A cloud edge decoupling heterogeneous cluster management and control system and method

Abstract

The application relates to a cloud edge end decoupling heterogeneous cluster management and control system and method, and belongs to the technical field of robot control. The system comprises the following: a management and control and digital twin module for realizing global planning and visual management of the heterogeneous cluster; a general master container module based on unified scheduling of a master container and targeted control of different devices by a sub-container, which is used for realizing task analysis, concurrent control of tasks based on device load prediction and comprehensive adaptation of the tasks and devices, device cooperation and safety backup; a dynamic data adaptation module for realizing unified analysis and semantic conversion of heterogeneous hardware data; and a heterogeneous communication bus module for supporting data interaction between the modules and directly bridging protocols and transmitting data with the devices. The system can realize flexible scheduling, stable operation, intelligent cooperation and safe and reliable operation of a robot mobile 4S store, and comprehensively improves operation and maintenance management and control and service efficiency of the heterogeneous cluster.

Description

Technical Field

[0001] This application relates to the field of robot control technology, and in particular to a cloud-edge-device decoupled heterogeneous cluster management and control system and method. Background Technology

[0002] With the increasing prevalence of heterogeneous clusters in large-scale applications, the demands for unified equipment management and flexible deployment across scenarios are constantly rising. In the scenario of mobile robot 4S stores, the core requirement is to enable robots and IoT (Internet of Things) devices within heterogeneous clusters to be deployed and stopped immediately, scenarios to be rapidly reconfigured, and exhibits to be frequently replaced, in order to adapt to flexible operational needs.

[0003] However, current heterogeneous cluster control technology has significant shortcomings: the communication protocols and control interfaces of devices from different manufacturers are incompatible, and cluster management requires the separate development of adaptation programs and modification of underlying code, resulting in low adaptation efficiency, high technical barriers, and poor system compatibility. At the same time, traditional control methods rely on fixed terminals and sites, resulting in high deployment costs and insufficient flexibility, making it difficult to meet the dynamic task scheduling needs of mobile scenarios, which seriously restricts the promotion and multi-scenario expansion of heterogeneous clusters in mobile robot 4S stores. Summary of the Invention

[0004] Therefore, it is necessary to provide a cloud-edge-device decoupled heterogeneous cluster management system and method to address the above-mentioned technical problems.

[0005] A cloud-edge-device decoupled heterogeneous cluster management and control system, the system being applied to a mobile robot 4S store, the system comprising:

[0006] The control and digital twin module, based on a dual-terminal deployment of cloud and edge, is used to generate a 3D visualization scene of the showroom by using digital twin modeling based on the physical showroom data and equipment information of the robot mobile 4S store. It also generates standardized task behavior description files by receiving user input commands and real-time equipment status and task execution feedback. The equipment includes heterogeneous robots and IoT devices. The general-purpose main control container module includes a main container deployed at the edge and sub-containers deployed on the device. The main container is used to parse the task behavior description file, execute task concurrency control based on device load prediction and comprehensive task device adaptation, generate a global task execution scheduling table, and decompose it into individual device sub-execution instructions. These instructions are then distributed to the corresponding sub-containers on the device for unified management. At the same time, the main container collects the device status and task execution feedback uploaded by all sub-containers, writes them into the global status area built into the dynamic data adaptation module, and provides global security fallback. Each sub-container is used to convert the received individual device sub-execution instructions into control instructions that the device can recognize. Based on the control instructions, the sub-container drives the device to execute task actions and monitors the device status and task execution feedback in real time, which are then uploaded to the main container while providing local security fallback. The dynamic data adaptation module deployed at the edge is used to dynamically allocate a dedicated script execution sandbox based on the device ID without modifying the underlying hardware code. It automatically loads the corresponding lightweight dynamic adaptation script to perform device access, data collection and parsing, and unified semantic data conversion. It also writes the unified semantic data of each device into the global status area. Based on the dynamic updates of the global status area, it provides real-time data support for digital twin modeling and heterogeneous cluster management. The heterogeneous communication bus module is used to support data interaction between modules using the MQTT-SN+TCP / IP dual-protocol unified data bus, and directly bridges protocols and transmits data with each device. It is the only channel for devices to access the system.

[0007] In one embodiment, the control and digital twin module includes a digital twin modeling unit, an instruction parsing and semantic conversion unit, and a task generation unit; The digital twin modeling unit is used to adapt the basic data of the 2D physical showroom of the robot mobile 4S store to the dynamic data. The device basic information is uploaded by the heterogeneous communication bus module. The digital twin modeling is used to generate a 3D visualization scene of the showroom. The ID of each device in the robot mobile 4S store is bound to the 3D digital twin model to realize the coordinate alignment and status linkage between the physical device and the digital twin model. It supports the visualization display of the real-time position and operating parameters of the device in the 3D visualization scene. The instruction parsing and semantic conversion unit is used to perform semantic parsing on natural language instructions input by users, using a large language model to extract core business requirements; for business semantic instructions triggered by users through preset buttons, it calls the business logic template library for matching and filtering the corresponding business logic templates; for scene events triggered by sensors in the exhibition hall, it uses semantic conversion to generate core business requirements. The task generation unit is used to generate standardized task behavior description files based on core business requirements or business logic templates, combined with the device status and task execution feedback uploaded in real time by the main container through the heterogeneous communication bus module. The task behavior description files include task ID, task type, task priority, task execution time limit, schedulable devices corresponding to each task, task triggering conditions, task termination conditions, and associated rules of linked devices.

[0008] In one embodiment, the main container includes a file parsing unit, a task concurrency control unit, a detailed instruction issuing unit, a resource conflict detection unit, a task interruption and recovery unit, and a global security fallback unit. The file parsing unit is used to receive the task behavior description file issued by the control and digital twin module through the heterogeneous communication bus module, and parse it to obtain the task ID, task type, task priority, task execution time limit, schedulable devices corresponding to each task, task triggering conditions, task termination conditions, and linkage device association rules. The task concurrency control unit is used to filter the optimal execution device for each task and generate a global task execution schedule table based on the parsing results of the task behavior description file, through device load prediction and comprehensive task device adaptation calculation, thus clarifying the execution order, optimal execution device, execution start time and execution end time of each task. The subdivided instruction issuance unit is used to split the global task execution schedule table into single-device subdivided execution instructions according to the device ID, and send them to the sub-containers of the corresponding devices for unified management; the single-device subdivided execution instructions include standardized execution actions, action parameters, execution duration, and linkage triggering conditions; The resource conflict detection unit is used to detect resource conflicts when multiple devices execute tasks concurrently. If a conflict is detected, the execution order of each task in the global task execution schedule table is automatically adjusted, and a conflict-free task execution window is allocated to each sub-container. The task interruption and recovery unit is used to immediately trigger the task interruption mechanism when the file parsing unit receives a new task behavior description file or receives a manual emergency stop command / emergency intervention command issued by the user through the control and digital twin module. It records the current task execution progress and device status, parses the new task behavior description file and issues new single-device subdivision execution commands. If it is necessary to recover the original task, it continues to issue single-device subdivision execution commands from the breakpoint according to the recorded execution progress, so as to achieve seamless switching. The global security fallback unit is used to collect the device status and task execution feedback uploaded by all sub-containers through the heterogeneous communication bus module in real time, write them into the global status area in the dynamic data adaptation module, and upload the global task execution progress to the control and digital twin module. At the same time, by reading the unified semantic data, device status and task execution feedback of each device in the global status area, it performs global security risk detection. If the global security risk exceeds the preset threshold, it immediately issues a global emergency stop command to the sub-container to suspend the task execution of all devices.

[0009] In one embodiment, the task concurrency control unit includes a device load prediction mechanism, a task device comprehensive adaptation calculation mechanism, and an optimal execution device selection mechanism; The equipment load prediction mechanism first parses and obtains the current load data from the sub-containers of all schedulable devices uploaded via the heterogeneous communication bus module. Simultaneously, it calls the local historical load database to extract the historical average load and historical load fluctuation data for each schedulable device under similar task scenarios. Then, based on the current load data, historical average load data, and historical load fluctuation data of each schedulable device, it estimates the predicted load value for each schedulable device, expressed as: ; in, To predict load values; The current load data is as follows: for heterogeneous robots, the current load data is obtained by collecting CPU utilization, memory usage, queue length of pending tasks, and remaining time of currently executing tasks, and then normalizing and weighting the data; for IoT devices, the current load data is obtained by collecting port utilization, command response latency, and online status percentage, and then normalizing and weighting the data. This represents the historical average load across similar task scenarios. The standard deviation of historical load fluctuation data under similar task scenarios. This is the load deviation correction factor. This is the load fluctuation impact coefficient; Then, based on the predicted load value, the current load redundancy calculation result of each schedulable device is corrected. At the same time, the basic compatibility between the device and the task is combined, and devices with load redundancy correction results lower than the preset threshold or basic compatibility are eliminated to obtain candidate devices for each task. The task-device integrated adaptation computing mechanism first assigns basic scheduling weights to different types of tasks. Simultaneously, it introduces task correlation factors and dynamically adjusts the basic scheduling weights of each task based on task priority, urgency of task execution time, and the severity of losses from task execution errors, thus obtaining the optimized scheduling weights for each task. Finally, it integrates the optimized scheduling weights of all tasks. Result of load redundancy correction for corresponding candidate devices and basic device task adaptability The overall fit between each task and its corresponding candidate device is calculated and expressed as: ; in, , and These are the weighting coefficients for the three linear terms; This is used to quantify the interaction between the three linear terms, avoiding the dominance of a single factor in the overall fit result; The optimal execution device selection mechanism sorts all candidate devices for each task in descending order of overall suitability, prioritizes the allocation of the device with the highest overall suitability as the optimal execution device for each task, and generates a global task execution scheduling table.

[0010] In one embodiment, the sub-container includes an instruction parsing and hardware adaptation unit, a real-time task execution and status reporting unit, a local security fallback unit, and a linkage triggering execution unit. The instruction parsing and hardware adaptation unit is used to parse the single-device subdivision execution instructions issued by the main container through the heterogeneous communication bus module, and to convert the standardized execution actions in the single-device subdivision execution instructions into control instructions that can be recognized by the underlying hardware of the device. The real-time task execution and status reporting unit is used to drive the device to perform task actions based on control commands, collect device status and task execution feedback at a fixed frequency, and upload them to the main container through the heterogeneous communication bus module. The main container then writes the global status area in the dynamic data adaptation module. The local security fallback unit is used to detect local security risks based on the data collected in real time by the real-time task execution and status reporting unit. If the local security risk exceeds the preset threshold, the local emergency stop / obstacle avoidance mechanism is triggered immediately without waiting for the main container to issue the instruction. At the same time, the local security alarm information is uploaded to the main container and the control and digital twin module through the emergency data channel of the heterogeneous communication bus module. The linkage trigger execution unit is used to monitor the unified semantic data of each device in the global state area in real time. When it is detected that the unified semantic data meets the linkage trigger conditions in the single device subdivision execution instruction, it immediately executes the linkage action according to the linkage device association rules in the task behavior description file, and reports the linkage execution status to the main container, and then writes it into the global state area.

[0011] In one embodiment, the dynamic data adaptation module includes an adaptation script loading and management unit, a raw data dynamic parsing unit, a unified semantic data conversion unit, a global status area writing and updating unit, and a device basic information reporting unit. The adaptation script loading and management unit is used to assign a dedicated script execution sandbox to a new device based on its device ID without modifying the underlying hardware code when a new device is connected. It automatically loads the corresponding lightweight dynamic adaptation script. The script execution sandbox enables the independent execution of adaptation scripts for each device, avoiding script conflicts. It also supports online updates and uninstallation of scripts without restarting the system. The raw data dynamic parsing unit is used to perform real-time parsing and preprocessing of the raw hardware data uploaded by the device through the adaptation script, including removing redundant data, correcting abnormal data, and converting data format. The unified semantic data conversion unit is used to convert preprocessed device data according to the system's preset minimalist semantic data specification to generate unified semantic data for each device. The semantic data specification is universal across the entire system and has no device / protocol differences. The unified semantic data is represented as: Data Identifier = Value / Status. The global status area writing and updating unit is used to write the unified semantic data of each device into the global status area in the form of key-value pairs. The key is the device ID-data identifier, and the value is the corresponding data / status. At the same time, the global status area is updated according to the device status and task execution feedback collected by the sub-container at a fixed frequency to ensure the real-time performance of the data in the global status area. The device basic information reporting unit is used to upload the device's access status, data collection items, and unified semantic data list to the management and control and digital twin module, providing real-time data support for digital twin modeling and heterogeneous cluster management.

[0012] In one embodiment, the heterogeneous communication bus module includes a device access protocol bridging unit, a network segment unification and address allocation unit, a data classification transmission and routing forwarding unit, and a weak network data fault tolerance processing unit. The device access protocol bridging unit has a built-in multi-protocol bridging proxy plugin, which is used to automatically identify the communication protocol type of the device when receiving a device access request, allocate a dedicated bridging proxy channel for it, and encapsulate non-standard communication protocols into the bus-unified MQTT-SN protocol / TCP / IP protocol to realize bus access for cross-protocol devices. The network segment unification and address allocation unit is used to assign a unified local area network private IP address and bus device identifier to all devices connected to the bus, and to build a bus device address mapping table; the bus device identifier consists of network segment code + device ID; The data classification, transmission, and routing forwarding unit establishes three independent transmission channels: a downlink command channel, an uplink status channel, and an emergency data channel, and classifies and labels the input data. Downlink command data includes task behavior description files and detailed execution instructions for individual devices, which are accurately forwarded through the downlink command channel based on the address mapping table. Uplink status data includes basic device information, device status, task execution feedback, and global task execution progress, which are forwarded through the uplink status channel. Emergency data includes manual emergency stop / emergency intervention instructions, global emergency stop instructions, and local safety alarm information, which are given the highest transmission priority and transmitted through the emergency data channel block to ensure the real-time nature of emergency information. The weak network data fault tolerance processing unit is used to enable QoS2 level service of MQTT-SN protocol in weak network environment, realize the one-time transmission of data, and perform fragmentation and compression of transmitted data packets to reduce packet loss rate; if data transmission is interrupted, it will automatically resume transmission after the network is restored.

[0013] A cloud-edge-device decoupled heterogeneous cluster management method, the method being implemented based on the aforementioned cloud-edge-device decoupled heterogeneous cluster management system, comprising: S1 utilizes a dynamic data adaptation module deployed at the edge to dynamically allocate dedicated script execution sandboxes based on device IDs without modifying the underlying hardware code. It automatically loads corresponding lightweight dynamic adaptation scripts for device access, data collection and parsing, and unified semantic data conversion. Through a heterogeneous communication bus module, it writes the unified semantic data of each device into the global state area. The dynamic updates of the global state area provide real-time data support for digital twin modeling and heterogeneous cluster management. The devices include heterogeneous robots and IoT devices. S2 utilizes a control and digital twin module deployed on both cloud and edge platforms. Based on the physical showroom data and dynamic data adaptation module of the robot mobile 4S store, and the device basic information uploaded by the module through the heterogeneous communication bus, it generates a 3D visualization scene of the showroom using digital twin modeling. By receiving user input commands and the device status and task execution feedback uploaded in real time by the main container through the heterogeneous communication bus module, it generates a standardized task behavior description file and distributes it to the general main control container module through the heterogeneous communication bus module. S3 utilizes the main container to parse the task behavior description file, executes task concurrency control based on device load prediction and comprehensive task device adaptation, generates a global task execution schedule table, and decomposes it into individual device sub-execution instructions. These instructions are then distributed to the corresponding device sub-containers via the heterogeneous communication bus module for unified management. Simultaneously, it collects device status and task execution feedback uploaded by all sub-containers, writes them into the global status area built into the dynamic data adaptation module, and provides global security fallback. Each sub-container converts the received individual device sub-execution instructions into device-recognizable control instructions, drives the device to execute task actions based on the control instructions, and monitors device status and task execution feedback in real time. These are then uploaded to the main container via the heterogeneous communication bus module while providing local security fallback.

[0014] The aforementioned cloud-edge-device decoupled heterogeneous cluster management system and method has the following advantages compared to existing technologies: 1. In the dynamic data adaptation module, the traditional underlying hardware hard-coding mode has been abandoned. When a new device is connected, only a lightweight dynamic adaptation script needs to be loaded, without modifying the underlying control code. The device connection time is greatly reduced, which greatly shortens the construction and equipment replacement delivery cycle of the robot mobile 4S store and enables flexible customization and expansion. Moreover, the dynamic data adaptation module realizes the unified semantic conversion of all device data. Combined with the 3D visualization scene of the control and digital twin module, operators can monitor the device status and task execution progress in real time, realizing intelligent and visualized control of the showroom.

[0015] 2. By separating the control and digital twin module responsible for remote management and digital twin modeling from the general main control container module responsible for local real-time control, efficient centralized management and control of various heterogeneous devices can be achieved. It can achieve overall planning and visual management based on remote management and digital twin modeling, and ensure the stability and response speed of real-time control through the local general main control container module. At the same time, with the unified scheduling of the main container and the targeted control of different devices by the sub-containers, the overall management and control architecture is more flexible and adaptable, and the efficiency and reliability of heterogeneous device collaborative operation are greatly improved.

[0016] 3. By executing task concurrency control based on device load prediction and comprehensive task device adaptation in the main container, the rationality and execution efficiency of global task scheduling can be effectively improved, avoiding abnormal operation of devices due to load imbalance or task conflict. Furthermore, by executing global / local security fallbacks in the main container and sub-containers, a multi-layered security protection mechanism is built to ensure that the system can operate stably and reliably under various working conditions.

[0017] 4. The heterogeneous communication bus module can support bridging of multiple communication protocols for heterogeneous devices, solving the problem of interconnection between heterogeneous devices; at the same time, by using the MQTT-SN+TCP / IP dual-protocol unified data bus, reliable data transmission can still be guaranteed in weak network environments, effectively meeting the network requirements of mobile 4S stores for outdoor deployment. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structure of a heterogeneous cluster management and control system with cloud-edge-device decoupling in one embodiment; Figure 2 This is a schematic diagram of the workflow of a cloud-edge-device decoupled heterogeneous cluster management and control system in one embodiment; Figure 3 This is a schematic diagram of the protocol bridging of heterogeneous communication bus modules in one embodiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] In one embodiment, such as Figure 1 and Figure 2As shown, a cloud-edge-device decoupled heterogeneous cluster management and control system is provided. This system is applied to a robot mobile 4S store. The system includes: a management and control and digital twin module, a general main control container module, a dynamic data adaptation module, and a heterogeneous communication bus module. The modules work together to realize unified management and control and dynamic reconstruction of business logic for heterogeneous robots and IoT devices. Among them, IoT devices include automatic doors, lights, sensors, etc.

[0021] 1. The control and digital twin module is deployed on a dual-end basis: cloud and edge. It is deployed on a cloud server (remote control) and a mobile 4S store edge server (local core control). Data is synchronized in real time between the two ends. The local edge server is the default control end and runs independently when the network is weak or offline. This module is used to generate a 3D visualization scene of the showroom using digital twin modeling based on the physical showroom data and equipment information of the mobile 4S store. It also generates standardized task behavior description files by receiving user input commands and real-time equipment status and task execution feedback.

[0022] Specifically, the control and digital twin module includes a digital twin modeling unit, an instruction parsing and semantic conversion unit, and a task generation unit.

[0023] The digital twin modeling unit is used to generate a 3D visualization scene of the exhibition hall based on the basic data of the 2D physical exhibition hall of the robot mobile 4S store (including the preset deployment coordinates of the equipment, the range of the passage, the boundary of the display area, and the physical size parameters of the exhibition hall) and the basic equipment information (equipment ID, equipment type, basic function list, and status collection items) uploaded by the dynamic data adaptation module through the heterogeneous communication bus module. The unit binds the ID of each equipment in the robot mobile 4S store with the 3D digital twin model, realizes the coordinate alignment and status linkage between the physical equipment and the digital twin model, and supports the visualization display of the real-time position and operating parameters of the equipment in the 3D visualization scene.

[0024] The instruction parsing and semantic conversion unit is used to perform semantic parsing on natural language instructions input by users (such as voice / text "VIP customer guided tour demonstration"), and extract core business requirements (such as "VIP guided tour" which needs to be parsed as "welcoming robot guidance + explanation robot function demonstration + light focusing display area + automatic door always open"). For business semantic instructions triggered by users through preset buttons, it calls the business logic template library for matching and filtering the corresponding business logic template. For scene events triggered by sensors in the exhibition hall (such as "customer enters the exhibition area"), it uses semantic conversion to generate core business requirements.

[0025] The task generation unit generates standardized task behavior description files based on core business requirements or business logic templates, combined with device status and task execution feedback uploaded in real-time by the main container via the heterogeneous communication bus module. These files are then distributed to the main container via the heterogeneous communication bus module. The task behavior description files include task ID, task type, task priority, task execution time limit, schedulable devices for each task, task triggering conditions, task termination conditions, and associated rules for linked devices. The task behavior description files can be in JSON format.

[0026] 2. The general-purpose main control container module includes a main container deployed at the edge and sub-containers deployed on the device. The main and sub-containers establish a dedicated communication link through a heterogeneous communication bus module, and the main container provides unified management and control over all sub-containers. This module is the core of the system's execution, realizing task parsing, concurrency control, device collaboration, and security fallback. It is the specific execution unit of the task behavior description file.

[0027] The main container is used to parse the task behavior description file, execute task concurrency control based on device load prediction and comprehensive task device adaptation, generate a global task execution scheduling table, and decompose it into individual device sub-execution instructions. These instructions are then distributed to the corresponding sub-containers for unified management. At the same time, the container collects the device status and task execution feedback uploaded by all sub-containers, writes them into the global status area built into the dynamic data adaptation module, and provides global security fallback. Each sub-container is used to convert the received individual device sub-execution instructions into control instructions that the device can recognize. Based on the control instructions, the container drives the device to execute task actions and monitors the device status and task execution feedback in real time, which are then uploaded to the main container while providing local security fallback.

[0028] Specifically, the main container includes a file parsing unit, a task concurrency control unit, a detailed instruction issuance unit, a resource conflict detection unit, a task interruption and recovery unit, and a global security fallback unit.

[0029] The file parsing unit receives task behavior description files from the control and digital twin module via the heterogeneous communication bus module, and parses them to obtain the task ID, task type, task priority, task execution time limit, schedulable devices corresponding to each task, task triggering conditions, task termination conditions, and associated rules for linked devices. The task behavior description file receiving frequency is set to 100ms / time.

[0030] The task concurrency control unit is used to filter the optimal execution device for each task and generate a global task execution schedule table based on the parsing results of the task behavior description file, through device load prediction and comprehensive task device adaptation calculation, thus clarifying the execution order, optimal execution device, execution start time and execution end time of each task.

[0031] The subdivided instruction distribution unit is used to split the global task execution schedule table into single-device subdivided execution instructions according to the device ID, and send them to the sub-containers of the corresponding devices for unified management; the single-device subdivided execution instructions include standardized execution actions, action parameters, execution duration and linkage trigger conditions, which are also in JSON format.

[0032] The resource conflict detection unit is used to detect resource conflicts when multiple devices execute tasks concurrently (such as avoiding path conflicts caused by two robots moving simultaneously in the same channel). If a conflict is detected, the execution order of each task in the global task execution schedule table is automatically adjusted, and a conflict-free task execution window is allocated to each sub-container.

[0033] The task interruption and recovery unit is used to immediately trigger the task interruption mechanism when the file parsing unit receives a new task behavior description file (such as switching from "nighttime security patrol" to "daytime VIP reception") or receives a manual emergency stop / emergency intervention command issued by the user through the control and digital twin module. It records the current task execution progress and device status, parses the new task behavior description file, and issues new single-device subdivision execution commands. If it is necessary to recover the original task, it continues to issue single-device subdivision execution commands from the breakpoint according to the recorded execution progress, so as to achieve seamless switching. The global security fallback unit is used to collect the device status and task execution feedback uploaded by all sub-containers through the heterogeneous communication bus module in real time, write them into the global status area in the dynamic data adaptation module, and upload the global task execution progress to the control and digital twin module. At the same time, by reading the unified semantic data, device status and task execution feedback of each device in the global status area, it performs global security risk detection. If the global security risk exceeds the preset threshold, it immediately issues a global emergency stop command to the sub-container to suspend the task execution of all devices.

[0034] Specifically, the task concurrency control unit includes a device load prediction mechanism, a task device comprehensive adaptation calculation mechanism, and an optimal execution device selection mechanism.

[0035] The equipment load prediction mechanism first parses and obtains the current load data from the sub-containers of all schedulable devices uploaded via the heterogeneous communication bus module. Simultaneously, it calls the local historical load database to extract the historical average load and historical load fluctuation data for each schedulable device under similar task scenarios. Then, based on the current load data, historical average load data, and historical load fluctuation data of each schedulable device, it estimates the predicted load value for each schedulable device, expressed as: ; in, To predict load values; The current load data is as follows: for heterogeneous robots, the current load data is obtained by collecting CPU utilization, memory usage, queue length of pending tasks, and remaining time of currently executing tasks, and then normalizing and weighting the data; for IoT devices, the current load data is obtained by collecting port utilization, command response latency, and online status percentage, and then normalizing and weighting the data. This represents the historical average load across similar task scenarios. The standard deviation of historical load fluctuation data under similar task scenarios. This is the load deviation correction factor. The load fluctuation impact coefficient is used; then, the current load redundancy calculation result of each schedulable device is corrected based on the predicted load value. At the same time, the basic compatibility between the device and the task is combined, and devices with load redundancy correction results lower than the preset threshold or basic compatibility are eliminated to obtain candidate devices for each task.

[0036] The task-device integrated adaptation computing mechanism first assigns basic scheduling weights to different types of tasks. Simultaneously, it introduces task correlation factors and dynamically adjusts the basic scheduling weights of each task based on task priority, urgency of task execution time, and the severity of losses from task execution errors, thus obtaining the optimized scheduling weights for each task. Finally, it integrates the optimized scheduling weights of all tasks. Result of load redundancy correction for corresponding candidate devices and basic device task adaptability The overall fit between each task and its corresponding candidate device is calculated and expressed as: ; in, , and These are the weighting coefficients for the three linear terms; This is used to quantify the interaction between the three linear terms, avoiding the dominance of a single factor in the overall fit result.

[0037] The optimal execution device selection mechanism sorts all candidate devices for each task in descending order of overall suitability, prioritizes the allocation of the device with the highest overall suitability as the optimal execution device for each task, and generates a global task execution scheduling table.

[0038] Specifically, the sub-container includes an instruction parsing and hardware adaptation unit, a real-time task execution and status reporting unit, a local security fallback unit, and a linkage trigger execution unit.

[0039] The instruction parsing and hardware adaptation unit is used to parse the single-device subdivision execution instructions issued by the main container through the heterogeneous communication bus module, and convert the standardized execution actions in the single-device subdivision execution instructions (such as "move forward 0.5m" and "adjust the light to 80% brightness") into control instructions that can be recognized by the underlying hardware of the device (such as the robot's motor pulse signal and the PWM control signal of the IoT device).

[0040] The real-time task execution and status reporting unit is used to drive the device to perform task actions based on control commands. It collects device status and task execution feedback at a fixed frequency (e.g., 50ms / time) and uploads it to the main container through the heterogeneous communication bus module. The main container then writes the data into the global status area of ​​the dynamic data adaptation module.

[0041] The local security fallback unit is used to detect local security risks based on the data collected in real time by the real-time task execution and status reporting unit. If the local security risk exceeds the preset threshold, the local emergency stop / obstacle avoidance mechanism is triggered immediately without waiting for the main container to issue instructions. At the same time, the local security alarm information is uploaded to the main container and the control and digital twin module through the emergency data channel of the heterogeneous communication bus module.

[0042] The linkage trigger execution unit is used to monitor the unified semantic data of each device in the global status area in real time. When it detects that the unified semantic data meets the linkage trigger condition in the single device subdivision execution instruction (such as "customer enters the display area"), it immediately executes the linkage action according to the linkage device association rules in the task behavior description file, and reports the linkage execution status to the main container, and then writes it into the global status area.

[0043] It should be understood that by separating the control and digital twin module responsible for remote management and digital twin modeling from the general-purpose main control container module responsible for local real-time control, efficient centralized management and control of various heterogeneous devices can be achieved. This allows for global coordination and visualized management through remote control and digital twin modeling, while ensuring the stability and response speed of real-time control through the local general-purpose main control container module. Furthermore, the unified scheduling of the main container and the targeted control of different devices by sub-containers make the overall management and control architecture more flexible and adaptable, significantly improving the efficiency and reliability of heterogeneous device collaborative operation. In addition, by executing task concurrency control based on device load prediction and comprehensive task device adaptation in the main container, the rationality and execution efficiency of global task scheduling can be effectively improved, preventing device malfunctions due to load imbalance or task conflicts. Finally, by implementing global / local security safeguards in the main and sub-containers, a multi-layered security protection mechanism is built to ensure stable and reliable system operation under various working conditions.

[0044] 3. A dynamic data adaptation module deployed at the edge shares local storage resources with the general-purpose main control container module and has a built-in global state area (a memory-based key-value database supporting high-concurrency read and write operations with a read / write frequency ≤10ms / time). This module enables unified parsing and semantic transformation of heterogeneous hardware data. Without recompiling the underlying system, it allows the system to identify the raw data of various heterogeneous devices, providing unified semantic data support for the general-purpose main control container module. Specifically, this module dynamically allocates dedicated script execution sandboxes based on device IDs without modifying the underlying hardware code. It automatically loads corresponding lightweight dynamic adaptation scripts for device access, data collection and parsing, and unified semantic data transformation, and writes the unified semantic data of each device into the global state area. The dynamic updates of the global state area provide real-time data support for digital twin modeling and heterogeneous cluster management.

[0045] Specifically, the dynamic data adaptation module includes an adaptation script loading and management unit, a raw data dynamic parsing unit, a unified semantic data conversion unit, a global status area writing and updating unit, and a device basic information reporting unit.

[0046] The adaptation script loading and management unit is used to assign a dedicated script execution sandbox to a new device based on its device ID when a new device is connected, without modifying the underlying hardware code. It automatically loads the corresponding lightweight, dynamic adaptation script. The script execution sandbox enables independent execution of adaptation scripts for each device, avoiding script conflicts. It also supports online updates and uninstallation of scripts without requiring a system restart. The adaptation scripts are written in Python / Lua, requiring no compilation and can be directly parsed and executed.

[0047] The raw data dynamic parsing unit is used to perform real-time parsing and preprocessing of raw hardware data uploaded by devices, such as topic data of the ROS1 robot, hexadecimal data of the Modbus sensor, and binary data of the ZigBee temperature and humidity sensor, through the adaptation script. This includes removing redundant data, correcting abnormal data (such as invalid data that exceeds the device's measurement range), and converting data formats (such as converting hexadecimal data to decimal and binary data to character type).

[0048] The unified semantic data conversion unit is used to convert preprocessed device data according to the system's preset minimalist semantic data specification, generating unified semantic data for each device. The semantic data specification is universal across the entire system, with no device / protocol differences. The unified semantic data is represented as: Data Identifier = Value / Status. For example: Robot LiDAR raw data → "Obstacle distance = 1.2m", "Obstacle direction = Directly in front"; Face recognition camera raw data → "Customer identity = VIP", "Customer location = Display area A"; Lighting equipment raw data → "Light brightness = 50%", "Light status = On".

[0049] The global status area writing and updating unit is used to write the unified semantic data of each device into the global status area in the form of key-value pairs. The key is the device ID-data identifier, and the value is the corresponding data / status. At the same time, the global status area is updated according to the device status and task execution feedback collected by the sub-container at a fixed frequency to ensure the real-time performance of the data in the global status area. The device basic information reporting unit is used to upload the device's access status, data collection items, and unified semantic data list to the management and control and digital twin module, providing real-time data support for digital twin modeling and heterogeneous cluster management.

[0050] It should be understood that the dynamic data adaptation module abandons the traditional underlying hardware hard-coding mode. When a new device is connected, only a lightweight dynamic adaptation script needs to be loaded, without modifying the underlying control code. The device connection time is greatly reduced, which greatly shortens the construction and equipment replacement delivery cycle of the robot mobile 4S store and enables flexible customization and expansion. Moreover, the dynamic data adaptation module realizes the unified semantic conversion of all device data. Combined with the 3D visualization scene of the control and digital twin module, operators can monitor the device status and task execution progress in real time, realizing intelligent and visualized control of the showroom.

[0051] 4. Heterogeneous Communication Bus Module: Utilizing a unified data bus with MQTT-SN and TCP / IP dual protocols, this module supports data interaction between modules and directly bridges protocols and transmits data with each device, serving as the sole channel for device access to the system. Specifically, this module can be deployed on the core switch of the mobile 4S store's local area network, equipped with a bus management program and bridging agent plugin, acting as the unified data transmission hub for the system. This module can construct a unified data bus based on MQTT-SN and TCP / IP dual protocols covering the entire mobile robot 4S store, resolving issues related to device interconnection, protocol heterogeneity, and data transmission in cross-network segment and weak network environments (network bandwidth ≤1Mbps, packet loss rate ≤20%).

[0052] Specifically, the heterogeneous communication bus module includes a device access protocol bridging unit, a network segment unification and address allocation unit, a data classification, transmission and routing forwarding unit, and a weak network data fault tolerance processing unit. The heterogeneous communication bus module protocol bridging process is as follows: Figure 3 As shown.

[0053] The device access protocol bridging unit has a built-in multi-protocol bridging proxy plugin. When it receives a device access request containing the device ID and communication protocol type, it automatically identifies the device's communication protocol type, assigns it a dedicated bridging proxy channel, and encapsulates non-standard communication protocols into a unified bus protocol (MQTT-SN, for weak networks) / TCP / IP, for normal networks, enabling bus access for devices with different protocols. The multi-protocol bridging proxy plugin supports mainstream communication protocols for robots / IoT devices, including ROS1 / ROS2, Modbus, TCP / IP, UDP, ZigBee, and Bluetooth 5.0.

[0054] The network segment unification and address allocation unit is used to assign a unified local area network private IP address and bus device identifier to all devices connected to the bus, and to build a bus device address mapping table, which can solve the problem that devices across network segments cannot communicate directly; the bus device identifier consists of network segment code + device ID.

[0055] The data classification, transmission, and routing forwarding unit establishes three independent transmission channels: a downlink command channel, an uplink status channel, and an emergency data channel, and classifies and labels the input data. Downlink command data includes task behavior description files and detailed execution instructions for individual devices, which are accurately forwarded through the downlink command channel based on the address mapping table. Uplink status data includes basic device information, device status, task execution feedback, and global task execution progress, which are forwarded through the uplink status channel. Emergency data includes manual emergency stop / emergency intervention instructions, global emergency stop instructions, and local safety alarm information, which are given the highest transmission priority and transmitted through the emergency data channel block to ensure the real-time nature of emergency information. The weak network data fault tolerance processing unit is used to enable QoS2 level service of MQTT-SN protocol in weak network environment, realize the one-time transmission of data, and perform fragmentation and compression of transmitted data packets to reduce packet loss rate; if data transmission is interrupted, it will automatically resume transmission after the network is restored.

[0056] It should be understood that the heterogeneous communication bus module can support the bridging of multiple communication protocols of heterogeneous devices, solving the problem of interconnection of heterogeneous devices; at the same time, by using the MQTT-SN+TCP / IP dual-protocol unified data bus, reliable data transmission can still be guaranteed in weak network environments, effectively meeting the network requirements of mobile 4S stores for outdoor deployment.

[0057] In summary, the system proposed in this application provides a complete centralized management and control capability for mobile robot stores. Through the management and control and digital twin modules, it achieves global coordination and visualized operation and maintenance of multiple mobile robots. Relying on the general main control container module, it completes intelligent task scheduling, load balancing, multi-device collaborative operation, and multi-level safety protection. With the help of the dynamic data adaptation module, it uniformly processes the heterogeneous hardware data of various mobile robots. Then, through the heterogeneous communication bus module, it realizes efficient interaction between modules and real-time data communication between robot devices. Thus, the entire mobile robot store has the ability to flexibly schedule, operate stably, intelligently collaborate, and operate safely and reliably, comprehensively improving operation and maintenance management and service efficiency.

[0058] In one embodiment, a cloud-edge-device decoupled heterogeneous cluster management and control method is provided. The method is implemented based on the aforementioned cloud-edge-device decoupled heterogeneous cluster management and control system, and includes: S1 utilizes a dynamic data adaptation module deployed at the edge to dynamically allocate dedicated script execution sandboxes based on device IDs without modifying the underlying hardware code. It automatically loads corresponding lightweight dynamic adaptation scripts for device access, data collection and parsing, and unified semantic data conversion. Through a heterogeneous communication bus module, it writes the unified semantic data of each device into the global state area. The dynamic updates of the global state area provide real-time data support for digital twin modeling and heterogeneous cluster management. The devices include heterogeneous robots and IoT devices. S2 utilizes a control and digital twin module deployed on both cloud and edge platforms. Based on the physical showroom data and dynamic data adaptation module of the robot mobile 4S store, and the device basic information uploaded by the module through the heterogeneous communication bus, it generates a 3D visualization scene of the showroom using digital twin modeling. By receiving user input commands and the device status and task execution feedback uploaded in real time by the main container through the heterogeneous communication bus module, it generates a standardized task behavior description file and distributes it to the general main control container module through the heterogeneous communication bus module. S3 utilizes the main container to parse the task behavior description file, executes task concurrency control based on device load prediction and comprehensive task device adaptation, generates a global task execution schedule table, and decomposes it into individual device sub-execution instructions. These instructions are then distributed to the corresponding device sub-containers via the heterogeneous communication bus module for unified management. Simultaneously, it collects device status and task execution feedback uploaded by all sub-containers, writes them into the global status area built into the dynamic data adaptation module, and provides global security fallback. Each sub-container converts the received individual device sub-execution instructions into device-recognizable control instructions, drives the device to execute task actions based on the control instructions, and monitors device status and task execution feedback in real time. These are then uploaded to the main container via the heterogeneous communication bus module while providing local security fallback.

[0059] This cloud-edge-device decoupled heterogeneous cluster management method, in the core workflow of a robot mobile 4S store operation scenario, is divided into two stages: rapid deployment in the showroom and rapid device access, and dynamic business demonstration and multi-device collaborative execution. Each stage includes specific technical implementation steps to ensure the operability of the process, as detailed below: Phase 1: Rapid deployment in the exhibition hall and ultra-fast device integration. Time taken ≤ 5 minutes, no modification to underlying code required.

[0060] Step 1: After the robot mobile 4S store trailer arrives at the designated location, it completes physical deployment to form a fixed layout showroom space. The operations staff enters the 2D floor plan map and physical dimension parameters of the showroom at the edge of the control and digital twin module. Step 2: Turn on all heterogeneous display robots and IoT devices in the store. All devices send network access requests to the heterogeneous communication bus module, including the device ID and its own communication protocol type. Step 3: The heterogeneous communication bus module automatically identifies the protocol type of each device, assigns a bridging proxy channel and a unified LAN IP address to it, encapsulates non-standard protocols into a unified bus protocol, completes the bus access of the devices, generates and updates the bus device address mapping table, and notifies the control and digital twin module of the device access status. Step 4: The control and digital twin module sends the basic configuration information of each device to the dynamic data adaptation module based on the device access status, including device ID, type and raw data collection items; Step 5: The dynamic data adaptation module loads the corresponding lightweight dynamic adaptation scripts provided by the manufacturer / automatically generated by the system for each device, and creates an independent script execution sandbox; Step 6: Each device uploads raw hardware data to the dynamic data adaptation module through the heterogeneous communication bus module. The dynamic data adaptation module parses and performs unified semantic transformation on the raw data through the adaptation script, generates simplified semantic data, and writes it into the global state area. Step 7: The dynamic data adaptation module reports the access completion status of each device and the unified semantic data list to the control and digital twin module. Based on the 2D map of the exhibition hall and the device information, the control and digital twin module automatically generates a 3D digital twin scene of the exhibition hall, completes the coordinate alignment of physical devices and digital models, and enables all devices to be connected at high speed. The system then enters the ready-to-operate state.

[0061] Phase 2: Dynamic business demonstration and multi-device collaborative execution. Real-time response, supporting seamless switching.

[0062] Step 1: Face recognition cameras, infrared sensors, and other devices in the exhibition hall collect scene events and upload the raw data to the dynamic data adaptation module through the heterogeneous communication bus module; Step 2: The dynamic data adaptation module parses and semantically transforms the raw sensor data to generate unified semantic data such as "Customer Identity = VIP" and "Customer Location = Display Area B", which is then updated to the global status area in real time. Step 3: When the sub-containers of the general master control container module (the sub-containers corresponding to the sensors) detect changes in the semantic data of the global state area, they immediately report the scene event to the master container. The master container then reports the event to the control and digital twin module through the heterogeneous communication bus module. Step 4: After receiving the scene event, the control and digital twin module analyzes the current scene requirements through the big model (or retrieves the preset business logic template), and the intelligent orchestration gateway combines the real-time status of all devices (read from the global status area) to generate a standardized JSON format task behavior description file (such as the "VIP customer tour demonstration" task), and distributes it to the main container of the general main control container module through the heterogeneous communication bus module. Step 5: The main container parses the task behavior description file at a frequency of 100ms / time, generates a global task execution schedule table, executes task concurrency control based on device load prediction and comprehensive adaptation of task devices, and then splits the global task into single-device sub-execution instructions, which are then sent to the corresponding device sub-containers through the heterogeneous communication bus module. Step 6: The sub-containers of each device parse and subdivide the execution instructions, convert them into control instructions that can be recognized by the underlying hardware of the device, and drive the device to perform actions, such as the welcoming robot moving forward, the explanation robot starting voice explanation, the display area lights brightening, and the automatic door keeping open. Step 7: Each sub-container collects the real-time operating status and safety status of the device at a frequency of 50ms / time, uploads it to the main container, and writes it to the global status area at the same time; the main container monitors the global task execution progress, and reports the summarized status to the control and digital twin module through the heterogeneous communication bus module to realize real-time status linkage of the 3D digital twin scene. The operators can monitor the entire business demonstration process in the visual interface. Step 8: If it is necessary to switch business logic, the control and digital twin module issues a new task behavior description file. The main container immediately triggers the task interruption mechanism, records the original task execution progress, issues new detailed execution instructions, and each sub-container seamlessly switches to the new task execution, realizing the dynamic reconstruction of business logic. Step 9: If a safety risk is detected during execution (such as the robot detecting an obstacle or equipment failure), the equipment sub-container immediately triggers the local safety fallback mechanism and uploads the alarm information to the main container. The main container triggers a partial / global emergency stop according to the risk level to ensure the safety of the equipment and the exhibition hall, and synchronizes the alarm information to the control and digital twin module for visual alarm.

[0063] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0064] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application.

Claims

1. A cloud-edge-device decoupled heterogeneous cluster management and control system, characterized in that, The system is applied to mobile robot stores and includes: The control and digital twin module, based on a dual-terminal deployment of cloud and edge, is used to generate a 3D visualization scene of the showroom by using digital twin modeling based on the physical showroom data and equipment information of the robot mobile 4S store. It also generates standardized task behavior description files by receiving user input commands and real-time equipment status and task execution feedback. The equipment includes heterogeneous robots and IoT devices. The general-purpose main control container module includes a main container deployed at the edge and sub-containers deployed on the device. The main container is used to parse the task behavior description file, execute task concurrency control based on device load prediction and comprehensive task device adaptation, generate a global task execution scheduling table, and decompose it into individual device sub-execution instructions. These instructions are then distributed to the corresponding sub-containers for unified management and control. At the same time, the main container collects the device status and task execution feedback uploaded by all sub-containers, writes them into the global status area built into the dynamic data adaptation module, and provides global security fallback. Each sub-container is used to convert the received individual device sub-execution instructions into control instructions that the device can recognize. Based on the control instructions, the sub-container drives the device to execute task actions and monitors the device status and task execution feedback in real time, which are then uploaded to the main container while providing local security fallback. The dynamic data adaptation module deployed at the edge is used to dynamically allocate a dedicated script execution sandbox based on the device ID without modifying the underlying hardware code. It automatically loads the corresponding lightweight dynamic adaptation script to perform device access, data collection and parsing, and unified semantic data conversion. It also writes the unified semantic data of each device into the global status area. Based on the dynamic updates of the global status area, it provides real-time data support for digital twin modeling and heterogeneous cluster management. The heterogeneous communication bus module is used to support data interaction between modules using the MQTT-SN+TCP / IP dual-protocol unified data bus, and directly bridges protocols and transmits data with each device. It is the only channel for devices to access the system. The main container includes a file parsing unit, a task concurrency control unit, a detailed instruction issuing unit, a resource conflict detection unit, a task interruption and recovery unit, and a global security fallback unit. The task concurrency control unit includes a device load prediction mechanism, a task device comprehensive adaptation calculation mechanism, and an optimal execution device selection mechanism; The device load prediction mechanism first parses and obtains the current load data from the device status uploaded by the sub-containers of all schedulable devices through the heterogeneous communication bus module. Simultaneously, it calls the local historical load database to extract the historical average load and historical load fluctuation data of each schedulable device under similar task scenarios. Then, based on the current load data, historical average load, and historical load fluctuation data of each schedulable device, it estimates the predicted load value of each schedulable device, expressed as: ； in, To predict load values; The current load data is as follows: for heterogeneous robots, the current load data is obtained by collecting CPU utilization, memory usage, queue length of pending tasks, and remaining time of currently executing tasks, and then normalizing and weighting the data; for IoT devices, the current load data is obtained by collecting port utilization, command response latency, and online status percentage, and then normalizing and weighting the data. This represents the historical average load across similar task scenarios. The standard deviation of historical load fluctuation data under similar task scenarios. This is the load deviation correction factor. This is the load fluctuation impact coefficient; Then, based on the predicted load value, the current load redundancy calculation result of each schedulable device is corrected. At the same time, the basic compatibility between the device and the task is combined, and devices with load redundancy correction results lower than the preset threshold or basic compatibility are eliminated to obtain candidate devices for each task. The task device comprehensive adaptation calculation mechanism is used to first assign basic scheduling weights to different types of tasks, and at the same time introduce task correlation factors, and dynamically adjust the basic scheduling weights of each task by combining task priority, urgency of task execution time limit and degree of loss of task execution failure, so as to obtain the optimized scheduling weights of each task; and combine the optimized scheduling weights of each task. Result of load redundancy correction for corresponding candidate devices and basic device task adaptability The overall fit between each task and its corresponding candidate device is calculated and expressed as: ； in, , and These are the weighting coefficients for the three linear terms; This is used to quantify the interaction between the three linear terms, avoiding the dominance of a single factor in the overall fit result; The optimal execution device screening mechanism is used to sort all candidate devices for each task in descending order of comprehensive suitability, prioritize the allocation of the device with the highest comprehensive suitability as the optimal execution device for each task, and generate a global task execution scheduling table.

2. The cloud-edge-device decoupled heterogeneous cluster management and control system according to claim 1, characterized in that, The control and digital twin module includes a digital twin modeling unit, an instruction parsing and semantic conversion unit, and a task generation unit; The digital twin modeling unit is used to generate a 3D visualization scene of the exhibition hall based on the basic data of the 2D physical showroom of the robot mobile 4S store and the dynamic data adaptation module uploaded by the heterogeneous communication bus module. It binds the ID of each device in the robot mobile 4S store with the 3D digital twin model, realizes the coordinate alignment and status linkage between the physical device and the digital twin model, and supports the visualization display of the real-time position and operating parameters of the device in the 3D visualization scene. The instruction parsing and semantic conversion unit is used to perform semantic parsing on natural language instructions input by users using a large language model to extract core business requirements; for business semantic instructions triggered by users through preset buttons, it calls the business logic template library for matching and filtering the corresponding business logic templates; for scene events triggered by sensors in the exhibition hall, it uses semantic conversion to generate core business requirements. The task generation unit is used to generate a standardized task behavior description file based on core business requirements or business logic templates, combined with the device status and task execution feedback uploaded in real time by the main container through the heterogeneous communication bus module, and then distribute it to the main container through the heterogeneous communication bus module. The task behavior description file includes task ID, task type, task priority, task execution time limit, schedulable devices corresponding to each task, task triggering conditions, task termination conditions, and linkage device association rules.

3. The cloud-edge-device decoupled heterogeneous cluster management and control system according to claim 1, characterized in that, The file parsing unit is used to receive the task behavior description file issued by the control and digital twin module through the heterogeneous communication bus module, and parse it to obtain the task ID, task type, task priority, task execution time limit, schedulable devices corresponding to each task, task triggering conditions, task termination conditions, and linkage device association rules. The task concurrency control unit is used to filter the optimal execution device for each task and generate a global task execution schedule table based on the parsing results of the task behavior description file, through device load prediction and comprehensive task device adaptation calculation, to clarify the execution order, optimal execution device, execution start time and execution end time of each task. The subdivided instruction issuing unit is used to split the global task execution scheduling table into single-device subdivided execution instructions according to the device ID, and send them to the sub-containers of the corresponding devices for unified management; the single-device subdivided execution instructions include standardized execution actions, action parameters, execution duration, and linkage triggering conditions; The resource conflict detection unit is used to detect resource conflicts when multiple devices execute tasks concurrently. If a conflict is detected, the execution order of each task in the global task execution schedule table is automatically adjusted, and a conflict-free task execution window is allocated to each sub-container. The task interruption and recovery unit is used to immediately trigger the task interruption mechanism when the file parsing unit receives a new task behavior description file or receives a manual emergency stop command / emergency intervention command issued by the user through the control and digital twin module. It records the current task execution progress and device status, parses the new task behavior description file and issues a new single-device subdivision execution command. If it is necessary to recover the original task, it continues to issue single-device subdivision execution commands from the breakpoint according to the recorded execution progress, so as to achieve seamless switching. The global security fallback unit is used to collect the device status and task execution feedback uploaded by all sub-containers through the heterogeneous communication bus module in real time, write them into the global status area in the dynamic data adaptation module, and upload the global task execution progress to the control and digital twin module. At the same time, by reading the unified semantic data, device status and task execution feedback of each device in the global status area, it performs global security risk detection. If the global security risk exceeds the preset threshold, it immediately issues a global emergency stop command to the sub-container to suspend the task execution of all devices.

4. The cloud-edge-device decoupled heterogeneous cluster management and control system according to claim 1, characterized in that, The sub-container includes an instruction parsing and hardware adaptation unit, a real-time task execution and status reporting unit, a local security fallback unit, and a linkage triggering execution unit. The instruction parsing and hardware adaptation unit is used to parse the single-device subdivision execution instructions issued by the main container through the heterogeneous communication bus module, and convert the standardized execution actions in the single-device subdivision execution instructions into control instructions that can be recognized by the underlying hardware of the device. The real-time task execution and status reporting unit is used to drive the device to perform task actions based on control commands, collect device status and task execution feedback at a fixed frequency, and upload them to the main container through the heterogeneous communication bus module, and the main container writes them into the global status area in the dynamic data adaptation module. The local security fallback unit is used to detect local security risks based on the data collected in real time by the real-time task execution and status reporting unit. If the local security risk exceeds the preset threshold, the local emergency stop / obstacle avoidance mechanism is triggered immediately without waiting for the main container to issue the instruction. At the same time, the local security alarm information is uploaded to the main container and the control and digital twin module through the emergency data channel of the heterogeneous communication bus module. The linkage trigger execution unit is used to monitor the unified semantic data of each device in the global status area in real time. When it detects that the unified semantic data meets the linkage triggering conditions in the single device subdivision execution instruction, it immediately executes the linkage action according to the linkage device association rules in the task behavior description file, and reports the linkage execution status to the main container, and then writes it into the global status area.

5. A cloud-edge-device decoupled heterogeneous cluster management and control system according to claim 1, characterized in that, The dynamic data adaptation module includes an adaptation script loading and management unit, a raw data dynamic parsing unit, a unified semantic data conversion unit, a global status area writing and updating unit, and a device basic information reporting unit. The adaptation script loading and management unit is used to allocate a dedicated script execution sandbox to a new device based on its device ID without modifying the underlying hardware code when the new device is connected. It automatically loads the corresponding lightweight dynamic adaptation script. The script execution sandbox enables the independent execution of the adaptation scripts of each device, avoids script conflicts, and supports online updates and uninstallation of scripts without restarting the system. The raw data dynamic parsing unit is used to perform real-time parsing and preprocessing of the raw hardware data uploaded by the device through an adaptation script, including removing redundant data, correcting abnormal data, and converting data formats. The unified semantic data conversion unit is used to convert the preprocessed device data according to the system's preset minimalist semantic data specification to generate unified semantic data for each device. The semantic data specification is universal across the entire system and has no device / protocol differences. The unified semantic data is represented as: Data identifier = value / status. The global status area writing and updating unit is used to write the unified semantic data of each device into the global status area in the form of key-value pairs. The key is the device ID-data identifier and the value is the corresponding data / status. At the same time, the global status area is updated according to the device status and task execution feedback collected by the sub-container at a fixed frequency to ensure the real-time performance of the data in the global status area. The device basic information reporting unit is used to upload the device's access status, data collection items, and unified semantic data list to the management and control and digital twin module, providing real-time data support for digital twin modeling and heterogeneous cluster management.

6. A cloud-edge-device decoupled heterogeneous cluster management and control system according to claim 1, characterized in that, The heterogeneous communication bus module includes a device access protocol bridging unit, a network segment unification and address allocation unit, a data classification transmission and routing forwarding unit, and a weak network data fault tolerance processing unit. The device access protocol bridging unit has a built-in multi-protocol bridging proxy plugin, which is used to automatically identify the communication protocol type of the device when receiving a device access request, allocate a dedicated bridging proxy channel for it, and encapsulate non-standard communication protocols into a bus-unified MQTT-SN protocol / TCP / IP protocol to realize bus access for cross-protocol devices. The network segment unification and address allocation unit is used to allocate a unified local area network private IP address and bus device identifier to all devices connected to the bus, and to build a bus device address mapping table; the bus device identifier consists of a network segment code and a device ID; The data classification, transmission, and routing forwarding unit is used to establish three independent transmission channels: a downlink command channel, an uplink status channel, and an emergency data channel, and to classify and label the input data. Downlink command data includes task behavior description files and detailed execution instructions for individual devices, which are accurately forwarded through the downlink command channel based on the address mapping table. Uplink status data includes basic device information, device status, task execution feedback, and global task execution progress, which are forwarded through the uplink status channel. Emergency data includes manual emergency stop / emergency intervention instructions, global emergency stop instructions, and local security alarm information, which are given the highest transmission priority and transmitted through the emergency data channel block to ensure the real-time nature of emergency information. The weak network data fault tolerance processing unit is used to enable QoS2 level service of MQTT-SN protocol in weak network environment to realize one-time transmission of data, and to fragment and compress the transmitted data packets to reduce packet loss rate; if data transmission is interrupted, it will automatically resume transmission after the network is restored.

7. A cloud-edge-device decoupled heterogeneous cluster management method, characterized in that, The method is implemented based on a cloud-edge-device decoupled heterogeneous cluster management system as described in any one of claims 1-6, including: S1 utilizes a dynamic data adaptation module deployed at the edge to dynamically allocate dedicated script execution sandboxes based on device IDs without modifying the underlying hardware code. It automatically loads corresponding lightweight dynamic adaptation scripts for device access, data collection and parsing, and unified semantic data conversion. Through a heterogeneous communication bus module, it writes the unified semantic data of each device into the global state area. The dynamic updates of the global state area provide real-time data support for digital twin modeling and heterogeneous cluster management. The devices include heterogeneous robots and IoT devices. S2 utilizes a control and digital twin module deployed on both cloud and edge platforms. Based on the physical showroom data and dynamic data adaptation module of the robot mobile 4S store, and the device basic information uploaded by the module through the heterogeneous communication bus, it generates a 3D visualization scene of the showroom using digital twin modeling. By receiving user input commands and the device status and task execution feedback uploaded in real time by the main container through the heterogeneous communication bus module, it generates a standardized task behavior description file and distributes it to the general main control container module through the heterogeneous communication bus module. S3 utilizes the main container to parse the task behavior description file, executes task concurrency control based on device load prediction and comprehensive task device adaptation, generates a global task execution schedule table, and decomposes it into individual device sub-execution instructions. These instructions are then distributed to the corresponding device sub-containers via the heterogeneous communication bus module for unified management. Simultaneously, it collects device status and task execution feedback uploaded by all sub-containers, writes them into the global status area built into the dynamic data adaptation module, and provides global security fallback. Each sub-container converts the received individual device sub-execution instructions into device-recognizable control instructions, drives the device to execute task actions based on the control instructions, and monitors device status and task execution feedback in real time. These are then uploaded to the main container via the heterogeneous communication bus module while providing local security fallback.

Images