A perception decision-making method based on edge computing

Abstract

The present invention belongs to the field of robot edge computing, and in particular, relates to a perception decision-making method based on edge computing, which first constructs a multi-node edge computing array diagram and configures a load migration model to obtain the state perception and task instruction information of a quadruped robot in real time, and generate a corresponding subtask sequence; secondly, according to the shortest distance from the quadruped robot to the edge node, the perception information and subtasks are transmitted to the corresponding edge node to construct a perception decision input pair; thirdly, a distributed reinforcement decision model is used to generate perception decision instructions, and they are shared in real time through a multi-node edge computing array diagram; when it is monitored that the edge node calculation delay exceeds a set threshold, the optimal load migration is achieved through the load migration model; finally, the load state of each edge node is evaluated and adjusted in real time to ensure that each edge node can meet the instruction calculation delay threshold, thereby effectively improving the task execution capability and response speed of the quadruped robot in a complex environment.

Description

Technical Field

[0001] The present invention belongs to the field of robot edge computing, and in particular relates to a perception decision-making method based on edge computing. Background Art

[0002] With the development of industrial robot technology, quadruped robots have been widely used in industrial automation, closed space environment inspection, security, exploration and other fields; the traditional approach is to concentrate perception data and decision-making calculations on the cloud or central server, but there are problems such as data transmission delay and network bandwidth limitation; edge computing effectively solves these problems by deploying computing and storage resources at the edge of the network to achieve rapid data processing and real-time response; however, when applying edge computing to the perception and decision-making system of quadruped robots, it still faces challenges such as requiring resource-constrained edge devices to efficiently process large amounts of heterogeneous sensor data, quickly formulate and optimize decision-making strategies based on real-time perception results, and establish effective feedback mechanisms.

[0003] For example, a Chinese patent with publication number CN117541025A discloses an edge computing method for dense transmission line inspection, including building a dense transmission line inspection edge computing model to achieve full coverage inspection of transmission lines by obstacle-crossing inspection robots; building an energy consumption balance model for obstacle-crossing inspection robots to achieve energy consumption balance and minimize overall energy consumption, thereby maximizing inspection benefits; building a digital twin network for the inspection area, and based on the information fed back by the digital twin network, using the DR-MADDPG algorithm to obtain the unloading decision of the obstacle-crossing inspection robot, and feeding it back to the corresponding obstacle-crossing inspection robot in the physical entity network.

[0004] For example, a Chinese patent with publication number CN114493164A discloses a robot task parsing method and system based on edge computing. This method issues basic task items to the intelligent robot on the edge side through a cloud-based backend server. The intelligent robot converts them into a time-sequential workflow, and then generates adaptation factor items for the workflow by perceiving and identifying work scene patterns and environmental parameters, and detecting its own status information. Based on these adaptation factor items, the time-sequential work task units are executed and scheduled through expert decision scheduling, and action and function control instructions are issued to the intelligent robot.

[0005] The above-mentioned prior art has the following problems: when the prior art performs edge computing, the computing load may be large due to the presence of multiple robots near the same edge node, causing the computing control process to slow down and reducing the real-time performance of the robot control. For this reason, the present invention provides a perception decision-making method based on edge computing. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes a perception decision-making method based on edge computing, which first constructs a multi-node edge computing array diagram and configures a load migration model to obtain the state perception and task instruction information of the quadruped robot in real time, and generate a corresponding subtask sequence through a task decomposition model; secondly, according to the shortest distance from the quadruped robot to the edge node, the perception information and subtasks are transmitted to the corresponding edge node to construct a perception decision input pair; thirdly, the distributed reinforcement decision model is used to generate perception decision instructions, and they are shared in real time through a multi-node edge computing array diagram; when it is monitored that the calculation delay of the edge node exceeds the set threshold, the optimal load migration is achieved through the load migration model; finally, an evaluation algorithm is used to evaluate and adjust the load status of each edge node in real time to ensure that each edge node can meet the instruction calculation delay threshold, thereby effectively improving the task execution capability and response speed of the quadruped robot in a complex environment.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A perception decision method based on edge computing, comprising the following steps:

[0009] S1. Construct a multi-node edge computing array diagram and configure a load migration model to obtain the current state perception information and task instruction information of n quadruped robots in real time. According to the task instruction information, the configured task decomposition model is used to obtain the n subtask sequences corresponding to the n quadruped robots. The multi-node edge computing array diagram is constructed by the information intra-interaction chain, load migration information chain and all edge nodes obtained through the edge node working state information and interaction information.

[0010] S2. Obtain the distance from the quadruped robot to each edge node in real time, construct the shortest distance information transmission pair between the quadruped robot and the edge node, and transmit the current state perception information of the quadruped robot and the corresponding subtask to the edge node corresponding to the multi-node edge computing array diagram according to the shortest distance information transmission pair, and construct the perception decision input pair of each quadruped robot;

[0011] S3, input each quadruped robot perception decision input pair into the distributed reinforcement decision model configured by the corresponding edge node, generate the perception decision instruction corresponding to each quadruped robot, and share the perception decision instruction generated by each edge node in real time through the multi-node edge computing array diagram, generate shared perception decision instruction, and send the generated shared perception decision instruction to the corresponding quadruped robot to control the operating state of the quadruped robot in real time;

[0012] S4. Set a single-node instruction calculation delay threshold, and monitor the calculation delay corresponding to the shared perception decision instruction of each edge node in real time. When the calculation delay of the corresponding edge node is greater than the calculation delay threshold, the optimal migration edge node and migration path and load migration amount are obtained through the load migration model, and the decision instruction generation task corresponding to the load migration amount of the edge node that does not meet the calculation delay threshold is migrated to the optimal migration edge node.

[0013] Specifically, the steps of constructing a multi-node edge computing array graph include:

[0014] S101, obtain edge node working status information and interaction information and pre-process them, and set each edge node as a graph node in a multi-node edge computing array graph ,in represents the i-th graph node in the i-th multi-node edge computing array graph, where M is the total number of graph nodes in the multi-node edge computing array graph;

[0015] S102: Input the edge node location information into the DBSCAN algorithm. The graph nodes that can fully cover each other in the medium signal transmission distance are grouped into one category to obtain the edge computing array graph node subset;

[0016] S103, using the working status information and interaction information corresponding to the edge computing array graph node subset, constructing the information interaction chain between the two graph nodes in the edge computing array graph node subset, and constructing the multi-node edge computing array subgraph by using the information interaction chain between the two graph nodes and the edge computing array graph node subset ,in represents the A-th multi-node edge computing array subgraph, represents the edge computing array graph node subset in the A-th multi-node edge computing array subgraph, represents the information intra-interaction chain set between edge computing array graph node subsets in the A-th multi-node edge computing array subgraph;

[0017] S104. Assumptions Graph nodes in and Graph nodes in The closest and graph node Can only be used with graph nodes The signal transmission distances overlap each other, so use and The corresponding interaction information is constructed and The load migration information chain between them; specifically:

[0018] ,

[0019] in and express Neutralization The graph nodes are closest and the graph nodes Can only be used with graph nodes The signal transmission distance of the graph node pairs that cover each other, arg represents the inverse function operator, Representing a graph node With graph nodes Mahalanobis distance, express and The maximum effective distance of signal transmission corresponding to all graph nodes;

[0020] S105. Repeat the process of S104 to construct the load migration information chain between all multi-node edge computing array subgraphs, and use all the obtained load migration information chains and multi-node edge computing array subgraphs to construct a multi-node edge computing array graph.

[0021] Specifically, the information intra-interaction chain in S103 is configured with an interaction quality weight indicator, and the step of constructing the interaction quality weight indicator includes:

[0022] S1031, assuming a multi-node edge computing array subgraph A subset of edge computing array graph nodes contained in , using the edge computing array graph node subsets, each pair of graph nodes is used to construct node pairs within the subset, and according to the corresponding interaction information of each node pair within the subset, the interaction quality weight index of the corresponding node pair within the subset is calculated, and the interaction quality weight index is built into the information interaction chain corresponding to the node pair within the corresponding subset; the interaction quality weight index is specifically: ,in express The interaction quality weight index between the node pairs in the subset constructed by the jth and gth graph nodes in The number of nodes in the graph, express The signal transmission bandwidth per unit time between node pairs in the subset constructed by the jth and gth graph nodes in , express The signal transmission delay per unit time between the node pairs in the subset constructed by the j-th and g-th graph nodes; express The signal transmission packet loss rate per unit time between the node pairs in the subset constructed by the j-th and g-th graph nodes; express The length of the signal transmission task queue per unit time between node pairs in the subset constructed by the j-th and g-th graph nodes; and Represents the multi-node edge computing array subgraphs respectively The remaining computing resources of the jth and gth graph nodes at the current moment. When the corresponding graph node has a computing delay exception, the remaining computing resources of the corresponding graph node at the current moment are set to 1. They represent the weighted coefficients of transmission delay, transmission packet loss rate and task queue length respectively.

[0023] Specifically, the steps of constructing the interaction quality weight index also include:

[0024] S1032. Repeat S1031 to build a multi-node edge computing array subgraph The interaction quality weight indicators between all pairs of nodes in the subset are built into the corresponding information interaction chain;

[0025] S1033. Repeat processes S1031 and S1032 to construct the interaction quality weight index corresponding to the interaction chain in the corresponding information of all multi-node edge computing array subgraphs, and arrange the sizes of the interaction quality weight index in each multi-node edge computing array subgraph in descending order, and feed back the arrangement results and the working status and instruction calculation delay corresponding to each graph node to the load migration model.

[0026] Specifically, a load migration weight indicator is configured in the load migration information chain; the step of obtaining the optimal migration edge node and migration path in S4 includes:

[0027] S401, according to S104 process, obtain A collection of directly or indirectly connected multi-node edge computing array subgraphs, assuming that at the current moment Through the S4 process, it is found that there are m graph nodes with abnormal calculation delays, and the current If the remaining computing resources corresponding to the nodes in the non-existent computing delay anomaly graph satisfy the load migration amount corresponding to the nodes in the existing computing delay anomaly graph obtained by the load migration model and the computing delay after migration is less than or equal to the computing delay threshold, then through the constructed information intra-interaction chain, according to the configured interaction quality weight index, the decision instruction generation task of the corresponding computing delay anomaly graph node is migrated to The corresponding optimal migration edge node in ;

[0028] S402, current time Through the S4 process, it is found that there are m graph nodes with abnormal calculation delays, and If the graph nodes corresponding to the non-computational delay anomaly cannot complete the decision instruction generation task of the graph nodes corresponding to the computational delay anomaly, the multi-node edge computing array subgraph and corresponding to the optimal migration edge node are obtained through the load migration model. There is a calculation of the load migration amount corresponding to the delay graph node;

[0029] S403, if the multi-node edge computing array subgraph and Directly connected and only one multi-node edge computing array subgraph meets the computing conditions, then directly The decision instruction generation task corresponding to the load migration amount is migrated to the corresponding multi-node edge computing array subgraph.

[0030] Specifically, the step of obtaining the optimal migration edge node and migration path also includes:

[0031] S404: If a multi-node edge computing array subgraph and If the number of multi-node edge computing array subgraphs that are directly connected and meet the computing conditions is greater than one, the load migration weight index value corresponding to all load migration information chains that meet the computing conditions at the current moment is calculated through the load migration model according to the load migration weight index configured in the corresponding load migration information chain;

[0032] S405, arrange all calculated load migration weight index values ​​in descending order, migrate the decision instruction generation task corresponding to the load migration amount to the multi-node edge computing array subgraph corresponding to the maximum weight index value, and distribute the decision instruction generation task to the corresponding optimal graph node according to the S401 process;

[0033] S406: If a multi-node edge computing array subgraph and If the number of multi-node edge computing array subgraphs that are indirectly connected and meet the computing conditions is greater than or equal to one, The corresponding load migration amount, The load migration weight index values ​​corresponding to all load migration information chains and the total amount of remaining available computing resources corresponding to each multi-node edge computing array subgraph that meets the computing conditions are indirectly connected to all multi-node edge computing array subgraphs that meet the computing conditions, and are input into the path generation algorithm to obtain the multi-node edge computing array subgraph corresponding to the optimal migration path;

[0034] S407. According to the multi-node edge computing array subgraph corresponding to the optimal migration path, the decision instruction generation task corresponding to the load migration amount is migrated to the multi-node edge computing array subgraph corresponding to the maximum weight index value, and the decision instruction generation task is allocated to the corresponding optimal graph node according to the S401 process.

[0035] Specifically, the step of obtaining the optimal migration edge node and migration path also includes:

[0036] S408: If there is no single multi-node edge computing array subgraph satisfying Calculate the load migration amount corresponding to all graph nodes with abnormal delay in The load migration amount corresponding to each calculation delay abnormal graph node, the interaction quality weight index value corresponding to all information intra-interaction chains, the load migration weight index value corresponding to all load migration information chains, the remaining computing resources and computing delay of the graph nodes without calculation delay abnormalities are input into the path generation algorithm to obtain the migration path of the optimal migration edge node corresponding to each calculation delay abnormal graph node;

[0037] S409, according to the migration path of the optimal migration edge node corresponding to each calculation delay abnormal graph node, generate the corresponding decision instruction task and migrate it to the corresponding edge node;

[0038] S410, monitor in real time the computing operation status generated by the decision instructions of each multi-node edge computing array subgraph. When there is a computing anomaly in at least one multi-node edge computing array subgraph, repeat the S401-S410 process to perform load migration, so that the computing delay corresponding to each edge node meets the computing delay threshold in real time.

[0039] Specifically, the steps of constructing the load migration weight index include:

[0040] Assuming a multi-node edge computing array subgraph The total available computing resources at the current moment are , multi-node edge computing array subgraph The total available computing resources at the current moment are , then the load migration weight index The corresponding calculation formula is:

[0041]

[0042] in, represents the load migration information chain connection factor, Indicates the corresponding graph node of S104 to obtain the load migration information chain With graph nodes The signal transmission bandwidth between Indicates the corresponding graph node of S104 to obtain the load migration information chain With graph nodes The frequency of interaction between Represents a multi-node edge computing array subgraph China Map Node The amount of computing resources remaining at the current moment, and In turn, it represents the multi-node edge computing array subgraph or multi-node edge computing array subgraph The corresponding load migration amount when there is a calculation delay exception.

[0043] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements a perception decision method based on edge computing when executing the computer program.

[0044] A computer-readable storage medium stores computer instructions, which, when executed, execute a perception decision-making method based on edge computing.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] In view of the deficiencies of the prior art, the present invention constructs a multi-node edge computing array diagram and configures a load migration model, thereby effectively solving the problem in the prior art that the computing load is too large and the computing control process is slowed down due to the presence of multiple robots near the same edge node. Specifically, by acquiring the distance from the quadruped robot to each edge node in real time, a shortest distance information transmission pair is constructed to ensure that each robot communicates with the nearest edge node, thereby reducing the delay and packet loss rate of data transmission. Secondly, a task decomposition model is used to decompose complex tasks into multiple subtasks to disperse the computing load and avoid overloading a single edge node. Thirdly, a distributed enhanced decision model is used to generate perception decision instructions and share them in real time to ensure the efficiency and consistency of the decision. Finally, a computing delay threshold is set to monitor the computing delay of each edge node in real time. When the threshold is exceeded, the load migration model is used to automatically select the optimal migration edge node and path to further balance the computing load and ensure the real-time and stability of the system. This process further improves the real-time control performance of the quadruped robot and the robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of a perception decision method based on edge computing according to Embodiment 1 of the present invention;

[0048] Figure 2 This is a flow chart of the generation structure of shared perception decision instructions according to Example 1 of the present invention. DETAILED DESCRIPTION

[0049] Example 1

[0050] See also Figure 1 , an embodiment provided by the present invention: a perception decision method based on edge computing, the steps comprising:

[0051] S1. Build a multi-node edge computing array diagram and configure a load migration model to obtain the current state perception information and task instruction information of n quadruped robots in real time. According to the task instruction information, obtain the n subtask sequences corresponding to the n quadruped robots through the configured task decomposition model;

[0052] Furthermore, in this embodiment, the multi-node edge computing array diagram is constructed by obtaining the information interaction chain, load migration information chain and all edge nodes through the edge node working status information and interaction information.

[0053] Furthermore, in this embodiment, the step of obtaining n subtask sequences corresponding to n quadruped robots includes:

[0054] a1. Obtain the total number of tasks, the maximum number of safe tasks that a single robot can perform at a time, the number of currently available quadruped robots, and the maximum working duration of the quadruped robots, and pre-process the acquired data;

[0055] a2. Build a task decomposition model based on the random forest algorithm, input the preprocessed data into the task decomposition model, obtain the decomposed subtask sequence and the urgency of each subtask, and label each subtask; further, the urgency of each subtask is defined by the order of the completion time points of the subtasks. The earlier the completion time point, the higher the urgency of the subtask.

[0056] a3. According to the urgency of the subtask, the number of available quadruped robots and the current maximum working duration of the quadruped robots, a matching algorithm is used to obtain the label of the subtask that each quadruped robot executes. Furthermore, in this embodiment, the subtask corresponding to the high urgency is assigned to the robot with the large current maximum working duration.

[0057] Furthermore, in this embodiment, the steps of constructing a multi-node edge computing array graph include:

[0058] S101, obtain edge node working status information and interaction information and pre-process them, and set each edge node as a graph node in a multi-node edge computing array graph ,in represents the i-th graph node in the i-th multi-node edge computing array graph, where M is the total number of graph nodes in the multi-node edge computing array graph;

[0059] S102: Input the edge node location information into the DBSCAN algorithm. The graph nodes that can fully cover each other in the medium signal transmission distance are grouped into one category to obtain the edge computing array graph node subset;

[0060] Furthermore, in this embodiment, the signal transmission distance corresponding to each edge node is the same;

[0061] For example, suppose there is Among the five edge nodes, the signal transmission distances corresponding to a, b and c can overlap each other, so a, b and c are clustered into one category as an edge computing array graph node subset, and d or e can only overlap with the signal transmission distance of one of a, b, c, so d or e is not clustered into the same category with a, b, c; if the signal transmission distances of d and e overlap each other, the two are clustered into one category as another edge computing array graph node subset.

[0062] S103, using the working status information and interaction information corresponding to the edge computing array graph node subset, constructing the information interaction chain between the two graph nodes in the edge computing array graph node subset, and constructing the multi-node edge computing array subgraph by using the information interaction chain between the two graph nodes and the edge computing array graph node subset ,in represents the A-th multi-node edge computing array subgraph, represents the edge computing array graph node subset in the A-th multi-node edge computing array subgraph, represents the information intra-interaction chain set between edge computing array graph node subsets in the A-th multi-node edge computing array subgraph;

[0063] Furthermore, in this embodiment, the information intra-interaction chain in S103 is configured with an interaction quality weight indicator, and the step of constructing the interaction quality weight indicator includes:

[0064] S1031, assuming a multi-node edge computing array subgraph A subset of edge computing array graph nodes contained in , using the edge computing array graph node subsets, each pair of graph nodes is used to construct node pairs within the subset, and according to the corresponding interaction information of each node pair within the subset, the interaction quality weight index of the corresponding node pair within the subset is calculated, and the interaction quality weight index is built into the information interaction chain corresponding to the node pair within the corresponding subset; the interaction quality weight index is specifically: ,in express The interaction quality weight index between the node pairs in the subset constructed by the jth and gth graph nodes in The number of nodes in the graph, express The signal transmission bandwidth per unit time between node pairs in the subset constructed by the jth and gth graph nodes in , express The signal transmission delay per unit time between the node pairs in the subset constructed by the j-th and g-th graph nodes; express The signal transmission packet loss rate per unit time between the node pairs in the subset constructed by the j-th and g-th graph nodes; express The length of the signal transmission task queue per unit time between node pairs in the subset constructed by the j-th and g-th graph nodes; and Represents the multi-node edge computing array subgraphs respectively The remaining computing resources of the jth and gth graph nodes at the current moment. When the corresponding graph node has a computing delay exception, the remaining computing resources of the corresponding graph node at the current moment are set to 1. The weighted coefficients representing transmission delay, transmission packet loss rate and task queue length respectively;

[0065] S1032, repeat the process of S1031 to construct the interaction quality weight index between the node pairs in all subsets of the multi-node edge computing array subgraph A and embed it into the corresponding information interaction chain;

[0066] S1033. Repeat processes S1031 and S1032 to construct the interaction quality weight index corresponding to the interaction chain in the corresponding information of all multi-node edge computing array subgraphs, and arrange the sizes of the interaction quality weight index in each multi-node edge computing array subgraph in descending order, and feed back the arrangement results and the working status and instruction calculation delay corresponding to each graph node to the load migration model.

[0067] S104. Assumptions Graph nodes in and Graph nodes in The closest and graph node Can only be used with graph nodes The signal transmission distances overlap each other, so use and The corresponding interaction information is constructed and The load migration information chain between them; specifically:

[0068] ,

[0069] in and express Neutralization The graph nodes are closest and the graph nodes Can only be used with graph nodes The signal transmission distance of the graph node pairs that cover each other, arg represents the inverse function operator, Representing a graph node With graph nodes Mahalanobis distance, express and The maximum effective distance of signal transmission corresponding to all graph nodes in .

[0070] Furthermore, in this embodiment, a load migration weight indicator is configured in the load migration information chain; the steps of constructing the load migration weight indicator include:

[0071] Assuming a multi-node edge computing array subgraph The total available computing resources at the current moment are , multi-node edge computing array subgraph The total available computing resources at the current moment are , then the load migration weight index The corresponding calculation formula is:

[0072]

[0073] in, represents the load migration information chain connection factor, Indicates the corresponding graph node of S104 to obtain the load migration information chain With graph nodes The signal transmission bandwidth between Indicates the corresponding graph node of S104 to obtain the load migration information chain With graph nodes The frequency of interaction between Represents a multi-node edge computing array subgraph China Map Node The amount of computing resources remaining at the current moment, and In turn, it represents the multi-node edge computing array subgraph or multi-node edge computing array subgraph The corresponding load migration amount when there is a calculation delay exception.

[0074] Furthermore, in this embodiment, when calculating During the process, when the multi-node edge computing array subgraph There are abnormal computation delays, and the multi-node edge computing array subgraph When there is no calculation delay exception, ;

[0075] When the multi-node edge computing array subgraph There is no abnormal computing delay, and the multi-node edge computing array subgraph When there is a calculation delay exception, ;

[0076] When the multi-node edge computing array subgraph and multi-node edge computing array subgraph When there are calculation delay exceptions, , disconnect the multi-node edge computing array subgraph and multi-node edge computing array subgraph Load migration information chain between them.

[0077] S105. Repeat the process of S104 to construct the load migration information chain between all multi-node edge computing array subgraphs, and use all the obtained load migration information chains and multi-node edge computing array subgraphs to construct a multi-node edge computing array graph.

[0078] This process realizes efficient decision-making calculation for multiple quadruped robots by constructing a multi-node edge computing array graph and configuring a load migration model. First, the task decomposition model is constructed through the random forest algorithm to decompose the overall task into multiple subtasks, and the subtasks are reasonably allocated according to the urgency of the subtasks and the current state of the quadruped robot, ensuring efficient decomposition and execution of the task. Secondly, the edge nodes are classified according to the signal transmission distance through the DBSCAN algorithm to form multiple edge computing array graph node subsets, which enhances the local collaboration ability and robustness of the system and can effectively cope with environmental changes and node failures. In addition, by constructing the information intra-interaction chain between the two graph nodes in the subset and introducing the interaction quality weight index, the signal transmission bandwidth, delay, packet loss rate and task queue length are comprehensively considered to ensure the efficiency and reliability of inter-node communication. When a node has abnormal calculation delay, part of the load is migrated to other nodes through the load migration model and the load migration weight index configured in the load migration information chain, balancing the system load, avoiding the performance degradation caused by the overload of a single node, and improving the overall performance of the system.

[0079] S2. Obtain the distance from the quadruped robot to each edge node in real time, construct the shortest distance information transmission pair between the quadruped robot and the edge node, and transmit the current state perception information of the quadruped robot and the corresponding subtask to the edge node corresponding to the multi-node edge computing array diagram according to the shortest distance information transmission pair, and construct the perception decision input pair of each quadruped robot;

[0080] S3, input each quadruped robot perception decision input pair into the distributed reinforcement decision model configured by the corresponding edge node, generate the perception decision instruction corresponding to each quadruped robot, and share the perception decision instruction generated by each edge node in real time through the multi-node edge computing array diagram, generate shared perception decision instruction, and send the generated shared perception decision instruction to the corresponding quadruped robot to control the operating state of the quadruped robot;

[0081] Furthermore, the distributed enhanced decision model in this embodiment includes a collaborative control model and G enhanced decision sub-models; each enhanced decision sub-model corresponds one-to-one to an edge node; and each enhanced decision sub-model is constructed by the SAC algorithm.

[0082] For further information, see Figure 2 In this embodiment, the steps of generating a shared perception decision instruction include:

[0083] S301, assuming that there are currently N robots executing N assigned subtasks, and the N assigned subtasks are assigned to U edge nodes, then using the N assigned subtasks and the current state perception information uploaded by the corresponding robots, the enhanced decision submodel of the corresponding edge node is used to generate decision instructions to obtain N perception decision instructions;

[0084] S302. Upload the decision parameters corresponding to N perception decision instructions to the collaborative control model, generate collaborative control instructions, and send the generated collaborative control instructions to each edge node to merge with each perception decision instruction to obtain the shared perception decision instructions corresponding to each quadruped robot.

[0085] The process obtains the distance from the quadruped robot to each edge node in real time, constructs the shortest distance information transmission pair, and transmits the current state perception information of the quadruped robot and the corresponding subtask to the edge node corresponding to the multi-node edge computing array diagram to construct the perception decision input pair of each quadruped robot. By inputting these input pairs into the distributed reinforcement decision model, the perception decision instructions of each quadruped robot are generated, and they are shared in real time through the multi-node edge computing array diagram to generate shared perception decision instructions, which are finally issued to the corresponding quadruped robot for real-time control. This method not only improves the real-time and accuracy of task execution, but also enhances the robustness and adaptability of the system. The collaborative control model in the distributed reinforcement decision model and the reinforced decision submodel constructed by the SAC algorithm ensure that each robot can make the best decision based on global information when performing tasks, avoiding system failure caused by single node failure. Through the generation and fusion of collaborative control instructions, efficient collaboration between multiple robots is achieved, and the task completion efficiency and overall performance of the system are improved.

[0086] S4. Set a single-node instruction calculation delay threshold, monitor the calculation delay corresponding to the shared perception decision instruction of each edge node in real time, and when the calculation delay of the corresponding edge node is greater than the calculation delay threshold, obtain the optimal migration edge node and migration path and load migration amount through the load migration model, and migrate the decision instruction generation task corresponding to the load migration amount of the edge node that does not meet the calculation delay threshold to the optimal migration edge node;

[0087] In this embodiment, the load migration amount refers to the amount of computing tasks migrated from the edge node with abnormal computing delay to other edge nodes; if the current edge node can only process up to 30 decision instruction generation tasks at the same time while meeting the computing delay threshold, and there are currently 40 decision instruction generation tasks, then the excess 10 decision instruction generation tasks need to be migrated to other edge nodes for calculation, and the computing task amount corresponding to the 10 decision instruction generation tasks is the load migration amount.

[0088] Furthermore, in this embodiment, the step of obtaining the optimal migration edge node and migration path in S4 includes:

[0089] S401, according to S104 process, obtain A collection of directly or indirectly connected multi-node edge computing array subgraphs, assuming that at the current moment Through the S4 process, it is found that there are m graph nodes with abnormal calculation delays, and the current If the remaining computing resources corresponding to the nodes in the non-existent computing delay anomaly graph satisfy the load migration amount corresponding to the nodes in the existing computing delay anomaly graph obtained by the load migration model and the computing delay after migration is less than or equal to the computing delay threshold, then through the constructed information intra-interaction chain, according to the configured interaction quality weight index, the decision instruction generation task of the corresponding computing delay anomaly graph node is migrated to The corresponding optimal migration edge node in ;

[0090] Furthermore, the decision instruction generation task of the corresponding calculation delay exception graph node is migrated to The corresponding optimal migration edge node steps include:

[0091] Calculated according to S1033 process The interaction quality weight index and load migration amount corresponding to the node in the graph without abnormal calculation delay in the absence of abnormal calculation delay in the presence of abnormal calculation delay in the presence of abnormal calculation delay in the absence ... The ratio of the interaction quality weight index corresponding to all the nodes in the non-existent calculation delay anomaly graph is used to allocate the load migration amount to the nodes in the non-existent calculation delay anomaly graph to generate decision instructions.

[0092] S402, current time Through the S4 process, it is found that there are m graph nodes with abnormal calculation delays, and If the graph nodes corresponding to the non-computational delay anomaly cannot complete the decision instruction generation task of the graph nodes corresponding to the computational delay anomaly, the multi-node edge computing array subgraph and corresponding to the optimal migration edge node are obtained through the load migration model. There is a calculation of the load migration amount corresponding to the delay graph node;

[0093] S403, if the multi-node edge computing array subgraph and Directly connected and only one multi-node edge computing array subgraph meets the computing conditions, then directly The decision instruction generation task corresponding to the load migration amount is migrated to the corresponding multi-node edge computing array subgraph;

[0094] S404: If a multi-node edge computing array subgraph and If the number of multi-node edge computing array subgraphs that are directly connected and meet the computing conditions is greater than one, the load migration weight index value corresponding to all load migration information chains that meet the computing conditions at the current moment is calculated through the load migration model according to the load migration weight index configured in the corresponding load migration information chain;

[0095] S405, arrange all calculated load migration weight index values ​​in descending order, migrate the decision instruction generation task corresponding to the load migration amount to the multi-node edge computing array subgraph corresponding to the maximum weight index value, and distribute the decision instruction generation task to the corresponding optimal graph node according to the S401 process;

[0096] S406: If a multi-node edge computing array subgraph and If the number of multi-node edge computing array subgraphs that are indirectly connected and meet the computing conditions is greater than or equal to one, The corresponding load migration amount, The load migration weight index values ​​corresponding to all load migration information chains and the total amount of remaining available computing resources corresponding to each multi-node edge computing array subgraph that meets the computing conditions are indirectly connected to all multi-node edge computing array subgraphs that meet the computing conditions, and are input into the path generation algorithm to obtain the multi-node edge computing array subgraph corresponding to the optimal migration path;

[0097] S407, according to the multi-node edge computing array subgraph corresponding to the optimal migration path, migrate the decision instruction generation task corresponding to the load migration amount to the multi-node edge computing array subgraph corresponding to the maximum weight index value, and distribute the decision instruction generation task to the corresponding optimal graph node according to the S401 process;

[0098] S408: If there is no single multi-node edge computing array subgraph satisfying Calculate the load migration amount corresponding to all graph nodes with abnormal delay in The load migration amount corresponding to each calculation delay abnormal graph node, the interaction quality weight index value corresponding to all information intra-interaction chains, the load migration weight index value corresponding to all load migration information chains, the remaining computing resources and computing delay of the graph nodes without calculation delay abnormalities are input into the path generation algorithm to obtain the migration path of the optimal migration edge node corresponding to each calculation delay abnormal graph node;

[0099] S409, according to the migration path of the optimal migration edge node corresponding to each calculation delay abnormal graph node, generate the corresponding decision instruction task and migrate it to the corresponding edge node;

[0100] S410, monitor in real time the computing operation status generated by the decision instructions of each multi-node edge computing array subgraph. When there is a computing anomaly in at least one multi-node edge computing array subgraph, repeat the S401-S410 process to perform load migration, so that the computing delay corresponding to each edge node meets the computing delay threshold in real time.

[0101] The process sets a single-node instruction calculation delay threshold and monitors the calculation delay of each edge node in real time. When the calculation delay exceeds the threshold, the decision instructions that do not meet the calculation delay threshold are migrated to the optimal migration edge node through the load migration model, achieving efficient load balancing and high availability of the system. Specifically, the system first monitors the calculation delay of each edge node in real time. When it is detected that the calculation delay exceeds the preset threshold, the optimal migration edge node and its migration path are determined through the load migration model. The process includes evaluating the computing resources and load migration weight indicators in the multi-node edge computing array subgraph, selecting the best migration path and target node, and reasonably allocating the load migration amount to ensure that the calculation delay of each node is kept within the threshold range, thereby improving the overall performance and stability of the system. In addition, the method also adopts a path generation algorithm to generate the optimal migration path for multiple indirectly connected multi-node edge computing array subgraphs by comprehensively considering the load migration amount, interaction quality weight indicators and remaining computing resources. This method not only improves the efficiency of load migration, but also ensures the high availability and robustness of the system. Through real-time monitoring and dynamic adjustment, the system can respond to environmental changes and node failures in a timely manner, ensuring that the calculation delay of each edge node always meets the requirements.

[0102] S5. The calculation delay corresponding to each edge node after migration is evaluated in real time through an evaluation algorithm, and the load state corresponding to each edge node is adjusted in real time according to the obtained calculation delay, so that each edge node meets the instruction calculation delay threshold in real time.

[0103] Example 2

[0104] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements a perception decision method based on edge computing when executing the computer program.

[0105] A computer-readable storage medium stores computer instructions, which, when executed, execute a perception decision-making method based on edge computing.

[0106] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are within the protection of the present invention.

[0107] If the disclosed technical solution involves personal information, the product using the disclosed technical solution has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the disclosed technical solution involves sensitive personal information, the product using the disclosed technical solution has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A perception decision method based on edge computing, characterized in that: include: S1. Construct a multi-node edge computing array diagram and configure a load migration model to obtain the current state perception information and task instruction information of n quadruped robots in real time. According to the task instruction information, the configured task decomposition model is used to obtain the n subtask sequences corresponding to the n quadruped robots; the multi-node edge computing array diagram is constructed by the information intra-interaction chain, load migration information chain and all edge nodes obtained through the edge node working state information and interaction information; S2. Obtain the distance from the quadruped robot to each edge node in real time, construct the shortest distance information transmission pair between the quadruped robot and the edge node, and transmit the current state perception information of the quadruped robot and the corresponding subtask to the edge node corresponding to the multi-node edge computing array diagram according to the shortest distance information transmission pair, and construct the perception decision input pair of each quadruped robot; S3, input each quadruped robot perception decision input pair into the distributed reinforcement decision model configured by the corresponding edge node, generate the perception decision instruction corresponding to each quadruped robot, and share the perception decision instruction generated by each edge node in real time through the multi-node edge computing array diagram, generate shared perception decision instruction, and send the generated shared perception decision instruction to the corresponding quadruped robot to control the operating state of the quadruped robot in real time; S4. Set a single-node instruction calculation delay threshold, and monitor the calculation delay corresponding to the shared perception decision instruction of each edge node in real time. When the calculation delay of the corresponding edge node is greater than the calculation delay threshold, the optimal migration edge node, migration path and load migration amount are obtained through the load migration model, and the decision instruction generation task corresponding to the load migration amount of the edge node that does not meet the calculation delay threshold is migrated to the optimal migration edge node.

2. The perception decision method based on edge computing according to claim 1, characterized in that: The steps of constructing the multi-node edge computing array graph include: S101, obtain edge node working status information and interaction information and pre-process them, and set each edge node as a graph node in a multi-node edge computing array graph ,in represents the i-th graph node in the multi-node edge computing array graph, where M is the total number of graph nodes in the multi-node edge computing array graph; S102: Input the edge node location information into the DBSCAN algorithm. The graph nodes that can fully cover each other in the medium signal transmission distance are grouped into one category to obtain the edge computing array graph node subset; S103, using the working status information and interaction information corresponding to the edge computing array graph node subset, constructing the information interaction chain between the two graph nodes in the edge computing array graph node subset, and constructing the multi-node edge computing array subgraph by using the information interaction chain between the two graph nodes and the edge computing array graph node subset ,in represents the A-th multi-node edge computing array subgraph, represents the edge computing array graph node subset in the A-th multi-node edge computing array subgraph, represents the information intra-interaction chain set between edge computing array graph node subsets in the A-th multi-node edge computing array subgraph; S104. Assumptions Graph nodes in and Graph nodes in The closest and graph node Can only be used with graph nodes The signal transmission distances overlap each other, so use and The corresponding interaction information is constructed and The load migration information chain between them; specifically: ， in and express Neutralization The graph nodes are closest and the graph nodes Can only be used with graph nodes The signal transmission distance of the graph node pairs that cover each other, arg represents the inverse function operator, Representing a graph node With graph nodes Mahalanobis distance, express and The maximum effective distance of signal transmission corresponding to all graph nodes; S105. Repeat the process of S104 to construct the load migration information chain between all multi-node edge computing array subgraphs, and use all the obtained load migration information chains and multi-node edge computing array subgraphs to construct a multi-node edge computing array graph.

3. The perception decision method based on edge computing as claimed in claim 2, characterized in that: In the S103, the information intra-interaction chain is configured with an interaction quality weight indicator, and the step of constructing the interaction quality weight indicator includes: S1031, assuming a multi-node edge computing array subgraph A subset of edge computing array graph nodes contained in , using the edge computing array graph node subsets, each pair of graph nodes is used to construct node pairs within the subset, and according to the corresponding interaction information of each node pair within the subset, the interaction quality weight index of the corresponding node pair within the subset is calculated, and the interaction quality weight index is built into the information interaction chain corresponding to the node pair within the corresponding subset; the interaction quality weight index is specifically: ,in express The interaction quality weight index between the node pairs in the subset constructed by the jth and gth graph nodes in The number of nodes in the graph, express The signal transmission bandwidth per unit time between node pairs in the subset constructed by the jth and gth graph nodes in , express The signal transmission delay per unit time between the node pairs in the subset constructed by the j-th and g-th graph nodes; express The signal transmission packet loss rate per unit time between the node pairs in the subset constructed by the j-th and g-th graph nodes; express The length of the signal transmission task queue per unit time between node pairs in the subset constructed by the j-th and g-th graph nodes; and Represents the multi-node edge computing array subgraphs respectively The remaining computing resources of the jth and gth graph nodes at the current moment. When the corresponding graph node has a computing delay exception, the remaining computing resources of the corresponding graph node at the current moment are set to 1. They represent the weighted coefficients of transmission delay, transmission packet loss rate and task queue length respectively.

4. The perception decision method based on edge computing as claimed in claim 3, characterized in that: The step of constructing the interaction quality weight index also includes: S1032, repeat S1031 process to build a multi-node edge computing array subgraph The interaction quality weight indicators between all pairs of nodes in the subset are built into the corresponding information interaction chain; S1033. Repeat processes S1031 and S1032 to construct the interaction quality weight index corresponding to the interaction chain in the corresponding information of all multi-node edge computing array subgraphs, and arrange the sizes of the interaction quality weight index in each multi-node edge computing array subgraph in descending order, and feed back the arrangement results and the working status and instruction calculation delay corresponding to each graph node to the load migration model.

5. The edge computing-based perception decision method according to claim 4, characterized in that: The load migration information chain is configured with a load migration weight indicator; the step of obtaining the optimal migration edge node and migration path in S4 includes: S401, according to S104 process, obtain A collection of directly or indirectly connected multi-node edge computing array subgraphs, assuming that at the current moment Through the S4 process, it is found that there are m graph nodes with abnormal calculation delays, and the current If the remaining computing resources corresponding to the nodes in the non-existent computing delay anomaly graph satisfy the load migration amount corresponding to the nodes in the existing computing delay anomaly graph obtained by the load migration model and the computing delay after migration is less than or equal to the computing delay threshold, then through the constructed information intra-interaction chain, according to the configured interaction quality weight index, the decision instruction generation task of the corresponding computing delay anomaly graph node is migrated to The corresponding optimal migration edge node in ; S402, current time Through the S4 process, it is found that there are m graph nodes with abnormal calculation delays, and If the graph nodes corresponding to the non-computational delay anomaly cannot complete the decision instruction generation task of the graph nodes corresponding to the computational delay anomaly, the multi-node edge computing array subgraph and corresponding to the optimal migration edge node are obtained through the load migration model. There is a calculation of the load migration amount corresponding to the delay graph node; S403, if the multi-node edge computing array subgraph and Directly connected and only one multi-node edge computing array subgraph meets the computing conditions, then directly The decision instruction generation task corresponding to the load migration amount is migrated to the corresponding multi-node edge computing array subgraph.

6. The perception decision method based on edge computing according to claim 5, characterized in that: The step of obtaining the optimal migration edge node and migration path also includes: S404: If a multi-node edge computing array subgraph and If the number of multi-node edge computing array subgraphs that are directly connected and meet the computing conditions is greater than one, the load migration weight index value corresponding to all load migration information chains that meet the computing conditions at the current moment is calculated through the load migration model according to the load migration weight index configured in the corresponding load migration information chain; S405, arrange all calculated load migration weight index values ​​in descending order, migrate the decision instruction generation task corresponding to the load migration amount to the multi-node edge computing array subgraph corresponding to the maximum weight index value, and distribute the decision instruction generation task to the corresponding optimal graph node according to the S401 process; S406: If a multi-node edge computing array subgraph and If the number of multi-node edge computing array subgraphs that are indirectly connected and meet the computing conditions is greater than or equal to one, The corresponding load migration amount, The load migration weight index values ​​corresponding to all load migration information chains and the total amount of remaining available computing resources corresponding to each multi-node edge computing array subgraph that meets the computing conditions are indirectly connected to all multi-node edge computing array subgraphs that meet the computing conditions, and are input into the path generation algorithm to obtain the multi-node edge computing array subgraph corresponding to the optimal migration path; S407. According to the multi-node edge computing array subgraph corresponding to the optimal migration path, the decision instruction generation task corresponding to the load migration amount is migrated to the multi-node edge computing array subgraph corresponding to the maximum weight index value, and the decision instruction generation task is allocated to the corresponding optimal graph node according to the S401 process.

7. The perception decision method based on edge computing according to claim 6, characterized in that: The step of obtaining the optimal migration edge node and migration path also includes: S408: If there is no single multi-node edge computing array subgraph satisfying Calculate the load migration amount corresponding to all graph nodes with abnormal delay in The load migration amount corresponding to each calculation delay abnormal graph node, the interaction quality weight index value corresponding to all information intra-interaction chains, the load migration weight index value corresponding to all load migration information chains, the remaining computing resources and computing delay of the graph nodes without calculation delay abnormalities are input into the path generation algorithm to obtain the migration path of the optimal migration edge node corresponding to each calculation delay abnormal graph node; S409, according to the migration path of the optimal migration edge node corresponding to each calculation delay abnormal graph node, generate the corresponding decision instruction task and migrate it to the corresponding edge node; S410, monitor in real time the computing operation status generated by the decision instructions of each multi-node edge computing array subgraph. When there is a computing anomaly in at least one multi-node edge computing array subgraph, repeat the S401-S410 process to perform load migration, so that the computing delay corresponding to each edge node meets the computing delay threshold in real time.

8. The perception decision method based on edge computing according to claim 7, characterized in that: The steps of constructing the load migration weight index include: Assuming a multi-node edge computing array subgraph The total available computing resources at the current moment are , multi-node edge computing array subgraph The total available computing resources at the current moment are , then the load migration weight index The corresponding calculation formula is: ； in, represents the load migration information chain connection factor, Indicates the corresponding graph node of S104 to obtain the load migration information chain With graph nodes The signal transmission bandwidth between Indicates the corresponding graph node of S104 to obtain the load migration information chain With graph nodes The frequency of interaction between Represents a multi-node edge computing array subgraph China Map Node The amount of computing resources remaining at the current moment, and In turn, it represents the multi-node edge computing array subgraph or multi-node edge computing array subgraph The corresponding load migration amount when there is a calculation delay exception.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the edge computing-based perception decision method described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, a perception decision method based on edge computing as described in any one of claims 1-8 is executed.

Images