Robot remote collaborative scheduling method and system based on artificial intelligence

Abstract

The present invention discloses a robot remote collaborative scheduling method and system based on artificial intelligence, including: dividing complex tasks into subtasks and classifying them, constructing correlation and dependency functions, and then establishing a topological relationship network model, taking the preferred task grouping as the initial state, designing a neural network model for the robot and calculating the next action, selecting the optimal joint action as the collaborative strategy after determining the collaborative goal, generating execution logic for the robot to perform subtasks, and finally learning and optimizing the robot's neural network model through an intelligent agent collaborative network. This method uses intelligent algorithms to improve task execution efficiency and accuracy, achieve efficient collaboration, adapt to complex task scenarios, and has good interpretability, and can be directly applied to a robot remote collaborative scheduling system based on artificial intelligence.

Description

Technical Field

[0001] The present invention relates to the field of multi-agent collaborative control, and in particular to a robot remote collaborative scheduling method and system based on artificial intelligence. Background Art

[0002] With the continuous advancement of science and technology, robots are increasingly used in various fields, from automated production in industrial manufacturing to auxiliary surgery and nursing work in the medical and health field; from cargo handling and sorting in logistics warehousing to detection and rescue missions in dangerous environments. Different types of robots have different functions and characteristics and can play an important role in specific scenarios.

[0003] In recent years, artificial intelligence technology has made great progress. The continuous development of machine learning, deep learning, computer vision, natural language processing and other technologies has provided strong support for the intelligence of robots. Artificial intelligence can enable robots to have autonomous perception, decision-making and learning capabilities, and improve the adaptability and flexibility of robots.

[0004] At the same time, artificial intelligence technology also provides new ideas and methods for the remote collaborative scheduling of robots. In order to meet the requirements of complex tasks and solve the collaboration problems of robots in remote working environments, a remote collaborative scheduling method and system of robots based on artificial intelligence is designed. Summary of the invention

[0005] The purpose of the present invention is to provide a robot remote collaborative scheduling method and system based on artificial intelligence.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0007] The first aspect of the present invention provides a robot remote collaborative scheduling method based on artificial intelligence, comprising:

[0008] S100 divides a complex task into multiple subtasks, performs coupling analysis, and obtains a coupled subtask set and an uncoupled subtask set;

[0009] S200 constructs a correlation function between coupled subtasks, analyzes the dependency relationship between uncoupled subtasks and coupled subtasks, and obtains a dependency function;

[0010] S300 constructs a topological relationship network model based on the correlation function and the dependency function, and constructs a neural network model for the robot according to the topological relationship network model;

[0011] S400 groups the robots into tasks based on preferences as an initial state, and the robots calculate a next action set according to the initial state through their respective neural network models;

[0012] S500 determines the coordination goal and selects the best action combination from the next action set as the coordination strategy;

[0013] S600 generates execution logic based on the collaborative strategy, and the robot executes the subtask according to the execution logic;

[0014] S700 builds an intelligent agent collaborative network to learn the execution logic of each task, and continuously optimizes the neural network model of the robot based on the intelligent agent collaborative network.

[0015] As a further method, the method of constructing a correlation function between coupled subtasks, analyzing the dependency relationship between uncoupled subtasks and coupled subtasks, and obtaining the dependency function includes:

[0016] The correlation function is constructed based on the characteristic parameters between the coupled subtasks, and the expression is:

[0017]

[0018] Among them, n and m are the number of features of coupled subtask A and coupled subtask B, respectively, and w ij for and The degree of correlation, and are the i-th eigenvalue in coupled subtask A and the j-th eigenvalue in coupled subtask B, respectively. and They are and The standard deviation of and The nonlinear influence factor of and The difference attenuation coefficient, and They are and The mean of

[0019] Extract feature data for uncoupled and coupled subtasks, train the model through machine learning algorithm, and obtain the dependency function, which is expressed as:

[0020] D=σ2(W2·σ1(W1·[F NC , F C ]+b1)+b2)

[0021] Among them, σ1 and σ2 are the nonlinear activation function of the hidden layer and the linear activation function of the output layer, W1 and W2 are the weights from the input layer to the hidden layer and the weights from the hidden layer to the output layer, b1 and b2 are the biases of the hidden layer and the output layer, respectively.

[0022] As a further method, the method for constructing a topological relationship network model based on the association function and the dependency function includes: taking each subtask as a node, establishing undirected edges between coupled subtasks, determining the weight of the edge according to the value of the association function, establishing directed edges between uncoupled subtasks and coupled subtasks, and determining the weight of the edge according to the dependency function.

[0023] As a further method, the method of constructing a neural network model for a robot according to the topological relationship network model includes:

[0024] The priority of the subtask is calculated based on the topological relationship network model, and the expression is:

[0025]

[0026] Among them, P(v i ) is the subtask v i The priority of U i v i The importance of I i v i The number of factors affecting the remaining subtasks, R i v i The resource requirements of C i v i The completion cost, T i v i The time required to complete, V is the subtask set, W is ij Represented as v i With subtask v j The weight between in (v i ) is v i In the topological relationship network model, the in-degree, D out (v j ) is represented by v j In the topological relationship network model, the out-degree, S ij v i and v j The degree of similarity in characteristic parameters;

[0027] Define decision rules based on subtask priorities, specifically: execute subtasks in descending order of priority, and allocate more resources to high-priority subtasks; execute high-priority tasks first in case of task conflicts; and immediately raise the priority of urgent tasks and execute them when urgent tasks occur;

[0028] A neural network model is constructed for the robot based on a feedforward neural network. The priority of the subtask is used as the weight. The number of nodes in the input layer corresponds to the number of features acquired by the robot, the number of nodes in the output layer corresponds to the robot action dimension, and the number of nodes in the hidden layer is set to half of the sum of the number of nodes in the input layer and the output layer.

[0029] According to the decision rule, the difference between the output and the expectation is obtained, and back propagation is performed to calculate the gradient from the output layer to the input layer and adjust the model parameters, and iterate until the model meets the decision rule.

[0030] As a further method, the method of grouping robots into tasks based on preference as an initial state includes:

[0031] Determine the capabilities and preference characteristics of the robots, build a preference model for each robot using a decision tree algorithm, and use a linear programming method to assign subtasks to each robot based on the robot's preference model and the attributes of the subtasks;

[0032] The expression of the objective function in the linear programming method is:

[0033]

[0034] Among them, R is the total number of robots, T is the total number of subtasks, Pr t represents the basic benefit of robot r when performing subtask t, x rt is the execution decision variable, when x rt =1, indicating that robot r is assigned to perform subtask t, otherwise x ij =0,q rt represents the efficiency of robot r when performing subtask t, y rt is a preference decision variable, when y rt =1, indicating that robot r prefers to perform subtask t, otherwise y rt =0, F is the robot's preferred feature set, u rf It is expressed as the preference strength of robot r for feature f, w rf is a characteristic decision variable, when w rf = 1, indicating that feature f is activated in the preference model of robot r, otherwise w rf =0, v r is the load level of robot r;

[0035] The constraints in the linear programming method include:

[0036]

[0037] Among them, h rt The amount of resources required for robot r to perform subtask t, H ris the total resources available to robot r, l r is the maximum task execution capability of robot r, L is the load balance coefficient, C r is the maximum task capacity of robot r;

[0038] Define the initial state for each robot, including the robot's subtask allocation, current location information, and current task state information.

[0039] As a further method, the method of selecting the optimal action combination from the next action set as the collaborative strategy includes:

[0040] The joint action set that achieves the collaborative goal is obtained from the next action of each robot. The optimal action combination is selected from the joint action set with the goal of maximizing the task completion benefit and maximizing the collaborative benefit. The evaluation function of the task completion benefit is expressed as:

[0041]

[0042] Where T is the total number of subtasks, n is the number of completed subtasks, and R max is the maximum subtask completion rate, T max is the longest subtask completion time, T min is the shortest completion time of the subtask, C min is the minimum subtask cost, C max is the highest subtask cost, R is the total number of robots, C rj is the operating cost of the jth robot, K is the number of resources, C rk is the cost of using the kth resource, Q is the subtask completion quality evaluation score, P is the total power consumption of the robot, η max For the ideal maximum system efficiency;

[0043] The evaluation function of the synergistic benefit is expressed as:

[0044]

[0045] Where E is the total number of time steps, γ e is the discount factor at time step e, R is the total number of robots, s e is the global state of the robot at time step e, is the action of robot i at time step e, For robot i in state s e Take action is the immediate benefit when , λ is the weight parameter of the synergistic benefit, Robots i and j are in state s at time step e The following collaborative groups take action The immediate benefit of E * is the final time step, For robot i at the final time step E * Actions, For robot i at the final time step E * terminal cost.

[0046] As a further method, the method for the robot to execute subtasks according to the execution logic includes: introducing a dynamic task allocation mechanism between robots, the robots regularly update status information, and adjust task allocation through a neural network model, wherein when a robot failure occurs and a new task is added, the subtask reallocation is triggered.

[0047] As a further method, the method of constructing an intelligent agent collaboration network to learn the execution logic of each task includes:

[0048] During the execution of the subtask, the state, action, observation and reward of each robot are collected;

[0049] Calculate the policy gradient of the robot's execution logic each time, build an agent collaboration network based on the decision tree, and train the agent collaboration network through reinforcement learning according to the policy gradient. The calculation formula of the policy gradient is:

[0050]

[0051] Where θ is the parameter of strategy π, J(π) is the cumulative return of strategy π, N is the total number of robots participating in the collaboration, is the action taken by robot i at time step t, is the state of robot i at time step t, is the state at time step t Next, select actions according to strategy π The probability distribution of is the expected value of strategy π, γ t is the discount factor at time step t, is the reward obtained by robot i at time step t.

[0052] As a further method, the method for continuously optimizing the neural network model of the robot based on the intelligent agent collaborative network includes:

[0053] Transfer learning is used to transfer the features learned by the agent collaboration network to the neural network model, where the expression of the output of the migrated neural network model is:

[0054]

[0055] Among them, A is the neural network model, f is the activation function of the model output layer, n is the number of layers of the model, σ is the activation function of the model hidden layer, α is the transfer learning rate, is the weight matrix of the i-th layer in the original neural network model, is the weight matrix feature transferred from the agent collaboration network, O C is the output of the agent collaboration network, h i -1 is the input from the i-1th layer to the ith layer of the model, is the bias vector of the i-th layer of the model, is the bias vector feature transferred from the agent collaboration network;

[0056] The neural network model is adjusted according to the structure of the decision tree in the intelligent agent collaborative network. Specifically, a neuron layer corresponding to the internal node characteristics of the decision tree is designed in the neural network model, different neuron paths are activated according to the characteristic values, and subsequent nodes are activated according to the output of the previous layer and the current characteristics, with the leaf node weight as the output.

[0057] The second aspect of the present invention provides a robot remote collaborative scheduling system based on artificial intelligence, comprising:

[0058] Task division module, used to divide a complex task into multiple subtasks, conduct coupling analysis, and obtain coupled subtask sets and uncoupled subtask sets;

[0059] The correlation building module is used to build the correlation function between the coupled subtasks, analyze the dependency relationship between the uncoupled subtasks and the coupled subtasks, and obtain the dependency function;

[0060] A network modeling module, used to construct a topological relationship network model based on the correlation function and the dependency function, and to construct a neural network model for the robot according to the topological relationship network model;

[0061] A preference grouping module, used to group robots into tasks based on preferences, as an initial state, and the robots calculate the next action set according to the initial state through their respective neural network models;

[0062] A strategy selection module is used to determine the coordination goal and select the best action combination from the next action set as the coordination strategy;

[0063] A logic execution module, used to generate execution logic based on the collaborative strategy, and the robot dynamically executes subtasks according to the execution logic;

[0064] The network optimization module is used to build an intelligent agent collaborative network to learn the execution logic of each task, and continuously optimize the neural network model of the robot based on the intelligent agent collaborative network.

[0065] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0066] (1) The present invention divides a complex task into multiple subtasks and performs coupling analysis to obtain a coupled subtask set and an uncoupled subtask set, and then constructs a correlation function for the coupled subtasks, which helps to better coordinate and manage these interrelated subtasks and ensure their collaborative work.

[0067] (2) By determining the collaborative goal and generating a set of joint actions, the present invention can find the optimal joint action as a collaborative strategy to ensure that the collaborative work between robots achieves the best effect. Based on the collaborative strategy, the execution logic is generated so that the robots can accurately perform the corresponding tasks, thereby improving the reliability and stability of task execution.

[0068] (3) The present invention can continuously optimize the neural network model of the agent by building an agent collaborative network to learn the execution logic of the optimal strategy for each task. This continuous optimization process can enable the agent to continuously improve its performance and adaptability when facing different tasks and environments, thereby improving the efficiency and reliability of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 The present invention is a flowchart of the steps of a robot remote collaborative scheduling method based on artificial intelligence in an embodiment of the present invention.

[0070] Figure 2 It is a logical schematic diagram of a robot remote collaborative scheduling system based on artificial intelligence in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0072] Reference Figure 1 As shown, the present invention provides a robot remote collaborative scheduling method based on artificial intelligence, comprising:

[0073] In a specific implementation case, the radiation area of ​​the nuclear medicine department of a large hospital needs to be cleaned up for equipment maintenance. The radiation area is about 70 square meters, and a total of 10 key equipment need to be inspected and maintained. Among them, 3 are judged to be likely to be severely contaminated by radiation, 5 are likely to be slightly contaminated, and the contamination status of 2 is not yet determined;

[0074] S100 divides a complex task into multiple subtasks, performs coupling analysis, and obtains a coupled subtask set and an uncoupled subtask set;

[0075] In actual evaluation, complex tasks are divided into multiple subtasks:

[0076] Subtask 1: Conduct comprehensive radiation monitoring of the radiation area and record the radiation value every 5 minutes, for a total of 100 data records;

[0077] Subtask 2: Use advanced sensor technology to identify contaminated equipment and determine the degree of contamination. It is estimated that it will take 3 minutes to identify each device.

[0078] Subtask 3: Clean the lightly contaminated equipment. The cleaning time for each equipment is about 20 minutes.

[0079] Subtask 4: dismantle and transport heavily contaminated equipment to a specific treatment area. The dismantling and transporting time for each piece of equipment is approximately 40 minutes.

[0080] In the actual evaluation, the coupling analysis of the subtasks was carried out to obtain coupled subtask sets and uncoupled subtask sets: Coupled subtask set: Subtask two and subtask four are highly coupled, because only by accurately identifying heavily polluted equipment can efficient disassembly and transportation be carried out; Uncoupled subtask set: Subtask one is relatively independent and has a low coupling with other subtasks.

[0081] S200 constructs a correlation function between coupled subtasks, analyzes the dependency relationship between uncoupled subtasks and coupled subtasks, and obtains a dependency function;

[0082] It needs to be explained that the correlation function of coupled subtasks can help us understand how closely related tasks affect each other. For example, the correlation between certain nursing tasks in medical care helps to arrange resources as a whole, while the dependency function of uncoupled and coupled subtasks can determine the degree of dependency of different types of tasks, which is convenient for reasonable scheduling of robots, providing an accurate basis for subsequent model building and decision-making, and achieving more efficient collaborative scheduling;

[0083] In the actual evaluation, for subtask 2 and subtask 4, a correlation function is constructed; subtask 1 and subtask 2 have a certain dependency relationship, and a dependency function is constructed.

[0084] S300 constructs a topological relationship network model based on the correlation function and the dependency function, and constructs a neural network model for the robot according to the topological relationship network model;

[0085] It should be explained that building a neural network model provides a tool for robots to make intelligent decisions. Building a neural network based on a topological relationship network model enables the robot to fully understand the complex relationship between various subtasks, including the degree of tightness of coupled subtasks and the direction and degree of dependence between uncoupled subtasks and coupled subtasks. Through this model, the robot can calculate the next set of actions based on various input feature information, so as to perform tasks and conduct collaborative scheduling more efficiently;

[0086] In the actual evaluation, a topological relationship network model is constructed based on the correlation function and the dependency function. The four subtasks are used as nodes, and the connection relationship and weight between the nodes are determined according to the function. The connection weight from subtask 2 to subtask 4 is higher, while the connection weight from subtask 1 to other subtasks is relatively low; a neural network model is constructed for three different types of robots. The input includes the current task status, the position and action information of other robots, etc., and the output is the probability distribution of the next action.

[0087] S400 groups the robots into tasks based on preferences as an initial state, and the robots calculate a next action set according to the initial state through their respective neural network models;

[0088] It needs to be explained that different robots have different capabilities and preferences. If tasks are randomly assigned without considering these factors, the robots may not be able to fully utilize their advantages, and may even perform tasks poorly. Calculating the next action set based on the initial state can give the robot a clear direction and plan during the task execution process, better adapt to the dynamically changing task environment, and achieve more intelligent collaborative scheduling;

[0089] In the actual evaluation, there are three robots, A, B, and C. Robot A is good at radiation monitoring and is responsible for subtask one; Robot B is good at equipment identification and is responsible for subtask two; Robot C is good at equipment cleaning and handling and is responsible for subtask four, and assists in subtask three when necessary; According to the neural network model, the next action is calculated, and the action of robot A is to move to different locations for monitoring and send monitoring data. The actions of robot B include scanning different equipment and determining the degree of contamination. The actions of robot C include approaching the equipment, disassembling or carrying.

[0090] S500 determines the coordination goal and selects the best action combination from the next action set as the coordination strategy;

[0091] In the actual evaluation, the collaborative goal is to complete equipment maintenance and cleaning work in the radiation area within 4 hours, while ensuring that the total amount of radiation received by the robot does not exceed the safety threshold of 500 units; through the optimization algorithm, the optimal action combination is selected from the robot's next action set. Among them, when robot B identifies a heavily contaminated equipment, it immediately notifies robot C to go for disassembly and transportation. At the same time, robot A adjusts the monitoring position to better monitor the radiation changes in the area.

[0092] S600 generates execution logic based on the collaborative strategy, and the robot executes the subtask according to the execution logic;

[0093] In the actual evaluation, the execution logic is generated according to the optimal action combination, including the action sequence and timing of each robot. Robot A first monitors at the edge of the radiation area and then gradually moves to the center; Robot B scans the equipment in a certain order; Robot C immediately goes to the designated equipment for processing after receiving the notification; The robot dynamically executes subtasks according to the execution logic. During the execution process, if there is an unexpected situation, such as equipment failure or a sudden increase in radiation values, the robot will recalculate the next action according to the new situation and adjust the execution logic.

[0094] S700 builds an intelligent agent collaborative network to learn the execution logic of each task, and continuously optimizes the neural network model of the robot based on the intelligent agent collaborative network.

[0095] It should be explained that during the task execution, the status, action, observation and reward information of each robot are collected, and these data are used to understand the robot's behavior and task completion in different situations. In this way, we can continuously summarize experience and find better execution strategies;

[0096] In the actual evaluation, the intelligent agent collaboration network: records the execution logic and results of each task, forms an intelligent agent collaboration network, and records the robot's action selection, task completion time, radiation exposure and other data in different situations; model optimization: uses the data in the intelligent agent collaboration network to continuously optimize the robot's neural network model. Through continuous training and adjustment, the robot's collaborative ability and efficiency in future tasks will be improved;

[0097] Specifically, by introducing an artificial intelligence-based robot remote collaborative scheduling method and system, efficiency has been significantly improved, specifically: task completion time has been shortened by 30%. Before the introduction of the system, it took an average of 10 hours to complete all medical tasks. After the introduction of the system, the average completion time was shortened to 7 hours; resource utilization has increased by 25%. Through reasonable task allocation and dynamic adjustment, the utilization rate of medical resources has been effectively improved, reducing resource waste; the robot fault response time has been shortened by 40%. When a robot fault occurs, the system can quickly reallocate tasks, and the fault response time has been shortened from the original 30 minutes to 10 minutes.

[0098] In this embodiment, the method of constructing a correlation function between coupled subtasks, analyzing the dependency relationship between uncoupled subtasks and coupled subtasks, and obtaining a dependency function includes:

[0099] The correlation function is constructed based on the characteristic parameters between the coupled subtasks, and the expression is:

[0100]

[0101] Among them, n and m are the number of features of coupled subtask A and coupled subtask B, respectively, and w ij for and The degree of correlation, and are the i-th eigenvalue in coupled subtask A and the j-th eigenvalue in coupled subtask B, respectively. and They are and The standard deviation of and The nonlinear influence factor of and The difference attenuation coefficient, and They are and The mean of

[0102] Extract feature data for uncoupled and coupled subtasks, train the model through machine learning algorithm, and obtain the dependency function, which is expressed as:

[0103] D=σ2(W2·σ1(W1·[F NC , F C ]+b1)+b2)

[0104] Among them, σ1 and σ2 are the nonlinear activation function of the hidden layer and the linear activation function of the output layer, W1 and W2 are the weights from the input layer to the hidden layer and the weights from the hidden layer to the output layer, b1 and b2 are the biases of the hidden layer and the output layer, respectively.

[0105] In this embodiment, the method for constructing a topological relationship network model based on the association function and the dependency function includes: taking each subtask as a node, establishing undirected edges between coupled subtasks, determining the weight of the edge according to the value of the association function, establishing directed edges between uncoupled subtasks and coupled subtasks, and determining the weight of the edge according to the dependency function.

[0106] In this embodiment, the method of constructing a neural network model for a robot according to the topological relationship network model includes:

[0107] The priority of the subtask is calculated based on the topological relationship network model, and the expression is:

[0108]

[0109] Among them, P(v i ) is the subtask v i The priority of U i v i The importance of I i v i The number of factors affecting the remaining subtasks, R i v i The resource requirements of C i v i The completion cost, T i v i The time required to complete, V is the subtask set, W is ij Represented as v i With subtask v j The weight between in (v i ) is v i In the topological relationship network model, the in-degree, D out (v j ) is represented by v j In the topological relationship network model, the out-degree, S ij v i and v j The degree of similarity in characteristic parameters;

[0110] Define decision rules based on subtask priorities, specifically: execute subtasks in descending order of priority, and allocate more resources to high-priority subtasks; execute high-priority tasks first in case of task conflicts; and immediately raise the priority of urgent tasks and execute them when urgent tasks occur;

[0111] A neural network model is constructed for the robot based on a feedforward neural network. The priority of the subtask is used as the weight. The number of nodes in the input layer corresponds to the number of features acquired by the robot, the number of nodes in the output layer corresponds to the robot action dimension, and the number of nodes in the hidden layer is set to half of the sum of the number of nodes in the input layer and the output layer.

[0112] According to the decision rule, the difference between the output and the expectation is obtained, and back propagation is performed to calculate the gradient from the output layer to the input layer and adjust the model parameters, and iterate until the model meets the decision rule.

[0113] In this embodiment, the method of grouping robots into tasks based on preferences as an initial state includes:

[0114] Determine the capabilities and preference characteristics of the robots, build a preference model for each robot using a decision tree algorithm, and use a linear programming method to assign subtasks to each robot based on the robot's preference model and the attributes of the subtasks;

[0115] The expression of the objective function in the linear programming method is:

[0116]

[0117] Among them, R is the total number of robots, T is the total number of subtasks, Pr t represents the basic benefit of robot r when performing subtask t, x rt is the execution decision variable, when x rt =1, indicating that robot r is assigned to perform subtask t, otherwise x ij =0,q rt represents the efficiency of robot r when performing subtask t, y rt is a preference decision variable, when y rt =1, indicating that robot r prefers to perform subtask t, otherwise y rt =0, F is the robot's preferred feature set, u rf It is expressed as the preference strength of robot r for feature f, w rf is a characteristic decision variable, when w rf = 1, indicating that feature f is activated in the preference model of robot r, otherwise w rf =0, v r is the load level of robot r;

[0118] The constraints in the linear programming method include:

[0119]

[0120] Among them, h rt The amount of resources required for robot r to perform subtask t, Hr is the total resources available to robot r, l r is the maximum task execution capability of robot r, L is the load balance coefficient, C r is the maximum task capacity of robot r;

[0121] Define the initial state for each robot, including the robot's subtask allocation, current location information, and current task state information.

[0122] In this embodiment, the method of selecting the optimal action combination from the next action set as the collaborative strategy includes:

[0123] The joint action set that achieves the collaborative goal is obtained from the next action of each robot. The optimal action combination is selected from the joint action set with the goal of maximizing the task completion benefit and maximizing the collaborative benefit. The evaluation function of the task completion benefit is expressed as:

[0124]

[0125] Where T is the total number of subtasks, n is the number of completed subtasks, and R max is the maximum subtask completion rate, T max is the longest subtask completion time, T min is the shortest completion time of the subtask, C min is the minimum subtask cost, C max is the highest subtask cost, R is the total number of robots, C rj is the operating cost of the jth robot, K is the number of resources, C rk is the cost of using the kth resource, Q is the subtask completion quality evaluation score, P is the total power consumption of the robot, η max For the ideal maximum system efficiency;

[0126] The evaluation function of the synergistic benefit is expressed as:

[0127]

[0128] Where E is the total number of time steps, γ e is the discount factor at time step e, R is the total number of robots, s e is the global state of the robot at time step e, is the action of robot i at time step e, For robot i in state s e Take action is the immediate benefit when , λ is the weight parameter of the synergistic benefit, Robots i and j are in state s at time step e The following collaborative groups take action The immediate benefit of E * is the final time step, For robot i at the final time step E * Actions, For robot i at the final time step E * terminal cost.

[0129] In this embodiment, the method for the robot to execute subtasks according to the execution logic includes: introducing a dynamic task allocation mechanism between robots, the robots regularly update status information, and adjust task allocation through a neural network model, wherein when a robot failure occurs and a new task is added, the subtask reallocation is triggered.

[0130] In this embodiment, the method of constructing an agent collaboration network to learn the execution logic of each task includes:

[0131] During the execution of the subtask, the state, action, observation and reward of each robot are collected;

[0132] Calculate the policy gradient of the robot's execution logic each time, build an agent collaboration network based on the decision tree, and train the agent collaboration network through reinforcement learning according to the policy gradient. The calculation formula of the policy gradient is:

[0133]

[0134] Where θ is the parameter of strategy π, J(π) is the cumulative return of strategy π, N is the total number of robots participating in the collaboration, is the action taken by robot i at time step t, is the state of robot i at time step t, is the state at time step t Next, select actions according to strategy π The probability distribution of is the expected value of strategy π, γ t is the discount factor at time step t, is the reward obtained by robot i at time step t.

[0135] In this embodiment, the method for continuously optimizing the neural network model of the robot based on the intelligent agent collaborative network includes:

[0136] Transfer learning is used to transfer the features learned by the agent collaboration network to the neural network model, where the expression of the output of the migrated neural network model is:

[0137]

[0138] Among them, A is the neural network model, f is the activation function of the model output layer, n is the number of layers of the model, σ is the activation function of the model hidden layer, α is the transfer learning rate, is the weight matrix of the i-th layer in the original neural network model, is the weight matrix feature transferred from the agent collaboration network, O C is the output of the agent collaboration network, h i -1 is the input from the i-1th layer to the ith layer of the model, is the bias vector of the i-th layer of the model, is the bias vector feature transferred from the agent collaboration network;

[0139] The neural network model is adjusted according to the structure of the decision tree in the intelligent agent collaborative network. Specifically, a neuron layer corresponding to the internal node characteristics of the decision tree is designed in the neural network model, different neuron paths are activated according to the characteristic values, and subsequent nodes are activated according to the output of the previous layer and the current characteristics, with the leaf node weight as the output.

[0140] The second aspect of the present invention provides a robot remote collaborative scheduling system based on artificial intelligence, comprising:

[0141] Task division module, used to divide a complex task into multiple subtasks, conduct coupling analysis, and obtain coupled subtask sets and uncoupled subtask sets;

[0142] The correlation building module is used to build the correlation function between the coupled subtasks, analyze the dependency relationship between the uncoupled subtasks and the coupled subtasks, and obtain the dependency function;

[0143] A network modeling module, used to construct a topological relationship network model based on the correlation function and the dependency function, and to construct a neural network model for the robot according to the topological relationship network model;

[0144] A preference grouping module, used to group robots into tasks based on preferences, as an initial state, and the robots calculate the next action set according to the initial state through their respective neural network models;

[0145] A strategy selection module is used to determine the coordination goal and select the best action combination from the next action set as the coordination strategy;

[0146] A logic execution module, used to generate execution logic based on the collaborative strategy, and the robot dynamically executes subtasks according to the execution logic;

[0147] The network optimization module is used to build an intelligent agent collaborative network to learn the execution logic of each task, and continuously optimize the neural network model of the robot based on the intelligent agent collaborative network.

[0148] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A robot remote collaborative scheduling method based on artificial intelligence, characterized in that: The following steps are involved: Divide a complex task into multiple subtasks, perform coupling analysis, and obtain a coupled subtask set and an uncoupled subtask set; Construct a correlation function between coupled subtasks, analyze the dependency between uncoupled subtasks and coupled subtasks, and obtain a dependency function; Building a topological relationship network model based on the correlation function and the dependency function, and building a neural network model for the robot according to the topological relationship network model; The robots are grouped into tasks based on preferences as initial states, and the robots calculate the next action set according to the initial states through their respective neural network models; Determine the coordination goal, and select the best action combination from the next action set as the coordination strategy; Generate execution logic based on the collaborative strategy, and the robot executes the subtask according to the execution logic; An intelligent agent collaboration network is constructed to learn the execution logic of each task, and the neural network model of the robot is continuously optimized based on the intelligent agent collaboration network.

2. The method for remote collaborative scheduling of robots based on artificial intelligence according to claim 1, characterized in that: The method of constructing a correlation function between coupled subtasks, analyzing the dependency relationship between uncoupled subtasks and coupled subtasks, and obtaining a dependency function includes: The correlation function is constructed based on the characteristic parameters between the coupled subtasks, and the expression is: Among them, n and m are the number of features of coupled subtask A and coupled subtask B, respectively, and w ij for and The degree of correlation, and are the i-th eigenvalue in coupled subtask A and the j-th eigenvalue in coupled subtask B, respectively. and They are and The standard deviation of and The nonlinear influence factor of and The difference attenuation coefficient, and They are and The mean of Extract feature data for uncoupled and coupled subtasks, train the model through machine learning algorithm, and obtain the dependency function, which is expressed as: D=σ2(W2·σ1(W1·[F NC ,F C ]+b1)+b2) Among them, σ1 and σ2 are the nonlinear activation function of the hidden layer and the linear activation function of the output layer, W1 and W2 are the weights from the input layer to the hidden layer and the weights from the hidden layer to the output layer, b1 and b2 are the biases of the hidden layer and the output layer, respectively.

3. The method for remote collaborative scheduling of robots based on artificial intelligence according to claim 1, characterized in that: The method for constructing a topological relationship network model based on the association function and the dependency function includes: taking each subtask as a node, establishing undirected edges between coupled subtasks, determining the weight of the edge according to the value of the association function, establishing directed edges between uncoupled subtasks and coupled subtasks, and determining the weight of the edge according to the dependency function.

4. The method for remote collaborative scheduling of robots based on artificial intelligence according to claim 1, characterized in that: The method for constructing a neural network model for a robot according to the topological relationship network model comprises: The priority of the subtask is calculated based on the topological relationship network model, and the expression is: Among them, P(v i ) is the subtask v i The priority of U i v i The importance of I i v i The number of factors affecting the remaining subtasks, R i v i The resource requirements of C i v i The completion cost, T i v i The time required to complete, V is the subtask set, W ij Represented as v i With subtask v j The weight between in (v i ) is v i In the topological relationship network model, the in-degree, D out (v j ) is represented by v j In the topological relationship network model, the out-degree, S ij v i and v j The degree of similarity in characteristic parameters; Define decision rules based on subtask priorities, specifically: execute subtasks in descending order of priority, and allocate more resources to high-priority subtasks; execute high-priority tasks first in case of task conflicts; and immediately raise the priority of urgent tasks and execute them when urgent tasks occur; A neural network model is constructed for the robot based on a feedforward neural network. The priority of the subtask is used as the weight. The number of nodes in the input layer corresponds to the number of features acquired by the robot, the number of nodes in the output layer corresponds to the robot action dimension, and the number of nodes in the hidden layer is set to half of the sum of the number of nodes in the input layer and the output layer. According to the decision rule, the difference between the output and the expectation is obtained, and back propagation is performed to calculate the gradient from the output layer to the input layer and adjust the model parameters, and iterate until the model meets the decision rule.

5. The method for remote collaborative scheduling of robots based on artificial intelligence according to claim 1, characterized in that: The method of grouping robots into tasks based on preference as an initial state includes: Determine the capabilities and preference characteristics of the robots, build a preference model for each robot using a decision tree algorithm, and use a linear programming method to assign subtasks to each robot based on the robot's preference model and the attributes of the subtasks; The expression of the objective function in the linear programming method is: Among them, R is the total number of robots, T is the total number of subtasks, and p rt represents the basic benefit of robot r when performing subtask t, x rt is the execution decision variable, when x rt =1, indicating that robot r is assigned to perform subtask t, otherwise x ij =0,q rt represents the efficiency of robot r when performing subtask t, y rt is a preference decision variable, when y rt =1, indicating that robot r prefers to perform subtask t, otherwise y rt =0, F is the robot's preferred feature set, u rf It is expressed as the preference strength of robot r for feature f, w rf is a characteristic decision variable, when w rf = 1, indicating that feature f is activated in the preference model of robot r, otherwise w rf =0,v r is the load level of robot r; The constraints in the linear programming method include: Among them, h rt The amount of resources required for robot r to perform subtask t, H r is the total resources available to robot r, l r is the maximum task execution capability of robot r, L is the load balance coefficient, C r is the maximum task capacity of robot r; Define the initial state for each robot, including the robot's subtask allocation, current location information, and current task state information.

6. The method for remote collaborative scheduling of robots based on artificial intelligence according to claim 1, characterized in that: The method of selecting the optimal action combination from the next action set as the collaborative strategy includes: The joint action set that achieves the collaborative goal is obtained from the next action of each robot. The optimal action combination is selected from the joint action set with the goal of maximizing the task completion benefit and maximizing the collaborative benefit. The evaluation function of the task completion benefit is expressed as: Where T is the total number of subtasks, n is the number of completed subtasks, and R max is the maximum subtask completion rate, T max is the longest subtask completion time, T min is the shortest completion time of the subtask, C min is the minimum subtask cost, C max is the highest subtask cost, R is the total number of robots, C rj is the operating cost of the jth robot, K is the number of resources, C rk is the cost of using the kth resource, Q is the subtask completion quality evaluation score, P is the total power consumption of the robot, η max For the ideal maximum system efficiency; The evaluation function of the synergistic benefit is expressed as: Where E is the total number of time steps, γ e is the discount factor at time step e, R is the total number of robots, s e is the global state of the robot at time step e, is the action of robot i at time step e, For robot i in state s e Take action is the immediate benefit when , λ is the weight parameter of the synergistic benefit, Robots i and j are in state s at time step e The following collaborative groups take action The immediate benefit of E * is the final time step, For robot i at the final time step E * Actions, is the terminal cost E* of robot i at the final time step.

7. The method for remote collaborative scheduling of robots based on artificial intelligence according to claim 1, characterized in that: The method for the robot to execute subtasks according to the execution logic includes: introducing a dynamic task allocation mechanism between robots, the robots regularly update status information, and adjust task allocation through a neural network model, wherein when a robot failure occurs and a new task is added, subtask reallocation is triggered.

8. The method for remote collaborative scheduling of robots based on artificial intelligence according to claim 1, characterized in that: The method of constructing an intelligent agent collaborative network to learn the execution logic of each task includes: During the execution of the subtask, the state, action, observation and reward of each robot are collected; Calculate the policy gradient of the robot's execution logic each time, build an agent collaboration network based on the decision tree, and train the agent collaboration network through reinforcement learning according to the policy gradient. The calculation formula of the policy gradient is: Where θ is the parameter of strategy π, J(π) is the cumulative return of strategy π, N is the total number of robots participating in the collaboration, is the action taken by robot i at time step t, is the state of robot i at time step t, is the state at time step t Next, select actions according to strategy π The probability distribution of is the expected value of strategy π, γ t is the discount factor at time step t, is the reward obtained by robot i at time step t.

9. The method for remote collaborative scheduling of robots based on artificial intelligence according to claim 1, characterized in that: The method for continuously optimizing the neural network model of the robot based on the intelligent agent collaborative network includes: Transfer learning is used to transfer the features learned by the agent collaboration network to the neural network model, where the expression of the output of the migrated neural network model is: Among them, A is the neural network model, f is the activation function of the model output layer, n is the number of layers of the model, σ is the activation function of the model hidden layer, α is the transfer learning rate, is the weight matrix of the i-th layer in the original neural network model, is the weight matrix feature transferred from the agent collaboration network, O C is the output of the agent collaboration network, h i-1 is the input from the i-1th layer to the ith layer of the model, is the bias vector of the i-th layer of the model, is the bias vector feature transferred from the agent collaboration network; The neural network model is adjusted according to the structure of the decision tree in the intelligent agent collaborative network. Specifically, a neuron layer corresponding to the internal node characteristics of the decision tree is designed in the neural network model, different neuron paths are activated according to the characteristic values, and subsequent nodes are activated according to the output of the previous layer and the current characteristics, with the leaf node weight as the output.

10. A robot remote collaborative scheduling system based on artificial intelligence, used to execute the robot remote collaborative scheduling system based on artificial intelligence according to any one of claims 1 to 9, characterized in that: The system comprises: Task division module, used to divide a complex task into multiple subtasks, conduct coupling analysis, and obtain coupled subtask sets and uncoupled subtask sets; The correlation building module is used to build the correlation function between the coupled subtasks, analyze the dependency relationship between the uncoupled subtasks and the coupled subtasks, and obtain the dependency function; A network modeling module, used to construct a topological relationship network model based on the correlation function and the dependency function, and to construct a neural network model for the robot according to the topological relationship network model; A preference grouping module, used to group the robots into tasks based on preferences, as an initial state, and the robots calculate the next action set according to the initial state through their respective neural network models; A strategy selection module is used to determine the coordination goal and select the best action combination from the next action set as the coordination strategy; A logic execution module, used to generate execution logic based on the collaborative strategy, and the robot dynamically executes subtasks according to the execution logic; The network optimization module is used to build an intelligent agent collaborative network to learn the execution logic of each task, and continuously optimize the neural network model of the robot based on the intelligent agent collaborative network.

Images