An intelligent scheduling method, device and equipment for an AGV and a medium

Abstract

The application discloses an intelligent scheduling method, device and equipment of an AGV and a medium, and belongs to the technical field of AGV scheduling. The method comprises the following steps: obtaining an order to be processed at a current time, reading a starting point coordinate and an ending point coordinate in transportation information in the order to be processed; identifying state information of all AGVs in a current scheduling system, taking at least one AGV with the state information as an idle state as an AGV to be scheduled, and obtaining real-time coordinates of the AGV to be scheduled; inputting the starting point coordinate, the ending point coordinate and the real-time coordinates into a pre-constructed AGV scheduling model, determining a scheduling strategy for the AGV to be scheduled according to an output result of the AGV scheduling model; and wherein the AGV scheduling model is a proxy hybrid neural network model. Through the technical solution, the applicability, reliability and scheduling calculation efficiency of AGV scheduling can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of AGV scheduling, and particularly relates to an intelligent scheduling method, device and equipment of AGV and a medium. BACKGROUND

[0002] AGV (Automated Guided Vehicle) refers to an unmanned and automated vehicle with an automatic guiding device such as a magnetic strip, a track or a laser, which travels along a planned path, is powered by a battery, and is equipped with safety protection and various auxiliary mechanisms. Usually, a plurality of AGVs, a control computer (console), a navigation device, a charging device and peripheral auxiliary devices form an AGV system, and the main working principle is that the AGV can accurately travel according to the specified path under the monitoring and task scheduling of the control computer, and complete a series of work tasks after reaching the designated position.

[0003] In related technologies, the scheduling mode of AGV is mainly global scheduling. Global scheduling mainly calculates the transportation paths of all AGVs before the AGVs execute tasks, and schedules according to the transportation paths of the AGVs and the idle conditions. In order to solve the scheduling problem with multiple target points, group intelligence algorithms such as ant colony algorithm, genetic algorithm and particle swarm algorithm are mainly used.

[0004] Since global scheduling needs to be performed in a scenario with a known and fixed map, and the material points and transportation destinations may change during the actual execution of the AGV tasks, the application scope of the prior art is small, and since the group intelligence algorithms have the disadvantages of slow convergence speed, poor local search ability, many control variables and no definite termination criterion, the use of the prior art has the problems of large computing cost and low computing efficiency. SUMMARY

[0005] The embodiments of the application provide an intelligent scheduling method, device, equipment and medium of AGV, solve the problems of poor applicability and low computing efficiency of the prior art, obtain the starting point coordinates, the end point coordinates and the real-time coordinates of the AGVs to be scheduled in the orders to be processed, input the coordinates into a pre-constructed AGV scheduling model to determine the scheduling strategy for the AGVs to be scheduled, and can improve the applicability, reliability and scheduling computing efficiency of AGV scheduling.

[0006] In a first aspect, the embodiments of the application provide an intelligent scheduling method of AGV, and the method comprises:

[0007] obtaining an order to be processed at a current time, reading starting point coordinates and end point coordinates in transportation information in the order to be processed;

[0008] Identify the state information of all AGVs in the current scheduling system, and at least one AGV with the state information as an idle state as a to-be-scheduled AGV, and obtain the real-time coordinates of the to-be-scheduled AGV;

[0009] Input the start point coordinates, end point coordinates and real-time coordinates into a pre-constructed AGV scheduling model, and determine the scheduling strategy for the to-be-scheduled AGV according to the output result of the AGV scheduling model; wherein the AGV scheduling model is a proxy hybrid neural network model.

[0010] Further, the AGV scheduling model includes a proxy network model and a hybrid network model.

[0011] The proxy network model is used to:

[0012] Input at least two AGV state observation values of the external environment, output the action value function of each AGV, and obtain the joint action value function by combining the action value function; wherein the at least two AGVs share the same proxy network, and the state observation value is the local observation value of a single AGV to the external environment;

[0013] The hybrid network model is used to:

[0014] Input the state observation value and the joint action value function, and output the next step action value function related to each AGV.

[0015] Further, after inputting at least two AGV state observation values of the external environment, the method further includes:

[0016] Combining the local observation value of the single AGV to the external environment by the proxy network model to obtain a global observation value;

[0017] Correspondingly, the hybrid network model is used to:

[0018] Input the global observation value and the joint action value function, and output the next step action value function related to each AGV.

[0019] Further, inputting the state observation value and the joint action value function, and outputting the next step action value function related to each AGV, includes:

[0020] Calculate the similarity of the global observation value and the local observation value by using the MLP network, and normalize the similarity by using the softmax function to obtain the weight corresponding to the local observation value;

[0021] The weights are multiplied by the action value functions of the AGVs and summed to obtain a next step action value function related to each AGV.

[0022] Further, before obtaining the to-be-processed order at the current time, the method further comprises:

[0023] drawing a scheduling map according to the actual transportation scenario of the AGV; wherein the scheduling map is a node mesh map; and the node mesh map at least comprises a transportation destination, a material position, and an AGV stayable position;

[0024] obtaining an initial position of the AGV and a transportation destination, and determining a shortest path from the initial position of the AGV to the transportation destination according to Manhattan distance; wherein the initial position of the AGV is one of the AGV stayable positions.

[0025] Further, the method further comprises:

[0026] identifying action state coordinate values of all AGVs in the current scheduling system, determining an AGV with an action state coordinate value of 0 as an idle AGV, and taking at least one AGV in the idle state as a to-be-scheduled AGV.

[0027] Correspondingly, after determining the scheduling strategy of the to-be-scheduled AGV according to the transportation starting point coordinate, the ending point coordinate, and the pre-constructed AGV scheduling model, the method further comprises:

[0028] identifying task state coordinate values of all AGVs in the current scheduling system, and determining an AGV with a task state coordinate value of 0 as a task-completed AGV.

[0029] identifying whether there is an unprocessed order at the current time;

[0030] if there is, obtaining the unprocessed order and assigning the unprocessed order to the task-completed AGV;

[0031] if there is not, issuing a stay-in-place instruction to the task-completed AGV.

[0032] Further, after determining the AGV with the task state coordinate value of 0 as the task-completed AGV, the method further comprises:

[0033] obtaining a task execution duration of the task-completed AGV, and taking the inverse of the task execution duration as a reward value of the task-completed AGV;

[0034] The reward values of all AGVs in the current scheduling system are traversed, a loss function of each AGV in the current scheduling system is determined according to the reward values, and the task execution time of the AGV is optimized according to the loss function.

[0035] In a second aspect, the embodiments of the present application provide an intelligent scheduling device of an AGV, and the device comprises:

[0036] A transportation information acquisition module is configured to acquire a to-be-processed order at a current time, and read a start point coordinate and an end point coordinate in transportation information in the to-be-processed order.

[0037] A state information acquisition module is configured to identify state information of all AGVs in a current scheduling system, take at least one AGV with an idle state as a to-be-scheduled AGV, and acquire a real-time coordinate of the to-be-scheduled AGV.

[0038] A scheduling strategy determination module is configured to input the start point coordinate, the end point coordinate, and the real-time coordinate into a pre-constructed AGV scheduling model, determine a scheduling strategy for the to-be-scheduled AGV according to an output result of the AGV scheduling model, and the AGV scheduling model is a proxy hybrid neural network model.

[0039] In a third aspect, the embodiments of the present application provide an electronic device, which comprises a processor, a memory, and a program or instruction stored on the memory and executable on the processor, and the program or instruction is executed by the processor to implement the steps of the method in the first aspect.

[0040] In a fourth aspect, the embodiments of the present application provide a readable storage medium, and the readable storage medium stores a program or instruction, and the program or instruction is executed by a processor to implement the steps of the method in the first aspect.

[0041] In a fifth aspect, the embodiments of the present application provide a chip, and the chip comprises a processor and a communication interface, the communication interface is coupled with the processor, the processor is configured to run a program or instruction to implement the method in the first aspect.

[0042] In the embodiment of the present application, the order to be processed at the current time is obtained, the starting point coordinates and the end point coordinates in the transportation information in the order to be processed are read; the state information of all AGVs in the current scheduling system is identified, at least one AGV with the state information as an idle state is taken as a to-be-scheduled AGV, and the real-time coordinates of the to-be-scheduled AGV are obtained; the starting point coordinates, the end point coordinates and the real-time coordinates are input into a pre-constructed AGV scheduling model, a scheduling strategy for the to-be-scheduled AGV is determined according to the output result of the AGV scheduling model; wherein the AGV scheduling model is a proxy hybrid neural network model. Through the above-mentioned intelligent scheduling method of AGV, the problems of poor applicability and low calculation efficiency in the prior art are solved. By obtaining the starting point coordinates, the end point coordinates in the order to be processed and the real-time coordinates of the to-be-scheduled AGV, and inputting the coordinates into the pre-constructed AGV scheduling model to determine the scheduling strategy for the to-be-scheduled AGV, the applicability, reliability and scheduling calculation efficiency of AGV scheduling can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is a flowchart of the intelligent scheduling method of AGV provided in the first embodiment of the present application;

[0044] Figure 2 is a flowchart of the intelligent scheduling method of AGV provided in the second embodiment of the present application;

[0045] Figure 3 is a flowchart of the intelligent scheduling method of AGV provided in the third embodiment of the present application;

[0046] Figure 4 is a proxy network model structure diagram provided in the embodiment of the present application;

[0047] Figure 5 is a hybrid network model structure diagram provided in the embodiment of the present application;

[0048] Figure 6 is a structure diagram of the intelligent scheduling device of AGV provided in the fourth embodiment of the present application;

[0049] Figure 7 is a structure diagram of the electronic device provided in the fifth embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the purposes, technical solutions and advantages of the present application clearer, the specific embodiments of the present application are further described in detail below in conjunction with the drawings. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only parts related to the present application are shown in the drawings, but not all. Before discussing the example embodiments in more detail, it should be mentioned that some example embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations can be implemented in parallel, concurrently or simultaneously. In addition, the order of the operations can be rearranged. The processes can be terminated when the operations are completed, but can also have additional steps not included in the drawings. The processes can correspond to methods, functions, procedures, subroutines, etc.

[0051] The technical solutions in the embodiments of the present application will be described clearly in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0052] The terms "first", "second" and the like in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second" and the like are usually a class, not limited to the number of objects, for example, the first object can be one or more. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / ", generally means that the objects before and after are in a "or" relationship.

[0053] The AGV intelligent scheduling method, device, equipment and medium provided by the embodiments of the present application will be described in detail below in conjunction with the drawings and specific embodiments and their application scenarios.

[0054] Embodiment one

[0055] Figure 1 is a flowchart of the AGV intelligent scheduling method provided by the first embodiment of the present application. As shown in Figure 1 , it specifically includes the following steps:

[0056] S101, obtaining the to-be-processed order at the current time, reading the starting point coordinates and the ending point coordinates in the transportation information in the to-be-processed order;

[0057] Firstly, the use scenario of the scheme can be a scenario of transporting by using an AGV, especially a scenario of scheduling the AGV when transporting by using the AGV.

[0058] Based on the above use scenario, it can be understood that the execution subject of the scheme can be an intelligent terminal with data reading, data processing and AGV scheduling capabilities, such as a mobile phone, a tablet computer and a desktop computer, etc., which is not limited here.

[0059] AGV (Automated Guided Vehicle) refers to a transport vehicle equipped with an automatic guidance device such as an electromagnetic or optical device, which can travel along a specified guidance path, has safety protection and various transfer functions. The AGV trolley with automatic transfer device can automatically complete the whole process of taking, placing and horizontal running of goods under the command of the control system. The order to be processed can be a task to be executed by the AGV. For example, a goods moving task and a material transportation task. The transportation information can be the transportation path, the transportation endpoint and the transportation speed information required by the AGV in the process of completing the order to be processed. The start point coordinate can be the coordinate of the handling start point in the AGV scheduling coordinate system in the transportation task. The end point coordinate can be the coordinate of the handling destination in the AGV scheduling coordinate system in the transportation task. The AGV scheduling coordinate system can be a coordinate system established for the actual operation or scheduling range of the AGV.

[0060] In one embodiment, the order to be processed at the current time can be obtained by reading the order state in the memory, the transportation information in the order to be processed is read, and the start point coordinate and the end point coordinate of the order to be processed are determined according to the transportation information.

[0061] In a feasible embodiment, optionally, before obtaining the order to be processed at the current time, the method further comprises:

[0062] drawing a scheduling map according to the actual transportation scenario of the AGV; wherein the scheduling map is a node mesh graph; the node mesh graph at least includes: a transportation destination, a material position and an AGV stayable position;

[0063] obtaining the initial position of the AGV and the transportation destination, and determining the shortest path from the initial position of the AGV to the transportation destination according to the Manhattan distance; wherein the initial position of the AGV is one of the AGV stayable positions.

[0064] The scheduling map can be a map drawn according to an actual transportation scene of the AGV. The scheduling map is a node network map, and the node network map at least includes a transportation destination, a material position, and an AGV stayable position. The purpose of the node network map is mainly to accurately describe an AGV scheduling problem in a scheduling system. In the node network, positions of different machines are represented by nodes, and a connection line between different nodes represents a road. According to a real scene, a length of each connection line in the network is given in advance. The road between nodes can meet the requirements of two AGVs traveling in opposite directions at the same time, and therefore, a collision problem in the process of AGV transportation is not considered. The initial position of the AGV can be one of a stayable position of the AGV before the AGV is scheduled, including an AGV garage position and an AGV stayable position. The transportation destination can be a material transportation end point in the order to be processed. The Manhattan distance is a geometry term used in a geometry measurement space, and is used to indicate an absolute axis distance sum of two points in a standard coordinate system.

[0065] In the scheme, the starting positions of all materials are in a material warehouse; the initial positions of all AGVs can be random; the capacities of the input buffer and the output buffer of each AGV are sufficient, and after completing a current transportation task, the AGV can receive a next transportation task, and if there is no subsequent task, the AGV can wait at a current position; the failure or charging problem of the AGV is not considered; at a scheduling initial moment, all AGVs and machines are idle; and the AGV completes a transportation task when the AGV transports a material from a starting position to a destination position.

[0066] In one embodiment, a scheduling map can be drawn according to an actual transportation scene of the AGV. The scheduling map is a node network map, and the node network map at least includes a transportation destination, a material position, and an AGV stayable position. The initial position of the AGV and the transportation destination are obtained, and a shortest path from the initial position of the AGV to the transportation destination is determined according to the Manhattan distance.

[0067] In the scheme, the scheduling map is drawn according to the actual transportation scene of the AGV, the initial position of the AGV and the transportation destination are obtained, and the shortest path from the initial position of the AGV to the transportation destination is determined according to the Manhattan distance. The transportation path of the AGV can be shortened, and therefore, the transportation efficiency of the AGV is improved, and the reliability of AGV scheduling is improved.

[0068] In S102, state information of all AGVs in a current scheduling system is recognized, at least one AGV with the state information as an idle state is taken as a to-be-scheduled AGV, and real-time coordinates of the to-be-scheduled AGV are obtained.

[0069] The scheduling system can be a system for centrally supervising, state monitoring, path planning, and scheduling of multiple AGVs. The state information of the AGVs can be running state information of the AGVs, including: working state, idle state, current location, and running speed, etc. The AGV to be scheduled can be an AGV currently in an idle state. The real-time coordinates of the AGV to be scheduled can be real-time coordinates of the AGV to be scheduled in the AGV scheduling coordinate system.

[0070] In one embodiment, the state information of all AGVs in the current scheduling system can be identified according to the running speed and position change of the AGVs in the current scheduling system. AGVs with a running speed of 0 and / or no position change over time are regarded as AGVs in an idle state. At least one AGV with the state information in an idle state is regarded as an AGV to be scheduled, and the real-time coordinates of the AGV to be scheduled are obtained according to the actual position of the AGV to be scheduled.

[0071] S103, inputting the start point coordinates, end point coordinates, and real-time coordinates into a pre-constructed AGV scheduling model, and determining a scheduling strategy for the AGV to be scheduled according to an output result of the AGV scheduling model; wherein the AGV scheduling model is a proxy hybrid neural network model.

[0072] The AGV scheduling model can be a model for calculating an optimal scheduling strategy of AGVs. The AGV scheduling model is a proxy hybrid neural network model. The optimal scheduling strategy can be a strategy that enables AGVs to have the advantages of shortest running time, highest task execution efficiency, and shortest empty path during task execution. The proxy hybrid neural network model includes a proxy network model and a hybrid network model. The proxy network model can enable all AGVs in the scheduling system to share the same proxy network, and take the observation value of the external environment of the AGVs as input and the action value function related to each AGV as output. The hybrid network model can be used to calculate the value generated by the next action of all AGVs in the scheduling system. The hybrid network model can take the action value function related to each AGV output by the proxy network model as input, calculate the weight of each action value function using an attention mechanism, and then perform weighted summation on each action value function to obtain the value generated by the next action of all AGVs in the scheduling system.

[0073] In one embodiment, the start point coordinates, end point coordinates, and real-time coordinates are input into a pre-constructed AGV scheduling model, the value generated by the next action of all AGVs in the scheduling system is determined according to an output result of the AGV scheduling model, and the scheduling strategy for the AGV to be scheduled is determined according to the value. The AGV scheduling model is a proxy hybrid neural network model.

[0074] In an embodiment, the AGV scheduling model comprises an agent network model and a hybrid network model.

[0075] The agent network model is configured to:

[0076] input state observation values of at least two AGVs on an external environment, output action value functions of the AGVs, and combine the action value functions to obtain a joint action value function; wherein the at least two AGVs share the same agent network, and the state observation values are local observation values of a single AGV on the external environment.

[0077] The hybrid network model is configured to:

[0078] input the state observation values and the joint action value function, and output a next step action value function related to each AGV.

[0079] The agent network model can be a model that combines local information of each agent into global information and guides actions of each agent to achieve optimal benefits in a collaborative task. The agent network model in the present solution can be a model that combines local observation values of each AGV on the external environment and calculates action value functions corresponding to the observation values. In order to improve the calculation efficiency, all AGVs share the same agent network. The action value function represents an action taken by the AGV according to the observed local observation value. The hybrid network model can be a model for calculating the value generated by the next step action of all AGVs in the scheduling system. The next step action value function can be a value function that can be generated by the next step action of the AGV. The value function can be a reward function, and can also be a penalty function. The AGV scheduling model can adjust the action of the AGV with the purpose of maximizing the reward function.

[0080] In an embodiment, the AGV scheduling model comprises an agent network model and a hybrid network model.

[0081] The agent network model is configured to: according to input state observation values of at least two AGVs on an external environment, calculate and output action value functions corresponding to the AGVs, and combine the action value functions to obtain a joint action value function. Wherein the at least two AGVs share the same agent network, and the state observation values are local observation values of a single AGV on the external environment. The hybrid network model is configured to: according to input state observation values and the joint action value function, calculate and output a next step action value function related to each AGV.

[0082] In an embodiment, after inputting the state observation values of the at least two AGVs with respect to the external environment, the method further comprises:

[0083] combining the local observation values of the single AGVs with respect to the external environment by the agent network model to obtain global observation values;

[0084] Correspondingly, the hybrid network model is configured to:

[0085] inputting the global observation values and the joint action value function, and outputting a next step action value function related to each AGV.

[0086] The global observation values can be observation values of comprehensive observation of the external environment in the scheduling system. The global observation values can be obtained by combining the local observation values of all the single AGVs with respect to the external environment in the scheduling system.

[0087] In an embodiment, the global observation values can be obtained by combining the local observation values of the single AGVs with respect to the external environment by the agent network model. The next step action value function related to each AGV can be calculated and outputted by the hybrid network model according to the inputted global observation values and the joint action value function.

[0088] According to the scheme, the global observation values can be obtained by combining the local observation values of the single AGVs with respect to the external environment by the agent network model, and the next step action value function related to each AGV can be outputted by the hybrid network model according to the global observation values and the joint action value function. The local observation values can be combined in a nonlinear manner, the performance of the algorithm is improved, and the effectiveness and rationality of AGV scheduling in a complex environment can be ensured.

[0089] In an embodiment, inputting the state observation values and the joint action value function, and outputting the next step action value function related to each AGV, comprises:

[0090] calculating the similarity between the global observation values and the local observation values by using the MLP network, and normalizing the similarity by using the softmax function to obtain the weight corresponding to the local observation values;

[0091] multiplying the weight and the action value function of each AGV and summing up to obtain the next step action value function related to each AGV.

[0092] The MLP (Multilayer Perceptron) is a feedforward artificial neural network model that maps a plurality of input data sets to a single output data set. The MLP includes three layers: an input layer, an intermediate layer, and an output layer. The first layer is referred to as the input layer, the last layer is referred to as the output layer, and the intermediate layer is referred to as the hidden layer. The MLP neural network is fully connected between different layers (i.e., any neuron in the previous layer is connected to all neurons in the next layer). The MLP does not limit the number of hidden layers, and the number of neurons in the output layer is also not limited. The number of hidden layers can be selected according to the respective requirements. The MLP maps the non-linearly distributable space of the input space to the linearly distributable space of the hidden space through the nonlinear transformation of the hidden layer. The softmax function can be a normalized exponential function. The softmax function can "compress" a K-dimensional vector z containing any real number into another K-dimensional real vector σ(z), so that the range of each element is between 0 and 1, and the sum of all elements is 1.

[0093] In one embodiment, the similarity between the global observation value and the local observation value can be calculated by using the MLP network, and the weight corresponding to the local observation value can be obtained by normalizing the similarity by using the softmax function. The weight is multiplied by the action value function of each AGV and summed to obtain the next action value function related to each AGV.

[0094] The present scheme can improve the efficiency and reliability of the calculation of the next action value function related to each AGV by calculating the similarity between the global observation value and the local observation value by using the MLP network, obtaining the weight corresponding to the local observation value by normalizing the similarity by using the softmax function, and multiplying the weight by the action value function of each AGV and summing to obtain the next action value function related to each AGV.

[0095] The technical scheme provided in the embodiments of the present application obtains a to-be-processed order at a current time, reads a starting point coordinate and an ending point coordinate in transportation information in the to-be-processed order, identifies state information of all AGVs in a current scheduling system, determines at least one AGV in an idle state as a to-be-scheduled AGV according to the state information, and obtains a real-time coordinate of the to-be-scheduled AGV; the starting point coordinate, the ending point coordinate, and the real-time coordinate are input into a pre-constructed AGV scheduling model, a scheduling strategy for the to-be-scheduled AGV is determined according to an output result of the AGV scheduling model, and the AGV scheduling model is a proxy hybrid neural network model. Through the above intelligent scheduling method of the AGV, the problems of poor applicability and low calculation efficiency in the prior art are solved. The starting point coordinate, the ending point coordinate, and the real-time coordinate of the to-be-scheduled AGV in the to-be-processed order are obtained, and the coordinates are input into the pre-constructed AGV scheduling model to determine the scheduling strategy for the to-be-scheduled AGV, so that the applicability, reliability, and scheduling calculation efficiency of AGV scheduling can be improved.

[0096] Embodiment Two

[0097] Figure 2 is a flowchart of an intelligent scheduling device of an AGV provided in Embodiment Two of the present application. As shown in Figure 2 , the specific steps include the following steps:

[0098] S201, a to-be-processed order at a current time is obtained, and a starting point coordinate and an ending point coordinate in transportation information in the to-be-processed order are read;

[0099] S202, action state coordinate values of all AGVs in a current scheduling system are identified, AGVs with the action state coordinate values of 0 are determined as AGVs in an idle state, at least one AGV in the idle state is determined as a to-be-scheduled AGV, and a real-time coordinate of the to-be-scheduled AGV is obtained;

[0100] The action state coordinate value of the AGV can be a movement coordinate value of the AGV, which is used to indicate whether the AGV has moved. The action state coordinate value of the AGV can be represented by a displacement amount in a horizontal coordinate direction and / or a vertical coordinate direction in an AGV scheduling coordinate system. For example, the action state coordinate value of the i th AGV is represented by ai x and ai y, ai x and ai y are respectively the horizontal coordinate value and the vertical coordinate value of the i th AGV, and ai x and ai y are respectively the displacement amount in the horizontal coordinate direction and the displacement amount in the vertical coordinate direction of the i th AGV. x y When a

[0101] ​​In one embodiment, the action state coordinate values of all AGVs in the current scheduling system can be identified, the AGV with the action state coordinate value of 0 is determined as an idle AGV, at least one AGV in the idle state is determined as a to-be-scheduled AGV, and the real-time coordinate of the to-be-scheduled AGV is obtained.

[0102] S203, inputting the start point coordinate, the end point coordinate and the real-time coordinate into a pre-constructed AGV scheduling model, and determining a scheduling strategy for the to-be-scheduled AGV according to an output result of the AGV scheduling model; wherein the AGV scheduling model is a proxy hybrid neural network model.

[0103] S204, identifying task state coordinate values of all AGVs in the current scheduling system, and determining an AGV with a task state coordinate value of 0 as a task-completed AGV.

[0104] The task state coordinate value of the AGV can be a start point coordinate value and an end point coordinate value of a to-be-executed task, and is used to indicate whether the AGV has a to-be-executed task. For example, the jthtask state coordinate value of the AGV is represented as b x y The start point to which the AGV needs to go in the task is represented as a x y The end point to which the AGV needs to go in the task is represented as e, and when the above four coordinate values are all 0, it indicates that the task has been completed by the AGV.

[0105] In one embodiment, after determining the scheduling strategy for the to-be-scheduled AGV according to the transportation start point coordinate, the end point coordinate and the pre-constructed AGV scheduling model, the task state coordinate values of all AGVs in the current scheduling system can be identified, and the AGV with the task state coordinate value of 0 is determined as a task-completed AGV.

[0106] S205, identifying whether there is an unprocessed order at the current time;

[0107] S2051, if there is, obtaining the unprocessed order and assigning the unprocessed order to the task-completed AGV;

[0108] S2052, if there is not, issuing a stay-in-place instruction to the task-completed AGV.

[0109] ​​​In one embodiment, whether there is an unprocessed order at the current time can be identified according to the order processing state or order increment or decrement in the memory, if there is, the unprocessed order is obtained, and the unprocessed order is assigned to the task-completed AGV; if not, an on-the-spot stay instruction is sent to the task-completed AGV.

[0110] In a feasible embodiment, after the AGV with the task state coordinate value of 0 is determined as the task-completed AGV, the method further comprises:

[0111] The task execution duration of the task-completed AGV is obtained, and the inverse of the task execution duration is taken as the reward value of the task-completed AGV;

[0112] The reward values of all AGVs in the current scheduling system are traversed, the loss function of each AGV in the current scheduling system is determined according to the reward value, and the task execution duration of the AGV is optimized according to the loss function.

[0113] The task execution duration can be the operation function used to measure the difference between the predicted value and the true value of the model, which is a non-negative real value function, the smaller the loss function, the better the robustness of the model, and is mainly used in the training stage of the model. After each batch of training data is input into the model, the predicted value is output through forward propagation, and then the loss function calculates the difference between the predicted value and the true value, that is, the loss value. After obtaining the loss value, the model updates each parameter through back propagation to reduce the loss between the true value and the predicted value, so that the predicted value generated by the model approaches the true value, thereby achieving the purpose of learning. The loss function in the present scheme can be a function for reducing the transportation time loss of the AGV. The task execution duration, transportation path and scheduling strategy of the AGV can be optimized through the loss function. The loss function of the AGV can be determined according to the reward value of the AGV. When the reward value is maximized, the waiting time of the AGV executing the task will be minimized. When the AGV is assigned to a task, the timing starts, and when all tasks are completed, the task execution duration can be obtained. The inverse of the task execution duration is taken as the reward value of the task-completed AGV.

[0114] In one embodiment, the task execution duration of the task-completed AGV can be obtained, and the inverse of the task execution duration is taken as the reward value of the task-completed AGV. The reward values of all AGVs in the current scheduling system are traversed, the loss function of each AGV in the current scheduling system is determined according to the reward value, and the task execution duration of the AGV is optimized according to the loss function.

[0115] The technical scheme provided in the embodiment of the application can improve the task execution efficiency of the AGV, and further optimize the scheduling of the AGV.

[0116] The technical scheme provided in the embodiment of the application can improve the task execution efficiency of the AGV, and further optimize the scheduling of the AGV.

[0117] Embodiment Three

[0118] Figure 3 is a flowchart of the intelligent scheduling method of the AGV provided in Embodiment Three of the application. As shown in Figure 3 the specific steps include the following steps:

[0119] S301, a node network of a destination, a material warehouse and AGVs is constructed;

[0120] S302, a shortest path of the AGV is planned;

[0121] S303, a mathematical symbol is predefined, a real scenario is assumed and processed, and a node network is constructed;

[0122] S304, a material handling problem of the AGV is modeled as a multi-agent reinforcement learning problem;

[0123] S305, all agents in the improved model share an agent network, the input of the network is an observation value o, and the output is an action value function Q(o, u) related to each agent;

[0124] S306, the action value function of each agent obtained is combined as the input of a Mixing network;

[0125] S307, a state s is taken as the input of an attention mechanism model, and the final weight w is obtained by comparing the similarity of s and the observation results of each agent and performing some operations;

[0126] S308, the w obtained is multiplied by the input in S306 to obtain a final output Qtot.

[0127] In one embodiment, the construction of the node network is mainly to accurately describe the AGV scheduling problem in the system. In the node network, the positions of different machines are represented by nodes, and the edges between different nodes represent roads. According to the real scene, the length of each edge in the network is given in advance.

[0128] In one embodiment, the planning of the shortest path of the AGV refers to generating a path with the shortest distance from the current position of the AGV to the destination of the task. Here, the Manhattan distance is selected as the shortest path.

[0129] In one embodiment, the assumption processing of the real scene mainly includes: (1) the starting positions of all materials are in the material warehouse. (2) the initial positions of all AGVs are random. (3) the capacities of the input buffer and the output buffer of each machine are sufficient. After completing the current transportation task, the AGV can receive the next transportation task, and if there is no subsequent task, the AGV will wait at the current position. (4) the edges of the node network can meet the requirements of two AGVs traveling in opposite directions at the same time. Therefore, the collision occurring during the transportation of the AGV is not considered. (5) the failure or charging problem of the AGV is not considered. (6) at the beginning, all AGVs and machines are idle. (7) completing a task means transporting materials from the starting position to the destination position.

[0130] In one embodiment, the modeling of the AGV scheduling problem mainly includes: (1) the maximum number of AGVs and orders to be processed are defined as and respectively. (2) the state of the th AGV is defined as, which represents the coordinate point of the AGV at that time, and represents that the AGV is in an idle state. (3) the state of the th task is defined as, where represents the starting point that the AGV should go to, and represents the end point that the AGV should go to. When all four values are 0, it means that the task has been completed. (4) and are binary vectors of and lengths respectively. For example, the first bit represents the state of the first AGV, that is, when the AGV is idle, the value of the first bit will change from 1 to 0, and vice versa. In this way, the state of each task is marked. (5) each task has its own reward, which is equal to (-1) x (the waiting time of the task). When the reward is maximized, the waiting time is minimized. The faster the task is completed, the less time it takes to complete. At the beginning of each stage, the AGV is assigned a task. When all tasks are completed, the current stage ends. Finally, the definition of the loss function is obtained.

[0131] In one embodiment, the AGV intelligent scheduling algorithm based on the agent hybrid network is characterized in that an agent network corresponding to each agent in the traditional agent hybrid network algorithm is designed as an agent network shared by all agents, mainly to improve the calculation efficiency and reduce the implementation difficulty.

[0132] In one embodiment, the AGV intelligent scheduling algorithm based on the agent hybrid network is characterized in that the action function values generated by all agents are combined as inputs of the hybrid network.

[0133] In one embodiment, the AGV intelligent scheduling algorithm based on the agent hybrid network is characterized in that the state s is taken as an input of the attention mechanism model, the similarity between the global state s and the observation of each agent is compared, the similarity is calculated by using an MLP network, and then the weight is obtained by normalization.

[0134] In one embodiment, the AGV intelligent scheduling algorithm based on the agent hybrid network is characterized in that the weight corresponding to each agent obtained above is multiplied by the agent outputs Q1, Q2, Q3, etc., and finally added to obtain Qtot.

[0135] In one embodiment, the trained model is applied to other maps of different sizes to verify the universality of the method.

[0136] The agent hybrid network model in the scheme includes an agent network model and a hybrid network model.

[0137] Figure 4 The agent network model structure diagram provided for the embodiments of the application is shown.

[0138] As shown in Figure 4 , the agent network plays an important role in the agent hybrid network algorithm, which combines the local information of each agent into global information and guides the action of each agent to achieve the optimal benefit in the collaborative task. In the present application, all agents share an agent network in order to improve the calculation efficiency.

[0139] Figure 5 The hybrid network model structure diagram provided for the embodiments of the application is shown.

[0140] As shown in Figure 5As shown, the mixed network model takes the global state s and the output of the agent network as input, uses the MLP network to calculate the similarity of the global state s and the observation of each agent, and normalizes the obtained similarity weight using softmax. The obtained value is multiplied by the output observation value of each agent and summed up to obtain the next action value function Qtot related to each agent.

[0141] In one embodiment, the main way to optimize AGV scheduling is to minimize the maximum completion time of AGV, which is expressed by the following formula:

[0142] T = minimize(maxEA k );

[0143] The maximum completion time of AGV k is calculated as follows:

[0144]

[0145] The formula that a task can only be passed by one AGV is as follows:

[0146]

[0147] There is only one connection point after the starting point of AGV k, as shown in the formula:

[0148]

[0149] The formula for limiting the flow balance of the starting point and the ending point is as follows:

[0150]

[0151] The formula for ensuring that there is only one connection point before the end point of AGV k is as follows:

[0152]

[0153] The formula for AGV k at a specific location at a specific time t is as follows:

[0154]

[0155] In order to extract the decentralized strategy from the Qtot action value function, the argmax operation can be performed on the global action value function Qtot, which can ensure the consistency of individual action values and global action values. The mathematical expression is shown in equation (9) as follows:

[0156]

[0157] To ensure this consistency, we can constrain the monotonicity between Qtot and Q of each individual agent, which can be described by the following expression:

[0158]

[0159] All the above restrictions can be achieved by the hybrid network.

[0160] The technical scheme provided by the embodiment of the application can improve the applicability, reliability and scheduling calculation efficiency of AGV scheduling by drawing a node network of AGV scheduling and calculating the optimal scheduling strategy of AGV through the agent hybrid model.

[0161] Embodiment four

[0162] Figure 6 is a structural schematic diagram of the intelligent scheduling device of the AGV provided by the embodiment four of the application. As shown in the figure, it specifically comprises the following: Figure 6

[0163] The transportation information acquisition module 601 is configured to acquire the to-be-processed order at the current time, and read the start point coordinates and the end point coordinates in the transportation information in the to-be-processed order.

[0164] The state information acquisition module 602 is configured to identify the state information of all AGVs in the current scheduling system, and take at least one AGV with the state information as the idle state as the AGV to be scheduled, and acquire the real-time coordinates of the AGV to be scheduled.

[0165] The scheduling strategy determination module 603 is configured to input the start point coordinates, the end point coordinates and the real-time coordinates into a pre-constructed AGV scheduling model, and determine the scheduling strategy for the AGV to be scheduled according to the output result of the AGV scheduling model; wherein the AGV scheduling model is an agent hybrid neural network model.

[0166] Further, the AGV scheduling model comprises an agent network model and a hybrid network model.

[0167] The agent network model is configured to:

[0168] input the state observation values of at least two AGVs to the external environment, output the action value function of each AGV, and obtain the joint action value function by jointing the action value functions; wherein the at least two AGVs share the same agent network, and the state observation values are the local observation values of a single AGV to the external environment.

[0169] The hybrid network model is configured to:

[0170] ​Input the state observation value and the joint action value function, and output a next step action value function related to each AGV.

[0171] Further, the agent network model is also used for:

[0172] The local observation value of the single AGV to the external environment is combined by the agent network model to obtain a global observation value.

[0173] Correspondingly, the hybrid network model is used for:

[0174] Input the global observation value and the joint action value function, and output a next step action value function related to each AGV.

[0175] Further, the hybrid network model is specifically used for:

[0176] The similarity between the global observation value and the local observation value is calculated by using the MLP network, and the weight corresponding to the local observation value is obtained by normalizing the similarity by using the softmax function.

[0177] The weight is multiplied by the action value function of each AGV and summed to obtain a next step action value function related to each AGV.

[0178] Further, the device further comprises:

[0179] The scheduling mapping module is used for drawing a scheduling map according to the actual transportation scene of the AGV; wherein the scheduling map is a node net map; the node net map at least includes: a transportation destination, a material position and an AGV stayable position.

[0180] The path determination module is used for obtaining an initial position of the AGV and a transportation destination, and determining a shortest path from the initial position of the AGV to the transportation destination according to Manhattan distance; wherein the initial position of the AGV is one of the AGV stayable positions.

[0181] Further, the state information acquisition module 602 is specifically used for:

[0182] The action state coordinate value of all AGVs in the current scheduling system is identified, the AGV with the action state coordinate value of 0 is determined as an idle state AGV, and at least one AGV in the idle state is determined as a to-be-scheduled AGV.

[0183] Correspondingly, the device further comprises:

[0184] A task state recognition module is configured to recognize task state coordinate values of all AGVs in the current scheduling system, and determine an AGV with a task state coordinate value of 0 as a task-completed AGV.

[0185] An unprocessed order recognition module is configured to recognize whether there is an unprocessed order at the current time.

[0186] An unprocessed order distribution module is configured to, when it is recognized that there is an unprocessed order at the current time, acquire the unprocessed order and distribute the unprocessed order to the task-completed AGV.

[0187] A stay-in-place instruction issuing module is configured to, when it is recognized that there is no unprocessed order at the current time, issue a stay-in-place instruction to the task-completed AGV.

[0188] Further, the device further comprises:

[0189] A reward value calculation module is configured to acquire a task execution duration of the task-completed AGV, and take the inverse of the task execution duration as a reward value of the task-completed AGV.

[0190] A task execution duration optimization module is configured to traverse reward values of all AGVs in the current scheduling system, determine a loss function of each AGV in the current scheduling system according to the reward values, and optimize a task execution duration of the AGV according to the loss function.

[0191] The technical scheme provided by the embodiment of the application comprises: a transportation information acquisition module configured to acquire a to-be-processed order at the current time, and read start point coordinates and end point coordinates in transportation information in the to-be-processed order; a state information acquisition module configured to recognize state information of all AGVs in a current scheduling system, and acquire real-time coordinates of at least one AGV in an idle state as a to-be-scheduled AGV; and a scheduling strategy determination module configured to input the start point coordinates, the end point coordinates, and the real-time coordinates into a pre-constructed AGV scheduling model, and determine a scheduling strategy for the to-be-scheduled AGV according to an output result of the AGV scheduling model. The AGV scheduling model is a proxy hybrid neural network model. The above-mentioned intelligent scheduling device for AGVs solves the problems of poor applicability and low calculation efficiency in the prior art. By acquiring the start point coordinates, the end point coordinates, and the real-time coordinates of the to-be-scheduled AGV in the to-be-processed order, and inputting the coordinates into the pre-constructed AGV scheduling model to determine the scheduling strategy for the to-be-scheduled AGV, the applicability, reliability, and scheduling calculation efficiency of AGV scheduling can be improved.

[0192] The intelligent scheduling device of the AGV in the embodiments of the present applicationapplicationbe a device, or a component, an integrated circuit, or a chip in a terminal. The deviceapplicationbe a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic deviceapplicationbe a mobile phone, a tablet computer, a notebook computer, a palm computer, a vehicle-mounted electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc., and the non-mobile electronic deviceapplicationbe a server, a network attached storage (NAS), a personal computer (PC), a television (TV), a teller machine, or a self-service machine, etc., and the embodiments of the present application do not make a specific limitation.

[0193] The intelligent scheduling device of the AGV in the embodiments of the present applicationapplicationbe a device with an operating system. The operating systemapplicationbe an Android operating system, an ios operating system, or other possible operating systems, and the embodiments of the present application do not make a specific limitation.

[0194] The intelligent scheduling device of the AGV provided in the embodiments of the present applicationapplicationimplement the various processes implemented by the method embodiments, and thus repeated details are not described herein.

[0195] Embodiment Five

[0196] As shown in Figure 7 The embodiments of the present application further provide an electronic device 700, which includes a processor 701, a memory 702, and a program or instruction stored in the memory 702 and executable on the processor 701. The program or instruction is executed by the processor 701 to implement the various processes of the above-mentioned embodiments of the intelligent scheduling device of the AGV, and achieve the same technical effects. Thus, repeated details are not described herein.

[0197] It should be noted that the electronic device in the embodiments of the present applicationapplicationinclude the above-mentioned mobile electronic device and non-mobile electronic device.

[0198] Embodiment Six

[0199] The embodiments of the present application further provide a readable storage medium having a program or instruction stored thereon. The program or instruction is executed by a processor to implement the various processes of the above-mentioned embodiments of the intelligent scheduling device of the AGV, and achieve the same technical effects. Thus, repeated details are not described herein.

[0200] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0201] Embodiment seven

[0202] The embodiment of the application further provides a chip, which comprises a processor and a communication interface, the communication interface is coupled with the processor, the processor is used for running programs or instructions to realize the processes of the intelligent scheduling device of the AGV and achieve the same technical effects. To avoid repetition, details are not described here.

[0203] It should be understood that the chip mentioned in the embodiment of the application can also be referred to as a system-level chip, a system chip, a chip system, or a system-on-chip chip.

[0204] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, articles, or devices including a series of elements not only include those elements, but also include other elements not explicitly listed, or further include elements inherent to such processes, methods, articles, or devices. Without more limitations, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article, or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the application is not limited to the order of performing the functions as shown or discussed, but can also include performing the functions in a substantially simultaneous manner or in reverse order, for example, the described method can be performed in an order different from the described order, and various steps can be added, omitted, or combined. In addition, the features described with reference to certain examples can be combined in other examples.

[0205] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, it can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a plurality of instructions for making a terminal (which can be a mobile phone, computer, server, or network equipment, etc.) execute the method described in each embodiment of the present application.

[0206] The embodiments of the present application are described above with reference to the drawings, but the present application is not limited to the specific embodiments described above, and the specific embodiments described above are merely illustrative, but not restrictive, and those of ordinary skill in the art can make many forms without departing from the purpose of the present application and the scope protected by the claims under the inspiration of the present application, which all belong to the protection of the present application.

[0207] The above are only the preferred embodiments of the present application and the technical principles used. The present application is not limited to the specific embodiments described herein, and various obvious changes, re-adjustments and replacements made by those skilled in the art will not deviate from the scope of protection of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and more other equivalent embodiments can be included without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. An intelligent scheduling method of AGVs, characterized in that, The method comprises: acquiring a to-be-processed order at a current time, reading a starting point coordinate and an ending point coordinate in transportation information in the to-be-processed order; identifying state information of all AGVs in a current scheduling system, and taking at least one AGV in an idle state as a to-be-scheduled AGV, comprising: identifying a motion state coordinate value of all AGVs in the current scheduling system, determining an AGV with a motion state coordinate value of 0 as an AGV in an idle state, and taking at least one AGV in the idle state as a to-be-scheduled AGV, and acquiring a real-time coordinate of the to-be-scheduled AGV; inputting the starting point coordinate, the ending point coordinate and the real-time coordinate into a pre-constructed AGV scheduling model, determining a scheduling strategy for the to-be-scheduled AGV according to an output result of the AGV scheduling model, identifying a task state coordinate value of all AGVs in the current scheduling system, determining an AGV with a task state coordinate value of 0 as a task-completed AGV, identifying whether there is an unprocessed order at the current time, if there is, acquiring the unprocessed order and assigning the unprocessed order to the task-completed AGV, and if not, issuing a stay-in-place instruction to the task-completed AGV; The AGV scheduling model comprises: an agent network model and a hybrid network model; The agent network model is configured to: input state observation values of at least two AGVs on an external environment, output an action value function of each AGV, and obtain a joint action value function by jointing the action value functions; wherein the at least two AGVs share the same agent network, and the state observation values are local observation values of a single AGV on the external environment; The hybrid network model is configured to: input the state observation values and the joint action value function, and output a next-step action value function related to each AGV.

2. The intelligent scheduling method of AGVs according to claim 1, characterized in that, After inputting the state observation values of at least two AGVs on an external environment, the method further comprises: combining the local observation values of the single AGV on the external environment by the agent network model to obtain a global observation value; Correspondingly, the hybrid network model is configured to: input the global observation value and the joint action value function, and output a next-step action value function related to each AGV.

3. The intelligent scheduling method of AGVs according to claim 2, characterized in that, Inputting the state observation values and the joint action value function, and outputting a next-step action value function related to each AGV, comprises: calculating a similarity of the global observation value and the local observation value by using an MLP network, and normalizing the similarity by using a softmax function to obtain a weight corresponding to the local observation value; multiplying the weight and the action value function of each AGV and summing to obtain a next-step action value function related to each AGV.

4. The intelligent scheduling method of AGVs according to claim 1, characterized in that, Before acquiring the to-be-processed order at the current time, the method further comprises: drawing a scheduling map according to an actual transportation scene of the AGV; wherein the scheduling map is a node mesh graph; the node mesh graph at least comprises: a transportation destination, a material position and an AGV stayable position. The initial position of the AGV and a transportation destination are acquired, and a shortest path from the initial position of the AGV to the transportation destination is determined according to Manhattan distance.

5. The intelligent scheduling method of AGVs according to claim 1, characterized in that, After the AGV with the task state coordinate value of 0 is determined as the task-completed AGV, the method further comprises: The task execution duration of the task-completed AGV is acquired, and the reciprocal of the task execution duration is taken as a reward value of the task-completed AGV; The reward values of all AGVs in the current scheduling system are traversed, a loss function of each AGV in the current scheduling system is determined according to the reward values, and the task execution duration of the AGV is optimized according to the loss function.

6. An intelligent scheduling device for AGVs, characterized in that, The device comprises: The transportation information acquisition module is configured to acquire a to-be-processed order at a current time, and read a start point coordinate and an end point coordinate in transportation information in the to-be-processed order. The state information acquisition module is configured to identify state information of all AGVs in a current scheduling system, and take at least one AGV with an idle state as a to-be-scheduled AGV, including: identifying an action state coordinate value of all AGVs in the current scheduling system, determining an AGV with the action state coordinate value of 0 as the AGV in the idle state, taking the at least one AGV in the idle state as the to-be-scheduled AGV, and acquiring a real-time coordinate of the to-be-scheduled AGV. The scheduling strategy determination module is configured to input the start point coordinate, the end point coordinate and the real-time coordinate into a pre-constructed AGV scheduling model, determine a scheduling strategy for the to-be-scheduled AGV according to an output result of the AGV scheduling model, identify a task state coordinate value of all AGVs in the current scheduling system, determine an AGV with the task state coordinate value of 0 as a task-completed AGV, identify whether there is an unprocessed order at the current time, if there is, acquire the unprocessed order and assign the unprocessed order to the task-completed AGV, and if there is not, issue a stay-in-place instruction to the task-completed AGV. The AGV scheduling model comprises: an agent network model and a hybrid network model. The agent network model is configured to input state observation values of at least two AGVs on an external environment, output an action value function of each AGV, and obtain a joint action value function by jointing the action value functions; wherein the at least two AGVs share the same agent network, and the state observation values are local observation values of a single AGV on the external environment. The hybrid network model is configured to input the state observation values and the joint action value function, and output a next-step action value function related to each AGV.

7. An electronic device, comprising: The device comprises a processor, a memory and a program or instructions stored on the memory and executable on the processor, and the program or instructions are executed by the processor to implement the steps of the intelligent scheduling method of the AGV according to any one of claims 1-5.

8. A readable storage medium, characterized by, The readable storage medium stores programs or instructions, and the programs or instructions are executed by the processor to realize the steps of the intelligent scheduling method of the AGV according to any one of claims 1-5.

Images