] Energy consumption thresholds within flexible constraint sets are crucial for energy conservation and emission reduction in factories. Take AMR (Automatic Mobile Transporter) robots in a factory as an example: each robot consumes electricity during operation. To reduce overall energy consumption, the cloud platform sets energy consumption thresholds. For instance, it stipulates that the energy consumption of a certain model of AMR cannot exceed a certain value when completing a transport task. If, during task planning, a transport plan is found to potentially cause the AMR's energy consumption to exceed the threshold, the cloud platform will replan the path or adjust the transport strategy, selecting a lower-energy-consumption option to achieve energy-saving goals.

[0070] Path conflict identification is crucial for ensuring smooth logistics within a factory. In a factory production workshop, AMR robots need to navigate within limited space when transporting goods. Without proper planning, path conflicts between robots can easily occur, leading to congestion and impacting production efficiency. Path conflict identification acts like traffic lights, analyzing and planning the path for each task to prevent different AMRs from entering the same area at the same time. For example, when two AMRs need to pass through a narrow passage, the cloud platform will use path conflict identification to rationally arrange their passage order, ensuring safe and efficient logistics transportation.

[0071] The data interaction interface plays a crucial role when the task scheduling module divides job requirement data based on preset task decomposition rules. Various devices in the factory, such as production equipment, handling robots, and sensors, interact with the cloud platform through a physical device network. The task scheduling module obtains mixed data packets in real time through the data interaction interface; these packets contain various job requirement information.

[0072] The protocol tags of the mixed data packets are matched against identifiers in the rigid constraint set to separate the basic task segments. For example, a mixed data packet may contain information on various tasks such as raw material handling, component processing, and finished product assembly. By matching these with rigid constraints such as time window identifiers, equipment capacity identifiers, and job priority identifiers, basic tasks with fixed time requirements, clear equipment needs, and high priority can be accurately identified. For instance, if a production line needs to replenish raw materials at a fixed time each day according to its production plan, this task will be identified as a basic task segment and separated from the mixed data packet.

[0073] The extended tags of the mixed data packets are traversed based on the identifiers in the flexible constraint set to extract flexible task segments. The extended tags record information related to flexible constraints such as energy consumption and route planning. For example, for some non-urgent handling tasks, their execution time and path can be adjusted according to actual conditions. By traversing the extended tags and combining energy consumption threshold identifiers and path conflict identifiers, these flexible task segments can be extracted. For instance, if a handling task is found to be able to choose a route with lower energy consumption but a slightly longer path without affecting the overall production schedule, this task will be extracted as a flexible task segment.

[0074] After synchronizing and aligning the basic task segments and elastic task segments according to their task trigger times, they are written to the task storage area and the extended task buffer, respectively. In the factory production process, task trigger times vary. For example, a raw material handling task might be triggered when raw materials arrive, while a parts processing task might be triggered after raw material handling is completed. The cloud platform synchronizes and aligns the separated basic task segments and elastic task segments according to their trigger times to ensure a reasonable task execution order. Then, the basic task segments are written to the task storage area, which stores tasks that require priority processing and have clear execution rules; the elastic task segments are written to the extended task buffer, awaiting further scheduling and optimization based on actual conditions to better adapt to various changes in the production process and improve production efficiency and resource utilization.

[0075] Example 3:

[0076] In the cluster status analysis module, different types of preset constraint matrices lead to different scheduling level generation methods. When the preset constraint matrix uses a static analysis model, the scheduling level is an interval mapping result of the linear superposition value of the first task topology model and the second task topology model. In this case, the first and second task topology models are first linearly superimposed. Linear superposition is a simple and direct fusion method, adding the relevant data from the two models to obtain a comprehensive data value. Then, this linear superposition value is interval mapped. Interval mapping maps the linear superposition value to the corresponding scheduling level interval according to a pre-defined interval range, thereby determining the current scheduling level. This method is suitable for scenarios where tasks and resources are relatively stable and can quickly generate scheduling levels.

[0077] When the preset constraint matrix adopts a dynamic adjustment model, the scheduling level is a continuous set of states that are continuously corrected in real time based on the correlation data between the first and second task topology models using an incremental learning algorithm. The incremental learning algorithm can gradually update and optimize the model as new data is continuously input. During this process, the correlation data between the first and second task topology models is continuously analyzed and learned, and the scheduling level is corrected in real time according to new circumstances. This approach can better adapt to the dynamic changes in tasks and resources, improving the accuracy and rationality of task allocation.

[0078] Suppose a large intelligent logistics park exists, with numerous cargo handling and storage tasks, which are executed by an AMR cluster. Within this logistics park, cargo handling and warehouse storage tasks continuously generate new demand data. This data, after being processed by the task scheduling module and the twin modeling module, yields a first task topology model M1 and a second task topology model M2.

[0079] When the preset constraint matrix adopts a static analysis model: During a certain time period, the logistics park receives a batch of goods for a major e-commerce promotion, requiring handling and storage. The first task topology model M1 mainly reflects the distribution and resource requirements of basic handling tasks in three-dimensional space, such as the positional relationship between each goods storage point and the target storage point, as well as the demand for AMRs for different handling paths; the second task topology model M2 reflects the dynamic task adjustment and resource matching, such as the impact of temporary expedited orders on the priority and resource allocation of handling tasks.

[0080] First, M1 and M2 are linearly superimposed. The formula for linear superposition is S = M1 + M2, where S represents the value after linear superposition, and M1 and M2 are the first and second task topology models, respectively. This formula means simply adding the various data information contained in the two models. For example, the resource requirement value corresponding to each task node in M1 is added to the resource requirement value of the corresponding node in M2 to obtain a new comprehensive resource requirement value; the correlation weight values ​​between tasks in the two models are also added to combine the features of the two models.

[0081] After obtaining the linear superposition value S, interval mapping is performed. Assuming that based on long-term operational data and experience of the logistics park, three scheduling level intervals are set: low-priority tasks in the range [0, 100), medium-priority tasks in the range [100, 300), and high-priority tasks in the range [300, +∞). The value of S is compared and mapped with these intervals. If S = 150, then according to the interval mapping rules, the scheduling level of this batch of tasks is determined to be medium priority. In this way, the logistics park can prioritize resource allocation for medium-priority tasks based on this scheduling level, such as deploying an appropriate number of AMR robots to perform the handling and storage tasks of this batch of e-commerce promotional goods, ensuring the orderly progress of the tasks.

[0082] When the preset constraint matrix adopts a dynamic adjustment model: In the daily operation of a logistics park, order status changes at any time, and warehouse layout may also change due to adjustments in the types of goods. For example, on a certain workday, a warehouse area originally planned for storing daily necessities may temporarily need to store a large amount of fresh goods, which will cause changes in warehouse storage tasks and handling routes, and the corresponding task topology model will also change.

[0083] At this point, the cloud platform uses an incremental learning algorithm to correct the correlation data between M1 and M2 in real time. The incremental learning algorithm continuously receives new task data and equipment status information, such as the weight and volume of goods in new orders, and the battery level and location of the AMR robots. For example, if at some point a new batch of heavy industrial equipment handling tasks is received, this will increase the demand for large AMR robots and may change the handling route. The incremental learning algorithm will adjust the relevant task nodes, task relationships, and resource requirements in M1 and M2 based on this new information.

[0084] In this dynamic adjustment process, the scheduling level is no longer a fixed interval mapping result, but a continuous set of states that is constantly updated through an incremental learning algorithm. Each new data input causes a corresponding change in the scheduling level to adapt to the ever-changing task and resource conditions within the logistics park. For example, as new tasks are continuously added and equipment status is updated in real time, the scheduling level may gradually evolve from an initial state to a new state that better reflects the current situation. This new state comprehensively considers all the latest task and resource information, thereby achieving precise allocation and efficient scheduling of AMR cluster tasks and ensuring the efficient operation of the logistics park.

[0085] Example 4:

[0086] The cloud platform also includes a device coordination module connected to the core computing unit. This module connects to the physical device resource pool via a data interaction interface. Based on the device scheduling requirements in the final scheduling instruction, the device coordination module selects available device queues from the physical device resource pool and generates task execution sequences to optimize the cluster job process.

[0087] When generating the task execution sequence, a 3D spatial topology model is first loaded. This model visually displays the spatial layout and positional relationships of the physical equipment. The spatial coordinates of each device in the available equipment queue are located within the topology model, accurately obtaining the device's current location information. Then, the optimal movement trajectory from each device's current location to the target work area is calculated using a dynamic programming algorithm. Dynamic programming is a highly efficient optimization algorithm capable of finding the optimal path in complex spatial environments. Simultaneously, the available equipment queue is prioritized according to task urgency, ensuring that urgent tasks are processed first. Finally, the optimal movement trajectory and performance ranking are integrated into the topology model to generate a visualized task execution sequence. This visualized sequence allows operators to more intuitively understand the task execution process and equipment operation, facilitating monitoring and management.

[0088] Take a large electronics manufacturing plant as an example. The plant has multiple production lines, and every day a large number of tasks such as raw material handling, component assembly, and finished product transportation need to be completed by the AMR cluster. This relies heavily on the AMR cluster task planning cloud platform based on digital twins for efficient scheduling.

[0089] After receiving task requirement data, the factory's cloud platform processes the data to generate final scheduling instructions. The device coordination module, connected to the core computing unit, then comes into play. In this factory, the physical equipment resource pool contains different types and functions of Automatic Mobile Responders (AMRs). For example, some AMRs are responsible for moving materials between the raw material warehouse and the production line, while others are specifically used to transfer parts between different workstations on the production line. Based on the equipment scheduling requirements in the final scheduling instructions, the device coordination module selects available equipment queues from the physical equipment resource pool.

[0090] Suppose there is an urgent production task requiring the rapid transfer of a specific batch of raw materials from the warehouse to a designated workstation on the production line. The equipment coordination module first filters the Active Mobile Replicas (AMRs) in the physical equipment resource pool, considering factors such as the AMR's current location, battery level, and load capacity. For example, AMRs that are close to the raw material warehouse, have sufficient battery power, and whose load capacity meets the requirements of this transfer task will be prioritized and formed into an available equipment queue.

[0091] The equipment coordination module generates task execution sequences to optimize cluster workflows. It loads a 3D spatial topology model, which details the layout of various areas within the factory, including raw material warehouses, production lines, temporary storage areas, and the location of each piece of equipment. Within this model, the equipment coordination module can accurately locate the spatial coordinates of each AMR in the available equipment queue, clearly understanding the current position of each AMR.

[0092] Dynamic programming is used to calculate the optimal movement trajectory from the current position of each AMR to the target work area (i.e., the production line station where raw materials need to be transported). The dynamic programming algorithm comprehensively considers the factory layout, such as avoiding obstacles, choosing the shortest path, or the smoothest route. For example, if a passageway in a certain area of ​​the factory is temporarily impassable due to equipment maintenance, the dynamic programming algorithm will automatically avoid this area and plan an alternative suitable route for the AMR.

[0093] The equipment coordination module prioritizes available equipment based on task urgency. For the aforementioned urgent production tasks, AMRs that can complete the handling tasks faster will be prioritized. For example, faster and more efficient AMRs will be given priority to ensure that raw materials are delivered to the production line as quickly as possible, without delaying production schedules.

[0094] The equipment coordination module integrates the calculated optimal movement trajectory and efficiency ranking into a 3D spatial topology model, generating a visualized task execution sequence. This visualized task execution sequence intuitively displays the movement route and execution order of each AMR, facilitating real-time monitoring of task progress by factory managers. Through this visualization interface, managers can clearly see which AMR is heading to the raw material warehouse, which AMR has loaded raw materials and is being transported to the production line, and the estimated completion time of the entire handling task. If any abnormalities occur during task execution, such as an AMR malfunctioning, managers can promptly detect them and readjust task allocation based on the visualized information to ensure the smooth operation of production tasks.

[0095] Example 5:

[0096] When the core computing unit performs logical verification of task allocation benchmarks and navigation parameters, it employs a dual verification mode consisting of an anomaly detection mechanism and a constraint conflict verification mechanism. The anomaly detection mechanism is used to confirm the integrity of equipment operation. During task execution, various faults or abnormal situations may occur. The anomaly detection mechanism determines whether the equipment is in normal operating condition by monitoring and analyzing the equipment's operating data in real time. Once an anomaly is detected, corresponding measures are taken promptly, such as adjusting task allocation or performing equipment maintenance, to ensure the smooth progress of the task.

[0097] Constraint conflict verification mechanisms are used to resolve spatiotemporal inconsistencies between tasks. In multi-task execution scenarios, different tasks may experience spatiotemporal conflicts, such as two tasks needing to use the same device at the same time, or conflicting task execution paths. Constraint conflict verification mechanisms analyze and coordinate the time and space requirements of tasks to avoid conflicts and ensure that tasks can be executed in an orderly manner.

[0098] In addition, the cloud platform includes an instruction conversion module and a task archiving module connected to the core computing unit. The instruction conversion module converts the final scheduling instructions into device control codes and sends these codes to the designated device cluster via a data interaction interface to initiate the task program. The task archiving module stores job requirement data, the first task topology model, the second task topology model, the scheduling level, and the final scheduling instructions, and generates a complete job record chain according to the task lifecycle. These modules work together to ensure the efficient operation of the cloud platform and the smooth execution of tasks.

[0099] Taking a large express sorting center as an example, tens of thousands of packages are sorted and delivered here every day. In this express sorting center, the AMR cluster is responsible for tasks such as package handling, sorting, and loading. After generating the final scheduling instructions, the core computing unit performs logical verification on the task allocation benchmark and navigation parameters, adopting a dual verification mode of anomaly detection mechanism and constraint conflict verification mechanism.

[0100] An anomaly detection mechanism is used to confirm the integrity of equipment operation. In express sorting centers, AMR robots need to work continuously for long periods, and equipment failures can occur at any time. For example, during the process of an AMR handling packages, its battery power monitoring system sends data to the cloud platform. The anomaly detection mechanism analyzes this data in real time. If it detects an abnormal drop in battery power—for example, under normal circumstances, the battery consumption for completing a handling task should be within 10%, but the AMR's battery power drops by more than 15%—the system will immediately determine that the equipment has malfunctioned. Simultaneously, the anomaly detection mechanism also monitors the AMR's sensor data, such as LiDAR data and vision sensor data. If the LiDAR data shows abnormal fluctuations in its scanning range, it may indicate that the sensor is being interfered with or malfunctioning, which will also trigger the anomaly detection mechanism. Once an anomaly is detected, the cloud platform will immediately issue an alarm and notify technicians for repairs. Simultaneously, it will reassign the tasks originally performed by the AMR to other normally operating equipment to ensure that package sorting operations are not significantly affected.

[0101] The constraint conflict verification mechanism is used to resolve spatiotemporal conflicts between tasks. In a parcel sorting center, numerous automated guided vehicles (AMRs) operate simultaneously within a limited space. If task planning is inadequate, spatiotemporal conflicts can easily occur. For example, at a certain moment, two AMRs receive separate tasks: one to move a batch of packages from sorting area A to loading area X, and the other to move another batch of packages from sorting area B to loading area Y. Their planned paths intersect at a narrow passageway. The constraint conflict verification mechanism analyzes the task paths and execution times of all AMRs before task execution. When a conflict is detected, the system adjusts its path based on factors such as the urgency of the packages and the current location of the AMRs. If the task of moving packages from sorting area A is more urgent, the system adjusts the path of the other AMR, making it take a slightly longer route to avoid the conflict area, ensuring that both tasks can be executed smoothly and preventing AMRs from waiting for each other or colliding in the passageway.

[0102] In addition, the cloud platform includes an instruction conversion module and a task archiving module. The instruction conversion module converts the final scheduling instruction into device control codes and sends these codes to the designated device cluster via a data interaction interface to initiate the task program. In a parcel sorting center, the final scheduling instruction might be "Instruct AMR number 001 to move 50 packages from sorting area C to loading area Z." The instruction conversion module converts this instruction into a control code that the AMR can recognize, such as a series of numbers and letters, and sends it to AMR number 001 via the data interaction interface. After receiving the control code, the AMR initiates the handling task according to the instruction, accurately drives to sorting area C, picks up 50 packages, and then transports them to loading area Z.

[0103] The task archiving module stores job requirement data, the first task topology model, the second task topology model, the scheduling level, and the final scheduling instruction, generating a complete job record chain according to the task lifecycle. In the daily operation of a courier sorting center, every package handling task is meticulously recorded. For example, for a task involving moving packages from sorting area D to loading area W, the task archiving module records job requirement data, including the number, weight, and destination of the packages; the relevant parts of the first and second task topology models, such as the task's layout and resource matching in three-dimensional space; the scheduling level, i.e., the task's priority; and the final scheduling instruction. This information is organized according to the task lifecycle, from task generation, allocation, execution to completion, forming a complete job record chain. By reviewing these records, managers can trace the task execution process, analyze the rationality of task planning, and summarize lessons learned to optimize subsequent task scheduling and management strategies, thereby improving the overall operational efficiency of the courier sorting center.

[0104] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0105] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud platform for AMR cluster task planning based on digital twins, characterized in that, include: The task scheduling module is used to receive multi-source job requirement data and identify the task type, and divide the job requirement data into basic task data stream and extended task data stream based on preset task decomposition rules. The twin modeling module is used to perform three-dimensional spatial meshing processing on the basic task data stream to generate a first task topology model, and to perform dynamic resource matching processing on the extended task data stream to generate a second task topology model. The cluster status analysis module is used to map the first task topology model and the second task topology model to the corresponding scheduling level after data fusion according to the preset constraint condition matrix, and use the scheduling level as the task allocation benchmark for the current cluster. The path planning module is used to call the optimal path scheme from the preset path database based on the scheduling level, and use the optimal path scheme as the navigation parameter of the current cluster. The core computing unit is used to perform logical verification on the task allocation benchmark and the navigation parameters to generate the final scheduling instruction; The preset task decomposition rules include a set of rigid constraints and a set of flexible constraints; the set of rigid constraints includes a time window identifier, a device capacity identifier, and a job priority identifier; the set of flexible constraints includes an energy consumption threshold identifier and a path conflict identifier, and each identifier corresponds to an independent task processing channel. The platform also includes a data interaction interface, which is used to enable the task scheduling module, the twin modeling module, the cluster status analysis module and the path planning module to communicate with the physical device network respectively. The task scheduling module divides the job requirement data based on the preset task decomposition rules, including: The data interaction interface is used to obtain mixed data packets from the physical device network in real time, and the protocol tags of the mixed data packets are matched with the identifiers in the rigid constraint set to separate the basic task segments. The extended tags of the hybrid data packets are traversed according to the identifiers in the set of flexible constraints to extract the elastic task segments; The basic task segment and the elastic task segment are synchronized and aligned according to the task trigger time, and then written into the task storage area and the extended task buffer respectively.

2. The AMR cluster task planning cloud platform based on digital twins according to claim 1, characterized in that, The twin modeling module performs dynamic resource matching processing on the extended task data stream, including: The task sequence in the extended task data stream is divided into fixed task groups and elastic task groups. Based on a preset dynamic load balancing algorithm, the resource utilization rate of the elastic task groups is calculated to generate a load feature set. The device status data in the extended task data stream is processed hierarchically, and the energy consumption characteristics of each level of device are extracted and an energy consumption map is constructed. The load feature set is dynamically associated and matched with the energy consumption map to generate the second task topology model.

3. The AMR cluster task planning cloud platform based on digital twins according to claim 1, characterized in that, When the preset constraint matrix adopts a static analysis model, the scheduling level is the interval mapping result of the linear superposition value of the first task topology model and the second task topology model; When the preset constraint matrix adopts a dynamic adjustment model, the scheduling level is a continuous set of states that are corrected in real time by using an incremental learning algorithm to modify the associated data of the first task topology model and the second task topology model.

4. The AMR cluster task planning cloud platform based on digital twins according to claim 1, characterized in that, It also includes a device collaboration module connected to the core computing unit, which is connected to the physical device resource pool through the data interaction interface; The device coordination module is used to select available device queues from the physical device resource pool according to the device scheduling requirements in the final scheduling instruction, and generate a task execution sequence to optimize the cluster operation process.

5. The AMR cluster task planning cloud platform based on digital twins according to claim 4, characterized in that, The device collaboration module generates a task execution sequence including: Load a three-dimensional spatial topology model, and locate the spatial coordinates of each device in the available device queue within the topology model; The optimal movement trajectory from the current position of each device to the target work area is calculated based on the dynamic programming algorithm, and the available device queue is sorted by performance according to the urgency of the task. The optimal movement trajectory and the performance ranking are integrated into the topology model to generate a visual task execution sequence.

6. The AMR cluster task planning cloud platform based on digital twins according to claim 1, characterized in that, When the core computing unit performs logical verification on the task allocation benchmark and navigation parameters, it adopts a dual verification mode of an anomaly detection mechanism and a constraint conflict verification mechanism. The anomaly detection mechanism is used to confirm the integrity of the equipment operation, and the constraint conflict verification mechanism is used to resolve spatiotemporal contradictions between tasks.

7. The AMR cluster task planning cloud platform based on digital twins according to claim 1, characterized in that, It also includes an instruction conversion module connected to the core computing unit. The instruction conversion module is used to convert the final scheduling instruction into device control code and send the device control code to the designated device cluster through the data interaction interface to start the task program.

8. The AMR cluster task planning cloud platform based on digital twins according to claim 1, characterized in that, It also includes a task archiving module connected to the core computing unit. The task archiving module is used to store the job requirement data, the first task topology model, the second task topology model, the scheduling level and the final scheduling instruction, and to generate a complete job record chain according to the task lifecycle.

Images