A robot multi-modal resource intelligent scheduling method and system

Abstract

The application discloses a kind of robot multimodal resource intelligent scheduling method and system, method includes: according to the multimodal data collected by robot, constructs the request message carrying robot unique identification and current analysis task type;According to the analysis result data generated after default edge analysis of request message by leaf gateway, the device identification of network node is attached to the analysis result data and then continues to be forwarded to back-end server;According to the analysis result data received by back-end server carrying the network node identification along the way, the root routing device corresponding to each robot and the number distribution of different analysis task types under each root routing device are extracted and statistically analyzed;According to the statistical result of task number distribution, the execution position of specific analysis task type under the root routing device is adjusted. By using the embodiment of the application, the execution position of the computing task can be adaptively adjusted according to the distribution, effectively balancing the computing load of the center and the edge, reducing the network transmission pressure and processing delay.

Description

Technical Field

[0001] This invention belongs to the field of intelligent robot technology, and in particular to a method and system for intelligent scheduling of multimodal resources of robots. Background Technology

[0002] With the widespread application of robotics, the real-time processing of multimodal data (such as vision, laser, and inertial measurement unit data) poses a severe challenge to network architecture and computing resources. While traditional cloud computing possesses powerful computing capabilities, transmitting all data back to a central server for processing easily leads to network bandwidth congestion and increased processing latency, making it difficult to meet the low latency and high reliability requirements of collaborative robot operations. To alleviate the pressure on the central server, edge computing has been introduced, offloading some computing tasks to the network edge closer to the data source. However, existing solutions mostly employ static task allocation strategies, failing to flexibly adjust the execution location of computing tasks based on dynamic changes in network topology and the task load in different areas. When tasks surge in a certain area, fixed edge nodes may become processing bottlenecks; conversely, if edge nodes are idle, the central server still needs to process a large amount of data, resulting in resource waste. Therefore, there is an urgent need for a resource scheduling method that can sense network topology, statistically analyze task distribution, and dynamically adjust computing locations to achieve global optimization of network and computing resources. Summary of the Invention

[0003] The purpose of this invention is to provide a robot multimodal resource intelligent scheduling method and system to overcome the shortcomings of the prior art. It can adaptively adjust the execution position of computing tasks according to the distribution, effectively balance the computing load between the center and the edge, and reduce network transmission pressure and processing latency.

[0004] One embodiment of this application provides a robot multimodal resource intelligent scheduling method, the method comprising: Multimodal data transmission and identification carrying: Based on the multimodal data collected by the robot, construct a request message carrying the robot's unique identifier and the current analysis task type, and send it to the leaf gateway it is connected to; Network node awareness and topology information appending: Based on the analysis result data generated after the request message is analyzed by the default edge analysis of the leaf gateway, each network node along the way extracts the robot's unique identifier and analysis task type during the forwarding process, records the mapping relationship between the robot's unique identifier and the forwarding interface, and appends the device identifier of the network node to the analysis result data before forwarding it to the backend server. Task distribution statistics and root route identification: Based on the analysis results data carrying the network node identifiers along the route received by the backend server, extract and statistically analyze the root route devices corresponding to each robot and the number distribution of different analysis task types under each root route device; Dynamic scheduling decision and instruction issuance: Based on the statistical results of the task quantity distribution, the backend server determines whether the conditions for centralized scheduling or edge decentralization are met, and issues scheduling instructions to the corresponding root routing device to adjust the execution position of a specific analysis task type under that root routing device.

[0005] Optionally, the multimodal data transmission and identifier carrying includes: Multimodal data acquisition: Collect multimodal data such as visual, environmental, or traffic flow data based on the robot's operating environment to determine the types of data that need to be intelligently analyzed; Message Construction: Based on the data type and the robot's own stored unique identifier, a request message is constructed. The message includes at least the robot's unique identifier, the analysis task type identifier, and the data payload to be analyzed. Message transmission: Based on the wireless network currently connected to the robot, the request message is sent to the directly connected leaf gateway.

[0006] Optionally, the network node awareness and topology information appending includes: Default edge analysis: Based on the request message received by the leaf gateway, the analysis task is assigned to the directly connected edge analysis device for execution. After the edge analysis device completes the analysis, it generates analysis result data. The analysis result data includes at least the robot's unique identifier, the leaf gateway's own identifier, the analysis task type, and the analysis result. Mapping relationship record: Based on the analysis result data received by the upstream routing device, extract the robot's unique identifier, leaf gateway identifier, and analysis task type, and record the mapping relationship between the robot's unique identifier, leaf gateway identifier, analysis task type, and the ingress interface, egress interface, and its own device identifier locally; Device Identifier Attachment: The upstream routing device attaches its own device identifier to the extended field of the analysis result data, and the attached data continues to be forwarded to the upper layer. Each subsequent network node returns the execution step mapping relationship record and the attached device identifier until the data reaches the backend server.

[0007] Optionally, the task distribution statistics and root route identification include: Information extraction and aggregation: Based on the final analysis results data received from the backend server, extract the robot's unique identifier, leaf gateway identifier, device identifiers of each network node along the way, and analysis task type carried in the data; Root routing device identification: Based on the leaf gateway identifier and the network node device identifier along the way, determine the common upstream device for the convergence of each robot's data flow and define it as the root routing device of the robot; Task distribution statistics: Based on the root routing device and analysis task type of each robot, count the number of robots with different analysis task types under each root routing device to form a task distribution statistics table.

[0008] Optionally, the dynamic scheduling decision and instruction issuance include: Centralized scheduling condition judgment: According to the task distribution statistics table, when the number of robots of a certain analysis task type under a certain root routing device reaches or exceeds the preset centralized scheduling threshold, the backend server generates a centralized scheduling instruction. The instruction includes the root routing device identifier, analysis task type, centralized scheduling mode, and backend analysis server address. Centralized scheduling and execution: Based on the centralized scheduling instruction received by the root routing device, a local task scheduling policy table is established or updated, and the forwarding target of the corresponding analysis task type is pointed to the backend analysis server. When the same type of analysis task is received in the future, it is directly forwarded to the backend analysis server for centralized processing. Edge deployment condition judgment: According to the task distribution statistics table, when the number of robots of a certain analysis task type under a certain root routing device drops to 1, the backend server generates an edge deployment scheduling instruction. The instruction includes the root routing device identifier, analysis task type, scheduling mode as edge, and optional backend analysis server address. Edge-deployed execution: Based on the edge-deployed scheduling instruction received by the root routing device, the local task scheduling policy table is updated, and the processing method for the corresponding analysis task type is changed to local edge execution. When such analysis tasks are received in the future, they will no longer be forwarded to the backend analysis server.

[0009] Optionally, the method further includes a robot movement state processing mechanism, specifically including: State initialization and detection: The robot's initial state is assumed to be unstable, and the system continuously monitors whether the leaf gateway it connects to changes. State switching determination: If the leaf gateway connected to the robot does not change for more than a preset time, its state is switched to stable; if a change in the gateway is detected in the stable state, it is immediately switched back to unstable. State-based scheduling strategy: If the robot is in a stable state, the backend server is allowed to make scheduling decisions based on the task distribution statistics and issue scheduling instructions to the corresponding root routing device to adjust the execution position of the analysis task; if the robot is in an unstable state, the robot's analysis task is forced to be fixed to the leaf gateway it is currently connected to and is not involved in the centralized scheduling of the backend server. Status information transmission: When a robot sends a request message, it carries its current status identifier, which is used by the backend server to distinguish between stable and unstable robots when calculating task distribution, ensuring that scheduling decisions are based on robots in a stable state.

[0010] Another embodiment of this application provides a robot multimodal resource intelligent scheduling system, the system comprising: The sending module is used for multimodal data transmission and identification carrying: based on the multimodal data collected by the robot, it constructs a request message carrying the robot's unique identifier and the current analysis task type, and sends it to the leaf gateway it is connected to. The perception module is used for network node perception and topology information appending: based on the analysis result data generated after the request message is analyzed by the default edge analysis of the leaf gateway, each network node along the way extracts the robot's unique identifier and analysis task type during the forwarding process, records the mapping relationship between the robot's unique identifier and the forwarding interface, and appends the device identifier of the network node to the analysis result data before forwarding it to the backend server. The statistics module is used for task distribution statistics and root route identification: based on the analysis result data carrying the network node identifiers along the route received by the backend server, it extracts and statistically analyzes the root route devices corresponding to each robot and the number distribution of different analysis task types under each root route device; The decision module is used for dynamic scheduling decisions and instruction issuance: based on the statistical results of the task quantity distribution, the backend server determines whether the conditions for centralized scheduling or edge decentralization are met, and issues scheduling instructions to the corresponding root routing device to adjust the execution position of a specific analysis task type under that root routing device.

[0011] Another embodiment of this application provides a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any of the preceding claims when running.

[0012] Another embodiment of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any of the preceding claims.

[0013] Compared with existing technologies, the intelligent scheduling method for multimodal resources of robots provided by this invention can adaptively adjust the execution position of computing tasks according to the distribution, effectively balance the computing load between the center and the edge, and reduce network transmission pressure and processing latency. Attached Figure Description

[0014] Figure 1 A hardware structure block diagram of a computer terminal for a robot multimodal resource intelligent scheduling method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating a robot multimodal resource intelligent scheduling method provided in an embodiment of the present invention; Figure 3This is a network diagram of a robot multimodal resource intelligent scheduling system provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a robot multimodal resource intelligent scheduling system provided in an embodiment of the present invention. Detailed Implementation

[0015] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0016] With the widespread deployment of robotic systems in smart manufacturing, smart parks, unmanned delivery, and smart warehousing, collaborative operations among multiple robots have become commonplace. In actual operation, robots need to analyze and process the collected multimodal data (such as visual data, environmental sensor data, and voice data). In existing technologies, multimodal data analysis typically employs the following methods: All files are uploaded to the central server for processing. Processing is performed by fixed deployment on edge nodes; Tasks are distributed using a simple load balancing strategy.

[0017] However, the above method has the following problems: 1. Unable to perceive the distribution of robots within the network topology; 2. It is impossible to determine the degree of clustering of similar analytical tasks within the network; 3. It cannot dynamically adjust the execution location at the edge or center based on task distribution; 4. The robot's scheduling changes frequently when it is in motion, resulting in poor stability; 5. Computing resources cannot be optimized based on the network structure.

[0018] Therefore, there is an urgent need for an intelligent scheduling method that combines the robot's unique identifier with the network node's perception capabilities, enabling the backend server to: Robot topological position; Analyze the distribution of task types; Root routing structure relationship; The robot is in a stable moving state; This enables intelligent centralized or decentralized scheduling of similar tasks.

[0019] This invention first provides a robot multimodal resource intelligent scheduling method, which can be applied to electronic devices, such as computer terminals, specifically ordinary computers.

[0020] The following detailed explanation uses a computer terminal as an example. Figure 1 This is a hardware structure block diagram of a computer terminal for a robot multimodal resource intelligent scheduling method provided in an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0021] See Figure 2 The present invention provides a robot multimodal resource intelligent scheduling method, which may include the following steps: S201, Multimodal Data Transmission and Identification Carrying: Based on the multimodal data collected by the robot, a request message carrying the robot's unique identifier and the current analysis task type is constructed and sent to the leaf gateway it accesses; specifically, the multimodal data transmission and identification carrying includes: Multimodal data acquisition: Collect multimodal data such as visual, environmental, or traffic flow data based on the robot's operating environment to determine the types of data that need to be intelligently analyzed; Message Construction: Based on the data type and the robot's own stored unique identifier, a request message is constructed. The message includes at least the robot's unique identifier, the analysis task type identifier, and the data payload to be analyzed. Message transmission: Based on the wireless network currently connected to the robot, the request message is sent to the directly connected leaf gateway.

[0022] S202, Network Node Awareness and Topology Information Addition: Based on the analysis result data generated after the request message undergoes default edge analysis by the leaf gateway, each network node along the route extracts the robot's unique identifier and analysis task type during forwarding, records the mapping relationship between the robot's unique identifier and the forwarding interface, and adds the network node's device identifier to the analysis result data before forwarding it to the backend server; specifically, the network node awareness and topology information addition includes: Default edge analysis: Based on the request message received by the leaf gateway, the analysis task is assigned to the directly connected edge analysis device for execution. After the edge analysis device completes the analysis, it generates analysis result data. The analysis result data includes at least the robot's unique identifier, the leaf gateway's own identifier, the analysis task type, and the analysis result. Mapping relationship record: Based on the analysis result data received by the upstream routing device, extract the robot's unique identifier, leaf gateway identifier, and analysis task type, and record the mapping relationship between the robot's unique identifier, leaf gateway identifier, analysis task type, and the ingress interface, egress interface, and its own device identifier locally; Device Identifier Attachment: The upstream routing device attaches its own device identifier to the extended field of the analysis result data, and the attached data continues to be forwarded to the upper layer. Each subsequent network node returns the execution step mapping relationship record and the attached device identifier until the data reaches the backend server.

[0023] S203, Task Distribution Statistics and Root Route Identification: Based on the analysis result data carrying network node identifiers along the route received by the backend server, extract and statistically analyze the root route devices corresponding to each robot and the quantity distribution of different analysis task types under each root route device; specifically, the task distribution statistics and root route identification include: Information extraction and aggregation: Based on the final analysis results data received from the backend server, extract the robot's unique identifier, leaf gateway identifier, device identifiers of each network node along the way, and analysis task type carried in the data; Root routing device identification: Based on the leaf gateway identifier and the network node device identifier along the way, determine the common upstream device for the convergence of each robot's data flow and define it as the root routing device of the robot; Task distribution statistics: Based on the root routing device and analysis task type of each robot, count the number of robots with different analysis task types under each root routing device to form a task distribution statistics table.

[0024] S204, Dynamic Scheduling Decision and Instruction Issuance: Based on the statistical results of the task quantity distribution, the backend server determines whether the conditions for centralized scheduling or edge deployment are met, and issues scheduling instructions to the corresponding root routing device to adjust the execution position of a specific analysis task type under that root routing device. Specifically, the dynamic scheduling decision and instruction issuance includes: Centralized scheduling condition judgment: According to the task distribution statistics table, when the number of robots of a certain analysis task type under a certain root routing device reaches or exceeds the preset centralized scheduling threshold, the backend server generates a centralized scheduling instruction. The instruction includes the root routing device identifier, analysis task type, centralized scheduling mode, and backend analysis server address. Centralized scheduling and execution: Based on the centralized scheduling instruction received by the root routing device, a local task scheduling policy table is established or updated, and the forwarding target of the corresponding analysis task type is pointed to the backend analysis server. When the same type of analysis task is received in the future, it is directly forwarded to the backend analysis server for centralized processing. Edge deployment condition judgment: According to the task distribution statistics table, when the number of robots of a certain analysis task type under a certain root routing device drops to 1, the backend server generates an edge deployment scheduling instruction. The instruction includes the root routing device identifier, analysis task type, scheduling mode as edge, and optional backend analysis server address. Edge-deployed execution: Based on the edge-deployed scheduling instruction received by the root routing device, the local task scheduling policy table is updated, and the processing method for the corresponding analysis task type is changed to local edge execution. When such analysis tasks are received in the future, they will no longer be forwarded to the backend analysis server.

[0025] Furthermore, the method also includes a robot movement state processing mechanism, specifically including: State initialization and detection: The robot's initial state is assumed to be unstable, and the system continuously monitors whether the leaf gateway it connects to changes. State switching determination: If the leaf gateway connected to the robot does not change for more than a preset time, its state is switched to stable; if a change in the gateway is detected in the stable state, it is immediately switched back to unstable. State-based scheduling strategy: If the robot is in a stable state, the backend server is allowed to make scheduling decisions based on the task distribution statistics and issue scheduling instructions to the corresponding root routing device to adjust the execution position of the analysis task; if the robot is in an unstable state, the robot's analysis task is forced to be fixed to the leaf gateway it is currently connected to and is not involved in the centralized scheduling of the backend server. Status information transmission: When a robot sends a request message, it carries its current status identifier, which is used by the backend server to distinguish between stable and unstable robots when calculating task distribution, ensuring that scheduling decisions are based on robots in a stable state.

[0026] In light of the above context, this invention aims to provide an intelligent scheduling method and system based on the combination of robot unique identifier (Robot ID) and network node perception. By using network devices to perceive and record the multimodal task types of robots, the backend server can statistically analyze the task distribution under the "root route," thereby achieving dynamic intelligent scheduling in a multi-robot environment. Furthermore, by combining the robot's stable movement state, it enables adaptive switching between centralized and edge analysis.

[0027] Technical solution: 1. Core idea: This technical solution provides an intelligent scheduling method and system based on the combination of robot unique identifier and network node perception. By sensing and reporting the robot's multimodal data stream through network nodes, the backend server can grasp the topological relationship between the robot and network nodes as well as the distribution of current analysis tasks, thereby realizing intelligent and flexible task scheduling.

[0028] To implement this solution, the robot must first include its unique identifier (Robot ID) in the message header when transmitting multimodal data to the backend. This unique identifier is then recorded by its directly connected switches and upper-layer routers as the message travels through the network. These network nodes extract and record the robot's unique identifier from the message through listening. Simultaneously, network devices also obtain the type of multimodal data analysis task from the message (e.g., network traffic data analysis, visual task analysis, etc.; different types of analysis tasks require different analysis algorithms). The network devices then perform overall classification and coordination based on the different robot unique identifiers and the types of analysis tasks they perform. The specific steps are as follows: Step 1: The robot sends the multimodal data it has collected to the directly connected leaf gateway it has connected to. The message must carry at least the following: the robot's unique identifier and the identifier of the current analysis task type.

[0029] Step 2: The leaf gateway assigns the task to the edge analysis device directly connected to it by default. The analysis results of the analysis device are then transmitted upward by the leaf gateway - the analysis results carry the robot ID.

[0030] Step 3: During the forwarding of the analysis result data, the upstream routing device extracts the robot's unique identifier and the analysis task type, then records the mapping relationship between the robot's unique identifier and the corresponding outgoing interface, and then appends the mapping relationship information to the analysis result data and forwards it to the backend server. In this way, the analysis result data contains the outgoing interface information of all network nodes along the way, thus understanding the transmission path and the specific location of the analyst.

[0031] In this way, the backend server can perceive the robot's position in the network topology and its analysis task type. When multiple robots are performing the same analysis task in the network, the backend server aggregates the network topology positions and analysis task types of all robots using the above method. This provides the foundation for the backend server to analyze and schedule tasks, and the backend server then performs analysis and scheduling according to the following steps: Step 4: The backend server statistically analyzes the distribution of the robot's "root routing device" (here, the root routing device is defined as the routing device through which multiple robots ultimately exchange data upstream) and task types. The backend server first determines the root routing device corresponding to each robot, then statistically analyzes the distribution of different types of analysis tasks across these root routing devices to determine whether multiple robots are executing the same type of analysis task under a given routing device.

[0032] Step 5: When the backend server determines that multiple robots under a certain root routing device have the same type of analysis task, the server issues a scheduling instruction to the corresponding root routing device to uniformly schedule the same type of analysis task to the backend analysis server for unified analysis, and at the same time notifies the relevant devices to stop the analysis task. (For simplification, it is assumed that the root routing device is a converged routing device with artificial intelligence analysis capabilities, but it can also be an intelligent analysis server directly connected to the root routing device.)

[0033] At this point, the backend server has achieved intelligent scheduling and centralized analysis of similar tasks based on the analysis task type and the "traffic" analysis of the routing devices. However, since the robots are mobile and their analysis tasks are not fixed and need to change frequently with environmental changes, the scheduling of the above centralized analysis and processing also needs to be flexible. When the backend server determines that a certain analysis task under a certain root routing device corresponds to only one robot, it also needs to reassign the task to the root routing device for edge analysis, as follows: Step 6: When the backend server determines that a certain analysis task type on the root routing device corresponds to only a single robot, it sends a scheduling instruction to the leaf gateway to which the robot is connected, based on the robot's unique identifier, and assigns the analysis task to the leaf gateway for execution.

[0034] Considering that the robot is in a mobile state, the leaf gateway it connects to changes frequently when the robot moves. Therefore, a state determination mechanism is introduced here. When the system determines that the robot's state is stable, the gateway connected to the robot is fixed, and the process of steps one to six described above is followed. When the system determines that the robot's state is unstable, the gateway connected to the robot is constantly changing. In this case, the robot's analysis task is fixed at its leaf gateway (which switches naturally as the robot moves) and executed until it reaches a stable state, at which point the above process is applied again.

[0035] Here, the robot's state can be defaulted to an unstable state (state value set to 0). When the system detects that the gateway it is connected to has not changed for more than 5 minutes (this is an example time, which can be set by the system), it switches to a stable state (state value set to 1). When a change in the gateway is detected, it immediately switches back to 0. Therefore, the above steps one to six are all for processing when the state value is 1 (i.e., stable state). That is, the robot's own state value must also be included in the messages it sends to the network, and the backend server also needs to filter the state value when counting task types.

[0036] Thus, a closed-loop process is achieved for the intelligent scheduling and management method of intelligent analysis tasks for robot multimodal data in a multi-robot network environment. By judging the robot's movement status, based on the robot's unique identifier, and recording relevant information of each key forwarding device in the network, the backend server can statistically analyze the load status of different analysis tasks on the "root routing device" and intelligently schedule similar tasks from the edge to the center. It also supports dynamic scheduling adjustments based on the number of tasks and the robot's status. This realizes an intelligent scheduling method and system based on the combination of robot unique identifier and network node perception. This method can effectively improve the scheduling flexibility and scalability of intelligent analysis of robot multimodal data and improve the overall analysis efficiency of the system.

[0037] 2. The complete technical implementation process is as follows: (1) Pre-set conditions: The system has established a Wi-Fi network, achieving full-area networking through core switches, aggregation switches, access switches, and wireless access points (APs). This includes multiple robot terminals, all using anycast networking (i.e., identical IP addresses). Each robot has a unique coded ID (Robot_1, Robot_2…Robot_N). Robots connect to the Wi-Fi network and participate in network communication through their respective APs. Intelligent analysis services are integrated into the switch devices or participate in network communication through individually assigned IP addresses. A schematic diagram of the entire network topology is shown below. Figure 3 As shown.

[0038] (2) Detailed implementation process: Step 1: The robot sends multimodal data; When a robot collects multimodal data and needs to send it to the backend analysis service for intelligent analysis, the robot constructs a request message and sends it to the leaf gateway it connects to. The message must contain at least the robot's unique ID, data type (analysis type), and data content. An example message format is as follows: {Robot_ID: RID_0001, Task_Type: VISION_ANALYSIS, Payload: video frame data}. Here, VISION_ANALYSIS is used as an example to represent visual data that needs to be analyzed visually. Similarly, this data could also be FLOW_ANALYSIS to represent network traffic analysis, or ENV_ANALYSIS to represent environmental perception analysis, etc.

[0039] Step 2: Leaf Gateway Default Edge Analysis; After receiving the above data, the leaf gateway performs analysis on its own edge by default. It first parses the Robot_ID and Task_Type, assigns the task to the directly connected edge analysis device, and the edge device executes the analysis task and sends the analysis results to the leaf gateway for uploading to the upper-layer network. Here, the message content of the analysis result must at least include the robot's unique ID, its own gateway ID, data type (analysis type), and the analysis result data content. An example message format is as follows: {Robot_ID: RID_0001, Gateway_ID: GW_0001, Task_Type: VISION_ANALYSIS, DataResult: Analysis result data}. This data is then forwarded to the upper layer.

[0040] Step 3: Upper-layer routing devices record topology mappings; When the upper-layer routing device receives the above result data, it extracts the corresponding fields from the data and records the mapping relationship table. The table records the mapping relationship between the received Robot_ID, Gateway_ID, and Task_Type and the corresponding ingress interface, egress interface, and its own routing node's device ID. An example of this table entry is shown in Table 1 below: Table 1 ; Then, the Router_ID information is appended to the packet extension field and forwarded upwards. Subsequent routing devices repeat this step until the data is forwarded to the backend storage server.

[0041] Step 4: The backend server analyzes the root routing devices and task distribution; After receiving the data, the backend storage server, in addition to storing and responding to the analysis results normally (the normal storage and response process is irrelevant to this invention and will not be described further; the specific method can be found in the method described in the second application of this patent on the same application date), needs to perform additional task statistical analysis. Based on the information carried in the message, such as Robot_ID, Gateway_ID, Task_Type, and Router_ID, it can statistically determine the root routing device for each robot and the distribution of different task types under each root routing device. (Here, a root routing device is defined as: if multiple robots complete the analysis task on the same leaf gateway device, then that device is the root routing device for that group of robots). The server's statistical examples are shown in Table 2 below. Table 2 ; Step 5: Centralized dispatch mechanism; According to the above statistics, when the number of Task_Types under a certain root router device is greater than or equal to a preset threshold (3 in this example), the server sends a task scheduling instruction to that root router device to centrally schedule the intelligent analysis tasks under that root router device to the backend analysis server for unified processing. This task scheduling instruction includes information such as the root router device ID, task type, scheduling mode, and backend analysis server address, as shown in the example below: {Root_Router_ID: GW_0001, Task_Type: VISION_ANALYSIS, Dispatch_Mode: CENTRAL, Analysis_Server_IP: 10.10.20.10}. After receiving this message, the root router device establishes a task dispatch table locally. This table records the mapping relationship between analysis task types and target analysis servers, as well as the scheduling type. The dispatch strategy table is shown in Table 3 below: Table 3 ; Subsequently, when the routing device receives an analysis task of the aforementioned type, it forwards the task to the backend analysis server according to the table, specifically the IP address corresponding to Target_Server in the table. The backend analysis server receives the analysis task, processes it normally, and responds to the robot with a normal reply (here, the normal analysis and response process is irrelevant to this invention and will not be described in detail; the specific method can also be referred to the method described in the second application of this patent on the same application date). Thus, when the number of analysis tasks of the same type processed at the same leaf gateway exceeds a preset threshold, the above scheduling method can schedule these tasks to the backend for centralized processing.

[0042] Step 6: Task delegation mechanism; Similar to step five, when the backend storage server counts the number of tasks, if it finds that the number of a certain Task_Type under a certain root router is equal to 1, meaning that there is only one task left for that type of analysis task, the task should be scheduled back to the root router for processing. Therefore, the server sends a task scheduling instruction to the root router, delegating the backend analysis task to the root router for processing. This task scheduling instruction includes information such as the root router ID, task type, scheduling mode (changing from CENTRAL to EDGE), and backend analysis server address, as shown in the example below: {Root_Router_ID: GW_0001, Task_Type: VISION_ANALYSIS, Dispatch_Mode: EDGE, Analysis_Server_IP: 10.10.20.10}. After receiving this message, the root router establishes a task dispatch table locally. This table records the mapping relationship between analysis task types and target analysis servers, as well as the dispatch type. The dispatch strategy table is shown in Table 4 below: Table 4 ; Subsequently, when the routing device receives an analysis task of the aforementioned type, it will change the task, which originally needed to be forwarded to the backend analysis server, to be processed locally according to the Dispatch_Mode, based on the table.

[0043] Thus, a standard method for intelligent scheduling of robot analysis tasks is completed in a closed loop. This method utilizes routing devices to record topology mapping relationships and backend servers to statistically analyze task distribution. By employing a strategy of automatically scheduling similar tasks to the backend for centralized analysis when there are many of them, and automatically delegating similar tasks to the edge for separate analysis when there are few, this enables intelligent autonomous scheduling of multimodal data analysis tasks for a multi-robot system in an anycast network environment. However, considering that robots are often in a mobile state, to avoid excessively frequent and unstable task scheduling due to frequent gateway switching caused by robot movement, this method additionally introduces a stability judgment mechanism for the robot's mobile state, as described below.

[0044] (3) Robot movement state processing mechanism: Considering that the robot is frequently in motion, a stability determination mechanism is introduced. This mechanism defines the following rules: By default, the robot is placed in an unstable state, i.e., the parameter State_Flag = 0 (unstable); if the gateway the robot is connected to remains unchanged for more than 5 consecutive minutes, the parameter State_Flag = 1 (stable); if the robot detects a gateway change while in a stable state, it immediately sets the parameter State_Flag = 0 (unstable) until it has been connected to the gateway for more than 5 consecutive minutes, at which point it switches to a stable state.

[0045] Under the aforementioned robot state mechanism, the intelligent scheduling rule prioritizes State_Flag. Specifically, when State_Flag = 1, steps one through six of the intelligent scheduling process are executed. When State_Flag = 0, the task is fixed at the leaf gateway and does not participate in centralized scheduling. This mechanism effectively prevents system oscillations caused by repeated task migrations when the robot moves frequently.

[0046] Beneficial effects: This invention achieves the following beneficial effects through a unique robot identifier and a network node perception mechanism: 1. Active perception of the robot's topological position by the network layer; 2. Structured statistics based on task type; 3. Task aggregation judgment based on "root route"; 4. Supports centralized backend processing of similar tasks; 5. Supports single-task edge execution; 6. Supports stability protection during movement; 7. Improve overall computing power utilization; 8. Reduce the overhead of redundant analysis; 9. Improve system scalability and scheduling flexibility.

[0047] New innovation point: The key process design and protection points in this invention are as follows: 1. A sensing mechanism that carries the robot's unique ID (Robot_ID) during multimodal data analysis and transmission, and is monitored by each node in the network; 2. When each node in the network detects robot data transmission, it records the method of mapping and associating Robot_ID with the inlet / outlet interface, thereby obtaining the topology path of different types of business for each robot; 3. Based on the statistical analysis of the above topology paths, the statistical model of the "root routing device" and its task distribution judgment mechanism are analyzed. 4. Based on the above statistics and task distribution, an automated task scheduling method is proposed to move similar analysis tasks from peripheral analysis to centralized analysis; 5. An automated scheduling method that, based on the above statistics and task distribution, reassigns individual tasks to the edge for further analysis; 6. Based on the robot's movement state, define the robot's stable and unstable states, and implement an automatic scheduling and control mechanism based on these states; 7. During the above state transition process, the autonomous dynamic switching mechanism between centralized task scheduling and decentralization does not rely on central scheduling.

[0048] Another embodiment of the present invention provides a robot multimodal resource intelligent scheduling system, see [link to relevant documentation]. Figure 4 The system may include: The sending module 401 is used for multimodal data transmission and identification carrying: based on the multimodal data collected by the robot, it constructs a request message carrying the robot's unique identifier and the current analysis task type, and sends it to the leaf gateway to which it is connected. The perception module 402 is used for network node perception and topology information appending: based on the analysis result data generated after the request message is analyzed by the default edge analysis of the leaf gateway, each network node along the way extracts the robot's unique identifier and analysis task type during the forwarding process, records the mapping relationship between the robot's unique identifier and the forwarding interface, and appends the device identifier of the network node to the analysis result data before continuing to forward it to the backend server. The statistics module 403 is used for task distribution statistics and root route identification: based on the analysis result data carrying the network node identifiers along the route received by the backend server, it extracts and statistically analyzes the root route devices corresponding to each robot and the number distribution of different analysis task types under each root route device; The decision module 404 is used for dynamic scheduling decisions and instruction issuance: based on the statistical results of the task quantity distribution, the backend server determines whether the conditions for centralized scheduling or edge decentralization are met, and issues scheduling instructions to the corresponding root routing device to adjust the execution position of a specific analysis task type under the root routing device.

[0049] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0050] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0051] Specifically, the aforementioned electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the aforementioned processor, and the input / output device is connected to the aforementioned processor.

[0052] The above description, based on the embodiments shown in the figures, details the structure, features, and effects of the present invention. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A method for intelligent scheduling of multimodal resources for robots, characterized in that, The method includes: Multimodal data transmission and identification carrying: Based on the multimodal data collected by the robot, construct a request message carrying the robot's unique identifier and the current analysis task type, and send it to the leaf gateway it is connected to; Network node awareness and topology information appending: Based on the analysis result data generated after the request message is analyzed by the default edge analysis of the leaf gateway, each network node along the way extracts the robot's unique identifier and analysis task type during the forwarding process, records the mapping relationship between the robot's unique identifier and the forwarding interface, and appends the device identifier of the network node to the analysis result data before forwarding it to the backend server. Task distribution statistics and root route identification: Based on the analysis result data received by the backend server carrying the identifiers of network nodes along the route, extract and statistically analyze the root route devices corresponding to each robot and the quantity distribution of different analysis task types under each root route device; wherein, the task distribution statistics and root route identification include: Information extraction and aggregation: Based on the final analysis result data received by the backend server, extract the robot's unique identifier, leaf gateway identifier, network node device identifiers along the route, and analysis task type carried in the data; Root route device identification: Based on the leaf gateway identifier and network node device identifiers along the route, determine the common upstream device for the aggregation of each robot's data flow and define it as the root route device of that robot; Task distribution statistics: Based on the root route device and analysis task type of each robot, count the number of robots with different analysis task types under each root route device and form a task distribution statistics table; Dynamic scheduling decision and instruction issuance: Based on the statistical results of the quantity distribution of the task types, the backend server determines whether the conditions for centralized scheduling or edge decentralization are met, and issues scheduling instructions to the corresponding root routing device to adjust the execution position of the specific analysis task type under the root routing device.

2. The method according to claim 1, characterized in that, The multimodal data transmission and identifier carrying include: Multimodal data acquisition: Collect multimodal data such as visual, environmental, or traffic flow data based on the robot's operating environment to determine the types of data that need to be intelligently analyzed; Message Construction: Based on the data type and the robot's own stored unique identifier, a request message is constructed. The message includes at least the robot's unique identifier, the analysis task type identifier, and the data payload to be analyzed. Message transmission: Based on the wireless network currently connected to the robot, the request message is sent to the directly connected leaf gateway.

3. The method according to claim 2, characterized in that, The network node perception and topology information appending includes: Default edge analysis: Based on the request message received by the leaf gateway, the analysis task is assigned to the directly connected edge analysis device for execution. After the edge analysis device completes the analysis, it generates analysis result data. The analysis result data includes at least the robot's unique identifier, the leaf gateway's own identifier, the analysis task type, and the analysis result. Mapping relationship record: Based on the analysis result data received by the upstream routing device, extract the robot's unique identifier, leaf gateway identifier, and analysis task type, and record the mapping relationship between the robot's unique identifier, leaf gateway identifier, analysis task type, and the ingress interface, egress interface, and its own device identifier locally; Device Identifier Attachment: The upstream routing device attaches its own device identifier to the extended field of the analysis result data, and the attached data continues to be forwarded to the upper layer. Each subsequent network node returns the execution step mapping relationship record and the attached device identifier until the data reaches the backend server.

4. The method according to claim 3, characterized in that, The dynamic scheduling decision-making and instruction issuance include: Centralized scheduling condition judgment: According to the task distribution statistics table, when the number of robots of a certain analysis task type under a certain root routing device reaches or exceeds the preset centralized scheduling threshold, the backend server generates a centralized scheduling instruction. The instruction includes the root routing device identifier, analysis task type, scheduling mode is centralized, and backend analysis server address. Centralized scheduling and execution: Based on the centralized scheduling instruction received by the root routing device, a local task scheduling policy table is established or updated, and the forwarding target of the corresponding analysis task type is pointed to the backend analysis server. When the same type of analysis task is received in the future, it is directly forwarded to the backend analysis server for centralized processing. Edge deployment condition judgment: According to the task distribution statistics table, when the number of robots of a certain analysis task type under a certain root routing device drops to 1, the backend server generates an edge deployment scheduling instruction. The instruction includes the root routing device identifier, analysis task type, scheduling mode as edge, and optional backend analysis server address. Edge-deployed execution: Based on the edge-deployed scheduling instruction received by the root routing device, the local task scheduling policy table is updated, and the processing method for the corresponding analysis task type is changed to local edge execution. When such analysis tasks are received in the future, they will no longer be forwarded to the backend analysis server.

5. The method according to claim 4, characterized in that, The method also includes a robot movement state processing mechanism, specifically including: State initialization and detection: The robot's initial state is assumed to be unstable, and the system continuously monitors whether the leaf gateway it connects to changes. State switching determination: If the leaf gateway connected to the robot does not change for more than a preset time, its state is switched to stable; if a change is detected in the stable state, it is immediately switched back to unstable. State-based scheduling strategy: If the robot is in a stable state, the backend server is allowed to make scheduling decisions based on the task distribution statistics and issue scheduling instructions to the corresponding root routing device to adjust the execution position of the analysis task; if the robot is in an unstable state, the robot's analysis task is forced to be fixed to the leaf gateway it is currently connected to and is not involved in the centralized scheduling of the backend server. Status information transmission: When a robot sends a request message, it carries its current status identifier, which is used by the backend server to distinguish between stable and unstable robots when calculating task distribution, ensuring that scheduling decisions are based on robots in a stable state.

6. A robot multimodal resource intelligent scheduling system, characterized in that, The system includes: The sending module is used for multimodal data transmission and identification carrying: based on the multimodal data collected by the robot, it constructs a request message carrying the robot's unique identifier and the current analysis task type, and sends it to the leaf gateway it is connected to. The perception module is used for network node perception and topology information appending: based on the analysis result data generated after the request message is analyzed by the default edge analysis of the leaf gateway, each network node along the way extracts the robot's unique identifier and analysis task type during the forwarding process, records the mapping relationship between the robot's unique identifier and the forwarding interface, and appends the device identifier of the network node to the analysis result data before forwarding it to the backend server. The statistics module is used for task distribution statistics and root route identification: based on the analysis result data received by the backend server carrying the identifiers of network nodes along the route, it extracts and statistically analyzes the root route devices corresponding to each robot and the quantity distribution of different analysis task types under each root route device; wherein, the task distribution statistics and root route identification include: information extraction and aggregation: based on the final analysis result data received by the backend server, extracting the robot's unique identifier, leaf gateway identifier, network node device identifiers along the route, and analysis task type carried in the data; root route device identification: based on the leaf gateway identifier and network node device identifiers along the route, determining the common upstream device for the data flow aggregation of each robot, and defining it as the root route device of that robot; task distribution statistics: based on the root route device of each robot and the analysis task type, counting the number of robots with different analysis task types under each root route device, forming a task distribution statistics table; The decision module is used for dynamic scheduling decisions and instruction issuance: based on the statistical results of the quantity distribution of the task types, the backend server determines whether the conditions for centralized scheduling or edge decentralization are met, and issues scheduling instructions to the corresponding root routing device to adjust the execution position of the specific analysis task type under the root routing device.

7. The system according to claim 6, characterized in that, The sending module is specifically used for: Multimodal data acquisition: Collect multimodal data such as visual, environmental, or traffic flow data based on the robot's operating environment to determine the types of data that need to be intelligently analyzed; Message Construction: Based on the data type and the robot's own stored unique identifier, a request message is constructed. The message includes at least the robot's unique identifier, the analysis task type identifier, and the data payload to be analyzed. Message transmission: Based on the wireless network currently connected to the robot, the request message is sent to the directly connected leaf gateway.

8. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-5 when it is run.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1-5.

Images