Task scheduling method of robot and related equipment

Abstract

Embodiments of the present application provide a robot task scheduling method and related equipment, which are used to improve the task allocation and scheduling efficiency of a multi-robot system. The method of the embodiments of the present application comprises: obtaining current state information of each to-be-scheduled robot in a robot cluster; generating a target task allocation scheme corresponding to each to-be-scheduled robot according to the current state information; generating a current task allocation instruction of the target task allocation scheme, and allocating the target task allocation scheme to each corresponding to-be-scheduled robot according to the current task allocation instruction, so as to control each to-be-scheduled robot to execute the target task allocation scheme; obtaining current task execution data of each to-be-scheduled robot in executing the target task allocation scheme, adjusting the target task allocation scheme of each to-be-scheduled robot according to the current task execution data, and obtaining a target task scheduling strategy, so that each to-be-scheduled robot executes the target task allocation scheme allocated by the target task scheduling strategy.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of intelligent device, in particular to a task scheduling method of robot and related equipment. BACKGROUND

[0002] In the field of modern industry and logistics, multi-robot systems have been widely used in task execution, warehouse management, automated production lines, etc. These robots usually interact with the control center through various communication protocols (such as hypertext transfer protocol (HTTP), transfer control protocol (TCP), user datagram protocol, etc.) to report their state information in real time and receive task instructions. However, the existing technology has many problems in task allocation and scheduling optimization, especially when dealing with complex multi-robot task scheduling, it is easy to fall into a local optimal solution and it is difficult to achieve global optimal scheduling. SUMMARY

[0003] Embodiments of the present application provide a task scheduling method of robot and related equipment for improving the task allocation and scheduling efficiency of multi-robot systems.

[0004] The first aspect of the embodiments of the present application provides a task scheduling method of robot, comprising:

[0005] obtaining current state information of each to-be-scheduled robot in a robot cluster;

[0006] generating a target task allocation scheme corresponding to each to-be-scheduled robot according to the current state information;

[0007] generating a current task allocation instruction of the target task allocation scheme, and allocating the target task allocation scheme to the corresponding each to-be-scheduled robot according to the current task allocation instruction, to control the each to-be-scheduled robot to execute the target task allocation scheme;

[0008] obtaining current task execution data of the each to-be-scheduled robot in executing the target task allocation scheme, and adjusting the target task allocation scheme of the each to-be-scheduled robot according to the current task execution data to obtain a target task scheduling strategy, so that the each to-be-scheduled robot executes the target task allocation scheme allocated by the target task scheduling strategy.

[0009] Optionally, the generating a target task allocation scheme corresponding to each to-be-scheduled robot according to the current state information comprises:

[0010] generating an initial task allocation scheme for each robot to be scheduled;

[0011] obtaining an initial fitness value corresponding to the initial task allocation scheme according to the current state information, wherein the initial fitness value is used to represent the current state information and task execution data when each robot to be scheduled executes the initial task allocation scheme;

[0012] controlling each robot to be scheduled to execute the initial task allocation scheme to obtain a current fitness value;

[0013] comparing the current fitness value with the initial fitness value, and adjusting the initial task allocation scheme of the current robot to be scheduled based on a particle swarm optimization (PSO) algorithm and a comparison result to obtain a target fitness value of the current robot to be scheduled executing the adjusted initial task allocation scheme, wherein the current robot to be scheduled is any robot to be scheduled executing the initial task allocation scheme;

[0014] if the target fitness value is an optimal fitness value, determining the adjusted initial task allocation scheme as the target task allocation scheme.

[0015] Optionally, after obtaining the target fitness value of the current robot to be scheduled executing the adjusted initial task allocation scheme, the method further comprises:

[0016] obtaining the target fitness value of all robots to be scheduled executing the adjusted initial task allocation scheme;

[0017] comparing all target fitness values, finding an optimal fitness value from the target fitness values, and determining a task allocation scheme corresponding to the optimal fitness value, wherein the optimal fitness value corresponds to an optimal robot to be scheduled;

[0018] adjusting the initial task allocation scheme of an effective robot to be scheduled according to the task allocation scheme corresponding to the optimal fitness value to obtain the target task allocation scheme corresponding to each robot to be scheduled, wherein the effective robot to be scheduled is any robot to be scheduled other than the optimal robot to be scheduled from all robots to be scheduled.

[0019] Optionally, the current task execution data of each robot to be scheduled in executing the target task allocation scheme comprises:

[0020] During execution of the target task allocation scheme, each robot to be scheduled is monitored based on a deep deterministic policy gradient (DDPG) algorithm to obtain real-time state information of all robots to be scheduled, wherein the real-time state information is used to represent real-time running states of the robots to be scheduled.

[0021] The real-time state information is taken as a state input of the DDPG algorithm to obtain a running state vector, wherein the running state vector includes running states of the robots to be scheduled.

[0022] The current task execution data is obtained according to the running state vector.

[0023] Optionally, the target task allocation scheme of each robot to be scheduled is adjusted according to the current task execution data to obtain a target task scheduling strategy, including:

[0024] The running state vector is evaluated to obtain a target state action expectation of the robots to be scheduled under the real-time running state when the target task allocation scheme is executed, wherein the target state action expectation is used to represent expectations of different actions taken by the robots to be scheduled when the target task allocation scheme is executed.

[0025] A historical state action expectation is obtained, and a target task evaluation data is determined according to the target state action expectation and the historical state action expectation, wherein the historical state action is used to represent expectations of different actions taken by the robots to be scheduled when any task allocation scheme is executed in history.

[0026] The target task scheduling strategy is generated according to the target task evaluation data.

[0027] Optionally, after the target task scheduling strategy is obtained, the method further includes:

[0028] A real-time task allocation instruction is generated according to the target task scheduling strategy, wherein the real-time task allocation instruction is used to instruct the robot to be scheduled to adjust one or more of a task priority, a task planning path, or a task start-stop state in the target task allocation scheme.

[0029] According to the real-time task allocation instruction, one or more of the task priority, the task planning path, or the task start-stop state in the target task allocation scheme corresponding to the robot to be scheduled is adjusted to obtain the allocated target task allocation scheme.

[0030] Optionally, after the target task scheduling strategy is obtained, the method further includes:

[0031] record target task execution data of the all target task allocation schemes when all the target task allocation schemes are executed;

[0032] analyze the target task execution data to adjust a next task allocation scheme according to the target task execution data to obtain a next task scheduling strategy; wherein the next task allocation scheme is a next task allocation scheme of the each robot to be scheduled, and the next task scheduling strategy is a next task scheduling strategy of the each robot to be scheduled after the next task allocation scheme is adjusted.

[0033] The second aspect of the embodiment of the present application provides a robot task scheduling system, comprising:

[0034] an acquisition unit configured to acquire current state information of each robot to be scheduled in a robot cluster;

[0035] a generation unit configured to generate a target task allocation scheme corresponding to the each robot to be scheduled according to the current state information;

[0036] an allocation unit configured to generate a current task allocation instruction of the target task allocation scheme, and allocate the target task allocation scheme to the corresponding each robot to be scheduled according to the current task allocation instruction, so as to control the each robot to be scheduled to execute the target task allocation scheme;

[0037] an adjustment unit configured to acquire current task execution data of the each robot to be scheduled in executing the target task allocation scheme, and adjust the target task allocation scheme of the each robot to be scheduled according to the current task execution data to obtain a target task scheduling strategy, so that the each scheduled robot executes the target task allocation scheme allocated by the target task scheduling strategy.

[0038] The robot task scheduling system provided by the second aspect of the embodiment of the present application is used to execute the robot task scheduling method of the first aspect.

[0039] The third aspect of the embodiment of the present application provides a robot task scheduling device, comprising:

[0040] a central processing unit, a memory, an input and output interface, a wired or wireless network interface and a power supply;

[0041] the memory is a transitory storage memory or a persistent storage memory;

[0042] the central processing unit is configured to communicate with the memory, and execute instruction operation in the memory to execute the robot task scheduling method of the first aspect.

[0043] The fourth aspect of the embodiment of the present application provides a computer readable storage medium, the computer readable storage medium comprises instructions, when the instructions are executed on a computer, the computer executes the task scheduling method of the robot in the first aspect.

[0044] The fifth aspect of the embodiment of the present application provides a computer program product, the computer program product comprises instructions, when the instructions are executed on a computer, the computer executes the task scheduling method of the robot in the first aspect.

[0045] From the above technical solutions, the embodiment of the present application has the following advantages: through the task scheduling method of the robot disclosed in the embodiment of the present application, the current state information of each to-be-scheduled robot in the robot cluster is first acquired; then, according to the current state information, a target task allocation scheme corresponding to each to-be-scheduled robot is generated; then, the current task allocation instruction of the target task allocation scheme is generated, and the target task allocation scheme is allocated to each corresponding to-be-scheduled robot according to the current task allocation instruction, so as to control each to-be-scheduled robot to execute the target task allocation scheme; finally, the current task execution data of each to-be-scheduled robot in executing the target task allocation scheme is acquired, and the target task allocation scheme of each to-be-scheduled robot is adjusted according to the current task execution data, to obtain a target task scheduling strategy, so that each to-be-scheduled robot executes the target task allocation scheme allocated by the target task scheduling strategy. Therefore, through the self-adaptive adjustment in the dynamic environment, the optimal task allocation scheme is found in the global range, and the strategy is adjusted according to the real-time state information of the robot, so that the efficient scheduling of the multi-robot system is realized, and good adaptability is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art according to these drawings.

[0047] Figure 1 The flowchart of the task scheduling method of the robot disclosed in the embodiment of the present application;

[0048] Figure 2 The flowchart of another task scheduling method of the robot disclosed in the embodiment of the present application;

[0049] Figure 3 The flowchart of another task scheduling method of the robot disclosed in the embodiment of the present application;

[0050] Figure 4A structural schematic diagram of a task scheduling system of a robot disclosed in an embodiment of the present application;

[0051] Figure 5 A structural schematic diagram of a task scheduling device of a robot disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In modern industry and logistics fields, multi-robot systems have been widely applied in task execution, warehouse management, automated production lines, etc. These robots usually interact with the control center through various communication protocols (such as HTTP, TCP, UDP, etc.) to report their state information in real time and receive task instructions. However, the existing technology has many problems in task allocation and scheduling optimization, especially when dealing with complex multi-robot task scheduling, it is easy to fall into a local optimal solution and it is difficult to achieve global optimal scheduling. With the development of intelligent algorithms, reinforcement learning and optimization algorithms have been gradually applied to the scheduling optimization of multi-robot systems. However, using reinforcement learning or traditional optimization algorithms alone often cannot balance the global search ability and dynamic environment adaptability. Specifically, the embodiments of the present application propose a scheme combining deep deterministic policy gradient (DDPG) and reinforcement learning algorithm with particle swarm optimization (PSO) algorithm, aiming to improve the task allocation and scheduling efficiency of multi-robot systems.

[0053] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0054] To solve the above-mentioned technical problems, please refer to Figure 1 , Figure 1 A flowchart of a task scheduling method of a robot disclosed in an embodiment of the present application. It includes steps 101-104.

[0055] 101, obtain the current state information of each robot to be scheduled in the robot cluster.

[0056] It should be noted that the embodiment can be run in a multi-robot system. The multi-robot system includes a plurality of robots and a control device. The control device controls the movement or stop of the robots by sending instructions or other means, and the specific implementation is not limited here. Further, the robot mentioned in the foregoing is the robot to be scheduled described in the foregoing. The robot to be scheduled can be an intelligent transport robot (AGV, automated guided vehicle) or other types of robots, and the specific implementation is not limited here and will not be described here.

[0057] In the embodiment, the multi-robot system needs to obtain the current state information of each robot to be scheduled in the robot cluster.

[0058] In one specific embodiment, when the multi-robot system is ready to start a task, the task scheduling can be started to obtain the robot state. Specifically, the communication module in the multi-robot system first collects the state information (i.e., the current state information in the foregoing) of each robot. The state information includes but is not limited to the current coordinate position of the robot, the battery power, the execution progress of the allocated task, or the current execution state, etc. These state information provides key data support for task allocation and scheduling.

[0059] 102. According to the current state information, a target task allocation scheme corresponding to each robot to be scheduled is generated.

[0060] Based on step 101, a target task allocation scheme corresponding to each robot to be scheduled can be generated according to the current state information.

[0061] In one specific embodiment, the multi-robot system can combine the current state information of each robot to be scheduled as a condition or initial value. The multi-robot system first initializes the task allocation scheme using the PSO algorithm, and then simulates the target task allocation scheme corresponding to each robot to be scheduled by taking the current state information as the condition or initial value. Specifically, PSO determines a globally optimal task allocation scheme by simulating the process of a group of particles searching for the optimal solution in the search space.

[0062] In one specific embodiment, the multi-robot system regards each robot (AGV) to be scheduled as a "particle", and each "particle" represents a possible task allocation scheme. For example, task A can be completed by AGV1, task B can be completed by AGV2, and so on. Then, the multi-robot system gradually finds a task allocation scheme that gradually approaches the global optimal solution by adjusting the task allocation scheme of each robot to be scheduled by simulating the current state information of each robot to be scheduled, such as the current coordinate position, battery level, execution progress of the allocated task, or current execution state, and so on, and then adjusting the task allocation scheme of each robot to be scheduled, thereby ensuring that the overall efficiency of the task allocation is the highest and the energy consumption is the lowest. The task allocation scheme is the target task allocation scheme described above.

[0063] 103. Generate a current task allocation instruction of the target task allocation scheme, and allocate the target task allocation scheme to each corresponding robot to be scheduled according to the current task allocation instruction, to control each robot to be scheduled to execute the target task allocation scheme.

[0064] Based on step 102, after obtaining the target task allocation scheme, the multi-robot system can generate the current task allocation quality of the target task allocation scheme, and then allocate the target task allocation scheme to each corresponding robot to be scheduled according to the current task allocation instruction. Thus, each robot to be scheduled can be controlled to execute the target task allocation scheme through the current task allocation instruction.

[0065] In one specific embodiment, after obtaining the task allocation scheme of each robot to be scheduled, the task allocation needs to be executed, and the instruction of the task allocation needs to be sent. Specifically, according to the initial strategy provided by the PSO, the multi-robot system gives each robot the target task allocation scheme. The specific instruction (i.e., the current task allocation quality described above) can be sent. After the robot to be scheduled receives the instruction, it starts to execute the task according to the plan. This step marks the formal start of the task scheduling, and each robot to be scheduled is clear about its work target.

[0066] 104. Obtain the current task execution data of each robot to be scheduled in executing the target task allocation scheme, and adjust the target task allocation scheme of each robot to be scheduled according to the current task execution data, to obtain a target task scheduling strategy, so that each scheduled robot executes the target task allocation scheme allocated by the target task scheduling strategy.

[0067] Based on step 103, the current task execution data of each to-be-scheduled robot in the process of executing the target task allocation scheme can be obtained. Then, the target task allocation scheme of each to-be-scheduled robot is adjusted according to the current task execution data, and a target task scheduling strategy is obtained. Thus, each scheduled robot executes the target task allocation scheme after being allocated by the target task scheduling strategy.

[0068] In one specific embodiment, after sending the current task allocation instruction, the multi-robot system also monitors the state information of each to-be-scheduled robot in real time. During task execution, the multi-robot system uses the DDPG algorithm for real-time monitoring and adjustment. In the DDPG algorithm, the multi-robot system dynamically adjusts the task allocation strategy according to the real-time data fed back during robot task execution. Specifically, the current task execution data of the to-be-scheduled robot in the process of executing the target task allocation scheme is obtained, and then the current task execution data is input as real-time data into the DDPG algorithm, so as to evaluate the real-time data. Then, according to the subsequent evaluation result and in combination with the current state of each to-be-scheduled robot, the policy network of the multi-robot system outputs or generates a new action instruction, thereby obtaining a target task scheduling strategy. The target task scheduling strategy includes but is not limited to adjusting the order of task execution, changing the path, or transferring the task to other more suitable robots. Thus, in combination with the above description, the multi-robot system can execute the new target task allocation scheme after being re-allocated through the target scheduling strategy.

[0069] Through the robot task scheduling method disclosed in this embodiment, the current state information of each to-be-scheduled robot in the robot cluster is first obtained; then, according to the current state information, a target task allocation scheme corresponding to each to-be-scheduled robot is generated; then, a current task allocation instruction of the target task allocation scheme is generated, and the target task allocation scheme is allocated to each corresponding to-be-scheduled robot according to the current task allocation instruction, so as to control each to-be-scheduled robot to execute the target task allocation scheme; finally, the current task execution data of each to-be-scheduled robot in the process of executing the target task allocation scheme is obtained, and the target task allocation scheme of each to-be-scheduled robot is adjusted according to the current task execution data, thereby obtaining a target task scheduling strategy, so that each scheduled robot executes the target task allocation scheme after being allocated by the target task scheduling strategy. Thus, through adaptive adjustment in a dynamic environment, the optimal task allocation scheme is found in the global range, and the strategy is adjusted according to the real-time state information of the robot, thereby realizing efficient scheduling of the multi-robot system, and having good adaptability.

[0070] For the steps 101-103 shown in the above Figure 1 Please refer to Figure 2 ,Figure 2 A flowchart of another task scheduling method of a robot disclosed in an embodiment of the present application. It includes steps 201-207.

[0071] 201. Obtain the current state information of each robot to be scheduled in the robot cluster.

[0072] Step 201 in the present embodiment is similar to step 101 in the foregoing Figure 1 embodiment, and will not be described here. However, it should be noted that in the present embodiment, for the current state information, the urgency and complexity of the task and the priority of the current robot position can be used to quickly determine which robot is more suitable for executing a specific task. For example, a robot that is close to the target task and has sufficient power will be given priority. According to the coordinate position of the current robot to be scheduled, the multi-robot system can calculate the optimal path for the robot to go to the target task. For possible obstacles or path congestion, the multi-robot system will preferentially select robots that can avoid these problems. If the robot is low on power, the multi-robot system will select other robots with sufficient power to execute the task, or adjust the priority of the task of the current robot to be scheduled to avoid the situation of running out of power.

[0073] For the sake of convenience, this will be described in detail later.

[0074] 202. For each robot to be scheduled, generate an initial task allocation scheme, and obtain an initial fitness value corresponding to the initial task allocation scheme according to the current state information.

[0075] Based on step 201, before obtaining the target task allocation scheme, it is necessary to generate the initial task allocation scheme of each robot to be scheduled, so as to obtain the initial fitness value corresponding to the initial task allocation scheme according to the current state information. It should be noted that the initial fitness value is used to represent the current state information and the task execution data when each robot to be scheduled executes the initial task allocation scheme.

[0076] In one specific embodiment, the PSO algorithm can be used to initialize the task allocation scheme in the present embodiment. For the sake of convenience, the PSO algorithm is described as follows.

[0077] The PSO algorithm is an optimization algorithm that simulates the foraging behavior of bird flocks in nature. In this algorithm, "particles" are like individuals in a bird flock, each with its own position and velocity, which "fly" in the search space and seek the optimal solution (just like bird flocks looking for food). The global optimization of PSO is to find the optimal solution in the global range by letting these particles communicate with each other in the entire search space. In a multi-robot system, task allocation is a key issue. For example, there is a group of robots to complete different tasks, we need to decide which robot to perform which task. This process is not simple, because robots may have different capabilities, states, and the urgency of the task, the length of the path, etc. The role of PSO here is to help find the optimal task allocation method. Suppose there are three robots (A, B, C) and three tasks (T1, T2, T3). Without PSO, the allocation may be done by simple rules, such as assigning robots to the nearest task. But this method is easy to fall into local optimization, such as a robot may long-term execute a certain type of task, while ignoring other more suitable tasks. After the introduction of PSO, each robot will run into different task allocation schemes (i.e. try different task combinations), and according to the feedback (such as task completion time, path length, etc.), adjust and update the task allocation strategy. After several iterations, the robot eventually finds a more optimal allocation method overall, such as minimizing energy consumption and task completion time at the same time.

[0078] In the embodiment, specifically, in the initialization stage, the system regards each robot to be scheduled, i.e., AGV, as a "particle", and each "particle" represents a possible task allocation scheme. For example, task A can be completed by AGV1, task B can be completed by AGV2, and so on. Initially, these "particles" randomly generate different task allocation schemes, and each scheme has a corresponding "fitness value", which is determined by factors such as task completion time, energy consumption, etc. It should be noted that the fitness value is obtained according to the state information of the robot. The fitness value is calculated according to the state information of the robot and the feedback of the actual execution of the task. For example: task completion time: the faster the task is completed, the higher the fitness value. Energy consumption: the lower the energy consumption, the higher the fitness value. Path condition: if the path passed by the robot is less blocked, the fitness value will be higher. Robot state: if the power of a certain AGV is insufficient or the load is too large, its fitness value will be lower, because it may not be able to complete the task efficiently. Each "particle" represents a possible task allocation scheme. For example, particle 1 may indicate that "task A is completed by AGV1, and task B is completed by AGV2"; particle 2 may indicate that "task A is completed by AGV2, and task B is completed by AGV1". These particles are initially randomly generated, that is, each robot is randomly allocated a task. Subsequently, the system will evaluate the schemes based on the fitness values of the particles (task allocation schemes). The adjustment and optimization of the task allocation scheme will be based on the feedback of the execution of each robot, and will gradually adjust towards a better direction.

[0079] 203、control each robot to be scheduled to execute the initial task allocation scheme to obtain a current fitness value.

[0080] Based on step 202, after the initialization stage is completed, the adjustment and evaluation stage of the task allocation scheme is entered. The task allocation scheme of each AGV needs to be adjusted and evaluated. Specifically, control each robot to be scheduled to execute the initial task allocation scheme to obtain a current fitness value.

[0081] In one specific embodiment, since each robot to be scheduled (AGV) will execute tasks according to the current task allocation scheme and record the completion of the tasks (such as task execution time, energy consumption, etc.). Thus, by controlling each robot to be scheduled to execute the initial task allocation scheme, the multi-robot system can obtain the current fitness value of the AGV executing the initial task allocation scheme at this time.

[0082] 204、compare the current fitness value with the initial fitness value, and adjust the initial task allocation scheme of the current robot to be scheduled based on the particle swarm optimization (PSO) algorithm and the comparison result to obtain a target fitness value of the current robot to be scheduled executing the adjusted initial task allocation scheme.

[0083] Based on step 203, by comparing the current fitness value with the initial fitness value, the initial task allocation scheme of the current to-be-scheduled robot can be adjusted based on the particle swarm optimization (PSO) algorithm and the comparison result, so as to obtain the target fitness value of the current to-be-scheduled robot executing the adjusted initial task allocation scheme. It should be noted that the current to-be-scheduled robot is the to-be-scheduled robot executing the initial task allocation scheme, and can also be a to-be-scheduled robot AGV in another execution sequence arranged by the multi-robot system.

[0084] In one specific embodiment, the multi-robot system compares the current fitness value with the initial fitness value to determine whether the overall efficiency is improved when different to-be-scheduled robots execute a certain task allocation scheme. Thus, the task allocation scheme is continuously adjusted, and the result is compared with the previous scheme to select a better scheme as a reference. Further, in each iteration, each AGV finds a better scheme under certain conditions by fine-tuning its task allocation, such as allowing different robots to handle different tasks.

[0085] 205、Obtain the target fitness value of all to-be-scheduled robots executing the adjusted initial task allocation scheme, compare the target fitness values, find the optimal fitness value among all target fitness values, and determine the task allocation scheme corresponding to the optimal fitness value.

[0086] Then, the multi-robot system can obtain the target fitness value of all to-be-scheduled robots executing the adjusted initial task allocation scheme, then compare the target fitness values, find the optimal fitness value among all target fitness values, and determine the task allocation scheme corresponding to the optimal fitness value. It should be noted that the to-be-scheduled robot corresponding to the optimal fitness value is the best to-be-scheduled robot.

[0087] In one specific embodiment, in combination with step 204, the multi-robot system obtains the target fitness value of all AGVs executing the adjusted initial task allocation scheme, then compares all target fitness values, so as to filter out the overall optimal fitness value, i.e., the optimal fitness value. The optimal fitness value is a task allocation scheme corresponding to a specific condition of all to-be-scheduled robots.

[0088] Further, the multi-robot system finds a local optimal scheme through group cooperation (information sharing among AGVs), and gradually optimizes to a global optimal scheme. This means that the task allocation scheme finally found by the system can perform better under multiple indicators (such as time, energy consumption, path efficiency, etc.). Thus, after multiple iterations, the target fitness value of the current to-be-scheduled robot executing the adjusted initial task allocation scheme can be obtained.

[0089] It should be noted that steps 204-205 in this embodiment belong to the group cooperation process in the PSO algorithm. The group cooperation process will be described later.

[0090] 206、Adjust the initial task allocation scheme of the effective to-be-scheduled robot according to the task allocation scheme corresponding to the optimal fitness value, to obtain a target task allocation scheme corresponding to each to-be-scheduled robot.

[0091] Then, the initial task allocation scheme of the effective to-be-scheduled robot can be adjusted according to the task allocation scheme corresponding to the optimal fitness value, to obtain a target task allocation scheme corresponding to each to-be-scheduled robot. It should be noted that the effective to-be-scheduled robot is the to-be-scheduled robot other than the best to-be-scheduled robot among all to-be-scheduled robots. For example, the multi-robot system includes AGV1, AGV2 and AGV3. Among them, the best to-be-scheduled robot is AGV2, and the effective to-be-scheduled robot is AGV1 and AGV3.

[0092] In one specific embodiment, after the task allocation scheme corresponding to the optimal fitness value is found, the initial task allocation of the effective to-be-scheduled robot can be adjusted, which can adjust the task execution path or priority, etc. which will not be described here. Then, the target task allocation scheme corresponding to all to-be-scheduled robots can be determined.

[0093] Further, all AGVs can also share the optimal task allocation scheme they find. Assuming that a certain AGV finds a better task allocation method, other AGVs will also refer to this optimal scheme to adjust their task allocation accordingly. This step ensures that the globally optimal task allocation scheme is quickly disseminated and applied.

[0094] 207、When the target fitness value is the optimal fitness value, the adjusted initial task allocation scheme is determined as the target task allocation scheme.

[0095] Based on step 204, if the target fitness value at this time is the optimal fitness value, the adjusted initial task allocation scheme can be directly determined as the target task allocation scheme.

[0096] In one specific embodiment, when it is determined that the target fitness value is the optimal fitness value, it can be determined that the adjusted initial task allocation scheme at this time is the globally optimal task allocation scheme, so as to determine the globally optimal task allocation scheme as the target allocation scheme.

[0097] In other implementable technical solutions, global updating and optimization can also be performed. Specifically, the multi-robot records the global optimal task allocation scheme found at present (among all possible task allocation schemes, a scheme that performs best under certain conditions, such as minimizing task completion time, energy consumption, path optimization, etc., is found), and in the next round of allocation, all AGVs converge to this optimal scheme. Through continuous iteration and optimization, the task allocation scheme will gradually approach the global optimal solution, ensuring the highest overall efficiency and the lowest energy consumption of task allocation. After multiple iterations, the AGV finds a globally optimal task allocation scheme. Under this scheme, each AGV is allocated to the most suitable task, thereby completing the task in the shortest time and saving energy to the maximum extent.

[0098] It should be noted that steps 204-207 in this embodiment are the group cooperation stage in the PSO algorithm. In the group cooperation stage, all AGVs share the optimal task allocation scheme and attempt to adjust it. Then the multi-robot system performs global updating and optimization to determine whether the best task allocation scheme under certain conditions is reached. The best task allocation scheme is the target task allocation scheme described above.

[0099] Through the robot task scheduling method disclosed in this embodiment, the AGV system can effectively find and optimize the task allocation scheme through the PSO algorithm, ensuring efficient and intelligent task scheduling. This method combines the advantages of global search and local adjustment, ultimately enabling the AGV system to quickly find the best task allocation strategy in a complex task environment. This embodiment allocates each task to the most suitable robot and inputs it into the DDPG algorithm as an initial strategy. This step ensures that the basic task allocation scheme has been optimized and fully considered. Further, after combining the PSO algorithm, the multi-robot system can consider the cooperation relationship between multiple robots and find a better task allocation scheme in the global range. For example, the system may find that robot B performs task T1 instead of A, although A is closer to T1, but B has higher execution efficiency. This global optimization can significantly improve the overall efficiency of the system and reduce the time and energy consumption of task completion.

[0100] For the above Figure 1 Step 104, please refer to Figure 3 , Figure 3 is another flowchart of the robot task scheduling method disclosed in the embodiments of the present application. It includes steps 301-307.

[0101] 301. During the execution of the target task allocation scheme, the Deep Deterministic Policy Gradient (DDPG) algorithm is used to monitor each robot to be scheduled, and obtain the real-time status information of all robots to be scheduled.

[0102] It should be noted that steps 301-307 in this embodiment are mainly implemented using the DDPG algorithm. Specifically, during the execution of the target task allocation scheme, each robot to be scheduled can be monitored based on the Deep Deterministic Policy Gradient (DDPG) algorithm to obtain the real-time status information of all robots to be scheduled. It should be noted that the real-time status information is used to characterize the real-time operating status of all robots to be scheduled.

[0103] In one specific embodiment, during the execution of the target task allocation scheme, that is, corresponding to Figure 1 In step 103, based on the initial strategy provided by the PSO, the system assigns tasks to each robot and sends specific instructions. Upon receiving the instructions, the robots begin executing the tasks as planned. This step marks the formal start of task scheduling, with each robot clearly understanding its work objective. Then, the multi-robot system needs to adjust the task allocation strategy in real time. Specifically, during task execution, the system uses the DDPG algorithm for real-time monitoring and adjustment. In the DDPG algorithm, the system dynamically adjusts the task allocation strategy based on real-time data fed back by the robots during task execution. A detailed explanation of the DDPG algorithm follows.

[0104] DDPG (Directed Decision Processing) is a policy gradient-based reinforcement learning method suitable for problems with continuous action spaces. Compared to traditional Q-learning, DDPG can handle complex high-dimensional state and action spaces while exhibiting higher convergence speed and stability. Specific method descriptions: 1. Policy Network and Q-Network: DDPG uses a policy network (Actor) to generate continuous actions and a Q-network (Critic) to evaluate the quality of these actions. The policy network is responsible for deciding what action to take at each step, while the Q-network guides the adjustment of the policy network by evaluating the value of each action. 2. Target Network: To stabilize the training process, DDPG introduces a target policy network and a target Q-network. The parameters of these networks move closer to the parameters of the main network with small steps during each update, reducing training instability. 3. Experience Replay: DDPG also uses an experience replay mechanism, randomly sampling small batches of data from past experiences to update the network. This reduces the correlation between training data and improves the learning effect of the algorithm. Using DDPG techniques: It features adaptive task allocation and path optimization. Specifically, by using DDPG technology, the lifting AGV system can dynamically adjust task allocation and path planning through reinforcement learning. When executing tasks, the AGV not only considers the current location of the goods and the target location, but also adaptively adjusts based on the real-time warehouse environment (such as the status of other AGVs, current traffic conditions, and the presence of obstacles). For example, suppose an AGV is heading to the storage area to retrieve goods and encounters another AGV occupying its planned path. In this case, the DDPG algorithm allows the AGV to quickly evaluate multiple feasible paths and select the optimal alternative. Simultaneously, the DDPG algorithm can also reallocate tasks based on the real-time system status, such as assigning another urgent task to a closer AGV. Thus, through DDPG, the AGV system can continuously optimize its behavior during task execution, making task allocation more flexible and efficient, and reducing delays caused by path conflicts or traffic congestion. Ultimately, the overall task completion efficiency of the system is significantly improved, while resource utilization is also higher.

[0105] 302. Use the real-time status information as the status input of the DDPG algorithm to obtain the running status vector, and obtain the current task execution data based on the running status vector.

[0106] Based on step 301, the real-time data input stage of the DDPG algorithm begins. Specifically, the real-time status information is used as the status input of the DDPG algorithm to obtain the running status vector, and the current task execution data is obtained based on the running status vector.

[0107] In one specific embodiment, feedback data such as path congestion, insufficient power, etc. are used as state inputs. Specifically, in the DDPG algorithm, the system constantly obtains real-time state data from the feedback of the robot performing the task. These data include but are not limited to: power level (detecting whether the battery power of the robot is sufficient to complete the task), task progress (tracking the real-time status of task completion, whether it is within the predetermined time), and position coordinates (obtaining the accurate position of the robot in real time to ensure that it does not deviate from the path). These state data will be used as state inputs of the DDPG algorithm. Then, DDPG will mathematically model the current state of all AGVs to form a multi-dimensional state vector, i.e. a motion state vector. This motion state vector contains real-time information of all AGVs, such as current task, position, power, etc.

[0108] 303、evaluate the motion state vector to obtain the target state action expectation of all the scheduled robots in the real-time running state executing the target task allocation scheme.

[0109] After obtaining the motion state vector, it is necessary to evaluate the motion state vector to obtain the target state action expectation of all the scheduled robots in the real-time running state executing the target task allocation scheme. It should be noted that the target state action expectation is used to represent the expectation of different actions taken by all the scheduled robots when executing the target task allocation scheme.

[0110] In one specific embodiment, the DDPG algorithm can use a Q function in value evaluation. Specifically, DDPG uses a Q network to evaluate the value of each state and action combination. By evaluating the return that the robot can get after taking a certain action in the current state, the target state action expectation is obtained. The Q network can help the system select high-value actions. High return usually means shorter task completion time, less energy consumption or more efficient path.

[0111] 304、obtain historical state action expectation, and determine target task evaluation data according to the target state action expectation and the historical state action expectation.

[0112] Therefore, in combination with step 303, historical state action expectation is also needed, and target task evaluation data is determined according to the target state action expectation and the historical state action expectation. It should be noted that the historical state action is used to represent the expectation of different actions taken by all the scheduled robots when executing any task allocation scheme in history.

[0113] In one specific embodiment, the value network in DDPG predicts different actions that can be taken in the current state and estimates the cumulative rewards that can be obtained in the future under these actions. This evaluation is based on the current feedback data of the robot and combined with historical experience, i.e. historical state-action expectations, to make the evaluation more accurate and obtain target task evaluation data.

[0114] 305. Generating a target task scheduling strategy according to the target task evaluation data.

[0115] Then, a target task scheduling strategy can be generated according to the target task evaluation data.

[0116] In one specific embodiment, the multi-robot system generates a new task scheduling strategy, i.e. a target task scheduling strategy, according to the target task evaluation data by using the policy network in the DDPG algorithm.

[0117] 306. Generating real-time task allocation instructions according to the target task scheduling strategy, and adjusting one or more of the task priority, task planning path or task start-stop state in each robot to be scheduled corresponding to the target task allocation scheme to obtain the allocated target task allocation scheme.

[0118] Based on step 305, the multi-robot system generates real-time task allocation instructions according to the target task scheduling strategy by combining the DDPG algorithm. At the same time, according to the real-time task allocation instructions, one or more of the task priority, task planning path or task start-stop state in each robot to be scheduled corresponding to the target task allocation scheme is adjusted to obtain the allocated target task allocation scheme. It should be noted that the real-time task allocation instructions are used to indicate the adjustment of one or more of the task priority, task planning path or task start-stop state in each robot to be scheduled corresponding to the target task allocation scheme.

[0119] In one specific embodiment, DDPG first updates the policy network and then controls the robot to perform actions. Specifically, the policy network in DDPG generates real-time task allocation instructions, i.e. new action instructions, according to the current state, such as adjusting the order of task execution, changing the path, or transferring the task to other more suitable robots AGV. Then, the multi-robot system will output the new instructions (real-time task allocation instructions) to the robot according to the output of the policy network. This may include re-planning the path, modifying the task priority, or even stopping the task of a robot in an emergency to avoid failure. These supplements can make the entire process clearer and more detailed, helping to better understand how to use these data, how the algorithm works, and how the task allocation strategy is dynamically adjusted. Ensure that the system has a high degree of adaptability to environmental changes and ensure that the task can be completed smoothly.

[0120] 307、When all target task allocation schemes are executed, record the target task execution data of all target task allocation schemes, and analyze the target task execution data to adjust the next task allocation scheme according to the target task execution data, and obtain the next task scheduling strategy.

[0121] Based on step 305, it needs to be supplemented that after all target task allocation schemes are executed, the target task execution data of all target task allocation schemes can also be recorded, and the target task execution data is analyzed to adjust the next task allocation scheme according to the target task execution data, and obtain the next task scheduling strategy. It needs to be pointed out that the next task allocation scheme is the next task allocation scheme of each robot to be scheduled, and the next task scheduling strategy is the next task scheduling strategy of each robot to be scheduled after the next task allocation scheme is adjusted.

[0122] In one specific embodiment, the DDPG algorithm can also perform continuous learning and optimization. At the same time, record and analyze the execution data of the task allocation scheme in the execution process. Specifically, DDPG not only adjusts the task in real time, but also continuously learns and optimizes the strategy during the task execution process. Through repeated learning and adjustment, DDPG can gradually optimize the overall efficiency of task scheduling. For example, the multi-robot system can identify which paths are more prone to congestion and which tasks are more suitable for which robots, thereby avoiding inefficient decisions in future task scheduling. This process makes the system more intelligent and efficient in dealing with complex tasks.

[0123] Further, when all tasks are successfully completed, the multi-robot system will record detailed data of the entire task execution. These data include task completion time, resource consumption of robots, unexpected situations, and adjustments made by the system during the process. By analyzing these data, the system can find optimization space, so as to improve the strategy in the next task scheduling. In the long run, these records and analyses will significantly improve the overall performance of the system.

[0124] Through the task scheduling method of the robot disclosed in the embodiment, the entire process is more clear and detailed, helping to better understand how to use these data, how the algorithm works, and how the task allocation strategy dynamically adjusts. It ensures that the system has a high adaptability to environmental changes and ensures that the task can be completed smoothly. At the same time, DDPG not only adjusts the task in real time, but also continuously learns and optimizes the strategy during task execution. Through repeated learning and adjustment, DDPG can gradually optimize the overall efficiency of task scheduling. By analyzing these data, the system can find optimization space, thereby improving the strategy in the next task scheduling. In the long run, these records and analyses will significantly improve the overall performance of the system. Further, the embodiment combines the DDPG reinforcement learning algorithm with the particle swarm optimization (PSO) algorithm to achieve dynamic task allocation and optimization in a multi-robot system. PSO is responsible for initially finding the globally optimal task allocation scheme, while reinforcement learning (such as the DDPG algorithm) fine-tunes according to real-time data during task execution, ensuring that the system can adapt to dynamically changing environments. This combination not only improves the efficiency of global search, but also enhances the adaptability and real-time optimization capabilities of the system.

[0125] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or steps or stages in other steps.

[0126] Please refer to Figure 4 , Figure 4 The structure diagram of the task scheduling system of the robot disclosed in the embodiment of the application is shown.

[0127] The acquisition unit 401 is configured to acquire current state information of each to-be-scheduled robot in the robot cluster.

[0128] The generation unit 402 is configured to generate a target task allocation scheme corresponding to each to-be-scheduled robot according to the current state information.

[0129] The distribution unit 403 is configured to generate a current task distribution instruction of a target task distribution scheme, and distribute the target task distribution scheme to each corresponding robot to be dispatched according to the current task distribution instruction, so as to control each robot to be dispatched to execute the target task distribution scheme.

[0130] The adjustment unit 404 is configured to obtain current task execution data of each robot to be dispatched in executing the target task distribution scheme, and adjust the target task distribution scheme of each robot to be dispatched according to the current task execution data, to obtain a target task scheduling strategy, so that each dispatched robot executes the target task distribution scheme after being distributed by the target task scheduling strategy.

[0131] Exemplarily, the system further comprises a control unit 405 and a determination unit 406.

[0132] The generation unit 402 is specifically configured to generate, for each robot to be dispatched, an initial task distribution scheme.

[0133] The acquisition unit 401 is specifically configured to obtain, according to the current state information, an initial fitness value corresponding to the initial task distribution scheme; wherein the initial fitness value is used to represent the current state information and task execution data of each robot to be dispatched in executing the initial task distribution scheme.

[0134] The control unit 405 is configured to control each robot to be dispatched to execute the initial task distribution scheme, to obtain a current fitness value.

[0135] The adjustment unit 404 is specifically configured to compare the current fitness value with the initial fitness value, and adjust the initial task distribution scheme of the current robot to be dispatched based on a particle swarm optimization (PSO) algorithm and a comparison result, to obtain a target fitness value of the current robot to be dispatched in executing the adjusted initial task distribution scheme; wherein the current robot to be dispatched is any robot to be dispatched in executing the initial task distribution scheme.

[0136] The determination unit 406 is configured to determine the adjusted initial task distribution scheme as the target task distribution scheme if the target fitness value is an optimal fitness value.

[0137] Exemplarily, the system further comprises:

[0138] The acquisition unit 401 is further configured to obtain target fitness values of all robots to be dispatched in executing the adjusted initial task distribution scheme.

[0139] The determination unit 406 is further configured to compare all target fitness values, find an optimal fitness value from all target fitness values, and determine a task distribution scheme corresponding to the optimal fitness value; wherein the robot to be dispatched corresponding to the optimal fitness value is the best dispatched robot.

[0140] The adjusting unit 404 is further configured to adjust the initial task allocation scheme of the effective to-be-scheduled robot according to the task allocation scheme corresponding to the optimal fitness value, to obtain a target task allocation scheme corresponding to each to-be-scheduled robot; wherein the effective to-be-scheduled robot is a to-be-scheduled robot other than the optimal scheduling robot among all to-be-scheduled robots.

[0141] Exemplarily, the system comprises:

[0142] The obtaining unit 401 is configured to monitor each to-be-scheduled robot based on a deep deterministic policy gradient (DDPG) algorithm during execution of the target task allocation scheme, to obtain real-time state information of all to-be-scheduled robots; wherein the real-time state information is used to represent real-time running states of all to-be-scheduled robots.

[0143] The obtaining unit 401 is further configured to take the real-time state information as a state input of the DDPG algorithm, to obtain a running state vector; wherein the running state vector comprises running states of all to-be-scheduled robots.

[0144] The obtaining unit 401 is further configured to obtain current task execution data according to the running state vector.

[0145] Exemplarily, the system comprises:

[0146] The obtaining unit 401 is configured to evaluate the running state vector, to obtain a target state-action expectation of all to-be-scheduled robots in the real-time running state executing the target task allocation scheme; wherein the target state-action expectation is used to represent expectations of different actions taken by all to-be-scheduled robots when executing the target task allocation scheme.

[0147] The obtaining unit 401 is further configured to obtain a historical state-action expectation, and determine target task evaluation data according to the target state-action expectation and the historical state-action expectation; wherein the historical state-action expectation is used to represent expectations of different actions taken by all to-be-scheduled robots when executing any task allocation scheme in history.

[0148] The generating unit 402 is configured to generate a target task scheduling strategy according to the target task evaluation data.

[0149] Exemplarily, the system further comprises:

[0150] The generating unit 402 is further configured to generate a real-time task allocation instruction according to the target task scheduling strategy; wherein the real-time task allocation instruction is used to instruct each to-be-scheduled robot to adjust one or more of a task priority, a task planning path, or a task start-stop state of the target task allocation scheme.

[0151] The adjusting unit 404 is further configured to adjust one or more of the task priority, the task planning path, or the task start-stop state in each robot to be scheduled according to the real-time task allocation instruction, to obtain the target task allocation scheme after allocation.

[0152] Exemplarily, the system further comprises a recording unit 407.

[0153] The recording unit 407 is configured to record target task execution data of all target task allocation schemes when execution of all target task allocation schemes is completed.

[0154] The obtaining unit 401 is further configured to analyze the target task execution data, to adjust a next task allocation scheme according to the target task execution data, to obtain a next task scheduling strategy; the next task allocation scheme is a next task allocation scheme of each robot to be scheduled, and the next task scheduling strategy is a next task scheduling strategy of each robot to be scheduled after adjustment of the next task allocation scheme.

[0155] Hereinafter, please refer to Figure 5 The structure diagram of the task scheduling device of the robot disclosed in the embodiments of the present application comprises:

[0156] The central processing unit 501, the memory 505, the input-output interface 504, the wired or wireless network interface 503, and the power supply 502;

[0157] The memory 505 is a transitory storage memory or a persistent storage memory;

[0158] The central processing unit 501 is configured to communicate with the memory 505, and execute instruction operations in the memory 505 to perform the task scheduling method of the robot in any of the foregoing embodiments. Figures 1 to 3 The task scheduling method of the robot in any of the foregoing embodiments.

[0159] The embodiments of the present application further provide a chip system, which comprises at least one processor and a communication interface, the communication interface and the at least one processor are interconnected through a line, and the at least one processor is used to run a computer program or instruction to execute the task scheduling method of the robot in any of the foregoing embodiments. Figures 1 to 3 The task scheduling method of the robot in any of the foregoing embodiments.

[0160] The embodiments of the present application further provide a computer readable storage medium, which comprises an instruction, when the instruction is run on a computer, the computer executes the task scheduling method of the robot in any of the foregoing embodiments. Figures 1 to 3 The task scheduling method of the robot in any of the foregoing embodiments.

[0161] The embodiments of the present application further provide a computer program product comprising an instruction, when the computer program product is run on a computer, the computer executes the task scheduling method of the robot in any of the foregoing embodiments.Figures 1 to 3 The task scheduling method of the robot in any of the illustrated embodiments.

[0162] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the system, device and unit described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0163] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division, and actual implementation can have another division manner. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0164] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., they can be located in one place or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0165] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0166] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, read-only memory), a random access memory (RAM, random access memory), a magnetic disk or an optical disk, and various media that can store program codes.

Claims

1. A method for task scheduling of a robot, characterized in that, The method comprises: obtaining current state information of each robot to be scheduled in a robot cluster; generating a target task allocation scheme corresponding to each robot to be scheduled according to the current state information; generating a current task allocation instruction of the target task allocation scheme, and allocating the target task allocation scheme to the corresponding each robot to be scheduled according to the current task allocation instruction, to control the each robot to be scheduled to execute the target task allocation scheme; obtaining current task execution data of the each robot to be scheduled in executing the target task allocation scheme, and adjusting the target task allocation scheme of the each robot to be scheduled according to the current task execution data to obtain a target task scheduling strategy, so that the each robot to be scheduled executes the target task allocation scheme allocated by the target task scheduling strategy; the method further comprises: monitoring the each robot to be scheduled based on a deep deterministic policy gradient (DDPG) algorithm during execution of the target task allocation scheme, to obtain real-time state information of all robots to be scheduled, wherein the real-time state information is used to represent real-time running states of the all robots to be scheduled; taking the real-time state information as a state input of the DDPG algorithm to obtain a running state vector, wherein the running state vector comprises running states of the all robots to be scheduled; obtaining the current task execution data according to the running state vector.

2. The method of claim 1, wherein, the method further comprises: generating an initial task allocation scheme for the each robot to be scheduled; obtaining an initial fitness value corresponding to the initial task allocation scheme according to the current state information, wherein the initial fitness value is used to represent the current state information and task execution data when the each robot to be scheduled executes the initial task allocation scheme; controlling the each robot to be scheduled to execute the initial task allocation scheme to obtain a current fitness value; comparing the current fitness value with the initial fitness value, and adjusting the initial task allocation scheme of a current robot to be scheduled based on a particle swarm optimization (PSO) algorithm and a comparison result, to obtain a target fitness value of the current robot to be scheduled executing the adjusted initial task allocation scheme, wherein the current robot to be scheduled is any robot to be scheduled executing the initial task allocation scheme; if the target fitness value is an optimal fitness value, determining the adjusted initial task allocation scheme as the target task allocation scheme.

3. The method of claim 2, wherein, after obtaining the target fitness value of the current robot to be scheduled executing the adjusted initial task allocation scheme, the method further comprises: obtaining the target fitness value of all robots to be scheduled executing the adjusted initial task allocation scheme; Compare all target fitness values, find an optimal fitness value from the all target fitness values, and determine a task allocation scheme corresponding to the optimal fitness value; wherein the optimal fitness value corresponds to an optimal scheduling robot; According to the task allocation scheme corresponding to the optimal fitness value, adjust the initial task allocation scheme of the effective to-be-scheduled robot to obtain the target task allocation scheme corresponding to each to-be-scheduled robot; wherein the effective to-be-scheduled robot is a to-be-scheduled robot other than the optimal scheduling robot among all to-be-scheduled robots.

4. The method of claim 1, wherein, The method further comprises: According to the target state action expectation and the historical state action expectation, determine target task evaluation data; wherein the historical state action is used to represent the expectation of different actions taken by the all to-be-scheduled robots when executing any task allocation scheme in history. According to the target task evaluation data, generate the target task scheduling strategy. After obtaining the target task scheduling strategy, the method further comprises:

5. The method of claim 1, wherein, According to the target task scheduling strategy, generate real-time task allocation instructions; wherein the real-time task allocation instructions are used to instruct the to-be-scheduled robot to adjust one or more of the task priority, the task planning path, or the task start-stop state of the target task allocation scheme; According to the real-time task allocation instructions, adjust one or more of the task priority, the task planning path, or the task start-stop state corresponding to the target task allocation scheme in the to-be-scheduled robot to obtain the allocated target task allocation scheme. After obtaining the target task scheduling strategy, the method further comprises:

6. The method of claim 1, wherein, When all target task allocation schemes are executed, record the target task execution data of all target task allocation schemes; Analyze the target task execution data to adjust the next task allocation scheme according to the target task execution data to obtain a next task scheduling strategy; wherein the next task allocation scheme is the next task allocation scheme of each to-be-scheduled robot, and the next task scheduling strategy is the next task scheduling strategy of each to-be-scheduled robot after adjusting the next task allocation scheme. The system comprises:

7. A task scheduling system of a robot characterized by comprising: An acquisition unit is configured to acquire current state information of each to-be-scheduled robot in a robot cluster; A generation unit is configured to generate a target task allocation scheme corresponding to each to-be-scheduled robot according to the current state information. ​ The allocation unit is configured to generate a current task allocation instruction of the target task allocation scheme, and allocate the target task allocation scheme to each robot to be dispatched according to the current task allocation instruction, so as to control each robot to be dispatched to execute the target task allocation scheme. The adjustment unit is configured to obtain current task execution data of each robot to be dispatched in executing the target task allocation scheme, and adjust the target task allocation scheme of each robot to be dispatched according to the current task execution data, to obtain a target task scheduling strategy, so that each robot to be dispatched executes the target task allocation scheme after being allocated by the target task scheduling strategy. The system comprises: The acquisition unit is specifically configured to monitor each robot to be dispatched based on a deep deterministic policy gradient (DDPG) algorithm during execution of the target task allocation scheme, to obtain real-time state information of all robots to be dispatched; wherein the real-time state information is used to represent real-time running states of all robots to be dispatched. The acquisition unit is further configured to take the real-time state information as a state input of the DDPG algorithm, to obtain a running state vector; wherein the running state vector comprises running states of all robots to be dispatched. The acquisition unit is further configured to obtain the current task execution data according to the running state vector.

8. A task scheduling apparatus of a robot characterized by comprising: The device comprises: A central processor, a memory, an input-output interface, a wired or wireless network interface, and a power supply; The memory is a transitory storage memory or a persistent storage memory; The central processor is configured to communicate with the memory, and execute instruction operations in the memory to perform the robot task scheduling method in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium comprises instructions, when the instructions are executed on a computer, the computer executes the robot task scheduling method in any one of claims 1 to 6.

Images