A method for robot task load balancing

Through the computer robot's residual energy standard deviation and task transfer decision model, the problem of unbalanced task load in the robot team is solved, energy balance and team integrity are achieved, and it is suitable for a variety of task scenarios.

CN115080245BActive Publication Date: 2024-11-26THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202210809253.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-11
Publication Date
2024-11-26
Estimated Expiration
2042-07-11

AI Technical Summary

Technical Problem

The existing technology is difficult to achieve balance of task load in robot teams, resulting in the exhaustion of energy of a certain robot and the loss of team integrity.

Method used

Through the standard deviation of the remaining energy of the computer robot, the task is estimated and transferred, so that the robot with energy below the mean is selected as the task transfer robot, and the task transfer decision model is established using the convolutional neural network, and the task allocation is performed according to the energy and task state to ensure energy balanced.

Benefits of technology

It achieves balanced energy consumption within the robot team, avoids early energy exhaustion, and ensures the continuous integrity of the team. It is suitable for a variety of task scenarios such as robot football matches and robot handling.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115080245B_ABST
    Figure CN115080245B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for robot task load balancing, and belongs to the field of robot task collaborative scheduling. The present invention comprises the steps of: calculating the standard variance of the robot's remaining energy; judging whether the standard variance of the remaining energy exceeds a set threshold; determining the robot that needs to transfer out the task; determining the robot that undertakes the transferred task; and executing task transfer. The present invention can be used for collaborative scheduling of tasks between robots in a collaborative work scenario, achieving energy consumption balance between individuals in a robot team, and avoiding the loss of integrity of the robot team due to premature exhaustion of energy of a certain robot.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for robot task load balancing, which is used for collaborative scheduling of tasks between robots in a collaborative work scenario, so as to achieve energy consumption balance among individuals in a robot team and avoid the loss of integrity of the robot team due to premature exhaustion of energy of a certain robot. Background Art

[0002] With the rapid development of artificial intelligence technology, robots have been widely used in many industries such as product manufacturing, agricultural production, and robot games. In many cases, robots usually work under the drive of stored energy. Once the energy is exhausted, they lose the ability to work. On the other hand, during the application process, robots usually complete a task through division of labor and cooperation. In order to ensure the completion of the task, the robot team must be complete. For example, in a robot football game, the absence of an individual often leads to a decline in the ability of the entire team to perform the task, and even leads to the failure of the task.

[0003] Therefore, there is an urgent need in the prior art for a task load balancing method, which can ensure the energy consumption of each robot is balanced by balancing the task load, thereby ensuring the continuous integrity of the robot team. Summary of the invention

[0004] In view of this, the present invention proposes a method for balancing robot task loads. The method can achieve energy consumption balance of a robot team and ensure the continuous integrity of the robot team.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for robot task load balancing comprises the following steps:

[0007] (1) Calculate the standard deviation of the actual remaining energy of all robots;

[0008] (2) Determine whether the standard deviation exceeds a first threshold value. If so, execute step (3); otherwise, return to step (1).

[0009] (3) Estimate the estimated remaining energy of each robot after completing the remaining tasks, calculate the mean of the estimated remaining energy, and calculate the difference between the estimated remaining energy of each robot and the mean of the estimated remaining energy, and select the robot whose estimated remaining energy is lower than the mean of the estimated remaining energy by more than a second threshold as the task transfer-out robot;

[0010] (4) Sort all task transfer-out robots in ascending order of estimated remaining energy, and determine the task taking-over robot for each task transfer-out robot in sequence according to the task transfer strategy;

[0011] (5) The task-transferring robot performs the task transfer, and at the same time, the corresponding task-taking robot receives the transferred task. After all tasks are transferred in place, they start to execute the tasks according to the new task division and return to step (1).

[0012] Furthermore, in step (4), the specific method of determining the task taking-over robot of the outgoing task according to the task transfer strategy is as follows:

[0013] (4.1) Establish a task transfer decision model:

[0014] The task transfer decision model includes a task transfer decision agent and a multi-robot collaborative work scenario, wherein the task transfer decision agent sends the task offloading decision to the multi-robot collaborative work scenario, and each task-taking robot takes over the transferred task and carries out the work according to the arrangement of the task offloading decision, forming a new task execution state, and feeding back to the task transfer decision agent; in addition, the actual effect evaluation of the task offloading decision is also fed back to the task transfer decision agent, wherein if the standard deviation of the actual remaining energy is smaller than that before the task offloading is implemented, a positive evaluation is given, otherwise a negative evaluation is given;

[0015] The task transfer decision agent is a convolutional neural network. The input layer includes n sets of data, where n is the number of robots. Each set of data includes three values, namely the normalized value of the robot's remaining energy, the normalized value of the robot's remaining tasks, and the normalized value of the distance between the robot and the task-taking robot. The output layer includes the proportion of tasks shared by other robots to the remaining tasks of the task-transferring robot before the task is transferred out.

[0016] (4.2) Training the task transfer decision model:

[0017] According to the initial state of the multi-robot collaborative work scenario, the task transfer decision agent randomly gives an initial task offloading decision to drive the task transfer decision model training process to start working. The multi-robot collaborative work scenario regularly feeds back the task offloading decision reward information and task execution status information to the task transfer decision agent as the basis for its next task decision, and the process is repeated in this way. During the iteration process, the variance value of the actual remaining energy of the robot is used as a constraint requirement. When the variance value is lower than the third threshold, the iteration is stopped and the current round of training is completed. Through multiple rounds of training, the adaptability of the task transfer decision agent to various task scenarios is improved. In each round of training, the initial state of the multi-robot collaborative work scenario and the initial task offloading decision given by the task transfer decision agent are different.

[0018] (4.3) Test the trained task transfer decision model:

[0019] Use a multi-robot collaborative work scenario different from the training phase to test the task transfer decision model obtained through training to verify whether it can play a role in task load balancing in the test collaborative work scenario. If it cannot play a role, go to step (4.2) and train the task transfer decision model again.

[0020] (4.4) Use the tested task transfer decision model to determine the robot that will take over the outgoing task.

[0021] The beneficial effects of the present invention are:

[0022] 1. Based on the energy consumption model of local computing, remote computing and data transmission, the present invention uses artificial intelligence methods to find the optimal task load distribution strategy from the task load balancing strategy space under multiple constraints, and determines whether a task is completed locally or transmitted to other nodes to complete, thereby maintaining a basic balance in the energy consumption of each robot and effectively achieving the continuous integrity of the robot team.

[0023] 2. The present invention can be used for collaborative scheduling of tasks between robots in collaborative work scenarios, to achieve energy consumption balance among individuals within a robot team, and to avoid the loss of integrity of the robot team due to premature exhaustion of energy of a certain robot.

[0024] 3. The present invention is applicable to a variety of different task scenarios such as robot soccer games, robot transport, robot sweeping, etc. In robot soccer games, the first threshold can be set to 9-11, the second threshold can be set to 5-6, and the third threshold can be set to 0.1-0.2. The thresholds in other scenarios can be set according to actual conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of a method according to an embodiment of the present invention.

[0026] Figure 2 This is the task transfer decision principle diagram.

[0027] Figure 3 It is a neural network model diagram of the task transfer decision agent. DETAILED DESCRIPTION

[0028] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0029] A method for robot task load balancing, the technical problem to be solved by the method can be described as follows:

[0030] Assume that there are n robots, which have communication links between each other to form a fully connected network, G(V,E), where V=(1,…,n), and E represents the set of communication links between robots.

[0031] Assume that the task T undertaken by the robot can be decomposed into several subtasks, namely , k is the total number of subtasks.

[0032] It is necessary to determine whether task offloading should be performed, which robots to offload tasks to, and the amount of each task offloading based on the energy status, task status and distance between each robot, in order to achieve energy balance between the robots.

[0033] For this reason, Figure 1 As shown, this method adopts the following steps:

[0034] (1) Calculate the standard deviation of the robot's remaining energy:

[0035] Taking the remaining energy and tasks of the robots into consideration, the standard deviation of the remaining energy is calculated regularly after each member of the robot cluster completes its remaining tasks.

[0036] (2) Determine whether the standard deviation of the remaining energy exceeds the set threshold

[0037] Determine whether the standard deviation of the remaining energy of the robot cluster exceeds the set threshold. If it exceeds the set threshold, execute step (3); if it does not exceed the set threshold, execute step (1).

[0038] (3) Determine the robot that needs to transfer tasks

[0039] Estimate the remaining energy of each robot after completing the remaining tasks, calculate the mean of the remaining energy, calculate the difference between each robot and the mean, and select the robot with the most remaining energy below the mean as the task transfer robot.

[0040] (4) Determine the robot that will take over the outgoing task

[0041] According to the difference between each robot and the mean, the robots are sorted from large to small, and for each robot that needs to transfer tasks, the robot that will take over the task is determined according to the task transfer strategy, that is, which robots will be assigned to transfer the tasks.

[0042] (5) Execution task transfer

[0043] The task-transferring robot performs task migration, and accordingly, each task-taking robot receives the transferred tasks. After all tasks are transferred in place, they begin to execute tasks according to the new task division and execute step (1).

[0044] In step (4), the task transfer decision of determining the robot to undertake the task and the specific amount of task each robot undertakes is implemented based on the deep reinforcement learning method. Figure 2 As shown, the specific method is as follows:

[0045] (4.1) Task transfer decision model

[0046] The task transfer decision model includes a task transfer decision agent and a multi-robot collaborative work scenario. The task transfer decision agent sends the task offloading decision result to the multi-robot collaborative work scenario. Each task-taking robot takes over the transferred task and carries out the work according to the task offloading decision arrangement, forming a new task execution status, and then feeds back to the task transfer decision agent. At the same time, the feedback information also includes the task offloading decision reward, which is a quantitative evaluation of the actual effect of this task offloading decision.

[0047] like Figure 3 As shown in the figure, the task transfer decision agent is a convolutional neural network. The input layer includes n groups of data (n is the number of robots). Each group of data includes three values, namely the normalized value of the robot's remaining energy, the normalized value of the robot's remaining task volume, and the normalized value of the distance between the robot and the task unloading robot. The output layer includes the specific proportion of the robot's current remaining tasks that are shared by other robots.

[0048] (4.2) Task transfer decision method training

[0049] According to the initial state of the multi-robot collaborative work scenario, the task transfer decision agent randomly gives an initial task offloading decision to drive the task transfer decision model training process to start working. The multi-robot collaborative work scenario regularly feeds back the task offloading decision reward information and task execution status information to the task transfer decision agent as the basis for its next task decision. The above process is iterative, and the variance value of the remaining energy of multiple robots is used as a constraint requirement during the iteration process. When the energy variance is lower than the specified threshold, the current round of training is stopped. Through multiple rounds of training, the adaptability of the task transfer decision agent to various task scenarios is improved. The initial working scene state and initial task offloading decision of each round of training are different.

[0050] (4.3) Task transfer decision method test

[0051] In the task transfer decision method testing phase, the task transfer decision model obtained through training is tested using some task scenarios different from those in the training phase to verify whether it can meet the requirements. If it cannot meet the requirements, go to step (4.2) and train the task transfer decision method again.

[0052] (4.4) Application of task transfer decision method

[0053] During the application phase of the task transfer decision method, the tested task transfer decision model is used to actually support the task transfer decision work.

[0054] In short, the present invention can be used for collaborative scheduling of tasks between robots in collaborative work scenarios, achieving energy consumption balance among individuals in a robot team, and avoiding the loss of integrity of the robot team due to premature exhaustion of energy of a certain robot.

Claims

1. A method for robot task load balancing, characterized in that: The following steps are involved: (1) Calculate the standard deviation of the actual remaining energy of all robots; (2) determining whether the standard deviation exceeds a first threshold value, if so, executing step (3), otherwise returning to step (1); (3) estimating the estimated remaining energy of each robot after completing the remaining tasks, calculating the mean of the estimated remaining energy, and calculating the difference between the estimated remaining energy of each robot and the mean of the estimated remaining energy, and selecting the robot whose estimated remaining energy is lower than the mean of the estimated remaining energy by more than a second threshold as the task transfer-out robot; (4) All task transfer-out robots are sorted in ascending order of estimated residual energy, and for each task transfer-out robot, a task receiving robot for the transfer-out task is determined in sequence according to the task transfer strategy; wherein the specific method for determining the task receiving robot for the transfer-out task according to the task transfer strategy is as follows: (4.1) Establishing a task transfer decision model: The task transfer decision model includes a task transfer decision agent and a multi-robot collaborative work scenario, wherein the task transfer decision agent sends the task offloading decision to the multi-robot collaborative work scenario, and each task-taking robot takes over the transferred task and carries out the work according to the arrangement of the task offloading decision, forming a new task execution state, and feeding back to the task transfer decision agent; in addition, the actual effect evaluation of the task offloading decision is also fed back to the task transfer decision agent, wherein if the standard deviation of the actual remaining energy is smaller than that before the task offloading is implemented, a positive evaluation is given, otherwise a negative evaluation is given; The task transfer decision agent is a convolutional neural network. The input layer includes n sets of data, where n is the number of robots. Each set of data includes three values, namely the normalized value of the robot's remaining energy, the normalized value of the robot's remaining tasks, and the normalized value of the distance between the robot and the task-taking robot. The output layer includes the proportion of tasks shared by other robots to the remaining tasks of the task-transferring robot before the task is transferred out. (4.2) Training the task transfer decision model: According to the initial state of the multi-robot collaborative work scenario, the task transfer decision agent randomly gives an initial task offloading decision to drive the task transfer decision model training process to start working. The multi-robot collaborative work scenario regularly feeds back the task offloading decision reward information and task execution status information to the task transfer decision agent as the basis for its next task decision, and the process is repeated in this way. During the iteration process, the variance value of the actual remaining energy of the robot is used as a constraint requirement. When the variance value is lower than the third threshold, the iteration is stopped and the current round of training is completed. Through multiple rounds of training, the adaptability of the task transfer decision agent to various task scenarios is improved. In each round of training, the initial state of the multi-robot collaborative work scenario and the initial task offloading decision given by the task transfer decision agent are different. (4.3) Test the trained task transfer decision model: The task transfer decision model obtained through training is tested using a multi-robot collaborative work scenario different from that in the training phase to verify whether it can play a role in task load balancing in the test collaborative work scenario. If it cannot play a role, go to step (4.2) and train the task transfer decision model again. (4.4) Using the tested task transfer decision model to determine the task transfer robot; (5) The task-transferring robot performs the task transfer, and at the same time, the corresponding task-taking robot receives the transferred task; after all tasks are transferred in place, they start to execute the tasks according to the new task division and return to step (1).

Citation Information

Patent Citations

  • Method and device for realizing data center resource load balance in proportion to comprehensive allocation capability

    CN102185779A

  • Resource load balancing control method and cluster scheduler

    CN108009016A

  • Load balancing method and device

    CN109376013A

  • Logistics robot dynamic task allocation method and system

    CN113044458A

  • Edge server load balancing method and system

    CN113485826A